Materials

Processing
FUNDAMENTALS2023
FUNDAMENTALS

EDITED BY
Samuel Wagstaff
Alexandra Anderson
Adrian S. Sabau
The Minerals, Metals & Materials Series
Samuel Wagstaff · Alexandra Anderson ·
Adrian S. Sabau
Editors

Materials Processing
Fundamentals 2023
Editors
Samuel Wagstaff Alexandra Anderson
Oculatus Consulting Gopher Resource
Marietta, GA, USA Tampa, FL, USA

Adrian S. Sabau
Oak Ridge National Laboratory
Oak Ridge, TN, USA

ISSN 2367-1181 ISSN 2367-1696 (electronic)

The Minerals, Metals & Materials Series
ISBN 978-3-031-22656-4 ISBN 978-3-031-22657-1 (eBook)
[Link]

© The Minerals, Metals & Materials Society 2023

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Cover Illustration: Left: From Chapter “Toward Meso-scale Modelling of Slag Foaming Phenomena in
Pyrometallurgy”, Quinn G. Reynolds et al., Figure 5: Dispersed-phase marker phases and velocities for ug
= 0.02 m/s case. [Link] Right: From Chapter “Effect of Wearing
Impellers on Multiphase Flow and Desulfurization During KR Mechanical Stirring Process”, Wei Chen
et al., Figure 3: Effect of the number of employed heats of the impeller on the distribution of flow pattern,
a initial state, b after employed 120 heats, c after employed 220 heats. [Link]
031-22657-1_2. Bottom: From Chapter “Simulation of Ferroalloy Casting in Copper Moulds”, Haifei An
et al., Figure 8: Temperature field of the mould at 50 s before demoulding. [Link]
3-031-22657-1_18.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

The symposium Materials Processing Fundamentals is hosted at the Annual Meeting

of The Minerals, Metals & Materials Society (TMS) as the flagship symposium of the
Process Technology and Modeling Committee. It is a unique opportunity for inter-
disciplinary presentations and discussions about, among others, processing, sensing,
modeling, multi-physics, computational fluid dynamics, and thermodynamics.
The materials covered include ferrous and non-ferrous elements, and the processes
range from mining unit operations to jointing and surface finishing of materials.
Acknowledging that modern processes involve multi-physics, the symposium and
its proceedings allow the reader to learn the methods and outcome of other fields’
modeling practices, often enabling the development of practical solutions to common
problems. Modeling of basic thermodynamic and physical properties plays a key role,
along with computational fluid dynamics and multi-phase transport and interface
modeling.
Contributions to the proceedings include applications such as steel processing,
modeling of steel and non-ferrous alloys treatments for properties control, multi-
physics, and computational fluid dynamics modeling for molten metal processes
and properties measurement. Extractive, recovery, and recycling process modeling
is also presented, completing a broad view of the field and practices of modeling in
materials processing.
The engagement of TMS and committee members to chair sessions and review
manuscripts makes this symposium and its proceedings possible. The editor and co-
editors acknowledge the invaluable support and contribution of these volunteers as
well as TMS staff members, in particular, Patricia Warren, Trudi Dunlap, and Matt
Baker.

Samuel Wagstaff
Alexandra Anderson
Adrian S. Sabau

v
Contents

Part I Process Optimization

Modeling of Macro-scale Reaction Effects in a Secondary Lead
Reverberatory Furnace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Nicholas J. Walla, Emily A. Higley, Armin K. Silaen,
Alexandra Anderson, Joseph Grogan, and Chenn Q. Zhou
Effect of Wearing Impellers on Multiphase Flow
and Desulfurization During KR Mechanical Stirring
Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Wei Chen, Yanyu Zhao, and Lifeng Zhang
Post Processing Approach to Model Microsilica Formation . . . . . . . . . . . . 25
Kurian J. Vachaparambil, Kristian Etienne Einarsrud, Halvor Dalaker,
and Stefan Andersson
Thermodynamics and Kinetics of Coke Breeze Combustion Under
Different Oxygen Content in the Sintering Process . . . . . . . . . . . . . . . . . . . . 35
Dongqing Wang, Wen Pan, Zhixing Zhao, and Yapeng Zhang

Part II Continuous Casting

Inverse Calculation of Time-Spatial Varying Mold Heat Flux
During Continuous Casting from Fast Response Thermocouples . . . . . . . 47
Haihui Zhang, Huiqiang Shen, and Pengcheng Xiao
Quality Prediction of Hot Rolled Products and Optimization
of Continuous Casting Process Parameters Based on Big Data
Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Zibing Hou, Zhiqiang Peng, Qian Liu, and Guanghua Wen
Study on the Uniformity of Surface Temperature of Continuous
Casting Slab Based on Solidification and Heat Transfer Simulations . . . . 75
Yadong Wang and Lifeng Zhang

vii
viii Contents

How to Prevent Porosity Defects in Steel Casting Component . . . . . . . . . . 83

Izudin Dugic
Fluctuant Solidification Behavior in the Centerline of Continuous
Casting Billets Based on Numerical Simulation . . . . . . . . . . . . . . . . . . . . . . . 97
Dongwei Guo, Zihang Zeng, Kunhui Guo, and Zibing Hou

Part III Slag and Ladle Treatment

Toward Meso-scale Modelling of Slag Foaming Phenomena
in Pyrometallurgy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Quinn G. Reynolds and Oliver F. Oxtoby
Simulation on the Slag Desulfurization During the LF Refining
in a Gas-Blowing Ladle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Jujin Wang, Yuexin Zhang, Binyu Lyu, and Lifeng Zhang
A Modified Thermodynamic Software to Control the Composition
of Inclusions During Calcium Treatment Process . . . . . . . . . . . . . . . . . . . . . 135
Weijian Wang and Lifeng Zhang

Part IV New Processes and Insights

Reductant Formation Enthalpy in DC Ferrochrome Smelting:
Merely Academic or Fundamental to Operation? . . . . . . . . . . . . . . . . . . . . . 145
H. J. Oterdoom, M. A. Reuter, and J. H. Zietsman
Measuring and Processing of Electrical Parameters in a Submerged
Arc Furnace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Hákon Valur Haraldsson, Halldór Traustason,
Yonatan A. Tesfahunegn, Merete Tangstad, and Gurún Sævarsdóttir
Volatilization Behavior of Arsenic from a Hematite Ore During
Non-isothermal Heating in Argon Atmosphere: An Overview . . . . . . . . . . 171
E. K. Chiwandika and S.-M. Jung
Comprehensive Recovery of Elemental Sulfur and Sulfide Minerals
from Pressure Acid Leaching Residue of Zinc Sulfide Concentrate
with an Integrated Flocculation Flotation-Hot Filtration Process . . . . . . . 185
Guiqing Liu, Bangsheng Zhang, Zhonglin Dong, Fan Zhang,
Fang Wang, Tao Jiang, and Bin Xu

Part V Additive Manufacturing and Materials First Principles

Automatic Process Mapping for Ti64 Single Tracks in Laser
Powder Bed Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Toby Wilkinson, Massimiliano Casata, and Daniel Barba
Contents ix

Part VI Poster Session

Simulation of Ferroalloy Casting in Copper Moulds . . . . . . . . . . . . . . . . . . 213
Haifei An, Weijian Tian, Hao Chen, Shaojun Chu, Lihong Li,
and Hao Bai
Effects of Temperature and Density on Transition Slab Length
During Steel Grade Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Sicheng Song, Yanhui Sun, Yaoguang Li, and Chao Zhuo
Mathematical Simulation Study on the Effect of Nozzle Side Hole
Structure Parameters on the Behavior of Molten Steel in Stainless
Steel Mold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Si-kun Peng, Ming-mei Zhu, Kun-chi Jiang, and Cheng-hong Li

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
About the Editors

Samuel Wagstaff is currently a partner at Oculatus

Consulting, specializing in aluminum processing and
product development. He holds degrees in Mechanical
Engineering from Cornell University (B.S.) and Mate-
rials Science from MIT (M.S., Sc.D.). Previously as
the Lead Scientist at Novelis, he led new product and
process development for the entire R&D ecosystem
across three continents. Currently, Dr. Wagstaff focuses
on increasing profitability and productivities of non-
ferrous products by process improvement and funda-
mental research. He has helped to design over 1 million
tons of recycle capacity in the aluminum sector and is
the author of over 35 patents. Sam is currently serving
as the principal investigator on a $1M ReMADE grant
to develop technology to improve the recyclability of
organic laden aluminum scrap.

Alexandra Anderson Ph.D., PMP, is an R&D manager

at Gopher Resource, LLC, an environmental solutions
company specializing in lead battery recycling. Her
work focuses on driving furnace productivity and effi-
ciency initiatives through computational fluid dynamic
(CFD) modeling and implementing novel equipment
designs. Currently, she is also the principal investigator
for a DOE HPC4Manufacturing partnership between
Gopher Resource and Oak Ridge National Lab inves-
tigating high-fidelity multi-phase furnace modeling.
Alexandra obtained her B.S. in Mechanical Engineering
from Gonzaga University and her M.S. and Ph.D.
in Metallurgical and Materials Engineering from the

xi
xii About the Editors

Colorado School of Mines. Her dissertation investi-

gated fluid flow and thermal profiles within secondary
lead reverberatory furnaces using CFD techniques.
Alexandra is active in The Minerals, Metals & Materials
Society (TMS), where she serves as the vice-chair of
the Process Technology and Modeling Committee; she
was also the recipient of the 2021 TMS Extraction and
Processing Division (EPD) Young Leader Award. Her
scholarly activities include nine peer-reviewed publica-
tions, co-editorships of seven special topics for JOM,
as well as several podium presentations at national
conferences.

Adrian S. Sabau received an Engineer Diploma in

Mechanical and Materials Processing from the Univer-
sity of Craiova, Romania, and a Ph.D. degree in Mechan-
ical Engineering from Southern Methodist University in
1996. In 1999, Dr. Sabau joined Oak Ridge National
Laboratory as a Research Staff Member of Mate-
rials Science and Technology, where he worked as a
Senior Research Staff Member from 2008. Since 2018,
Dr. Sabau has been a Computational Materials Scien-
tist in the Computational Sciences & Engineering Divi-
sion. Dr. Sabau seeks to advance the materials
processing, metal casting, photonic processing, and
materials for energy applications through the develop-
ment of computational fluid dynamics and experimental
methodologies for the property measurement, process
analysis, and materials behavior in response to condi-
tions experienced in service. Dr. Sabau is an ASME
fellow, and ASM fellow, and the recipient of three R&D
100 awards in process sciences. He was granted 7 patents
and has published 169 technical papers.
Part I
Process Optimization
Modeling of Macro-scale Reaction
Effects in a Secondary Lead
Reverberatory Furnace

Nicholas J. Walla, Emily A. Higley, Armin K. Silaen, Alexandra Anderson,

Joseph Grogan, and Chenn Q. Zhou

Abstract A long-standing and effective way to recycle lead-acid battery materi-

als is through processing of lead compounds into lead product within a reverbera-
tory furnace. Exploration of process and design changes through unit modification
can be costly, time-consuming, and potentially harmful to operational efficiency.
Modeling of process behavior, including furnace heat transfer and material reduc-
tion/decomposition, can however be difficult. To this end, a method for reflecting
the production capabilities of a lead reverberatory furnace under various operational
conditions has been developed. Reactions of the lead compounds have been approx-
imated within a steady-state computational fluid dynamics simulation by adding or
removing heat from the domain depending on local thermal conditions. With this,
process and design changes can be explored in the simulated environment before
moving onto more-advanced stages of modeling or experimentation.

Keywords Reverberatory furnace · Lead recycling · Numerical modeling ·

Pyrometallurgy

N. J. Walla (B) · E. A. Higley · A. K. Silaen · C. Q. Zhou

Center for Innovation through Visualization and Simulation, Purdue University Northwest, 2200
169th Street, Hammond, IN 46323, USA
e-mail: njwalla@[Link]
E. A. Higley
e-mail: ehigley@[Link]
A. K. Silaen
e-mail: asilaen@[Link]
C. Q. Zhou
e-mail: czhou@[Link]
A. Anderson · J. Grogan
Research and Development, Gopher Resource, 6505 Jewel Avenue, Tampa 33619, FL, USA
e-mail: [Link]@[Link]
J. Grogan
e-mail: [Link]@[Link]

© The Minerals, Metals & Materials Society 2023 3

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
4 N. J. Walla et al.

Introduction

Nearly 80% of lead produced within the United States comes from the recycling
of scrap lead [1], with nearly half of global lead production being secondary lead
produced from lead-acid battery recycling [2]. The dominant method for producing
secondary lead is via pyrometallurgical processes in reverberatory furnaces, with
78% of secondary lead produced in the United States using reverberatory furnaces
[3]. The growth of secondary lead production brings with it efforts to optimize the
process and minimize any environmental impacts. For overall lead production (both
primary and secondary), the smelting of secondary lead is a strong contributor to the
global warming impact of the lead industry [4], hence the importance of optimization
and energy reduction.
Research into reverberatory furnaces has been ongoing for decades using a vari-
ety of approaches. Experimental furnaces have been built to examine the specifics
of operational behavior, such as the modular furnace design of King et al. [5]. Their
modular furnace, in this case an aluminum reverberatory furnace, allowed for explo-
ration of the material heat absorption based on surface area, the effects of combustion
space volume changes, the impact of wall emissivity on overall heat flux, and various
furnace input and output relationships. Kumar et al. [6] built an experimental furnace
to explore the effect of flame impingement on the solid charged material, working
to identify the effective heat transfer through the charge material under different
furnace conditions. This effort was combined with a computational fluid dynamics
(CFD) simulation of the furnace to further explore design and operation conditions.
Numerical modeling of pyrometallurgical furnaces is an attractive alternative
to experimentation. Constructing and operating experimental furnaces can be pro-
hibitively expensive or outright impossible in some instances. Experimentation on
production furnaces risks damage, lost time, and negative impacts on unit productiv-
ity. Using numerical approaches such as CFD to model the multiphase reacting flow
conditions offers the opportunity explore design and process decisions without risk
and at drastically-lower costs.
Golchert et al. [7] created a CFD analysis of a secondary aluminum reverbera-
tory furnace. The model focuses on the combustion gas space to simulate the fuel
combustion, heat transfer, and pollutants/gas flow. Exploration of burner configu-
ration using this model shows improvement to heat transfer when the burners were
angled slightly downwards to counter the natural lift of the flame. Also focusing on
the combustion gas space, Anderson et al. [8] performed simulations of a secondary
lead reverberatory furnace to examine the potential impacts of the burden shape on
potential productivity and refractory wear due to flame impingement, thus providing
insight into charging behavior and burner positioning.
Numerical exploration of melting and smelting presents inherent difficulties for
pyrometallurgical furnaces. Simplifications are often made to ensure numerical sta-
bility or to reduce complexity for the sake of computational resources or compute
time. Buchholz and Rodseth [9] implemented an artificial heat sink on the aluminum
product within the simulation of an aluminum reverberatory furnace. This allowed
Modeling of Macro-scale Reaction Effects in a Secondary … 5

for the approximation of the absorption of the heat of fusion that occurs in the actual
process while allowing for both a steady-state process and a static domain (i.e. no
material phase change from the melting product).
This work takes a similar approach to model a secondary lead reverberatory
furnace. However, the methodology has been adapted for additional considera-
tions needed to approximate the heat transfer of several melting/smelting reactions.
Extending off the previous work of this author [10], mechanisms are implemented
for the release of off-gases that occur during furnace operation. With these behaviors,
a model of the furnace is created that allows for exploration of design decisions and
their impact on productivity while still modeling only the combustion gas space.

Methodology

The simulation model is created in ANSYS Fluent 2022R1 as a 3D, multiphase,

steady-state simulation. Turbulence behavior is captured using the SST k-omega
RANS turbulence model. The burners are directly modeled, with combustion occur-
ring via the finite-rate/eddy-dissipation concept. Radiative heat transfer is considered
using the P1 radiation model. In addition to its use in combustion, the species trans-
port model is used for the off-gassing portion of the pseudo-reaction modeling. As
mentioned above, no melting model is considered in this study and only the com-
bustion gas space is included. Combustion of the natural gas fuel was modeled as
a two-step reaction, with burner inputs taken from furnace operational conditions
corresponding to the temperature and flue gas data used for model tuning.
Currently, not all the information needed to directly model the melting/smelting
reactions of the charged material is known. Thus, the primary focus of this work is
on the application of a “pseudo-reaction” condition that approximates a few macro-
scale behaviors that occur during these reactions. Specifically, the model seeks to
replicate the significant temperature changes and off-gas generation that occur.

Domain

With the simulation isolated to the combustion gas space alone, the numerical model
consists of mass flow inlets within the burners and a single pressure outlet at the
flue duct. The lead and slag tap holes are excluded from this study. A diagram of
the furnace domain in this study is shown in Fig. 1. The furnace front contains the
burners, the pile of charged burden material, and a small amount of slag/bath surface.
The rear of the furnace has the slag/bath surface and the flue. All sides excluding the
openings and bottom surfaces are composed of refractory bricks. These bricks are
modeled as flat surfaces with thicknesses and convective heat losses defined within
the simulation.
6 N. J. Walla et al.

Furnace Front Furnace Rear

Burner (side) Refractory Walls Flue

Burden Surface
Slag/Bath Surface

Fig. 1 Side view of furnace CFD domain

Side Burner
2x End Burners

Front Bath

Rear Bath

Flue
Burden Pile

T CB T CF lue
T CA
Side Burner

Fig. 2 Top-down view of domain lower surfaces and thermocouple locations

The bottom surface comprises three areas: the burden, the front bath, and the rear
bath. The transition from front bath to rear bath occurs at roughly halfway down the
length of the furnace. The diagram in Fig. 2 shows the extent of these regions. The
burden pile is the portion of charged material that is not submerged within the bath,
the shape of which is replicated from measurements of side slope angles, heights of
the “spine” along the central axis, and distance from the side walls at several points.
Also seen in the diagram are the locations of the three thermocouples used for tuning
the behavior the model.
The numerical grid consists of roughly six million polyhedral cells. The cell zone
for the combustion space within the furnace is divided into four zones: the overall area
combustion gas space of the furnace and a zone for each of the three lower surfaces
made by isolating the first three cells neighboring those respective surfaces. These
isolated cell zones are used for applying the pseudo-reaction source terms within the
simulation, allowing each zone to be treated separately (a necessary consideration
due to differences in their behavior).

Heat Transfer Subroutine

The pseudo-reaction methodology consists of using user-defined subroutines to mod-

ify the behavior of a base simulation such that the impacts of smelting and melting
reactions are included without the direct modeling of said reactions. For the study
of the reverberatory furnace, the two primary factors missed by not modeling the
Modeling of Macro-scale Reaction Effects in a Secondary … 7

Table 1 Charge material reactions and pseudo-reaction coefficients

i Reaction Ai (kJK) Bi (kJ) Tref (K)
1 PbCO3 → PbO(s) + CO2 (g) −0.084 39.01 588
2 PbSO4 + 2 C → PbS + 2 CO2 (g) −0.111 21.25 923
3 Pb(OH)2 → PbO + H2 O (g) 0.035 55.51 373
4 2 PbO + PbS → 3 Pb + SO2 (g) −0.294 237.7 1073
5 2 PbO + C → 2 Pb + CO2 (g) −0.156 43.94 873

reactions directly are the exo- and endo-thermic temperature changes caused by the
reactions and the off-gases produced by said reactions. The approximation of energy
changes within the model was explored by the authors in their previous work [10],
wherein surface conditions on the burden and bath were adjusted such that temper-
atures within the simulation matched measurements taken from furnace operations.
In short, the approach uses artificial heat sinks defined within the simulation using
user-defined functions (UDFs) as energy source terms. Each zone has separately-
defined heat sinks based on the expected physical behavior of the zone. For the
burden, the enthalpy changes and reaction temperatures for each of the five reactions
(seen in Table 1) are used to determine the strength of these heat sinks. These are
used to generate an approximate linear rate of enthalpy changes across a range
of temperatures. Within the UDF, the cell’s temperature is compared against each
reaction temperature, adding to the sum total of enthalpy change expected within
that cell for each matching condition. The equation of this is shown in Eqs. 1, where
αsink is a constant that is used to globally tune this behavior.
The rear bath does not have any reactions, thus its condition is a much-simpler
linear behavior based on the difference between the cell temperature and the lead
liquidus temperature (and also with a scaling factor ksink for adjustment) as seen in
Eq. 2. The front bath uses a blend of these two conditions to reflect the submerged
pile to the sides and front, with the front or rear behavior “fading” in based on cell
position, as seen in Eqs. 3 and 4.

5
qαsink (T ) = −αsink · Ai (T − Tref i ) + Bi (1)
i=1

qk (Tadj ) = −ksink · (T − Tliq ) (2)

x̄ − x̄max 3
qαsink,front = qαsink (T ) · max − + 1 ,0 (3)
x̄max

x̄max − x̄ 3
qksink,front = qksink (T ) · max − + 1 ,0 (4)
x̄max
8 N. J. Walla et al.

With the above behavior, the simulation of the reverberatory furnace can be given
different burner inputs and result in different temperature profiles on the lower surface
due to cells no longer meeting the requirements to “react”. However, upon analyzing
the resulting heat balance of the model, an issue was noticed. The relative amount
of heat leaving the furnace via the flue gas was far lower than expected. Further
analysis showed that the amount of mass leaving the flue was far less than what was
measured. While the pseudo-reaction heat condition could alter temperatures, it was
not generating the significant gaseous byproducts of the reactions.

Off-Gas Subroutine

To overcome this limitation, a new feature was implemented into the pseudo-reaction
code. Taking a similar approach as the heat-changing subroutine, off-gas generation
subroutine would use the temperature data of marked cells (this time only above the
burden pile) to determine if material needs to be generated. For each cell, the cell
temperature is compared to the reaction temperatures of the reactions in Table 1. If
the required condition is met, an amount of the resulting gas would be marked for
generation. There were only three species to consider for this: CO2 from reactions 1,
2, and 5, H2 O from reaction 3, and SO2 from reaction 4. The SO2 species is unique
in that it will only generate within the domain due to the burden reactions, as the
other two are also byproducts of the burner combustion.
As with the temperature data tuning via comparison with thermocouples, flue
gas measurements allowed for tuning of the off-gas condition. However, while the
temperature condition required only two tuning coefficients (α and k), the species
condition required four: βc for CO2 , βh for H2 O, βs for SO2 , and γ for the amount of
mass generated. The β values are used to adjust the amount of species “generated”
within the marked cells, as without the consideration of material consumption or
reaction rates there is no way to quantify the relative differences in reactant quantity.
With these β terms, each of the three separate off-gas species can be individually
adjusted as needed.
In ANSYS Fluent, defining a source term for a species forces a species fraction
but does not generate mass to match. Therefore, after determining how much of
each species is generated within a particular cell, a DEFINE_ADJUST subroutine
is executed to calculate a total amount of mass to be generated within the cell. This
value is multiplied by the γ constant to scale the mass generation up or down as
needed to match the known flue gas flow rate. A diagram of the UDF logic is shown
in Fig. 3.
Modeling of Macro-scale Reaction Effects in a Secondary … 9

Fig. 3 UDF procedure for

energy and off-gas behaviors

Results

As mentioned earlier, the pseudo-reaction behavior is tuned relative to data gathered

from operation of the reverberatory furnace. The energy-adjustment subroutine is
tuned such that temperature measurements within the simulation match measured
values from three thermocouples (two near the burden pile and one within the flue).
The previous efforts using only the energy-adjusting behavior [10] required tuning
of the α and k terms to match the thermocouples. The relationship between the
temperatures within the model and those parameters (mostly α) was nearly linear.
However, upon adding the off-gas generation behavior, differences were noted
in both the temperature profiles of the system and the simulation’s response to the
tuning coefficients. The two systems (energy and off-gas) are inherently linked; a
change in temperature results in an altered distribution/concentration of off-gasses
which in turn affect the thermal properties of the combustion gas thus affecting the
movement of heat within the domain. This leads to a more-difficult balancing act
between the coefficients and the target values.
Figure 4 shows contours of temperature (normalized relative to global minimum
and maximum values) on the lower surfaces of the furnace. Colder regions can be seen
in the back of the burden pile and in the front of the pile that is only indirectly exposed
to the heat from the burners. The spine of the burden pile is also at a slightly-lower
temperature. The highest temperatures can be seen on the sides of the pile directly
where the flames of the side burners impinge upon the burden material.
Plotting the strength of the pseudo-reaction’s energy removal, an interesting pat-
tern appears. Due to the nature of the enthalpy-based coefficients for each reaction,
the regions with the highest temperatures do not correspond to the areas with the
strongest degree of heat removal. As seen in Fig. 5, two “rings” of high energy
removal exist near the peripheries of the burner impingement zones. With the excep-
tion of the third reaction (which remains endothermic), the reactions transition from
endothermic behavior to exothermic behavior at various temperatures. At a temper-
ature of approximately 1600 K a transition point is reached where the endothermic
10 N. J. Walla et al.

Fig. 4 Locally-scaled contours of normalized temperatures on lower surfaces

Fig. 5 Contours of pseudo-reaction energy source term on lower surfaces

and exothermic behaviors of all five reactions cancel each other out. This leads to a
natural balancing effect in some regions of the burden pile.
As the SO2 in the system is only sourced from the pseudo-reaction source term,
contours of SO2 concentration along the center plane of the domain provide some
insight into the behavior approximated reactions. Figure 6 shows the off-gassing
of the SO2 from the burden surface, with the gas flow of the furnace driving the
emissions towards the flue. The rear of the burden pile produces very little SO2 ,
likely due to the lower temperatures as the reaction which produces the SO2 has the
highest reaction temperature of the five reactions in the model.
The resulting composition of the flue gas after tuning the coefficients was within
desired ranges for both the concentration of the target species as well as for the
overall mass flow rate. Table 2 shows the resulting error in the model results relative
to the measured data from furnace operations. The highest error seen is for the
oxygen content of the flue versus the simulation. One aspect of the reactions that is
not currently considered is the consumption of oxygen by the coke in Reaction 5.
In operational measurements, flue gas volume of oxygen was found to be between
Modeling of Macro-scale Reaction Effects in a Secondary … 11

Fig. 6 Side view center plane contours of normalized SO2 concentration

Table 2 Flue gas composition and flow rate error relative to operations
H2 O CO2 O2 SO2 ˙
m flue
Error % 9% 7% 69%+ 3% 5%

Table 3 Heat balance error relative to operations

Case Reactions (%) Walls (%) Cooling (%) Flue (%) Total (%)
No Off-Gas +65 −70 −57 −45 +7
Off-Gas <1 −33 −54 −9 +7

Table 4 Thermocouple temperature error relative to operations

Case T C A (%) T C B (%) T C F (%)
No off-gas 1 2 1
Off-gas 0 2 11

0–4% (averaged to 2% for this analysis) while the simulation shows roughly 5% of
flue output being oxygen.
Looking at the resulting heat removal within the furnace, Table 3 shows the impact
of the off-gas inclusion in the pseudo-reaction model. It is estimated that 30–40%
of the heat from the combustion sources is consumed by the melting and smelting
processes. Comparing the previous pseudo-reaction simulation (without off-gassing)
to a heat balance of the operating furnace, reactions and super-heating of the bath
went from an over-prediction of roughly 65% (62% of combustion input) to nearly
matching the estimation. The accuracy of heat removal via the walls also improved,
though the error in heat removed via cooling elements remained unchanged. The
heat removal via flue gasses is still slightly under-predicted but is much improved
over the simulation without off-gas consideration.
One noticeable impact of the off-gas inclusion was an increased sensitivity to the
adjustment of the αsink value for the burden and bath energy removal. Previously,
the αsink term was found to have an almost-linear impact on the temperatures at the
front and rear of the furnace. However, the addition of the off-gas species resulted
12 N. J. Walla et al.

in much greater changes at the front area of the furnace (at the two burden-adjacent
thermocouples) versus the rear (flue thermocouple) region, with flue temperatures
remaining close to the target value but front values plummeting at higher values of
αsink .
While the energy loss-only study used an αsink of 70 to obtain a temperature
match to the thermocouple data, the off-gas study required a lowering of the value
to 40. Even then, Table 4 shows errors in temperature measurements increased from
between 1–2% error up to as much as 11% error, with flue temperatures being much
higher. This behavior is currently being investigated by the authors, but may be due
to the “doubling-up” of heat removal via the source term and generation of new
mass in the cell, wherein added mass alters the impact of the energy source term as
applied by ANSYS Fluent as well as the increased heat capacity of the species-laden
gas convecting more heat away from the burden region.
Interestingly, the error for overall heat losses within the furnace remained the
same between the two simulations, indicating that the heat previously removed by
the energy subroutine is now being transported by the generated species to exit at
the flue.

Conclusions

In this work, a numerical subroutine was developed and applied to the steady-state
simulation of a secondary lead reverberatory furnace. Aiming to replicate the macro-
behaviors of the melting and smelting reactions from the conversion of the charge
materials into liquid lead, a pseudo-reaction was used to alter the heat balance within
the furnace as well as generate off-gasses that would occur during the process. The
current model has been adjusted to match the measured behaviors of the operational
furnace, with further work underway to validate and explore the model. Once vali-
dated, the simplifications made within this model will allow for faster modeling of
furnace conditions, including exploration of furnace design and burner set points.

Acknowledgements This work was funded by Research Award No. W911NF1920108 from the
Army Research Laboratories of the Department of Defense and the Worcester Polytechnic Institute.
The authors would also like to extend their gratitude for Gopher Resource in providing information
on design, operational conditions, measurement data, and process insight.

References

1. U.S. Geological Survey: Mineral commodity summaries 2021. Technical report, U.S. Geo-
logical Survey, Reston, VA (2021). [Link] [Link]
publication/mcs2021
2. Gottesfeld P, Pokhrel AK (2011) Review: lead exposure in battery manufacturing and recycling
in developing countries and among children in nearby communities. J Occupat Environ Hyg
8(9):520–532
Modeling of Macro-scale Reaction Effects in a Secondary … 13

3. Queneau PB, Leiby R, Robinson R (2015) Recycling lead and zinc in the United States. World
Metall ERZMETALL 68(3):149–162
4. Davidson AJ, Binks SP, Gediga J (2016) Lead industry life cycle studies: environmental impact
and life cycle assessment of lead battery and architectural sheet production. Int J Life Cycle
Assess 21(11):1624–1636
5. King PE, Hayes MC, Li T, Han Q, Hassan M, Golchert BM (2005) Design and operation of an
experimental reverberatory aluminum furnace (2005)
6. Kumar A, Venuturumilli R, Kiss L, Walter G (2008) Experimental and numerical study of
flame load heat transfer in an experimental furnace
7. Golchert BM, Zhou CQ, Quenette A, Han Q, King PE (2005) Combustion space modeling of
an aluminum furnace, pp 887–892
8. Anderson A, Grogan J, Bogin G, Taylor P (2018) Computational Modeling of a secondary
lead reverberatory furnace: effect of burden geometry. In: Davis BR, Moats MS, Wang S (eds)
Extraction 2018. Springer, Ottawa, Ontario, Canada, pp 881–890. [Link]
3-319-95022-8_70
9. Buchholz A, Rodseth J (2011) Investigation of heat transfer conditions in a reverberatory
melting furnace by numerical modeling. Light Metals 2011:1179–1184
10. Walla NJ, Anisiuba V, Silaen AK, Anderson A, Grogan J, Zhou CQ (2022) Boundary condi-
tions for modeling of a lead reverberatory furnace. In: ASME 2022 heat transfer summer
conference, p. 8. American Society of Mechanical Engineers, Philadelphia, Pennsylvania,
USA (2022). [Link] [Link]
HT/proceedings/HT2022/85796/V001T18A002/1146590
Effect of Wearing Impellers
on Multiphase Flow and Desulfurization
During KR Mechanical Stirring Process

Wei Chen, Yanyu Zhao, and Lifeng Zhang

Abstract It is of great significance to study the variation of the multiphase flow

during the entire service process of an impeller to improve the dispersion of the desul-
furizer and the desulfurization efficiency during the KR desulfurization process. In
the current study, the key dimensions of an actual KR impeller during the service
process were quantitatively measured first. Then, a three-dimensional model coupled
with the k-ε turbulence model, VOF multiphase flow model, DPM model, UDS
model, and unreacted core desulfurization model was established to predict the
multiphase flow and desulfurization during the KR mechanical stirring process with
different wearing impellers. The results show that with the increase of the wear
degree of the impeller, the stirring effect was gradually weakened, resulting in a
gradual weakening of the desulfurization efficiency. The desulfurization end sulfur
content was 58.2 ppm after the impeller employed 220 heats, which was more than
4 times higher than the 13.0 ppm with a new impeller.

Keywords Wearing impellers · Multiphase flow · Desulfurization · KR process

Introduction

The KR mechanical stirring method has obvious advantages in the desulfurization

effect and desulfurization cycle due to its superior kinetic conditions and is widely
used in the hot metal pretreatment desulfurization process [1, 2]. Many researches
about the fluid flow and desulfurization during the KR process have been published.
Ji [3] used the Euler-granular model to study the mixing of the desulfurizer and

W. Chen
School of Mechanical Engineering, Yanshan University, Qinhuangdao 066004, China
Y. Zhao
Beijing Shougang Co., Ltd, Qianan, Hebei 064404, China
L. Zhang (B)
School of Mechanical and Materials Engineering, North China University of Technology,
Beijing 100144, China
e-mail: zhanglifeng@[Link]

© The Minerals, Metals & Materials Society 2023 15

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
16 W. Chen et al.

high-sulfur hot metal under the variable-velocity stirring method. The effect of the
impeller geometry on the fluid flow and particle volume fraction was investigated
[4]. Li [5] developed a coupled k-ε model and VOF model to get insight into the flow
pattern and interface behavior taking place during the KR desulfurization process.
It shows that the interface profile and vortex depth strongly depend on the impeller
dimension. Wang [6] established a transient-coupled three-dimensional numerical
model to study the two-phase flow, heat transfer, desulfurizer motion, and desulfur-
ization behavior during the KR hot metal treatment. He [7] studied the distribution
and motion behavior of desulfurizers in a 170 t hot metal ladle with KR mechanical
stirring using the Eulerian–Lagrangian approach. Nakaoka [8] investigated the fluid
flow and particle transport accompanied by turbulent agglomeration using numer-
ical simulation. It can be seen that most of the research was concerned with the fluid
flow, vortex distribution, and particle dispersion. In addition, most of the published
studies on the hot metal desulfurization were theoretical analyses and experimental
observations applicable to special conditions [9, 10]. However, the desulfurization
model with a wider scope needed to further couple the numerical simulation of the
fluid flow and the desulfurization kinetic model, but this part of the research was still
relatively few. The effect of the wear degree of the impeller on the flow field and
desulfurization also needed further study [11].
In the current study, a three-dimensional model coupled with the k-ε turbulence
model, VOF multiphase flow model, DPM model, UDS model, and unreacted core
desulfurization model was established to investigate the effect of the impeller wear
degree on the multiphase flow and desulfurization in a 250 t KR hot metal ladle.

Mathematical Model

A three-dimensional model based on a 250 t KR hot metal ladle was established, as

shown in Fig. 1. The range of rotation speed and immersion depth of the impeller
in the actual production process was 90–110 rpm and 1500–1700 mm, respectively.
The upper and lower diameter of the hot metal ladle after removing the refractory
part was 4066 mm and 3738 mm. The total number of grids of the current mold was
about 0.5 million. Other specific model parameters and boundary conditions can be
found in the previous study [12].
The upper and lower rotation diameter of the initial unused impeller in Fig. 1 were
1400 mm and 1300 mm, respectively. The height and width were 950 and 480 mm.
As the stirring time increased, the employed impeller was gradually eroded by the
hot metal and slag. In addition, the desulfurization slag also adhered to the impeller.
Figure 2 shows the morphology of the same impeller with the initial state (left), after
employed 120 heats (middle), and after employed 220 heats (right). The service life
of the impeller was about 270 heats, and the results in Fig. 2 show that the impeller
was severely eroded after employed 220 heats. Therefore, the above impellers were
modeled and their effects on the flow field and desulfurization were investigated.
Effect of Wearing Impellers on Multiphase Flow and Desulfurization … 17

Fig. 1 Distribution of the computational domain

Fig. 2 Morphology of the same impeller with the initial state (left), after employed 120 heats
(middle), and after employed 220 heats (right)
18 W. Chen et al.

The VOF multiphase flow model was used to solve the volume fraction of the hot
metal and air phases. The continuity equation and momentum equation for phase q
are shown below.
∂
αq ρq + ∇ · αq ρq uq = 0 (1)
∂t

m
αq = 1 (2)
q=1

∂
(ρu) + ∇ · (ρuu) = −∇ P + ∇ · μ ∇u + ∇uT + ρg + F (3)
∂t

where α q is the volume fraction, ρ q is the density in kg/m3 , and F is the momentum
exchange between the phases due to interface force in kg/(m2 s2 ).
The turbulent flow inside the hot metal was solved using the standard k-ε model.
The equation of turbulent kinetic energy k and its dissipation rate ε is calculated as:

∂k ∂(ku i ) ∂ μt ∂k
αl ρl + = αl μ + + αl G k − αl ρl ε (4)
∂t ∂ xi ∂x j σk ∂ x j

∂ε ∂εu i ∂ μt ∂ε
αl ρl + = αl μ +
∂t ∂ xi ∂x j σε ∂ x j
ε ε2
+ αl C1ε G k − αl C2ε ρl (5)
k k

where α l is the volume fraction of hot metal; ρ l is the density of hot metal, kg/m3 ;
μt is the turbulent viscosity, kg/(m s); The C 1ε = 1.44, C 2ε = 1.92, σ k = 1.0, σ ε =
1.3.
The trajectory of the desulfurizer particles was solved using the DPM as follows:

du p ρ p − ρl 3μCD Re
= g+ ul − u p
dt ρp 4ρ p d 2p
ρl d ρl
+ CVM ul − u p + u p ∇ · ul (6)
ρ p dt ρp

where the four terms on the right side of the equation are gravity buoyancy force,
drag force, pressure gradient force, and virtual mass force, respectively. The details
can be found in the previous study [12].
A UDS transport equation was used for the convection and diffusion of the sulfur
in the hot metal.

∂ μt
(ρ[S]) + ∇ · (ρu[S]) = ∇ · ρ DS + ∇ · [S] + SCaO (7)
∂t Sct
Effect of Wearing Impellers on Multiphase Flow and Desulfurization … 19

where [S] is the mass fraction of sulfur in the hot metal; DS is the diffusion coefficient
m2 /s; Sct is the turbulent Schmidt number; SCaO is the desulfurization source in
kg/(m3 s).
The desulfurization source SCaO was calculated using the unreacted core model
as follows [12]:

4πr02 ρl ks
v1 = ([S] − [S]e ) (8)
100
DS
ks = 2.0 + 0.6Re1/2 Sc1/3 (9)
dp
μ
Sc = (10)
ρl DS

where each variable can be found in the previous study [12].

Multiphase Flow and Desulfurization in the Hot Metal Ladle

Figures 3 and 4 show the effect of the number of employed heats of the impeller on
the distribution of flow pattern and volume fraction of hot metal, respectively. The
rotation speed was 90 rpm, and the immersion depth was 1500 mm. It can be seen
that with the increase of the number of employed heats of the impeller, the stirring
effect on the hot metal gradually decreased and the overall velocity of the hot metal
also decreased accordingly. As the number of impeller employed heats increased,
two vortexes close to the shaft of the impeller gradually decreased and moved up
until nearly disappearing after employed 220 heats.
Figures 5 and 6 show the effect of the number of employed heats of the impeller
on the average sulfur content and desulfurization rate of the hot metal under the
conditions of a rotating speed of 90 rpm and an immersion depth of 1500 mm. The
desulfurization rate gradually decreased with the increase of the number impeller
employed heats. With the same amount of desulfurizer added, the sulfur content at
the desulfurization end increased from 13.0 ppm in the initial state of the impeller to
26.7 ppm and 58.2 ppm after the impeller employed 120 and 220 heats, respectively.

Conclusions

1. The key dimensions of an actual KR impeller with the initial state and after
employed 120 and 220 heats were quantitatively measured. The effect of
the number of employed heats of the impeller on the multiphase flow and
desulfurization in a 250 t hot metal was numerically studied.
20 W. Chen et al.

(a) Initial state (b) After employed 120 heats

(c) After employed 220 heats

Fig. 3 Effect of the number of employed heats of the impeller on the distribution of flow pattern,
a initial state, b after employed 120 heats, c after employed 220 heats

2. A three-dimensional model coupled with the k-ε turbulence model, VOF multi-
phase flow model, DPM model, UDS model, and unreacted core desulfurization
model was successfully developed to predict the variation of the sulfur content
and desulfurization rate.
3. The overall velocity of the hot metal significantly decreased with the increase of
the number of employed heats of the impeller. The sulfur content at the desulfur-
ization end increased from 13.0 ppm in the initial state of the impeller to 26.7 ppm
and 58.2 ppm after the impeller employed 120 and 220 heats, respectively.
Effect of Wearing Impellers on Multiphase Flow and Desulfurization … 21

(a) Initial state (b) After employed 120 heats

(c) After employed 220 heats

Fig. 4 Effect of the number of employed heats of the impeller on the distribution of hot metal
volume fraction, a initial state, b after employed 120 heats, c after employed 220 heats
22 W. Chen et al.

400

Average sulfur content (ppm) 350

300
After employed 220 heats
250
After employed 120 heats
200 Initial state

150

100

50

0
0 100 200 300 400 500 600
Time (s)

Fig. 5 Effect of the number of employed heats of the impeller on the sulfur content of hot metal

CaO addition
3.0
Average desulfurization rate (ppm/s)

2.5

Initial state
2.0

After employed 120 heats

1.5
After employed 220 heats

1.0

0.5

0.0
0 100 200 300 400 500 600
Time (s)

Fig. 6 Effect of the number of employed heats of the impeller on the desulfurization rate of hot
metal
Effect of Wearing Impellers on Multiphase Flow and Desulfurization … 23

Acknowledgements The authors are grateful for the funding and support of the Science and Tech-
nology Project of Hebei Province (20311004D, 20591001D, 20311005D), High Steel Center of
Yanshan University (HSC), and High Steel Center of North China University of Technology (HSC).

References

1. Nakai Y, Sumi I, Matsuno H, Kikuchi N, Kishimoto Y (2010) Effect of flux dispersion behavior
on desulfurization of hot metal. ISIJ Int 50(3):403–410
2. Kurokawa N, Matsuo S, Jouguchi H, Yamada K, Watanabe Y (1993) Development in hot metal
desulphurization process. Sumitomo Met 45(3):52–58
3. Ji J, Liang R, He J (2016) Simulation on mixing behavior of desulfurizer and high-sulfur hot
metal based on variable-velocity stirring. ISIJ Int 56(5):794–802
4. Ji J, Du H, Jiang Y, Li D, Zhang S (2021) Effect of impellers on particle mixing behaviour in
KR desulphurization process. Ironmaking Steelmaking 49(2):167–177
5. Li Q, Shen X, Guo S, Li M, Zou Z (2021) Computational investigation on effect of impeller
dimension on fluid flow and interface behavior for kanbara reactor hot metal treatment. Steel
Res Int 92(11)
6. Wang Q, Jia S, Tan F, Li G, Ouyang D, Zhu S, Sun W, He Z (2021) Numerical study on
desulfurization behavior during kanbara reactor hot metal treatment. Metall Mater Trans B
52(2):1085–1094
7. He M, Wang N, Chen M, Chen M, Li C (2020) Distribution and motion behavior of desulfurizer
particles in hot metal with mechanical stirring. Powder Technol 361:455–461
8. Nakaoka T, Taniguchi S, Matsumoto K, Johansen ST (2001) Particle-size-grouping method of
inclusion agglomeration and its application to water model experiments. ISIJ Int 41(10):1103–
1111
9. Nakai Y, Hino Y, Sumi I, Kikuchi N, Uchida Y, Miki Y (2015) Effect of flux addition method
on hot metal desulfurization by mechanical stirring process. ISIJ Int 55(7):1398–1407
10. Jeong B-K, Bang G-H, Kang Y-B (2022) Role of liquid and solid particles in solid-liquid mixed
fluxes on sulfur removal from molten iron under centrifugal stirring by an impeller. Powder
Technol 396:1–12
11. Xiao L, Chen K, Dan B, Rong Z, Wang R (2016) Numerical study of flow field on molten iron
desulfurization with mechanical stirring. Adv Mech Eng 8(7):1687814016649326
12. Zhao Y, Chen W, Cheng S, Zhang L (2022) Mathematical simulation of hot metal desulfur-
ization during KR process coupled with an unreacted core model. Int J Miner Metall Mater
29(4):758–766
Post Processing Approach to Model
Microsilica Formation

Kurian J. Vachaparambil, Kristian Etienne Einarsrud, Halvor Dalaker,

and Stefan Andersson

Abstract In this work an OpenFOAM-based framework to simulate the evolution

of microsilica, which is an important byproduct in the silicon/ferrosilicon industries,
is proposed. The framework decouples the combustion reaction of CO and SiO from
the microsilica generation based on the assumption that the combustion occurs in
an oxygen rich environment - SiO2 generated by combustion is much larger than its
depletion due to particle evolution. The combustion of the reactants in the furnace
hood is performed using rhoReactingBuoyantFoam, and its results are used as input
to a population balance solver that simulates the particle nucleation and growth (due
to mass transfer onto the particle surface) as well as depletion of SiO2 . The framework
predicts particles of size around 30 nm at the outlet which is approximately in the
smaller sizes of the particles observed in microsilica during experiments reported in
literature.

Keywords Microsilica · CFD · OpenFOAM

Supplementary Information The online version contains supplementary material available at

[Link]

K. J. Vachaparambil (B)
Process technology, SINTEF Industry, Trondheim, 7465, Norway
e-mail: [Link]@[Link]
K. E. Einarsrud
Department of materials science and engineering, Norwegian University of Science and
Technology (NTNU), Trondheim 7491, Norway
e-mail: [Link]@[Link]
H. Dalaker · S. Andersson
Metal production and processing, SINTEF Industry, Trondheim 7465, Norway
e-mail: [Link]@[Link]
S. Andersson
e-mail: [Link]@[Link]

© The Minerals, Metals & Materials Society 2023 25

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
26 K. J. Vachaparambil et al.

Introduction

During the production of silicon and ferrosilicon alloys, SiO reacts with oxygen
(above the furnace hood) to form SiO2 which nucleates to form particles which is
referred to as microsilica. These are essentially amorphous silicon dioxide particles,
typically around 0.1 µm in diameter (consisting of submicron and micron sized clus-
ters), which has applications in cements [1–3]. As the quality and yield of microsilica
are influenced by the conditions in the furnace, understanding how these particles
evolve is critical.
In the best knowledge of the authors, the work by González-Fariña et al. [4]
(and the associated doctoral thesis [5]) is the only one in literature to investigate the
evolution of microsilica. In the doctoral work, [5], steady state simulation (using
COMSOL) of the furnace hood (with a simplified 2D geometry) showed a narrow
region in the furnace hood where the reactants and air mixed are the region where
the microsilica was produced. Based on the simulation results of [4], [5] investigated
the formation microsilica by the fully coupled transient model for a 1D case which
investigated the phenomena when the reactants and air is well mixed and initially
separated in addition to exploring the dynamics at two physical limits. The first
limit: when O2 is plentiful - the model (with particle evolution not coupled to the
temperature and SiO2 distribution) predicted the particle growing perpetually as
SiO2 is not depleted. The other limit: when the particle evolution is fully coupled
to both the temperature field and SiO2 field resulting in particles evolving until
SiO2 was depleted in addition to temperature being higher due to the heat released
due to particle evolution that was hypothesized to cause the nucleation of more
particles. The main drawback of [4] is that effect of velocity and turbulence was
not treated in addition to the 1D simplification used. Additionally, the framework
used by González-Fariña et al. [4] couples the particle growth (but not the particle
nucleation) to the depletion of SiO2 .
In this work, inspired by González-Fariña et al. [4], we propose a OpenFOAM-
based framework to model microsilica evolution in the furnace hood. The model
assumes that the reactions occur in the oxygen rich environment to decouple the
microsilica evolution from the combustion. The model accounts for the ceasing of the
particle growth when SiO2 reaches saturation. The model considers the microsilica
evolution driven by just nucleation and growth by mass transfer to the surface, and that
particle evolution does not impact the temperature field obtained from combustion
reactions and that these particles do not agglomerate nor does it change the velocity
field. Due to the high Reynolds number of the flow in the furnace hood [5], advection
of the microsilica is treated and the flow is assumed to be dominated by turbulence.
Post Processing Approach to Model Microsilica Formation 27

Overview of the Proposed Framework

Modelling the Combustion of Gaseous Species

The main gaseous species found over the charge in the furnace react exothermically
based on: 2CO + O2 2CO2 and 2SiO + O2 2SiO2 with N2 assumed to be inert.
This is in theory a multi-component mixture undergoing turbulent combustion which
can be simulated using rhoReactingBuoyantFoam. The governing equations involv-
ing gaseous species in rhoReactingBuoyantFoam are the mass, momentum, energy,
and species conservation:

∂ρ
+ ∇ · ρ U = 0,
∂t
∂ρ U
+ ∇ · ρ U U = −∇ prgh − g · x∇ρ + ∇ · τ,
∂t (1)
∂ρYi
+ ∇ · ρ U Yi + ∇ · Ji = ẇi ,
∂t
∂ρh s ∂ρ K ∂p
+ ∇ · ρ U h s + + ∇ · ρ U K − + ∇ · Js = Q̇ r ,
∂t ∂t ∂t

where ρ is the density of gaseous mixture, U is the velocity, p is the pressure,

prgh = p − ρ g · x, τ is the viscous stress tensor, Yi is the mass fraction of ith species,
h s is the sensible enthalpy, and K is the specific kinetic energy. The diffusive flux
of species, Ji , is approximated based on Fick’s 1st law and the diffusion coefficient
computed based on momentum diffusivity under the assumption of Schmidt number
(Sc) is equal to one since turbulence is dominant. The heat flux, Js , is calculated
based on conductive heat transfer and by ignoring the effect of species diffusion.
The source terms in the energy equation, Q̇ r account for the heat production from
the chemical reactions. It should be mentioned that we have ignored the impact of
radiative heat transfer in this work. The source term in the species transport equation
(ẇi ), which takes into account the generation or consumption of species due to
chemical reactions, is computed based on Eddy Dissipation Concept (EDC) model.
Apart from the above conservation equations, there is governing equations related
to modelling the turbulent nature of these gaseous flow which in this work is based
on the k − ω SST model. The solver also treats the temperature dependent nature of
transport and thermodynamic properties of species as well as the equation of state
of the gas mixture (using perfect gas model).

Modelling Microsilica Evolution

To simulate the evolution of microsilica in the furnace hood, we employ quadrature

method of moment (QMOM)-based population balance modelling approach (see [6])
to solve for the evolution of a particle size distribution function (PSD) or number
28 K. J. Vachaparambil et al.

density function
∞ (NDF), n. The QMOM approach uses moments which are defined
N
as m k = L min n L k d L = i=1 wi L ik where L min is the smallest possible size of the
particles or the nucleation diameter for microsilica, wi and L i are the weight and
abscissa of ith node of the QMOM, and N is the total number of nodes used in the
quadrature approximation. It should be noted that the units of wi and L i are m−3 and
m respectively. The average size of particle, also known as Sauter mean diameter,
is computed as d32 = m 3 /m 2 . The governing equation for moments that describes
microsilica evolution is

∂m k N
+ ∇ · (U m k ) = ∇ · (D p ∇m k ) + L kmin J + kGwi L ik−1 . (2)
∂t i=1

where U is the flow velocity (which is assumed to be steady field which is not
impacted by the particle evolution), D p is the effective diffusion coefficient of par-
ticles in the turbulent flow (in m2 /s), J is the nucleation rate computed in Eq. 3, and
G is the growth rate of the microsilica particle size whose calculation is described
in Eq. 5.

Nucleation Rate

When the local SiO2 concentration reaches saturation, microsilica particles are nucle-
ated in the bulk—homogeneous nucleation. The critical size for nucleation or L min is
vC
calculated as L min = k4γ
B T lnS
where S is the supersaturation ratio (S = CSiO2 /Csat ), γ is
the surface energy of SiO2 molecule (= 3.2 × 10−2 J/m2 ), vC is the volume of a SiO2
molecule (= 4.536 × 10−29 m3 ), k B is the Boltzmann’s constant (= 1.3806 × 10−23
J/K), and T is the local temperature [4]. The nucleation rate (J ), which describes the
number of particles nucleated per unit volume per second, (based on [4]) is

G 16π γ 3 vC2
J = 10 exp
25
− = 10 exp
25
− . (3)
kB T 3(k B T )3 (lnS)2

Growth Rate

Once the microsilica particles nucleate, it essentially grows by mass transfer or

condensation of the SiO2 (which is in the gas phase) to the surface of the particle.
The driving flux for the growth of these particle (which has units of mol/m2 -s) is
given by
1/3
2 + 0.6Re1/2
p Sc
j = DSiO2 ,gas (CSiO2 − Csat ), (4)
Li

where Csat is the saturation concentration of SiO2 , CSiO2 is the local concentration
SiO2 , Sc is the Schmidt number is set equal to one (due to the turbulent nature of the
Post Processing Approach to Model Microsilica Formation 29

flow), DSiO2 ,gas is the diffusion coefficient of SiO2 in the gas (which is approximated
to be equal to the viscosity when Sc = 1), L i is the abscissa of ith node, and Re p
is the Reynolds number of particle whose diameter is L i . Using the driving flux
for particle growth ( j) defined in Eq. 4, particle growth rate (F) can be written as
F = jπ L i2 where π L i2 is the surface area of the particles assuming they are spherical.
M
The corresponding growth rate of the particle volume is equal to F ρSiOp 2 where molar
mass of SiO2 is represented by MSiO2 (= 6.008 × 10−2 kg/mol). The growth rate of
the particle size, G, used in Eq. 2 is computed based on dL i /dt [4] as

dL i MSiO2
G= = 2j . (5)
dt ρp

Coupling of Microsilica Evolution with Species Consumption

The evolution of microsilica is associated with the reduction of SiO2 from the gaseous
phase. If this coupling is not present, the microsilica particles grow perpetually. Since
the microsilica formation is decoupled from the combustion simulation, we treat the
SiO2 solely as a species which gets depleted based on

∂CSiO2 m SiO2 N
+ ∇ · (U CSiO2 ) = ∇ · (D∇CSiO2 ) − J − jπ wi L i2 , (6)
∂t MSiO2 i=1

where CSiO2 is the concentration of SiO2 (in units of mol/m3 ), D is the effective
diffusion coefficient of SiO2 in the gas which is computed as equal to the sum of
momentum and turbulent diffusivities (based on assumption that Sc = 1), m SiO2 is
the mass of a single SiO2 molecule (= 9.96 × 10−26 kg [4]). The second last term
in Eq. 6 takes into account the SiO2 which is depleted as a result of nucleation of
microsilica. The last term in Eq. 6 considers the depletion
∞ of SiO2 due to growth of
the microsilica particles which is computed from L min FndL.
Information on the implementation of the framework and its verification is avail-
able in Sections S-1 and S-2.

Simulation of Microsilica Evolution

Setup Description

The furnace hood, which is the region of interest in this work, is simplified to a two-
dimensional geometry as shown in Fig. 1 similar to the setup used by [5] to test the
proposed framework and get some preliminary insights into the process. The domain
used for the simulation is meshed using 19600 hexahedral cells. The boundary con-
ditions used in the combustion simulation are summarized in Tables S-1 and S-2.
30 K. J. Vachaparambil et al.

Fig. 1 Schematic of the simplified 2D geometry of the furnace hood (region within dashed green
line in the schematic of furnace) used in the simulations

Based on the inlet flow rates used in the simulation, the reactions in the furnace hood
occur in an oxygen rich environment. The species-specific thermophysical properties
and the data used for the reaction kinetics parameters to describe the combustion of
CO and SiO are set based on [7]. The inputs for the population balance simulations
are obtained from [4] except for the Csat which is set to value of 0.01 mol/m3 (since
it is expected to be in the range of 0–1 mol/m3 according to [4]). The population
balance solver is modelled with four moments as N = 2 is used for the simulations.
The combustion simulation is assumed to be converged when the variation of the
outflow of SiO2 (at the outlet boundary) is less than 5% with time and bottom right
region in the domain (qualitatively) reach a steady state. The combustion simulations
are solved using a first order pseudo-transient time stepping approach, and spatial
discretization is based on limited linear scheme (analogous to the setup used in [7]).
For the microsilica evolution simulation, a first order implicit time stepping scheme
along with second order spatial discretization scheme is used. Both these simulations
are run with convergence criteria of 10−6 .

Results

The distribution of CO, CO2 , SiO, SiO2 , T and flow velocity obtained from the com-
bustion simulations (at steady state) is plotted in Fig. S-4. The velocity streamlines
show that flow in this simple domain is quite non-uniform. It is observed that reac-
tions involving the reactants coming up from the charge surface and the colder air
coming from the sides occur only in the rather small region where reactants come
into contact with the oxygen due to the entrainment. Ensuing reactions results in
locally high temperatures and high concentration of CO2 and SiO2 . Also SiO seems
to be depleted much quickly when compared to CO - due to faster reaction kinetics
associated with SiO combustion, see [4, 7].
Post Processing Approach to Model Microsilica Formation 31

Fig. 2 Contour plots of the concentration of SiO2 (in mol/m3 ) at t = 0 s (left) and t = 30 µs (right)
associated with the microsilica evolution

The depletion of SiO2 due to the microsilica evolution is visualized in Fig. 2. The
distribution of moments that describes the microsilica evolution is shown in Fig.
S-5 and Fig. 3. As expected particles are generated in the high temperature where
the combustion reaction generates SiO2 . The average particle size predicted by the
framework, based on the peaks of m 3 and m 2 observed in Fig. 3, is calculated as
d32 = m 3 /m 2 to be around 30 nm. Interestingly the value of d32 obtained from the
model is close to lower limit of the microsilica size (typically in the range of 10–500
nm with a mean size that is about 0.1 μm) reported in literature, see [1]. The work
by Dingsøyr et al. [2] reported that the microsilica consists of submicron structures
consisting of primary agglomerates (or clusters) of size varying from 0.05–1 µm and
micron sized range of particles with diameter between 1–100 µm. Dingsøyr et al. [2]
also reported that the number and size of the spheres (constituting clusters or primary
agglomerates) depend on the production process and that these clusters consisted of
particles stabilized by ‘material bridges’ whereas the larger micron sized particles are
hypothesized to be formed by van der Waals forces. A later study, [3], investigated
the structures in a densified silica fume (a process used to increase the density of the
microsilica by increasing agglomeration [1]) and reported observing clusters made
of spheres (whose sizes range between 0.03–0.16 µm) are ‘fused at points of contact’
due to sintering. Interestingly, lower size of these individual spheres observed in the
experiments, which are in the range of tens of nanometers [2, 3], is around the same
range at the particle size predicted by the framework.

Conclusion

In this work, we have presented a OpenFOAM-based framework to model microsil-

ica in the furnace hood. The framework relies on a main assumption that the chemical
reactions occur in an oxygen rich environment which allows us to decouple the com-
bustion and microsilica evolution. The framework performs combustion simulation
using rhoReactingBuoyantFoam solver in OpenFOAM whose results are used as an
32 K. J. Vachaparambil et al.

m2
Fig. 3 Transient evolution of relative total area of particles (normalized second moment) or 2
m0 L
m3
(left) and relative total volume of particles (normalized third moment) or 3 (right) at the outlet
m0 L
boundary obtained from the population balance solver with m 0 = 10 m and L = 20 × 10−9 m
18 −3

input for the microsilica simulation based on population balance modelling based
on QMOM. The proposed framework is used to simulate a simplified 2D model of
the furnace hood. The results show that the SiO2 is generated in a relatively narrow
region where the reactants and oxygen mix due to entrainment. Despite the simpli-
fications used in this work, the size of particles predicted by the framework (around
30 nm) seems to be in the lower limit of the particle sizes typically reported in lit-
erature. Source of this error is most likely due to the model not treating the particle
sintering which has been reported in [2, 3] - which should be investigated in future
works. Additionally, future works should include the impact of microsilica evolution
on the temperature field as well as integrate the population balance model with the
combustion model into a fully coupled solver.

Acknowledgements The current work has been funded by SFI Metal Production (Centre for
Research-based Innovation, 237738). The authors gratefully acknowledge the financial support
from the Research Council of Norway and partners of the center.

References

1. ACI 234R-96 (reapproved 2000): Guide for the use of silica fume in concrete. Technical report,
American Concrete Institute (1996)
2. Dingsøyr E, Dåstøl E, Wedberg FRW, Kuntz M, Langtry R (1992) Particle size and particle size
distribution of Elkem microsilica and its relevance to technical application. In: 5th European
Symposium Particle Characterization, Nurnberg, Germany
3. Diamond S, Sahu S (2006) Densified silica fume: particle sizes and dispersion in concrete. Mater
Struct 39(9):849–859
4. González-Fariña R, Münch A, Oliver JM, Van Gorder RA (2020) Modeling microsilica particle
formation and growth due to the combustion reaction of silicon monoxide with oxygen. SIAM
J Appl Math 80(2):1003–1033
5. González-Fariña, R.: Modelling the mechanisms of microsilica particle formation and growth.
PhD thesis, University of Oxford (2021)
Post Processing Approach to Model Microsilica Formation 33

6. Passalacqua A, Heylmun J, Icardi M, Madadi E, Bachant P, Hu X (2019) Open-

QBMM/OpenQBMM: OpenQBMM 5.0.1 for OpenFOAM 7. [Link]
3541425
7. Vachaparambil KJ, Panjwani B, Olsen JE (2022) Gassing during tapping of silicon. JOM
74:3990–3998
Thermodynamics and Kinetics of Coke
Breeze Combustion Under Different
Oxygen Content in the Sintering Process

Dongqing Wang, Wen Pan, Zhixing Zhao, and Yapeng Zhang

Abstract In order to investigate the influence of oxygen content in sintering flue

gas on the heat generated by coke combustion and sintering flue gas, the thermo-
dynamic calculation was carried out with FactSage software. Combustion kinetics
under different oxygen content was studied with thermogravimetric analyzer. Ther-
modynamic research results showed that under the condition of sintering with iron
ore fines, the oxygen content in the combustion-supporting air was abundant. When
the oxygen content was more than 7%, the solid fuel in the sintered layer could
meet the combustion conditions and release all the heat. Kinetic experiment results
showed that the combustion rate of coke breeze in 5 min with 13% oxygen content in
flue gas was equivalent to combustion rate of coke breeze in 2.75 min 21% oxygen
content in flue gas. The research is helpful for energy recovery of sintering process
and reducing carbon emission of blast furnace ironmaking.

Keywords Low-oxygen gas · Sinter · Coke breeze combustion · Thermodynamics

and kinetics

Introduction

The sintering process is an important production link in the iron and steel industry,
and its energy consumption accounts for about 11–16% of the total energy consump-
tion of the industry, and the pollutants emitted by sintering account for about 50% of
the total emissions of steel enterprises [1, 2]. In April 2019, the Ministry of Ecology
and Environment of the People’s Republic of China [2019] No. 35 “Opinions on
Promoting the Implementation of Ultra-Low Emissions in the Iron and Steel Indus-
try”, the oxide emission limit of the sintering machine head flue gas, pellet roasting

D. Wang (B) · W. Pan · Z. Zhao · Y. Zhang

Beijing Key Laboratory of Green Recyclable Process for Iron and Steel Production Technology,
Beijing 100043, China
e-mail: wangdongqing1980@[Link]
Institute of Iron and Steel Department, Research Institute of Technology, Shougang Group
Corporation, Beijing 100043, China

© The Minerals, Metals & Materials Society 2023 35

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
36 D. Wang et al.

flue gas particulate matter, SO2 , and nitrogen was revised to 10, 35, and 50 mg/m3 ,
respectively, in order to achieve the ultra-low emission target of iron and steel enter-
prises [3]. Reusing part of the external exhaust gas can greatly reduce the emission of
pollutants. Sintering flue gas is characterized by large amount, low oxygen, and high
pollutant content. Typical atmospheric pollutants emitted include SO2 , NOx , smoke
dust, and dioxins. Many scholars have conducted researches on flue gas treatment
and desulfurization and denitrification, which have a very positive effect on reducing
pollutant emissions. Zhang et al. [4] studied the changes in the denitration perfor-
mance of activated coke in the sintering flue gas purification system and its influence
on the denitration rate of sintering flue gas. Yu et al. [5] proposed a multi-pollutant
ultra-low emission technology system that implements source emission reduction,
process control, and end treatment in the sintering process. Ye et al. [6] studied the
generation mechanism of flue gas and pollutants and the relationship between flue
gas volume and air leakage rate. Wang et al. [7] proposed a technology for syner-
gistic removal of multiple pollutants in the sintering process. Hu et al. [8] proposed to
reduce pollutant emissions through pre-reduction sintering. Zhang et al. [9] analyzed
the influence of solid fuel combustion and iron-containing raw materials on the gener-
ation and emission of NOx in sintering flue gas during the sintering process. Zhu et al.
[10] studied the greenhouse gas COx emission law of iron ore sintering process. Cai
[11] proposed to reduce air consumption and waste gas production to reduce the
total emissions of pollutants and improve the regional atmospheric environment.
Some scholars have found that: Adding an appropriate amount of Ca–Fe oxide in the
sintering raw material can promote the reduction of sintering NOx to a certain extent
[12–14]. Xu et al. [15] carried out a numerical simulation on the flue gas distributor
and the circulating fume hood of the flue gas circulation system and calculated the
velocity field and temperature field of the sintering circulation system. Pei et al. [16]
analyzed the reasonable selectivity of the sintering flue gas circulation process. Zhou
et al. [17] optimized the flue gas temperature, the active control of flue gas dust, and
the action time of flue gas by numerical simulation. Yang et al. [18] used FLUENT
simulation software to simulate the flow field of the flue gas distributor, mixer, and
circulation cover of the circulation system. Loo [19] proposed increasing the particle
size of coke breeze to increase the thermal efficiency of the sinter bed. As is known,
the mineralization characteristics of sintered ore at different positions of the sintering
machine are not the same. Therefore, to develop sintering flue gas recycling tech-
nology, it is necessary to give full play to the characteristics of sintering flue gas at
different stages to achieve energy saving, emission reduction, and improvement of
sintering quality. In order to investigate the influence of the change of oxygen content
in the sintering flue gas on the heat generated by the combustion of sintering flue
gas and coke breeze, this paper studies the characteristics of the combustion of coke
breeze under different oxygen content from both thermodynamics and kinetics, in
order to provide technology for the recovery and utilization of flue gas.
Thermodynamics and Kinetics of Coke Breeze Combustion Under … 37

Table 1 Compositions of coke breeze %

Sample FCd Vd Ad Ash
SiO2 Al2 O3 CaO MgO
Coke breeze 84.90 1.10 14.00 45.00 30.00 4.00 0.60

Table 2 Gas parameters for

Total gas flow rate O2 Ar
thermogravimetric
experiments (ml) 1 150 7.5 142.5
2 150 19.5 130.5
3 150 31.5 118.5

Research Methods

FactSage software was used to conduct thermodynamic research on flue gas combus-
tion with different oxygen content, and FactPS is selected as the database. Kinetic
experiments were conducted with the German Netzsch Simultaneous Thermal
Analyzer STA449C. The first program by thermobalance chamber is maintained
at 30 °C, subjected to vacuuming operation three times, and then, 5 mg (100 °C
air drying oven for 2 h) breeze (see Table 1 for composition) was placed in a hot
balance crucible. Purge the thermobalance with a flow rate of 50 mL/min argon.
After the entire purge process was carried out for 20 min, the impurity gas in the
reaction system was fully discharged. Then, an argon atmosphere (99.999% purity)
at 20 °C/min temperature was raised to 1280 °C, lasting for 2 min, through a mixed
gas (argon + oxygen) 150 ml (oxygen volume content accounting for the mixed gas:
5, 13, 21%), the experiment ended after 5 min. The specific experimental results are
shown in Table 2.

Discussion

Distribution and Characteristics of the Combustion Zone

Pan et al. [20] tested the material bed temperature of a 198 m2 sintering machine
in a plant, and the results are shown in Fig. 1. It can be seen from Fig. 1 that the
high temperature holding time of each layer of the combustion zone is different. The
upper sinter combustion zone lasts for a short time, about 2 min; the lower sinter
combustion zone lasts longer, close to 5 min; the middle combustion zone lasts in
the middle. The highest temperature of the upper combustion zone is about 100 °C
lower than that of the lower combustion zone, and its high temperature retention time
above 1200 °C is about one third of that of the lower combustion zone. The reason is
that the sintered material layer is self-storage, and the heat is gradually accumulated
38 D. Wang et al.

Fig. 1 Characteristics of combustion zone temperature

in the sintered material layer from top to bottom, resulting in less heat in the upper
layer and more heat in the lower layer.

Thermodynamics of Combustion Flue Gases with Different

Oxygen Contents

In order to investigate the influence of the change of oxygen content in the sintering
flue gas on the heat, thermodynamic calculations were carried out with FactSage7.2
software. The calculation process refers to the ratio of coke breeze and flue gas in
the actual sintering process. The calculation process of FactSage software is: 40 kg
coke breeze and 1200 m3 of 100 °C flue gas are mixed and burned, and the heat
released by the reaction is used to heat the system (adiabatic process). Balance the
thermodynamic parameters such as temperature and gas composition of the gas.
The result calculated by FactSage software thermodynamics was substituted into the
gas-specific heat capacity formula at constant pressure. Calculate the calorific value
released when the gas temperature of the equilibrium system was cooled to 125 °C.
The oxygen content was increased from 3 to 21%, CO2 was fixed at 3%, CO
was fixed at 0.5%, and the nitrogen content was adjusted accordingly to ensure
that the total volume percentage was 100%. It can be seen from Fig. 2 that as the
oxygen content increases, the heat released by the gas first increases and then slightly
decreases. The inflection point is that the oxygen content is 7%. Analyzing the
reasons, it is believed that when the oxygen content is low, the coke breeze is incom-
pletely burned and a large amount of CO is generated, so the combustion heat of the
Thermodynamics and Kinetics of Coke Breeze Combustion Under … 39

Fig. 2 Comparison of heat corresponding to the increase of oxygen content in flue gas (reduced to
125 °C)

system is low. As the oxygen content increases, the coke breeze is completely burned,
CO disappears, all carbon becomes CO2 , and the heat generated by the combustion
of the system remains basically unchanged.
Therefore, it can be considered from the perspective of thermodynamics that the
oxygen content in the combustion-supporting air is a large amount of excess under
the conditions of iron ore breeze sintering. When the oxygen content is greater than
7%, the solid fuel in the sintering layer can meet the requirements to complete the
combustion.
Adequate thermodynamic conditions do not mean that the fuel in the low-oxygen
content flue gas sinter layer can be completely burned. The fuel combustion process
is greatly affected by the high temperature duration of the combustion zone, oxygen
content, and moisture. Therefore, kinetics research is needed.

Kinetic Research

The thermogravimetry instrument automatically records the experimental data,

obtains the W-t (time) curve, and calculates the combustion conversion rate of coke
breeze using formula:

X = W/W0 (1 − Vd − Ad ) (1)

W —coal char participates in the gasification reaction and loses its quality; W 0 —
the initial quality of the sample; V d —the volatile content in coke breeze; Ad —the
ash content in coke breeze.
40 D. Wang et al.

Table 3 Combustion
Time/minute 5% O2 13% O2 21% O2
conversion of coke breeze at
different time 1.0 5.87 16.39 27.06
2.5 20.24 45.45 81.17
5.0 44.04 87.30 100.00

Table 4 Calculus
Model f (α) G(α)
expressions for common
gas–solid reactions Homogeneous 1−α − ln(1 − α)
reaction model
Unreacted shrinking 2(1 − α)2/3 1 − (1 − α)1/3
core model (sphere)
Two-dimensional −[ln(1 − α)]−1 α + (1 − α) ln(1 − α)
diffusion model

The coal coke conversion rate is first-order differentiated from the corresponding
gasification reaction time t, and the combustion conversion rate dx/dt of coke breeze
is obtained. The experiment was carried out for 1 min, 2.5 min, and 5 min. The data
of the combustion conversion rate of coke dust is shown in Table 3. It can be seen
from Table 4 that the higher the oxygen content in the atmosphere, the shorter the
time required to reach the same combustion rate. Probably the main reason is that
the decrease of oxygen volume fraction inhibits the diffusion between O2 and coke
breeze.
Figure 3 is a graph showing the influence of different oxygen content on the
combustion conversion rate of coke breeze. It can be seen from Fig. 3 that at the
beginning of the combustion reaction, the lower the oxygen content, the lower the
combustion conversion rate of coke breeze.
For the typical gas–solid reaction of coal char gasification, the reaction process can
be described by a homogeneous model, an unreacted shrinkage nucleus model, and
a two-dimensional diffusion model [21]. The calculus expressions of common gas–
solid reactions are shown in Table 4. If the model is selected appropriately, G(α) has a
good linear relationship with t. According to statistical regression analysis, according
to its maximum correlation coefficient, the appropriate response model f (α) can be
determined. Linear correlation coefficient of different models with different oxygen
content is shown in Table 5. Kinetic models of different reaction stages are shown
in Table 6 (Fig. 4).
According to the above test results, the combustion rate of coke breeze in 5 min
with 13% oxygen content in flue gas was equivalent to combustion rate of coke
breeze in 2.75 min 21% oxygen content in flue gas. Therefore, if the combustion
time is enough, the heat released by coke breeze combustion in low-oxygen flue gas
is basically equal to that of high-oxygen flue gas, and low-oxygen flue gas with an
oxygen content of 13% or above can also conform to the requirements of sintering
production.
Thermodynamics and Kinetics of Coke Breeze Combustion Under … 41

Fig. 3 Influence of different oxygen content at 1280 °C on the combustion conversion rate of coke
breeze

Table 5 Linear correlation coefficient of different models with different oxygen content
O2 , % Time/minute Model Correlation coefficient
5 0.25 < t ≤ 1.25 Homogeneous reaction model 0.996
Unreacted shrinking core model (sphere) 0.995
Two-dimensional diffusion model 0.942
1.25 < t < 5 Homogeneous reaction model 0.998
Unreacted shrinking core model (sphere) 0.999
Two-dimensional diffusion model 0.979
13 0.25 < t ≤ 1.75 Homogeneous reaction model 0.999
Unreacted shrinking core model (sphere) 0.998
Two-dimensional diffusion model 0.968
1.75 < t < 5 Homogeneous reaction model 0.985
Unreacted shrinking core model (sphere) 0.995
Two-dimensional diffusion model 0.985
21 0.25 < t ≤ 1.25 Homogeneous reaction model 0.999
Unreacted shrinking core model (sphere) 0.997
Two-dimensional diffusion model 0.968
1.25 < t < 2.75 Homogeneous reaction model 0.985
Unreacted shrinking core model (sphere) 0.995
Two-dimensional diffusion model 0.985
42 D. Wang et al.

Table 6 Kinetic models of different reaction stages

O2 (%) Time, minute Formula equation Kinetic models Correlation
coefficient
5 0.25 < t ≤ 1.25 x = 1−e0.022–0.082t Homogeneous reaction 0.996
13 0.25 < t ≤ 1.75 x = 1−e0.035–0.218t model 0.999
21 0.25 < t ≤ 1.25 x = 1−e0.108–0.41t 0.999
5 1.25 < t < 5 x= 1−(1.024–0.04t)3 Unreacted shrinking 0.999
13 1.75 < t < 5 x= core model (sphere) 0.995
1−(1.093–0.112t)3
21 1.25 < t < 2.75 x= 0.995
1−(1.195–0.254t)3

Fig. 4 Comparison of coke combustion rate and fitting curve value at 1280 °C with different oxygen
content

Conclusions

From the perspective of thermodynamics, under the sintering condition of iron ore
breeze, the oxygen content in the combustion-supporting air is abundant. When the
oxygen content is greater than 7%, the solid fuel in the sintered material layer can
meet the combustion conditions and release all the heat.
At the initial stage, the combustion reaction conforms to the characteristics of
the kinetic model of homogeneous reaction. In the middle and late period of fuel
combustion, the kinetic model is consistent with the characteristics of the unreacted
shrinking nucleus model. Kinetic experiment results showed that the combustion
rate of coke breeze in 5 min with 13% oxygen content in flue gas was equivalent
to combustion rate of coke breeze in 2.75 min 21% oxygen content in flue gas.
Therefore, as long as the combustion time is enough, the heat released by coke
breeze combustion in low-oxygen flue gas is basically equal to that of high-oxygen
Thermodynamics and Kinetics of Coke Breeze Combustion Under … 43

flue gas, and the low-oxygen flue gas with an oxygen content of 13% or above can
also conform to the requirements of sintering production.

References

1. Yan XM, Li YR, Zhu TY et al (2015) Review of emission and simultaneous control of multiple
pollutants from iron-steel sintering flue gas. J Environ Eng Technol 5(2):85–90
2. Liu J, Ma XG, Xia TY et al (2019) Study and application of recycled desulfurization ash of
sintering flue gas. Sinter Pelletizing 44(2):63–68
3. Ministry of Ecology and Environment of the People’s Republic of China. Opinions on
promoting the implementation of ultra low emissions in the steel industry [EB/OL]. [2019–
04–28]. [Link]
4. Zhang WL, Wu SL, Hu ZJ (2020) Analysis of affecting factors of activated coke denitration
efficiency for sintering flue gas. Iron Steel 55(05):109–115
5. Yu Y, Zhu TY, Liu XL (2019) Progress of ultra-low emission technology for key processes of
iron and steel industry in China. Iron Steel 54(9):1–11
6. Ye HD, Wang ZC, Liu Q et al (2019) Study on reduction technology of sintering flue gas and
pollutants. Sinter Pelletizing 44(06):60–67
7. Wang XD, Hou CJ, Tian JL (2020) Application and practice of multi-pollutant cooperative
control technology for flue gas in iron and steel industry. Chin J Process Eng 20(09):997–1007
8. Hu B, Gan M, Wang ZC (2017) Research status of pre-reduction sintering technology and
development on the new technology. Sinter Pelletizing 42(06):22–26, 28
9. Zhang DB, Zhou MJ, Wang YF (2015) Emission reduction technology of NOx process in
sintering flue gas of baosteel. Ironmaking 34(6):47–51
10. Zhu DQ, He AP, Pan J et al (2006) Rule of greenhouse Gas COx emission in iron ore sintering
process. Iron Steel 41(2):76–80
11. Cai JJ (2019) Air consumption and waste gas emission of steel industry. Iron Steel 54(04):1–11
12. Wu SL, Sugiyama T, Morioka K et al (1994) Elimination reaction of NO gas generated from
coke combustion in iron ore sinter bed. Tetsu-to-Hagané 80(4):276–281
13. Koichi M, Shinichi I, Masakata S et al (2000) Primary application of the “in-bed-de NOx ”
process using Ca–Fe oxides in iron ore sintering machines. ISIJ Int 40(3):280–285
14. Gan M, Fan XH, Lv W et al (2016) Fuel pre-granulation for reducing NOx emissions from the
iron ore sintering process. Breeze Technol 301:478–485
15. Xu Y, Mao R, Wang F et al (2020) Numerical simulation of sintering flue gas circulation system.
32(07):675–681
16. Pei YD, Zhang JJ, Qi CL (2019) Discussion on proper sintering flue gas recirculation process.
Sinter Pelletizing 44(2):74–77
17. Zhou ZA, Su D, Zhang SP et al (2019) Research and application of efficientcy clicutilization
technology of sintering residue energy. Sinter Pelletizing 44(2):69–73
18. Yang ZW, Wang ZC, Wen RY et al (2018) Simulation research of sintering flue gas recirculation
system. Sinter Pelletizing 43(3):63–68
19. Loo CE (1991) Role of coke size in sintering of a hematite ore blend. Ironmaking Steelmaking
18:33
20. Pan W, Zhao ZX, Feng L et al (2020) Research on mineralization characteristic of sinter. Sinter
Pelletizing 45(03):1–7, 12
21. Sun MM, Zhang JL et al (2018) Thermal and kinetic analysis on the co-combustion behaviors
of anthracite and PVC. Metall Res Technol 1–9
Part II
Continuous Casting
Inverse Calculation of Time-Spatial
Varying Mold Heat Flux During
Continuous Casting from Fast Response
Thermocouples

Haihui Zhang, Huiqiang Shen, and Pengcheng Xiao

Abstract This paper covers the determination of time-spatial varying mold heat flux
using mold temperatures measured by fast response thermocouples at a frequency up
to 60 Hz. A two-dimensional transient inverse heat conduction problem (2DIHCP)
is established, where 2DIHCP is developed based on the sequential function spec-
ification method implemented with the spatial regularization terms to reduce the
fluctuations in the estimated heat flux in both time and spatial domain. Then, the
inverse problem was validated using a designed numeric test-problem. Finally, the
inverse problem is applied to calculating the heat flux across the mold hot surface
for a continuous casting trial using the mold simulator.

Keywords Modeling and simulation · Iron and steel · Process technology ·

Continuous casting

Nomenclature

c Heat capacity (J/kg K)

f (∂4 , t) Temperature of the boundary ∂4 (K)
H Spatial regularization matrix
X Sensitivity matrix
M Numbers of the thermocouple in the calculated domain
nj Number of heat flux components at the boundary ∂j , j =
1, 2, 3.
N N = n1 + n2 + n3
q Heat flux (W/m2 )

H. Zhang (B) · H. Shen

Faculty of Materials Metallurgy and Chemistry, Jiangxi University of Science and Technology,
Ganzhou 341000, China
e-mail: zhanghaihuiemail@[Link]
P. Xiao
College of Metallurgy and Energy, North China University of Science and Technology University,
Tangshan 063210, China

© The Minerals, Metals & Materials Society 2023 47

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
48 H. Zhang et al.

r Future time step (−)

s Objective function
t, t j Time (s)
T Vector of estimated temperature (K)
Tm Estimated temperatures at the measurement location (x m , ym )
(K)
T ini Initial temperature (K)
x, y Cartesian spatial coordinates (m)
Y Vector of the measured temperature (K)
Ym Measured temperature at the measurement location (x m , ym )
(K)

Greek Symbols

α Regularization parameters (K2 m4 W4 )

λ Thermal conductivity (W m− 1 K− 1 )
ρ Density (kg m− 3 )
σ Standard deviation of the measurements
ω Random variable
Calculated domain
∂1 , ∂2 , ∂3 , ∂4 Boundary of the calculated domain

Introduction

Many surface defects in final rolled steel products originate from the initial solidi-
fication of molten steel inside continuous casting molds. This has been found to be
associated with the heat transfer behaviors of the mold itself [1–3]. The mold heat
flux, especially at the mold meniscus area, is extremely complex due to the tran-
sient nature of the infiltration of lubricant liquid mold flux, intensive fluid flow, and
mold oscillation, such that it would be very difficult to get a clear comprehensive
understanding of all dynamics within the system [4–8].
The heat flux of the mold is a space–time varying variable [9]. Usually, the mold
heat transfer is monitored by temperature sensors [10]. Hundreds of thermocou-
ples/sensors are employed to monitor the mold temperature during the continuous
casting process, with a temperature sampling rate of 1–65 Hz. The mold heat flux
cannot be calculated directly from the measured temperatures because of the lack of
information on the boundary conditions. Mathematically, the determination of the
mold heat flux from measured temperatures is an inverse problem that means the
solution is usually unstable, not unique or does not exist, and a small measured error
in the temperatures will result in a large error of heat flux [11, 12].
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 49

The inverse problem method has a broad application prospect in determining

the heat flux from measured temperatures. The principle of inverse problems for
restructuring the heat flux is to find out a heat flux with the maximum probability
making the sum of squared deviations between the calculated temperatures T and
the measured temperatures Y to be minimum [13].
The development of the use of inverse problem methods to calculate continuous
casting mold heat flux from temperature measurements can be traced back to the
works by Brimacomb et al. [14], followed by Thomas et al. [15], Wang et al. [16–
18], Griffiths [19], Rajaraman [20], and Talukdar et al. [21]. Those inverse problem
methods could be classified as the gradient-based type methods, such as the func-
tion specification method [9], the Tikhonov regularization [22] and the conjugate
gradient method [23, 24], and the stochastic-based type methods, such as the genetic
algorithm and the neural network algorithm [25]. However, the major weakness of
the inverse problem persists in that it is extremely sensitive to temperature measure-
ment errors, particularly as the time step is made smaller [12, 26]. As continuous
casting technology progresses, fast mold thermal monitoring systems are adopted that
could provide a more accurate detection precision supervising the fluctuation of mold
temperature [5, 6]. However, the faster temperature sampling rate of mold thermal
monitoring systems is inevitably accompanied by a higher intensity of the error/noise
in the measured temperatures [12]. As a result of the increase of the temperature
sampling rate, the use of small-time step frequently introduces instabilities in the
solution of the inverse heat conduction problem.
In the engineering community, the inverse problem of Beck’s sequential function
specification method is well-known and very successfully used in solving inverse
heat conduction problems for over 50 years [27–31]. For Beck’s sequential function
specification method, the heat flux form is assumed to be a constant function or a
linear function over several future time steps due to the fact that the temperature
response is lagging with respect to the boundary heat flux [23], then the stabilization
of the solution in the time domain can be improved by choosing an appropriate
number of future time steps [24, 32]. However, a common issue raised is how to
improve the stabilization of the solution in the time and spatial domain for extending
the sequential function specification method to the two-dimensional heat transfer
problem [33, 34].
Therefore, the purpose of this work is to establish a two-dimensional transient
inverse heat conduction problem (2DIHCP) for the estimation of the mold heat flux
from fast-sampled temperature data. The inverse problem is developed where the
function specification method with a spatial Tikhonov regularization [6, 35, 36] was
used to improve the stabilization of the solution in time and spatial domains, respec-
tively. Then, a numeric test-problem was designed to validate the inverse problem.
Finally, the inverse problem is applied to calculating the mold heat flux during liquid
steel casting.
50 H. Zhang et al.

The Experimental and Direct Problem Description

The continuous casting trial was conducted using a mold simulator. As Fig. 1 shows,
the mold simulator applied to this study is an inverse type water-cooled copper mold
(30 mm × 50 mm × 350 mm) with oscillation capability. A U-shaped type water-
cooling groove with a diameter of 10 mm is manufactured along the center line
of mold plate, 12.5 mm away from both ends and gets connected 20 mm above
the bottom, where the water flows from one end to the other. The copper mold is
equipped with an extractor that makes only one face of the mold exposed to the
liquid melt. The temperatures in the mold are measured by T-type thermocouples
at a frequency up to 60 Hz through a data acquisition system. The two columns of
thermocouples are spaced 3 and 8 mm away from the mold surface, respectively, as
shown in Fig. 1, the dots represent the locations of thermocouples. The arrangement
allows the first column of thermocouples to catch the temperature history of the
mold that will be used in the minimization of the objective function of the inverse
problem. The second column of thermocouple inside the mold wall on ∂4 (CD)
is to provide the temperature boundary condition f (t), where the temperature is
interpolated linearly from two near measured temperatures for the nodes in between
the two thermocouples at ∂4 . This arrangement of those thermocouples in the mold
wall was chosen based on the study of Badri et al. [37] so as to improve the signal-
to-noise ratio for the temperature measurements during the experiment. Then, the
rectangular area ABCD with the height (AB) of H (= 21 mm) and the width (BC)
of W (8 mm) is set as the heat transfer computational domain, which consists of four
boundary conditions ∂1 (DA), ∂2 (AB), ∂3 (BC), and ∂4 (CD).
The experiment runs as follows: Step I The oscillating water-cooling copper mold
and the extractor are dipped into the hot melt of liquid steel. After the mold and
extractor reached the target depth, the mold flux level and the liquid steel level would
be located in the mold thermocouple-measuring zone. Step II The mold and extractor
were held for several seconds to form an initial shell on the water-cooled copper mold
to ensure the initial shell is strong enough to prevent tearing during extraction. Step III
The extractor withdrew the solidifying shell downward at a constant speed to simulate
continuous casting. The mold moved upward at a certain speed to compensate for
the rise of the mold level, so that the liquid level could be kept in the same position
with respect to the mold. Step IV When the casting was completed for the desired
length, the mold and extractor were withdrawn out of the furnace and then cooled in
air.
By assuming the mold heat transfer is two-dimensional in the vertical section
ABCD perpendicular to the mold hot surface. The heat transfer within the mold
vertical section ABCD is assumed to be governed by Fourier partial differential
equations. The direct problem, namely the problem of 2D heat transfer within the
rectangular area ABCD, as shown in Fig. 1, is governed by the following equations.

∂T ∂ ∂T ∂ ∂T
cρ = λ + λ , in domain = [0, W ] × [0, H ] (1a)
∂t ∂x ∂x ∂y ∂y
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 51

Fig. 1 Computational area of the mold and locations of thermocouples for the casting experiment

∂ T (∂1 ∪ ∂2 ∪ ∂3 , t)

−λ = q to be estimated. (1b)
∂n

T (∂4 , t) = f (t) (1c)

T (x, y, 0) = Tini (1d)

where the symbol c represents the specific heat in J kg−1 , ρ is the density in kg m−3 ,
T is the temperature in K, t is the time in second, and λ is thermal conductivity in
W m−1 K−1 , q (W m−2 ) is the heat flux of boundary conditions ∂1 , ∂2, and ∂3 ,
n is the outer normal of boundary.

Inverse Problem Description

In this section, the inverse problem is developed based on the function specification
method implemented by a spatial Tikhonov regularization [6, 35, 36]. The sequential
function specification procedure is: (1) the heat fluxes are assumed to remain constant
over the r future time steps, qj = qj+1 = · · · = qj+r , owing to the temperature response
is lagging with respect to the boundary heat flux, and (2) By knowing the heat fluxes
for t < t j , namely q̃1 , q̃2 , …, q̃ j−1 are known, estimate q j so that makes the sum of the
squares of the deviations between the calculated temperatures T and the measured
52 H. Zhang et al.

temperatures Y in the time interval [t j , t j+r−1 ] to be minimized. Then, the inverse

heat conduction problem could be converted into a problem that is to estimate q j by
making the following objective function to be minimum. Therefore, the definition of
the inverse heat conduction problem is made as follows

r −1
M
2
s(q ) =
j
Ymj+k − Tmj+k (q j ) + α R(q j ) (2)
k=0 m=1

j j
where r is the number of future time steps. Ym and Tm are the measured tempera-
tures and the calculated temperatures at the time of t j and the measured position m,
respectively. M is the number of sensors. α is the spatial regularization parameter.
The finite difference method is used to solve the above direct problem, where the
rectangular area ABCD () will be divided into grids, and the boundaries of ∂1 ,
∂2, and ∂3 will be split into n 1 , n 2 , and n 3 divisions, respectively. Then, q has N
j
components, N = n 1 + n 2 + n 3 , and qk represents the kth component of heat flux q
at the time of t j .
T
j j j j j j
q j = [q 1 , q 2 , q 3 ]T , q 1 = q1 , q2 , . . . , qnj1 ,
T
j j j j
q 2 = qn 1 +1 , qn 1 +2 , . . . , qn 1 +n 2 and
T
j j j j
q 3 = qn 1 +n 2 +1 , qn 1 +n 2 +2 , . . . , q N .

Besides, a first-order of spatial Tikhonov regularization term is added to the objec-

tive function Eq. (2) so as to improve the spatial stabilization of the solution. It should
be mentioned that heat flux q might be subjected to discontinuity at the intersection
points between two adjacent boundaries of ∂1 and ∂2 (∂2 and ∂3 ). Thus, the
spatial Tikhonov regularizations for the heat flux of the boundaries ∂1 , ∂2, and
∂3 are described separately. That is

1 −1
n n 1 +n
2 −1 N −1

j j j
R(q j ) = (qnj − qn+1 )2 + (qnj − qn+1 )2 + (qnj − qn+1 )2 (3)
n=1 n=n 1 +1 n=n 1 +n 2 +1

The estimated temperature T j could be evaluated using a Taylor series expansion

around the current solution q̃ j .

T j = T̃(q̃ j ) + X j (q j − q̃ j ). (4)

T̃(q̃ j ) is the temperature calculated using the current solution q̃ j . X j is called as the
M × N sensitivity coefficient matrix at time t j and is defined as follows.
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 53
⎛ j j j
⎞
X 1,1 X 1,2 · · · X 1,N
⎜ j j j ⎟
∂T(q j )
T
⎜ X 2,1 X 2,2 · · · X 2,N ⎟
Xj = ⎜
=⎜ . . ⎟
.. .. ⎟, where
∂q j ⎝ .. . . .. ⎠
j j j
X M,1 X M,2 · · · X M,N
j
∂ Tm
j
X m,n = j
. (5)
∂qn

The least squares equation s(qj ) is minimized by differentiating it with respect to

each component of unknown heat flux and setting the resulting expressions equal to
zero, then a set of N equations is obtained for the estimation of heat flux.

∂s(q j ) ∂s(q j ) ∂s(q j )

j
= 0, j
= 0, . . . , j
=0 (6)
∂q1 ∂q2 ∂q N

The above equations yield to a matrix system to estimate the increment for a new
heat flux.
r −1 r

q =
j
XiT Xi + αH XiT (Y j+i−1 − T̃ j+i−1
) (7)
i=1 i=1

where H is the N × N block diagonal regularization matrix. Both Yj and Tj are M

× 1 vector, and M is the number of measurements.
j j j
Y j = [Y1 , Y2 , . . . , Y M ]T (8a)

j j j
T j = [T1 , T2 , . . . , TM ]T (8b)

H = blkdiag(H n 1 , H n 2 , H n 3 ) (8c)

⎡ ⎤
1 −1
⎢ −1 2 −1 ⎥
⎢ ⎥
⎢ .. .. .. ⎥
H ni =⎢ . . . ⎥ . (8d)
⎢ ⎥
⎣ −1 2 −1 ⎦
−1 1 n ×n
i i
54 H. Zhang et al.

Sensitivity Coefficient Matrix

j
Xm,n represents the temperature rise at the sensor location (x m , ym ) for a unit step
change in the surface heat flux at the point (x n , yn ) of boundaries ∂1 , ∂2, and
∂3 and the time t j . According to the definition of sensitivity coefficient [Eq. (5)],
the sensitivity coefficient problem for calculating sensitivity coefficient matrix is
obtained by taking the derivative of Eqs. [1a through 1d] with respect to a heat flux
j
component qn at the point (x n , yn ) of boundaries ∂1 , ∂2, and ∂3 , and n is 1, 2,
j
…, N. The sensitivity coefficient problem governing the sensitivity coefficients Xm,n
is

∂ X ,n ∂ ∂ X ,n ∂ ∂ X ,n
cρ = λ + λ , 0 < t ≤ tr
∂t ∂x ∂x ∂y ∂y
in domain = [0, W ] × [0, H ] (9a)

∂ X ,n (∂1 ∪ ∂2 ∪ ∂3 , t) 1, (x, y) = (xn , yn )
−λ = (9b)
∂n 0, others

X ,n (∂4 , t) = 0 (9c)

X ,n (x, y, 0) = 0 (9d)

The Determination of Regularization Parameter

The choice of the regularization parameter is required to balance the computational

cost and the stability of the inverse problem solution algorithm. The L-curve method
is adopted to optimize the regularization parameters α. L-curve criterion that plotted
the curve of {log(||Y–T||), log(R(qj ))} often takes on a characteristic L shape, and
the optimal regularization parameter was corresponding to the corner of maximum
curvature in L-curve [38, 39]. However, many tests should be executed for the plot
L-curve, which is very computationally expensive. Alternatively, we consider the
use of Arcangeli’s discrepancy principle to significantly speed up the selection of the
optimal regularization parameter [40], that is
σ
Y − T ≈ √ (10)
α
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 55

Stopping Criterion for the Iterative Procedure of Inverse

Problem

The stopping criterion for Eq. (2) is given by

s≤ε (11)

As temperatures contain measurement error, the temperature residual could be

approximated by

ε = (Y − T)T (Y − T) ∼
= Mσ 2 (12)

where σ is the standard deviation of temperature measurement error.

Algorithm of Inverse Problem

As shown in Fig. 2, specify the number of future time steps r. Step 1, 2DIHCP is
running with an initial guess heat flux qj and a guess regularization parameter α. Step
2, solve the direct problem given by Eq. (1) with the estimated heat flux qj for T
during the time from t j to t j+r . Step 3, solve the sensitivity coefficient problem given
by Eq. (9a–d) for the sensitivity coefficient matrix X in the time interval from 0 to t r .
Step 4, check the stopping criterion given by Eq. (11). If the stopping criterion Eq.
(11) is satisfied, let α old = α and renew α using Eq. (10). Otherwise, the heat flux
is updated using qj = qj + q and Eq. (7), and return to Step 2. Step 5, when |α −
α old |/α old < 0.05 is satisfied, qj could be regarded as the estimated heat flux at time
t j , then set j = j + l and return to Step 1 for the next time step calculation till the end
of the time steps. Otherwise, the heat flux is updated using qj = qj + q and Eq. (7),
and return to Step 2. The algorithm is achieved and programed using MATLABTM .
All the partial differential equations are solved using the finite difference method.

Model Verification

Validation for Solving Partial Differential Equations

by Numeric Solution

Partial differential equations of heat conduction problems Eqs. [1a through 1d] and
the sensitivity coefficient problem given by Eqs. [9a through 9d] are solved using
the finite difference method (FDM) with the classical Crank-Nicolson (CN) semi-
implicit scheme. The FDM code must be validated to be effective for solving partial
56 H. Zhang et al.

Fig. 2 Algorithm of the

two-dimensional inverse heat
conduction problem
(2DIHCP)

differential equations before the run of the inverse calculation. Herein, the FDM
code was also verified through comparison with the analytical solutions of the heat
transfer in the solid from a classical textbook [41]. It was observed that the results
of the FDM method were consistent with those analytic solutions, suggesting the
FDM method can solve the heat conduction partial differential equation.

Validation for Algorithm of the Inverse Problem

The numeric test-problem is designed to verify the 2DIHCP. The direct problem is
also a the rectangular area ABCD (made of copper, height H is 0.021 m and width
W is 0.008 m, see Fig. 1) with the initial temperature of 273.15 K. The boundaries
of ∂1 , ∂3, and ∂4 are insulated boundaries, and the heat flux f(y, t) is applied to
the ∂2 . The f(y, t) is
⎧
⎪
⎪ 1 × 106 , 0≤t <5
⎪
⎪
⎪
⎨ 1 × 106 1 −
1000
y t + 1 × 106 , 5 ≤ t < 10
21 5
f (y, t) = 1 × 106 1 − 1000
y + 1 × 106 , 10 ≤ t < 15 . (13)
⎪
⎪ 21
⎪ 1 × 106 1 −
⎪
1000
y 15−t + 1 × 106 , 15 ≤ t ≤ 20
⎪
⎩ 1 × 106 , 21 5
20 ≤ t < 25

The first column of eight virtual thermocouples was set 3 mm apart vertically
from each other and 3 mm away from the mold hot surface; while the second column
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 57

of other eight virtual thermocouples was spaced 3 mm apart vertically and 8 mm

horizontally away from the hot face of the mold. The temperature measurement is
taken every 0.1 s during the test-problem runs. The Gaussian noise signals σ ω (T
= T ture + σ ω, σ is the standard deviation of noise and ω is a random variable and
will be within −2.576 to 2.576 for the 99% confidence bounds) are added to the
temperatures to mimic the thermocouple measurement error. Then, the measured
temperatures are delivered to the 2DIHCP for the reconstruction of the heat flux in
2 . For the parameters used here, the number of future time steps is set as 15, the α
is 1.71 × 10–6 , respectively. The estimation results of heat flux in 2 with/without
noise (Gaussian noise signals, σ = 0.1, to simulate the temperature measurement
errors) and the exact value f(y, t) are shown in Fig. 3, where the heat fluxes at the
locations of y at 0 mm, 3, 6, 9, 15, and 21 mm are listed, respectively. It was found that
the heat fluxes calculated by 2DIHCP match well with the exact values, even for the
situation that the measured temperature was contaminated by noise. This suggests
that the 2DIHCP could reconstruct the boundary heat flux precisely and show the
ability to resist measurement noises.
The results are compared with those values calculated by a robust one-dimensional
inverse heat conduction problem (1D-IHCP) developed by Beck et al. [12], where the
future time step is 4. As shown in Fig. 3, although the heat fluxes obtained through 1D-
IHCP show the same variation tendency with the exact value, the difference between
the above two methods is obvious, especially when the measured temperature is
contaminated with noise.

Application to Continuous Casting Mold

A continuous casting experiment of liquid steel ([C]: 0.200 wt%, [Si]: 0.230 wt%,
[Mn]: 0.490 wt%, [P]: 0.012 wt%, and [S]: 0.03 wt%) was conducted, where the
mold was oscillated with a frequency of 2.17 Hz that helps to separate the metal
from the mold. With the progressive filling of the liquid steel (shown in Fig. 1),
the responding copper mold wall temperatures were measured by thermocouples
with a 60 Hz sampling rate, and the pouring temperature of steel was 1833 K. The
calculation domain size and the distribution of the thermocouples are the same as
the installation of the thermocouples in the numeric test-problem.
Figure 4 shows the measured in-mold temperature history of the first column
thermocouples at different positions in the vertical. The responding temperature can
be divided into two stages according to the process of the casting. In stage I (51.4–
52.5 s), the responding temperatures increase quickly as the progressive filling of
the liquid steel from the bottom to the top of the mold. After filling of the melt is
completed, the liquid level corresponds to the location S9 with y is 9 mm, where
the mold is exposed to air above this location (y > 9 mm). As it steps into stage II
(52.5–58 s), the liquid steel is solidified against the surface of the water-cooled mold,
the responding temperatures continue to increase and then keep relatively constant.
It was found that the steady state temperature values for Ytc16 (y is 0 mm) and Ytc14
58 H. Zhang et al.

2.0 2.0
(a) Exact
y= 21 mm (b) y= 15 mm
2D-IHCP
Heat Flux (MW/m2)

Heat Flux (MW/m2)

2D-IHCP: noise in data
1.6 1.6
1D-IHCP
1D-IHCP: noise in data

1.2 1.2

0.8 0.8

0 5 10 15 20 25 0 5 10 15 20 25
Time (s) Time (s)
2.0 2.0
(c) y= 9mm (d) y= 6 mm

Heat Flux (MW/m2)

Heat Flux (MW/m2)

1.6 1.6

1.2 1.2

0.8 0.8

0 5 10 15 20 25 0 5 10 15 20 25
Time (s) Time (s)
2.0
(e) y= 3mm 2.0 (f) y= 0 mm
Heat Flux (MW/m2)

Heat Flux (MW/m2)

1.6
1.6

1.2 1.2

0.8 0.8

0 5 10 15 20 25 0 5 10 15 20 25
Time (s) Time (s)
Time (s)

Fig. 3 Heat fluxes calculated by 2DIHCP and 1D-IHCP for the test-problem: a heat flux at y =
21 mm. b heat flux at y = 15 mm. c heat flux at y = 9 mm. d heat flux at y = 6 mm. e heat flux at y
= 3 mm. f heat flux at y = 0 mm

(y is 3 mm) are the highest, followed by Ytc12 (y is 6 mm), Ytc10 (y is 9 mm), Ytc8 (y
is 12 mm), Ytc6 (y is 15 mm), Ytc4 (y is 18 mm), and Ytc2 (y is 21 mm). It is suggested
that the thermocouples corresponding to mold surface exposed to the air are cooler
than those of thermocouples below the melt level.
Figure 5 shows the mold heat fluxes calculated by 2DIHCP from measured mold
temperatures. For the 2DIHCP runs, the future time step is configured as 8, and α is 7
× 10–9 , respectively. During stage I of filling, it was found that the local heat fluxes
increase rapidly and then reach their peak values and the highest value is around
1.03 MW/m2 , which is similar to industrial and other experimental results [3, 5–8].
During stage II, all the heat fluxes decrease rapidly at first due to the fast formation of
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 59

Fig. 4 Responding in-mold

temperatures
I II Ytc16
Ytc14

Measured Temperature (K)

330 Ytc12
Ytc10
Ytc8
320 Ytc6
Ytc4
Ytc2
310

300

290

51 52 53 54 55 56 57 58 59
Time (s)

the shell/mold air-gap, and then they continue to decrease gradually. From the above
results, it could be suggested that during the initial filling of the liquid steel, the heat
fluxes increased dramatically as the hot liquid directly contacted with the cold mold
and the latent heat was released greatly with the increase of heat flux in stage I. Then,
the initial shell is formed against the hot surface of the water-cooling mold; thus,
the air gap in between the mold-metal interface is formed due to the solidification of
liquid steel and consequently the interfacial thermal resistance increases dramatically,
which explains the reason why the heat flux reduces intensively at the initial stage
of stage II. With the further solidification of the liquid steel, the shell continues to
grow and correspondingly the total thermal resistance across the solidified shell and
mold-shell interface continues to increase. That is the reason why the heat fluxes
were reduced gradually throughout the rest of stage II.
In the latter part of stage II, the other interesting phenomenon that was observed
in this study is that the heat fluxes of locations at S3 (y is 3 mm) and S6 (y is 6 mm)
are the highest, followed by S9 (y is 9 mm) and S0 (y is 0 mm), then S12 (y is 12 mm),
S15 (y is 15 mm), S18 (y is 18 mm), and S21 (y is 21 mm). It might be explained
that the location of S0 below the liquid steel level S9 (y is 9 mm) has a thicker shell
thickness due to the longer solidification time, which resulted in a larger thermal
resistance than the locations of S3 (y is 3 mm) and S6 (y is 6 mm), which the finding
is consistent with other studies [5–8]. As for the position at the liquid level S9 (y
is 9 mm), there exists significant vertical heat transfer upward to the top part of the
mold, therefore, its heat flux is a little bit smaller than those of S3 and S6. As for the
heat fluxes at S12 (y is 12 mm), S15 (y is 15 mm), S18 (y is 18 mm), and S21 (y is
21 mm), they are decreasing because they are far away from the liquid melt and the
existing vertical heat transfer from the liquid to the upper mold. Therefore, the mold
heat transfer is two-dimensional from the location 3 to 6 mm below the liquid melt
vertically toward the upper part (may be a little downward) and horizontally to the
inside of the mold.
60 H. Zhang et al.

Fig. 5 Mold heat flux

calculated by 2DIHCP 1.0
I II
liquid-level at S9 with y=9mm
0.8

Heat Flux at Metal-mold

interface (MW/m2)
S6
0.6 S3
S9
S0
0.4 S12
S15
S18
0.2 S21

0.0

51 52 53 54 55 56 57 58 59
Time (s)

The mold wall temperature evolution during the casting process is reconstructed
by 2DIHCP and shown in Fig. 6. The temperatures of the lower part of the mold
wall are increased (turns to red) at first with the initial filling of the liquid steel, and
then the hot red area expands both horizontally and vertically with the progressive
filling of the liquid steel, which indicates the significant 2D heat transfer inside the
mold as suggested above. It was noticed that there was no significant evolution of
the temperature inside the mold after 54 s, which is consistent with Fig. 4, and the
maximum temperature of the mold wall reaches as high as 335 K.

Conclusions

In this work, based on the function specification method with first-order spatial regu-
larization, a two-dimensional transient inverse heat conduction problem (2DIHCP)
is established for the determination of the mold heat flux using mold temperatures
measured by fast response thermocouples at a frequency up to 60 Hz. The specific
conclusions are summarized as:
1. The inverse problem of the function specification method implemented with
the first-order spatial regularization could improve the temporal and spatial
stabilization of recovered mold heat flux.
2. The built two-dimensional transient inverse heat conduction problem (2DIHCP)
could reconstruct the boundary heat flux precisely and is capable of the ability to
resist measurement noises. It could also provide more precise heat flux calculation
than the one-dimensional inverse heat conduction problem (1D-IHCP).
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 61

Fig. 6 Temperature evolution of the mold wall during the casting

Acknowledgements This work was partially supported by NSFC (No. 52074135, 51904107), the
Natural Science Foundation of Jiangxi Province (No. 20202BAB214016), and the Youth Jinggang
Scholars Program in Jiangxi Province (QNJG2020049).

Conflicts of Interest
The authors declare that they have no conflicts of interest.

References

1. Balogun D, Roman M, Gerald RE, Huang J, Bartlett L, O’Malley R (2022) Shell measurements
and mold thermal mapping approach to characterize steel shell formation in peritectic grade
steels. Steel Res Int 93(1):2100455
2. Niu Z, Cai Z, Zhu M (2020) Heat transfer behaviour of funnel mould copper plates during
thin slab continuous casting and channel structure optimization. Ironmaking Steelmaking
47(10):1135–1147
3. Zhang H, Wang W (2017) Mold simulator study of heat transfer phenomenon during the initial
solidification in continuous casting mold. Metall Mater Trans B 48(2):779–793
4. Badri A, Natarajan TT, Snyder CC, Powers KD, Mannion FJ, Byrne M, Cramb AW (2005)
A mold simulator for continuous casting of steel: part II. The formation of oscillation marks
during the continuous casting of low carbon steel. Metall Mater Trans B 36(3):373–383
5. Zhang H, Wang W (2016) Mold simulator study of the initial solidification of molten steel in
continuous casting mold: part II. Effects of mold oscillation and mold level fluctuation. Metall
Mater Trans B 47(2):920–931
62 H. Zhang et al.

6. Lyu P, Wang W, Zhang H (2017) Mold simulator study on the initial solidification of molten
steel near the corner of continuous casting mold. Metall Mater Trans B 48(1):247–259
7. Zhang H, Wang W, Ma F, Zhou L (2015) Mold simulator study of the initial solidification of
molten steel in continuous casting mold. Part I: experiment process and measurement. Metall
Mater Trans B 46(5):2361–2373
8. Lopez PER, Mills KC, Lee PD, Santillana B (2012) A unified mechanism for the formation of
oscillation marks. Metall Mater Trans B 43(1):109–122
9. Arunkumar S, Rao KS, Kumar TP (2008) Spatial variation of heat flux at the metal–mold
interface due to mold filling effects in gravity die-casting. Int J Heat Mass Transf 51(11–
12):2676–2685
10. Roman M, Balogun D, Zhuang Y, Gerald RE, Bartlett L, O’Malle RJ, Huang J (2020) A
spatially distributed fiber-optic temperature sensor for applications in the steel industry. Sensors
20(14):3900
11. Cardiff M, Kitanidis PK (2008) Efficient solution of nonlinear, underdetermined inverse
problems with a generalized PDE model. Comput Geosci 34(11):1480–1491
12. Beck JV, Blackwell B, Clair CR Jr (1985) Inverse heat conduction: Ill-posed problems. James
Beck, New York
13. Yu Y, Luo X (2015) Estimation of heat transfer coefficients and heat flux on the billet surface
by an integrated approach. Int J Heat Mass Transf 90:645–653
14. Pinheiro CAM, Samarasekera IV, Brimacomb JK, Walker BN (2000) Mould heat transfer
and continuously cast billet quality with mould flux lubrication Part 1 Mould heat transfer.
Ironmaking Steelmaking 27(1):37–54
15. Thomas BG, Wells MA, Li D (2011) Monitoring of meniscus thermal phenomena with
thermocouples in continuous casting of steel. TC 216(416):8900
16. Wang X, Tang L, Zang X, Yao M (2012) Mold transient heat transfer behavior based
on measurement and inverse analysis of slab continuous casting. J Mater Process Technol
212(9):1811–1818
17. Chang CW, Liu CH, Wang CC (2018) Review of computational schemes in inverse heat
conduction problems. Smart Sci 6(1):94–103
18. Du F, Wang X, Fu J, Han X, Xu J, Yao M (2018) Inverse problem-based analysis on non-
uniform thermo-mechanical behaviors of slab during continuous casting. Int J Adv Manufact
Technol 94(1):1189–1196
19. Trovant M, Argyropoulos S (2000) Finding boundary conditions: a coupling strategy for the
modeling of metal casting processes: part I. Experimental study and correlation development.
Metall Mater Trans B 31(1):75–86
20. Griffiths WD (2000) Modelled heat transfer coefficients for Al–7 wt-% Si alloy cast-
ings unidirectionally solidified horizontally and vertically downwards. Mater Sci Technol
16(3):255–260
21. Chakraborty S, Ganguly S, Chacko EZ, Ajmani SK, Talukdar P (2017) Estimation of surface
heat flux in continuous casting mould with limited measurement of temperature. Int J Therm
Sci 118:435–447
22. Li Y, Wang G, Chen H (2015) Simultaneously estimation for surface heat fluxes of steel slab
in a reheating furnace based on DMC predictive control. Appl Therm Eng 80:396–403
23. Zhang H, Wang W, Zhou L (2015) Calculation of heat flux across the hot surface of continuous
casting mold through two-dimensional inverse heat conduction problem. Metall Mater Trans
B 46(5):2137–2152
24. Özisik MN, Orlande HR (2018) Inverse heat transfer: fundamentals and applications. Routledge
25. Zhang L, Li L, Ju H, Zhu B (2010) Inverse identification of interfacial heat transfer coef-
ficient between the casting and metal mold using neural network. Energy Convers Manage
51(10):1898–1904
26. Jayakrishna P, Chakraborty S, Ganguly S, Talukdar P (2021) Modelling of thermofluidic
behaviour and mechanical deformation in thin slab continuous casting of steel: an overview.
Can Metall Q 60(4):320–349
Inverse Calculation of Time-Spatial Varying Mold Heat Flux During … 63

27. Argyropoulos SA, Carletti H (2008) Comparisons of the effects of air and helium on heat
transfer at the metal-mold interface. Metall Mater Trans B 39(3):457–468
28. Kovačević L, Terek P, Kakaš D, Miletić A (2012) A correlation to describe interfacial heat
transfer coefficient during solidification of Al–Si alloy casting. J Mater Process Technol
212(9):1856–1861
29. Sun Z, Hu H, Niu X (2011) Determination of heat transfer coefficients by extrapolation and
numerical inverse methods in squeeze casting of magnesium alloy AM60. J Mater Process
Technol 211(8):1432–1440
30. Rajaraman R, Velraj R (2008) Comparison of interfacial heat transfer coefficient estimated by
two different techniques during solidification of cylindrical aluminum alloy casting. Heat Mass
Transf 44(9):1025–1034
31. Coates B, Argyropoulos SA (2007) The effects of surface roughness and metal temperature on
the heat-transfer coefficient at the metal mold interface. Metall Mater Trans B 38(2):243–255
32. Dour G, Dargusch M, Davidson C (2006) Recommendations and guidelines for the performance
of accurate heat transfer measurements in rapid forming processes. Int J Heat Mass Transf
49(11–12):1773–1789
33. Blanc G, Raynaud M, Chau TH (1998) A guide for the use of the function specification method
for 2D inverse heat conduction problems. Revue Générale Therm 37(1):17–30
34. Babu K, Prasanna Kumar TS (2010) Mathematical modeling of surface heat flux during
quenching. Metall Mater Trans B 41(1):214–224
35. Wang W, Long X, Zhang H, Lyu P (2018) Mold simulator study of effect of mold oscillation
frequency on heat transfer and lubrication of mold flux. ISIJ Int 58(9):1695–1704
36. Orlande HR (2012) Inverse problems in heat transfer: new trends on solution methodologies
and applications. J Heat Transf 134(3)
37. Badri A, Natarajan TT, Snyder CC, Powers KD, Mannion FJ, Cramb AW (2005) A mold
simulator for the continuous casting of steel: part I. The development of a simulator. Metall
Mater Trans B 36(3):355–71
38. Aster RC, Borchers B, Thurber CH (2018) Parameter estimation and inverse problems. Elsevier,
New York
39. Vogel CR (ed) (2002) Computational methods for inverse problems. Society for Industrial and
Applied Mathematics, Philadelphia
40. George S, Thamban Nair M (1998) On a generalized Arcangeli’s method for Tikhonov
regularization with inexact data. Numer Funct Anal Optim 19(7–8):773–787
41. Özısık MN (ed) (1993) Heat conduction, 2nd edn. Wiley, New York, pp 62–75
Quality Prediction of Hot Rolled
Products and Optimization
of Continuous Casting Process
Parameters Based on Big Data Mining

Zibing Hou, Zhiqiang Peng, Qian Liu, and Guanghua Wen

Abstract The continuous casting-rolling process is widely used due to the lower
energy consumption and its compact process. However, the defect in continuous
casting slab severely limits the development of the continuous casting-rolling tech-
nology. In this paper, based on big data mining technology, the quality prediction
model for hot rolled coils and the corresponding optimization method for continuous
casting parameters were proposed. Firstly, the GA (Genetic algorithm)-BP (Back-
propagation) neural network prediction model with high accuracy was constructed
according to the characteristics of actual production data. Then, the effect of contin-
uous casting parameters on the probability of defects occurrence was investigated
with the established model. The results show that the defects occurrence probability
decreases firstly and then increases with the casting speed, as well as the tempera-
ture of molten steel, which are consistent with metallurgical theory. Meanwhile, the
optimum casting speed and mold level are 1.3 m/min and 8200 mm, respectively.
When the flow rate of argon blowing for stopper and nozzle are restricted to 8.5 and
8 L/min, the defects occurrence probability will be lower. Furthermore, the optimum
critical values of temperature difference of mold cooling water and inlet temperature
are 8 °C and 35 °C, respectively. This paper can provide the guidance for narrow
range control of continuous casting parameters and contribute to the production of
high-quality steel.

Keywords Continuous casting · Hot rolled coils · Surface defects · Data mining ·
Parameter optimization

Z. Hou (B) · Z. Peng · Q. Liu · G. Wen

College of Materials Science and Engineering, Chongqing University, Chongqing 400044,
People’s Republic of China
e-mail: houzibing@[Link]
Chongqing Key Laboratory of Vanadium-Titanium Metallurgy and New Materials, Chongqing
University, Chongqing 400044, People’s Republic of China

© The Minerals, Metals & Materials Society 2023 65

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
66 Z. Hou et al.

Introduction

At present, the technology of continuous casting-rolling has been used widely due
to the low energy consumption and compact process [1]. However, it is severe to
avoid slab defects with current production conditions of continuous casting. These
defects will be inherited into the rolled products and manifested as surface warp [2],
inclusion [3], or crack [4], not only affecting the normal production process but also
leading to additional energy consumption. Therefore, it is necessary to judge the
slab’s quality accurately and then adjust the continuous casting process parameters
in advance. That will be conducive to the improvement of product’s quality and save
energy for steel industry.
The online defect recognition of slabs is mainly carried out from two aspects:
physical detection and expert system. Methods of physical detection include CCD
camera detection [5], eddy current detection [6], etc. However, the recognition accu-
racy of the physical detection is limited by iron oxide scale and harsh environment.
Meanwhile, this detection equipment can only identify surface defects but is not
suitable for internal defects. As for the expert system, experience accumulated from
the actual production conditions is used for defect recognition [7], mainly based
on sole parameter. Thus, it is difficult to meet the increasing demand for accurate
judgement since the generation of slab defects is affected by multiple parameters.
In recent years, technology of big data mining has been used in the prediction of
product performance and quality in the industrial field [8, 9]. Nevertheless, there
are few research on the prediction of slab quality due to lack of samples. With the
development of information collection technology, the defects can be distinguished
whether they are caused by the continuous casting process according to the quality
information of rolled products. Actually, the rich quality information provided by
the hot rolling mill is used for the traceability of some individual sample defects.
At the same time, optimization of continuous casting parameters is mostly carried
out through production experience or experimental research, which have problems
such as high cost and long development period. Hence, it is meaningful to establish
the quality prediction model for hot rolled coils based on the big data of production
process and optimize process parameters.
In this paper, the methods of data preprocessing and quality prediction model
based on BP (Backpropagation) neural network and GA (Genetic algorithm) were
established with continuous casting parameters, as well as the quality information
of hot rolled coils. Subsequently, the effect of parameters on surface defects of hot
rolled coils was investigated. According to this study, it can provide guidance for the
exact design of production process parameters, which is helpful to the fine control
of continuous casting process.
Quality Prediction of Hot Rolled Products and Optimization … 67

Methods

Database Construction

25 parameters of continuous casting process, which involve ladle, tundish, and mold,
were selected to establish the model. The parameters related to the second cooling
were not considered for the reason that the defects of slabs are mainly generated in the
initial solidification stage of the molten steel. Meanwhile, the production parameters,
defect types, and the number of defects vary with steel grades. Thus, the data were
classified into three categories, low carbon steel, peritectic steel, and medium carbon
steel. For the quality information of hot rolled coils, the defective and normal samples
are highly unbalanced, which will easily lead to the overfitting of the prediction
model and poor performance for the class with small samples. Figure 1 shows the
Pearson correlation coefficient among continuous casting parameters, and it can be
seen that there is a certain correlation among some continuous casting parameters.
The redundant data reduce the modeling efficiency of the prediction model, and it is
easy to make the model overfitting.

Fig. 1 Heat map of correlation coefficient among continuous casting parameters

68 Z. Hou et al.

Fig. 2 Flow chart of quality

prediction model for hot
rolled coils

Quality Prediction Model

Based on the data characteristics of continuous casting parameters and quality infor-
mation of hot rolled coils, the flow chart of prediction model was proposed, as shown
in Fig. 2. Firstly, the samples attributed to the continuous casting responsibility were
filtered from the database and merged with production data of continuous casting.
Then, data preprocessing was performed on the above database, including classifi-
cation for imbalanced data of defect information and data dimensionality reduction.
For convenience of modeling, defective and normal samples were marked as 1 and
0, respectively. The preprocessed data were used as the input of the hot rolled coil
quality prediction model, and the BP neural network algorithm was optimized with
the genetic algorithm in order to achieve effective prediction.

Analysis Method for Influence of Parameters on Defects

The GA-BP neural network can be used to investigate the effect of continuous
casting parameters on the defects of hot rolled coils. The prediction of new data
can be achieved by saving the constructed neural network model, which means that
the trained model is considered as a multivariable function. Thus, the values of
dependent variables with different values of the independent variable can be calcu-
lated. In this paper, a certain continuous casting parameter was taken as the inde-
pendent variable, the rest of the parameters were fixed, and the single parameter was
adjusted to investigate the relationship between the possibility of defect occurrence
and different parameters. Obviously, the probability of defects occurrence increases
with the calculated value of the model. Therefore, the optimal value or operating
interval of continuous casting parameters can be finally obtained.
Quality Prediction of Hot Rolled Products and Optimization … 69

Table 1 Prediction results of GA-BP model

Steel grade MAE MSE RMSE AR (%) FAR (%)
Low carbon steel 0.1067 0.0445 0.210 94.7 82.5
Peritectic steel 0.1299 0.0521 0.2236 93.3 87.5
Medium carbon steel 0.3063 0.1465 0.3827 85.4 78.3

Results and Discussion

Model Evaluation

Mean absolute error (MAE), mean squared error (MSE), and root mean squared
error (RMSE) were used to evaluate the performance of the prediction model. MAE
is one of the comprehensive evaluation methods of error analysis, which can avoid
the offset of positive and negative values. MSE is the mean absolute percentage
error. The smaller the MSE, the higher the prediction accuracy. RMSE represents the
dispersion degree of the predicted value, and lower value of which indicates higher
prediction accuracy. To better describe the prediction results of the model on the
quality of hot rolled coils, the accuracy rate for overall samples (AR) and defective
samples (FAR) were introduced to evaluate the accuracy rate of different models.
Table 1 shows the calculation results of different means of model evaluation. From
which, we can see that prediction model for medium carbon steel has the highest
model evaluation indexes (MAE, MSE, and RMSE) and relatively lower accuracy
rate compared to the other two steel grades, while the MAE, MSE, and RMSE
values for low carbon steel are the lowest, which means that the model has a higher
prediction accuracy. Meanwhile, it is also found that the prediction accuracy of the
model varied with steel grades, which may be caused by the difference in sample
size. Overall, the AR for each steel grade is larger than 85%, which indicates that the
prediction models for different steel grades are accurate enough to investigate the
influence of continuous casting parameters on the defects of hot rolled coils. Due to
the largest sample size and the best prediction effect among all steel grades, data of
low carbon steel was selected for further study.

Effect of Continuous Casting Parameters on Defects

Casting Speed, Temperature of Molten Steel, and Mold Level

Figure 3 shows the relationship between the quality of the hot rolled coils and three
continuous casting parameters (casting speed, temperature of molten steel, and mold
level). The horizontal axis represents the value of each continuous casting param-
eter, and the vertical axis represents the dimensionless output value calculated by
70 Z. Hou et al.

(a) (b) (c)

Fig. 3 Effect of a casting speed, b temperature tundish molten steel, and c mold level on possibility
of defects

the constructed neural network. The probability of defects occurrence increases with
dimensionless output value. As indicated in Fig. 3a, the possibility of defects occur-
rence first decreases and then gradually increases with casting speed. Meanwhile,
the hot rolled coils are not prone to defects when the casting speed is 1.3 m/min.
The reason for this result may be that the melting performance of the mold flux is
poor when the casting speed is low, thus affecting the surface quality of the slab.
High casting speed can cause the fluctuation of heat transfer on all sides of the mold,
which aggravates the generation of surface cracks [10]. Previous research has shown
that higher or lower casting speed may also lead to slag inclusions [11].
It can be seen from Fig. 3b that it is beneficial to the quality of the hot rolled
coils when the temperature of molten steel is in the range of 1550 ~ 1560 °C. Higher
temperature of molten steel leads to higher superheat, which causes thinner shell
thus resulting in breakout or crack. Meanwhile, lower superheat is not conducive to
the melting of slag [12], which will also lead to surface defects. Furthermore, the
possibility of defects occurrence is greatly affected by the mold level from Fig. 3c.
Mold level is an important parameter of the casting process, which can effectively
reflect the clogging of the nozzle [13]. This result shows that the probability of defect
occurrence is the lowest when the mold level is around 8200 mm, which is the setting
value in the plant.
The effect of casting speed, temperature of molten steel, and mold level on the
defects of hot rolled coils can be well explained by production experience and metal-
lurgical theory, which shows the feasibility of using neural networks to investigate
the effect of parameters on product defects. Moreover, argon blowing and the heat
transfer of the mold were also investigated.

Argon Blowing

Figure 4 shows the relationship between parameters of argon blowing and the possi-
bility of defects occurrence. It can be seen from Fig. 4a–c that the effects of argon
flow on defects of the hot rolled coils are nonlinear, and all of which have the same
trend of increasing first, then decreasing, and finally increasing again with flow rate.
The best value for stopper and nozzle is about 8.5 L/min and 8 L/min, respectively.
Quality Prediction of Hot Rolled Products and Optimization … 71

(a) (b) (c)

Fig. 4 Effect of on probability of defects: a argon flow of stopper; b argon flow of nozzle; c argon
flow of sliding plate

Higher or lower flow rate of argon blowing will easily lead to defects caused by large
fluctuation of mold level.

Heat Transfer of Mold

Figure 5 shows the relationship between the heat transfer parameters of mold and the
possibility of defects occurrence. From Fig. 5a–d, it can be seen that the influence
of cooling water flow at different positions on the possibility of defects is similar.
With the increase of water flow, it shows a trend of decreasing first, then increasing,
then decreasing, and finally increasing again, which indicates that the influence of
cooling water flow on slab defects is relatively complex. The results of Fig. 5e–h
show that the possibility of defects occurrence decreases first and then increases with
the temperature difference of cooling water. When the temperature difference is in the
range of 7–9 °C, the possibility of defects is relatively low, and it is recommended to
control the temperature difference to about 8 °C in actual production. From Fig. 5i–
k, it can be seen that optimum control values inlet water pressure and temperature
of the mold are around 1.08 Mpa and 35 °C, respectively. Within a certain range
(8600–8700 L/min), probability of defects occurrence decreases with increasing
total cooling water flow rate, which may be due to less fluctuations in heat transfer.

Conclusions

Based on the features of production data from the plant, the data preprocessing
method and the quality prediction model for hot rolled coils were firstly proposed.
Then, the relationships between product defects and continuous casting parameters
were investigated. The model performed well on the testing set. The prediction accu-
racies of overall samples to low carbon steel, peritectic steel, and medium carbon steel
were 94.7%, 93.3%, and 85.4%, respectively. As for the defective samples, prediction
accuracies for different steel grades were 82.8%, 87.5%, and 87.3%, respectively.
72 Z. Hou et al.

(a) (b) (c) (d)

(e) (f) (g) (h)

(i) (j) (k)

Fig. 5 Effect of the heat transfer of the mold on the probability of defects: a cooling water flow on
left narrow side; b cooling water flow on right narrow side; c cooling water flow on inner arc side;
d cooling water flow on outer arc side; e temperature difference on left narrow side; f temperature
difference on right narrow side; g temperature difference on inner arc side; h temperature difference
on outer arc side; i pressure of cooling water; j total flow of cooling water; k temperature of inlet
cooling water
Quality Prediction of Hot Rolled Products and Optimization … 73

Moreover, the optimal operating values of some parameters were obtained according
to the prediction model based on actual production data. For casting speed and mold
level, the optimal operating values are 1.3 m/min and 8200 mm, respectively. The
probability of defects occurrence is low when flow rate of argon blowing for stopper
and nozzle are restricted to around 8.5 L/min and 8 L/min, respectively. As to the heat
transfer of mold, the optimal temperature difference between inlet and outlet water
is 8 °C, and the temperature of inlet water should be controlled to around 35 °C.

Acknowledgements The authors are very grateful for the support from National Natural Science
Foundation of China (52274318).

References

1. Wang S, Zhao G, Bao S (2013) Surface cracks analysis on continuous casting slab Q345c
during hot delivery and hot charging. Adv Mat Res 785:1081–1086
2. Kainz A, Ilie S, Parteder E (2008) From slab corner cracks to edge-defects in hot rolled
strip—experimental and numerical investigations. Steel Res Int 79(11):861–867
3. Luo C, Stahlberg U (2001) Deformation of inclusions during hot rolling of steels. J Mater
Process Tech 114(1):87–89
4. Ervasti E, Stahlberg U (1999) Behaviour of longitudinal surface cracks in the hot rolling of
steel slabs. J Mater Process Tech 94(2–3):141–150
5. Zhao L, Qi O, Chen D, Zhang X (2011) CCD imaging definition controlling methods in
continuous casting slab surface defects acquisition. Adv Mat Res 233:2490–2494
6. Qiu X, Wei C, Cui X, Wei J (2013) Real-time pre-processing of the pulsed eddy current signal
from continuous casting slabs. Insight 55(3):136–141
7. Yan J, Bi X, Tian Z (2012) The on-line continuous casting slab surface quality prediction expert
system. Int J Educ Manag Eng 2(1):54–58
8. Li H, Li X, Liu X, Bu X, Li H, Lyu Q (2022) Prediction of blast furnace parameters using
feature engineering and stacking algorithm. Ironmak Steelmak 49(3):283–296
9. Shen C, Wang C, Wei X, Li Y, Zwaag S, Xu W (2019) Physical metallurgy-guided machine
learning and artificial intelligent design of ultrahigh-strength stainless steel. Acta Mater
179:201–214
10. Wang X, Tang L, Zang X, Yao M (2012) Mold transient heat transfer behavior based
on measurement and inverse analysis of slab continuous casting. J Mater Process Tech
212(9):1811–1818
11. Li S, Du X (2010) The effect of casting speed on slag-inclusion defects in 1050 × 200 mm
ultra-low-carbon automobile steel slabs. Metall Res Technol 118(6):611
12. Sun Y, Ni Y, Wang H, Xu Z, Cai K (2010) Longitudinal surface cracks of thin slabs. Int J Miner
Metall Mater 17(2):159–166
13. Nadif M, Lehmann J, Burty M, Domgin J (2007) Control of steel reoxidation and CC nozzle
clogging: an overview. Metall Res Technol 104(10):493–500
Study on the Uniformity of Surface
Temperature of Continuous Casting Slab
Based on Solidification and Heat
Transfer Simulations

Yadong Wang and Lifeng Zhang

Abstract In the current study, a two-dimension solidification and heat transfer

model was developed based on the slab continuous casting (CC) process in a domestic
steel plant. The layout of jet nozzles of segment 2 and segment 3 in secondary cooling
zones was adjusted to optimize the distribution of temperature along the width direc-
tion. The following main findings indicate that the temperature was more uniform. In
segment 2 and segment 3, middle jet nozzles were replaced by those in side positions,
and the distribution of water flow rate and temperature along the width direction was
more uniform. In the outlet of segment 2, except for the slab corner, the fluctuation
range of the temperature along the width direction decreased from 39.7 to 14.1 °C.
In the outlet of segment 3, except for the slab corner, the fluctuation range of the
temperature along the width direction decreased from 40.4 to 17.3 °C.

Keywords Continuous casting slab · Solidification · Heat transfer · Simulation

Introduction

In the slab CC process, the uniformity of heat transfer in the secondary cooling zones
has an important effect on the surface and internal quality of slabs. Transverse corner
cracks had a close relationship to secondary cooling which was reported by many
researchers [1–3]. At the bending and straightening segments during the CC process,
an appropriate cooling method to prevent the temperature of the slab corner from
avoiding the poor hot ductility temperature range is essential to avoid transverse
corner cracks. A method of optimizing the uniformity of solidified shell in the trans-
verse direction was introduced by Long et al. [4], and the center macro-segregation in
steel slabs was mitigated. A two-dimension heat transfer model was developed for the
optimization of cooling intensity along the transverse direction, which contributed
to regular solidification end in the CC slab, and the center macro-segregation was
significantly mitigated. Longitudinal cracks and serious shape defects originated

Y. Wang · L. Zhang (B)

School of Mechanical and Materials Engineering, North China University of Technology,
Beijing 100144, China
e-mail: zhanglifeng@[Link]

© The Minerals, Metals & Materials Society 2023 75

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
76 Y. Wang and L. Zhang

generally due to improper cooling patterns in the mold, but improper cooling in
secondary cooling zones may contribute to defect propagation [5–7].
In the current study, to obtain the uniform distribution of temperature along the
width direction, a two-dimension solidification and heat transfer model was estab-
lished to optimize the distribution of water flow rate. The layout of jet nozzles
of segment 2 and segment 3 in secondary cooling zones was adjusted, and the
distribution of temperature along the width direction was optimized.

Mathematical Formulation

In the current study, a two-dimensional heat transfer and solidification model was
established to calculate the thermal history of the CC slab. A detailed descrip-
tion of the two-dimensional heat transfer and solidification model can be found
in the previous study [1]. In the current model, crystallizer heat transfer, water spray
cooling, water evaporation cooling, roll contact cooling, and radiation cooling were
considered. Governing equations of boundary conditions in mold zone and secondary
cooling zones were given in the previous study [2]. Water spray cooling was the main
factor governing heat transfer in secondary cooling zones. The distribution of water
flow rate for all jet nozzles was measured, and the test equipment was given in the
previous work [2]. The main parameters used in the current study are listed in Table
1. The content of the IF steel is listed in Table 2. Water flow in mold and secondary
cooling zones is listed in Table 3. I represents the loose side of the wide face, and O
represents the fixed side of the wide face. IO means that I and O each account for
half of the water flow rate.

Table 1 Parameters of CC process

Parameters Values
Cross section of the slab (mm2 ) 240 × 1500
Effective mold length (mm) 800
Casting speed (m/min) 1.2
Pouring temperature (°C) 1555
Liquidus temperature of the steel (°C) 1530
Solidus temperature of the steel (°C) 1499

Table 2 Chemical compositions of IF steel

Elements C Mn Si S P Als Cr Ni Cu Nb V Ti
Content 0.001 0.109 0.002 0.002 0.009 0.045 0.008 0.006 0.011 0.001 0.003 0.072
(%)
Study on the Uniformity of Surface Temperature of Continuous Casting … 77

Table 3 Water flow in mold

Cooling zones Length (m) Water flow (L/min)
and secondary cooling zones
Wide face of mold 0.8 5300
Narrow face of mold 480
Wide face of foot zone 0.24 478
Narrow face of foot zone 0.9 342
1IO 1.15 1018
2IO 2.435 1226.6
3I 1.849 302.4
3O 334.8
4I 3.71 345.6
4O 457.2
5I 3.99 205.2
5O 340.2
6I 2.0 61.2
6O 136.8
7I 4.24 88.4
7O 238
8I 6.3 59.2
8O 123.9

Solidification and Heat Transfer Simulation of CC Slabs

When the superheat is 25 °C and the casting speed is 1.2 m/min, the shell thickness
and the temperature on the center of the loose side surface are shown in Fig. 1. With
the distance from the meniscus increasing, the temperature at the loose side surface
tended to decrease. At the inlet of each section, a temperature rise was generated
due to the secondary cooling water flow decreasing along the casting direction. The
shell thickness near the outlet of the mold was 18.0 mm, and the solidified end was
located at 22.9 m from the meniscus.
The temperature distribution along the width direction at the outlet of different
secondary cooling sections is shown in Fig. 2. In the foot zone and Sect. 1, the temper-
ature distribution of the CC slab along the width direction was uniform. Entering
Sect. 2, the temperature distribution became uneven. Near the outlet of Sect. 2, the
maximum temperature was 842.3 °C which was located at about 520 mm from the
slab center. Due to the two-dimensional cooling at the corner of the CC slab, the
temperature at the slab corner was minimum, and it was about 674 °C. A local
minimum point of temperature was 802.6 °C which was located at about 200 mm
from the slab center. Near the outlet of Sect. 3, there were two higher temperature
points, 808.5 °C, and 813.2 °C, and they were located at the slab center and 600 mm
from the slab center, respectively. A relatively low temperature was 772.8 °C, and it
was located about 250 mm from the slab center. Therefore, the layout of jet nozzles
78 Y. Wang and L. Zhang

Thickness (mm)
0 20 40 60 80 100 120
0
1600
Liquid
5
1400
Distance from meniscus (m)

Temperature (°C)
10 1200

Mush
1000
15
Solid

800
20

600
0 5 10 15 20 25 30 35
25 Distance from meniscus (m)

30
(b) Temperature at the center of the loose side
(a) Shell thickness
surface

Fig. 1 Shell thickness and temperature at the center of the loose side surface

of segment 2 and segment 3 in secondary cooling zones should be adjusted to obtain

more uniform distribution of temperature along the width direction.

Study on Optimizing the Distribution of Temperature Along

Width Direction

To obtain the uniform distribution of temperature along the width direction, the
layout of jet nozzles of segment 2 and segment 3 in secondary cooling zones should
be adjusted. The measured distribution of the water flow rate of segment 2 under
different gas pressure is shown in Fig. 3. For a jet nozzle in the middle, the distribution
of water flow rate was approximately parabolic. With the gas pressure increasing,
the water flow rate increased. For a jet nozzle in the sides, the distribution of water
flow rate was relatively uniform. Within 100 mm from the nozzle center, the water
flow rate was almost equal. When the distance from the nozzle center was higher
than 100 mm, the water flow rate decreased.
To obtain uniform distribution of temperature along the width direction, the layout
of jet nozzles should be adjusted. In the current study, middle jet nozzles were
replaced by those in side positions. The measured distribution of the water flow rate
of segment 2 in secondary cooling zones is shown in Fig. 4. Before optimization,
there were severe fluctuations along the width direction, and the fluctuation range
can reach 10 L/(m2 s). After optimization, more uniform distribution of water flow
rate was obtained, and the fluctuation range was within 5 L/(m2 s). The measured
Study on the Uniformity of Surface Temperature of Continuous Casting … 79

1000 1000

900 900
Temperature (°C)

Temperature (°C)
800 800

700 700

600 600

500 500
-800 -600 -400 -200 0 200 400 600 800 -800 -600 -400 -200 0 200 400 600 800
Distance from slab center (mm) Distance from slab center (mm)

(a) Outlet of foot zone (b) Outlet of section 1

1000 1000

900 900
Temperature (°C)

Temperature (°C)
800 800

700 700

600 600

500 500
-800 -600 -400 -200 0 200 400 600 800 -800 -600 -400 -200 0 200 400 600 800
Distance from slab center (m) Distance from slab center (m)

(c) Outlet of section 2 (d) Outlet of section 3

1000 1000

900 900
Temperature (°C)

Temperature (°C)

800 800

700 700

600 600

500 500
-800 -600 -400 -200 0 200 400 600 800 -800 -600 -400 -200 0 200 400 600 800
Distance from slab center (mm) Distance from slab center (mm)

(e) Outlet of section 4 (f) Outlet of section 5

1000 1000

900 900
Temperature (°C)

Temperature (°C)

800 800

700 700

600 600

500 500
-800 -600 -400 -200 0 200 400 600 800 -800 -600 -400 -200 0 200 400 600 800
Distance from slab center (m) Distance from slab center (mm)

(g) Outlet of section 6 (h) Outlet of section 7

Fig. 2 Temperature distribution of the CC slab along the width direction

80 Y. Wang and L. Zhang

60 50
0.1 MPa 0.1 MPa
0.23 MPa 0.23 MPa

Water flow rate (L· m-2·s -1)

50 40 0.35 MPa
0.35 MPa
Water flow rate (L· m-2·s -1)

0.6 MPa 0.6 MPa

40
30
30
20
20

10
10

0 0
-250 -150 -50 50 150 250 -300 -200 -100 0 100 200 300
Position (mm) Position (mm)

(a) Jet nozzles in the middle (b) Jet nozzles in side positions

Fig. 3 The measured distribution of water flow rate of segment 2 in secondary cooling zones

distribution of the water flow rate of segment 3 in secondary cooling zones is shown
in Fig. 5. In the original condition, the fluctuation range can reach 10 L/(m2 s). After
optimization, the fluctuation range was within 3 L/(m2 s).
After the layout of jet nozzles was adjusted, the temperature distribution of the
CC slab along the width direction was more uniform, which is shown in Fig. 6.
In the outlet of segment 2, except for the slab corner, the fluctuation range of the
temperature along the width direction decreased from 39.7 to 14.1 °C. In the outlet of
segment 3, except for the slab corner, the fluctuation range of the temperature along
the width direction decreased from 40.4 to 17.3 °C.

45 45

36 36
Water flow rate (L· m-2·s -1)
Water flow rate (L· m-2·s -1)

27 27

18 18

9 9

0 0
0 200 400 600 800 1000 0 200 400 600 800 1000 1200
Position (mm) Position (mm)

(a) Before optimization (b) After optimization

Fig. 4 The measured distribution of water flow rate of segment 2 in secondary cooling zones
Study on the Uniformity of Surface Temperature of Continuous Casting … 81

30 30

25 25

Water flow rate (L· m-2·s -1)

Water flow rate (L· m-2·s -1)

20 20

15 15

10 10

5 5

0 0
0 200 400 600 800 1000 0 200 400 600 800 1000
Position (mm) Position (mm)

(a) Before optimization (b) After optimization

Fig. 5 The measured distribution of water flow rate of segment 3 in secondary cooling zones

1000 1000
Optimized Optimized
Original Original
900 900
)

)
Temperature (

800 800
Temperature (

700 700

600 600

500 500
-800 -600 -400 -200 0 200 400 600 800 -800 -600 -400 -200 0 200 400 600 800
Distance from slab center (mm) Distance from slab center (mm)

(a) Outlet of segement 2 (b) Outlet of segement 3

Fig. 6 Temperature distribution of the CC slab along the width direction

Conclusions

The layout of jet nozzles of segment 2 and segment 3 in secondary cooling zones was
adjusted, and the distribution of temperature along the width direction was optimized.
The following conclusions can be drawn:
1. In segment 2 and segment 3, middle jet nozzles were replaced by those in side
positions, and the distribution of water flow rate and temperature along the width
direction was more uniform.
2. In the outlet of segment 2, except for the slab corner, the fluctuation range of the
temperature along the width direction decreased from 39.7 to 14.1 °C.
3. In the outlet of segment 3, except for the slab corner, the fluctuation range of the
temperature along the width direction decreased from 40.4 to 17.3 °C.
82 Y. Wang and L. Zhang

Acknowledgements The authors are grateful for support from the National Natural Science Foun-
dation of China (U22A20171, 52204333) and the High Steel Center (HSC) at North China University
of Technology, China.

References

1. Zhang L, Yang X, Li S (2014) Control of transverse corner cracks on low-carbon steel slabs.
JOM 66(9):1711–1720
2. Wang Y, Ren Q, Zhang L, Yang X, Yang W, Ren Y, Zhang H (2021) Formation and control of
transverse corner cracks in the continuous casting slab of a microalloyed steel. Steel Res Int
92(6):2000649
3. Xing L, Zhang Z, Bao Y, Chen B, Zheng F, Wang M (2021) Causes of corner cracks in
hypoperitectic microalloyed steel billets. Steel Res Int 92(7):2100035
4. Long M, Chen D (2011) Study on Mitigating center macro-segregation during steel continuous
casting process. Steel Res Int 82(7):847–856
5. Camisani-Calzolari FR, Craig IK, Pistorius PC (1999) Control strategies for the secondary
cooling zone in continuous casting. In: IEEE Africon, vol 1999, pp 1435–440
6. Wang P, Xiao H, Chen X, Li W, Yi B, Tang H, Zhang J (2022) Improved in-mold metallurgical
behavior for slab casting of IF steels by a novel multi-poles electromagnetic stirring. Metall
Mater Trans B 53(3):1691–1702
7. Zhang B, Xue Z, Liao Z, Wang S, Zheng H, Zhou J (2014) Analysis on surface longitudinal
cracking of Q345D slab. J Iron Steel Res Int 21:108–112
How to Prevent Porosity Defects in Steel
Casting Component

Izudin Dugic

Abstract The production of steel casting can in some cases be complicated and
difficult. To produce quality castings, it requires the casting to be clean and free from
any defects. One of the main casting defects is hot tearing, also known as a crack
or shortness. This phenomenon represents the formation of an irreversible failure
(crack) in the steel semisolid casting. The second defect that commonly occurs in
casting is porosity. Porosity is often used to describe any void or hole found in
a casting. To control the porosity, you need to understand its sources and causes.
Porosity can occur either by gas formation, solidification shrinkage, or non-metallic
compound formation, all while the metal is liquid. This experimental work is focused
on reducing the porosity defects in steel casting components at a production scale.
The produced casting component was strongly affected by porosity defects.

Keywords Steel casting · Gas porosity · Casting simulation · Solidification

Introduction

The production of quality castings requires the casting surface to be clean and free
from any defects. Steel castings production can be a complicated and difficult process.
There are different types of defects in steel casting produced in sand casting that
can occur. A majority of these defects are caused due the evolution of gases [1].
Furthermore, these different types of porosity may present both at the micro- and
macro-length scales.
Many authors have discussed the problem with gas porosity, and a lot of work
have been done in the foundry industries to understand how to avoid gas porosity
in the casting components [2–8]. An extensive review of the research in this area is
provided by Lee et al. [9].
One of the most important parts of the casting manufacturing process is the gating
system of the casting. They have an effective role in casting defects free. The gating

I. Dugic (B)
Faculty of Technology, Department of Mechanical Engineering, Linnaeus University, 35195
Vaxjo, Sweden
e-mail: [Link]@[Link]

© The Minerals, Metals & Materials Society 2023 83

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
84 I. Dugic

system’s purpose is to provide a smooth, uniform, and complete filling of the mold
with pure molten metal [10–12].
To give an increased knowledge of the gas porosity and the influence from different
factors, a series of test casting at a production scale using the sand casting process
were performed at the XY Foundry.

Experimental

Casting Component

In the experimental work, a typical casting component from production was selected,
see Fig. 1. The length of the component is 430 mm, the width is 354 mm, and the
weight is 93.80 kg. The material of the component is Cast steel, SS11-2225-24 (Steel
for hardening and tempering).
Serial production in the last three months showed a 23% scrappage due to internal
defects and parts not meeting standard requirements 6134K/12/E3. In series produc-
tion, there are 4 models on the pattern and one feeder per detail is used, see Fig. 2.
The circles in Fig. 2 show position for the feeders.
In order to investigate what types of defects occur in the casting component, a
scrap casting component from the serial production was used. The casting component
was sliced into small pieces to find any possible internal pores or other defects and
its location, see Fig. 3. Figure 4 shows one of the pieces with defects.

Fig. 1 The casting component

How to Prevent Porosity Defects in Steel Casting Component 85

Fig. 2 The geometry of the pattern plate

Fig. 3 The sliced casting

component

Casting Simulation

The whole casting process was simulated with the commercial software
MagmaSoft® , in order to investigate the solidification characteristics as well as
86 I. Dugic

Fig. 4 The pieces with the

defects

porosity formation in the casting studied. For this, a special module for cast steel
was used. Figures 5 and 6 show some of the 3-D results from the simulation.
As a result, from the simulation, the gating system was changed by putting three
extra feeders per detail, see Fig. 7. The circles in Fig. 7 show position for the extra
feeders.
Figure 8 shows the geometry of the one of the four models with the feeders. The
number 1 in Fig. 8 is the feeder that had used before changing the gating system.
Numbers 2, 3, and 4 in Fig. 8 show the extra added feeders.

Fig. 5 3-D results from the

simulation
How to Prevent Porosity Defects in Steel Casting Component 87

Fig. 6 3-D results from the simulation

Fig. 7 The geometry of the pattern plate after changing

Alloys and Other Materials

Melting was done in a high frequency furnace. After melting, the melt is transported
in a 1 ton pouring ladle to a production line. The light emission spectrometer—ARL
3460 was used to determine the chemical composition from cast coin specimens.
88 I. Dugic

Fig. 8 The geometry of one of the four models with the feeders

The coins were cast immediately before the melt was poured into the molds. The
chemical composition for the experiments is shown in Table 1.
Table 2 shows the pouring temperature and pouring time for the experiments.
The castings were made in furan sand molds. The cores are also made in furan
sand. For the experiments, 4 flasks were molded and casted. This yields a total of

Table 1 The chemical composition of the melt

Molds Element in wt %
C Si S P Mn Cr Ni Mo
I 0.250 0.430 0.018 0.023 0.780 1.070 0.030 0.210
II 0.245 0.430 0.019 0.023 0.775 1.070 0.030 0.208
III 0.247 0.425 0.018 0.022 0.785 1.068 0.030 0.210
IV 0.248 0.435 0.019 0.023 0.780 1.070 0.029 0.212
Req. min 0.220 0.300 – – 0.500 0.090 – 0.150
Req. max 0.290 0.600 0.035 0.035 0.800 1.200 0.300 0.250

Table 2 The pouring

Molds Pouring temperature, °C Pouring time, sec
temperature and pouring time
of the experiments I 1620 23
II 1618 24
III 1622 23
IV 1620 25
How to Prevent Porosity Defects in Steel Casting Component 89

Fig. 9 One of the lower

casting mold with the cores

16 castings, thereby providing a statistical basis for the evaluation of the effect of
changing gating systems on gas porosity. Figure 9 show one of the lower castings
mold with the cores.

Results and Discussion

Castings Inspection

After sand blasting, all casting components were investigated by an ocular inspection.
Only one casting (or 6.25%) was defected with castings defect, and it was sand
erosion, see Fig. 10. This detail is scrapped because it could not be used but this
detail is used for future mechanical measurements.
After ocular inspection, all castings are investigated by radiographic examination
according to ASTM E446. Some of the radiographic measurements are shown in
Appendix 1.

Measurements of Mechanical Properties

Four samples for tensile test and measurements of hardness were taken from the
scrapped detail, see Fig. 11. The rectangles with the number in Fig. 11 show position
for samples for the mechanical measurements. The test has been performed according
to the SS-EN 10 002-1. The results are shown in Table 3.
90 I. Dugic

Fig. 10 The casting

component with the defect

Fig. 11 The casting component used for the measurements of mechanical properties

Table 3 Results from mechanical measurements

Samples Hardness (HB) Rp0,2 (MPa) Rm (Mpa) A5 (%) Z (%)
1 265 762 875 16 52
2 261 743 869 14 50
3 267 755 872 13 51
4 265 760 870 14 50
Req. min 240 590 780 10 25
Req. max 290 930
Hardness (HB) = Brinell hardness
Rp0,2 (MPa) = Proof stress
Rm (MPa) = Tensile strength
A (%) = Elongation after fracture
Z (%) = Reduction of area after fracture
How to Prevent Porosity Defects in Steel Casting Component 91

Conclusions

After all the castings were inspected by an ocular inspection and investigated by
radiographic examination, it can be concluded that the changing of the gating systems
plays an important role in the formation of casting defects.
Of the 16 casting components, there was only one detail, or 6.25% that had casting
defect and it was due to sand erosion. The same detail is used for the measurements
of mechanical properties.
The measurements of the mechanical properties show that the casting meets the
requirements according to the standard for the material.
The casting component simulation was performed to understand the problems in
the casting. The feeder’s location and the number have found out the results from
the simulation. The simulation software increases the quality and applicability of the
casting product.

Acknowledgements The author would like to thank Linnaeus University, Faculty of Technology,
Department of Mechanical Engineering, Vaxjo, Sweden.

Appendix

See Fig. 12 and Tables 4 and 5.

92

Table 4 X-ray inspection report

Consign unit MMV AB Inspect unit BF-ray department
Part name XXX Part number XXXX Material 001, 122, 252 Quantity 15(Pc)
Radial type XXQ-3205 FFD 600 mm TS 2.2 × 2.2 mm Image quality indicator GB5618
Film type AGFAC7 Film dimension 80 × 180 Add influence screen thickness front 0.03 mm back 0.13 mm
100 × 360
Darkroom deal with Hand work Develop fixative temperature 20 °C Inspect scale 100%
Inspection standard ASTM E446-98 Accept standard ASTM E446
According to photo
I. Dugic
How to Prevent Porosity Defects in Steel Casting Component 93

Table 5 Irradiation parameter and inspection results

Irradiation Irradiation Tube Tube Irradiation FFD Image Film Result
position position voltage current time (Min) (mm) quality density (P/F)
number thickness (Kvp) (Ma) indicator
(mm) (mm)
1A 15 155 5.0 2.0 600 0.25 1.89 No
find
1B 15 155 5.0 2.0 600 0.20 2.52 No
find
1C 15 155 5.0 2.0 600 0.20 2.16 No
find
1D 15 155 5.0 2.0 600 0.20 2.30 No
find
2A 15 155 5.0 2.0 600 0.25 1.89 No
find
2B 15 155 5.0 2.0 600 0.20 2.52 No
find
2C 15 155 5.0 2.0 600 0.20 2.70 No
find
2D 15 155 5.0 2.0 600 0.20 2.22 B1
3A 15 155 5.0 2.0 600 0.25 1.89 No
find
3B 15 155 5.0 2.0 600 0.20 2.52 No
find
3C 15 155 5.0 2.0 600 0.20 3.15 No
find
3D 15 155 5.0 2.0 600 0.20 2.30 No
find
4A 15 155 5.0 2.0 600 0.25 1.89 No
find
4B 15 155 5.0 2.0 600 0.20 2.52 B1
4C 15 155 5.0 2.0 600 0.20 2.50 A1
4D 15 155 5.0 2.0 600 0.20 2.30 No
find
94 I. Dugic

Fig. 12 Irradiation position drawing

References

1. Campbell J (2003) Castings, 2nd edn. Butterworth-Heineman, Oxford

2. Catalina AV, Leon-Torres JF, Stefanescu DM, Johnson ML (2007) Prediction of shrinkage-
related defects in steel castings. In: Proceedings of the 5th decennial international conference
on solidification processing, Sheffield
3. Desai J, Parikh V, Shah T (2013) Reducing porosity for alloy steel casting. Int J Eng Res
Technol (IJERT) 2(11)
4. Carlson KD, Lin Z, Hardin R, Beckermann C (2002) Modeling of porosity formation and
feeding flow in steel casting. In: Proceedings of the 56th SFSA technical and operating
conference, Paper No. 4.4, Steel Founders’ Society of America, Chicago, IL
5. Griffin JA, Bates CE (1991) Ladle treating, pouring and gating for the production of clean steel
castings. In: Steel Founders’ Society of America Research Report No. 104, Technical Steering
Committee Report
6. Yang S-H, Lee Z-H (2006) A method for predicting nitrogen gas pores in nitrogen alloying
stainless steels. Mater Sci Eng 417(1–2):307–314
7. Stefanescu DM (2005) Computer simulation of shrinkage related defects in metal castings—a
review. Int J Cast Met Res 18(3):129–143. [Link]
8. Monroe R (2005) Porosity in castings. In: Paper 05–245(04).pdf, Page 1 of 28 AFS Transactions
2005 © American Foundry Society, Schaumburg, IL USA
9. Lee PD, Chirazi A, See D (2001) Modeling microporosity in aluminum–silicon alloys: a review.
J Light Met 1(1):15–30
10. Ducic N, Cojbasic Z, Manasijevic S, Radisa R, Slavkovic R, Milicevic I (2016) Optimization
of the gating system for sand casting using genetic algorithm. American Foundry Society, pp
1–11
How to Prevent Porosity Defects in Steel Casting Component 95

11. Dmitriev VS, Kostyuchenko TG (2018) Design of the gating system for production of casting
blanks for space device cases. Mater Sci Forum 1–7
12. Nandagopal M, Sivakumar K, Senthilkumar G, Sengottuvelan M (2017) Study of sand casting
gating system. Int J Recent Eng Res Dev 2(11):1–8
Fluctuant Solidification Behavior
in the Centerline of Continuous Casting
Billets Based on Numerical Simulation

Dongwei Guo, Zihang Zeng, Kunhui Guo, and Zibing Hou

Abstract The fluctuant solidification behavior of the solute-enriched liquid phase in

the centerline of the continuous casting billets, which is closely related to the center-
line segregation defect, will significantly affect the central quality uniformity along
the casting direction. Meantime, the fluctuation phenomenon of the solidification end
point during the continuous casting process is still unclear. In this work, the solid
fraction in the centerline of the billets was analyzed based on a three-dimensional
numerical model, and the fluctuation characteristics of the centerline solid fraction
were revealed. At the same time, through the analysis of the actual solidification
structure, the fluctuation of the centerline local cooling rate was also investigated,
which was associated to that of the solid fraction. On this basis, according to the peri-
odic fluctuation characteristics of centerline solid fraction and local cooling rate, the
periodic fluctuation mechanism of the solidification end point of continuous casting
was proposed.

Keywords Continuous casting · Segregation · Fluctuation of centerline solid

fraction · Fluctuation of local cooling rate · Fluctuation of solidification end point

D. Guo · Z. Zeng · K. Guo · Z. Hou (B)

College of Materials Science and Engineering, Chongqing University, No. 174 Shazhengjie,
Shapingba, Chongqing 400044, China
e-mail: houzibing@[Link]
D. Guo
e-mail: guodongwei@[Link]
Z. Zeng
e-mail: 202009131256@[Link]
K. Guo
e-mail: guokunhui@[Link]
Chongqing Key Laboratory of Vanadium-Titanium Metallurgy and New Materials, Chongqing
University, Chongqing 400044, China

© The Minerals, Metals & Materials Society 2023 97

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
98 D. Guo et al.

Introduction

Nowadays, the steel industry is increasingly demanding the uniformity and stability
of the internal quality of continuous casting billets [1, 2]. The centerline segregation
defect formed by the enrichment of solute elements during the continuous casting
process is a key factor affecting the central quality uniformity of billets [3]. The
fluctuation of centerline segregation along the casting direction is closely related to
the cooling process and solidification behavior of the solute-enriched liquid phase in
the central area of the billets at the end of solidification. Although previous studies
[4, 5] described the formation mechanism of centerline segregation, the continuous
cooling process and solidification behavior of the solute-enriched liquid phase along
the casting direction in the central area of the billets and its influence on the fluctuation
of centerline segregation are still unclear.
On the other hand, the final electromagnetic stirring (F-EMS) [6] and the mechan-
ical soft reduction (MSR) [7] are usually used during the actual continuous casting
process to improve the uniformity of solute distribution in the central area of the
billets and reduce the centerline segregation defect. During the application process
of the F-EMS and MSR, it is required that the central solid fraction [8] and the molten
pool width [9] of billets should be kept in a specific range. At this time, the fluctuation
of the solidification end point of continuous casting will lead to the fluctuation of
central solid fraction or molten pool width of billets, thus affecting the effect of the
F-EMS and MSR on centerline segregation. However, due to the high temperature
and “black box” characteristics of the continuous casting process, the fluctuation
characteristic of the solidification end point of continuous casting still needs to be
investigated.
Based on the above literature review, the typical medium-carbon steel continuous
casting billets were selected as the research object in this work. The main chemical
composition (mass %) is C 0.40, Si 0.25, Mn 0.7, Cr 0.95, Mo 0.18, and Fe balance.
Firstly, the fluctuation characteristic of solid fraction in the centerline of billets was
obtained through the three-dimensional numerical model of the continuous casting
process. Subsequently, based on the actual solidification structure on the longitudinal
section of selected billets, the fluctuation characteristic of the local cooling rate in the
centerline was also analyzed. On this basis, the periodic fluctuation mechanism of
the solidification end point during continuous casting was proposed combined with
the periodic fluctuation of centerline solid fraction and local cooling rate. This paper
can provide a theoretical reference for the cooling process and solidification behavior
of molten steel in the central area of continuous casting billets and the fluctuation
mechanism of the solidification end point of continuous casting.
Fluctuant Solidification Behavior in the Centerline of Continuous … 99

Experiment Method

Three-Dimensional Numerical Simulation of Continuous

Casting Process

In this paper, the fluctuation characteristics of solid fraction and local cooling rate
in the centerline of billets were used to reflect the continuous cooling process and
solidification behavior of the solute-enriched liquid phase in the centerline area.
However, the fluctuation of the centerline solid fraction cannot be obtained through
the actual solidification structure characteristics. Therefore, COMSOL Multiphysics
5.6 was used to model the solidification process of the selected continuous casting
billets under two sets of process parameters, and the fluctuation characteristic of
the centerline solid fraction along the direction was calculated. The computational
domain of the modeling process is shown in Fig. 1, and the specific continuous
casting process parameters are shown in Table 1.

Governing Equations

Based on the above-mentioned description and assumptions of the model, the flow
and solidification process of molten steel in billets is simulated by solving the conti-
nuity equation, Navier–Stokes equation, energy equation, and other supplementary
equations. The turbulent characteristic of the molten steel flow is defined by the k-ε
model, where k is the turbulent kinetic energy, m2 /s2 ; ε is the turbulent dissipation
rate, m2 /s3 . The equations are expressed as follows:

ρ∇u = 0 (1)

Fig. 1 Calculation domain

of three-dimensional
solidification model of
selected continuous casting
billets

Table 1 Main production

Sample No Superheat, °C Casting speed, Specific water
parameters of selected
m/min amount, L/kg
continuous casting billets
1 42 1.8 0.68
2 40 2.1 0.68
100 D. Guo et al.

ρ(u · ∇)u = −∇ p + ∇ μeff (∇u + (∇u)t + F + (ρ − ρref )g (2)

where ρ is the density, kg/m3 ; u is the vector velocity of molten steel, m/s; p is
pressure, Pa; F is volume force, N/m3 ; μeff is the effective viscosity of molten steel,
kg/(m s); ρ ref is the reference density of molten steel, kg/m3 ; is the acceleration of
gravity, m/s2 .
The effective viscosity of molten steel can be expressed as:

μeff = μ + μt (3)

k2
μt = ρCμ (4)
ε
where μ is the laminar viscosity, kg/(m s); μt is the turbulent viscosity, kg/(m s); and
C μ is a constant of 0.09.
The turbulence kinetic energy, k, and the turbulent dissipation rate, ε, are given
as follows:

μt
ρ(u.∇)k = ∇ · μ + ∇k + Pk − ρε (5)
σk

μt ε ε2
ρ(u · ∇)ε = ∇ · μ + ∇ε + Cε1 Pk − Cε2 ρ (6)
σε k k

∇u
Pk = μt (7)
∇u + (∇u)t

where σ k , σ ε , C ε1 , and C ε2 are empirical constants and the values are 1.0, 1.3, 1.44,
and 1.92, respectively.
The energy conservation equation can be described as follows:

ρCp u · ∇t + ∇ · q = Q + Q p + Q vd (8)

q = −λ∇t (9)

where C p is the heat capacity at constant pressure, J/(kg °C); λ is the thermal conduc-
tivity, W/(m °C); Q is the external heat source; Qp is the heat source by pressure works;
Qvd is the heat source by viscous dissipation.
During the solidification process, the physical parameters of molten steel in the
mushy zone, such as density, heat capacity, and thermal conductivity, are not constant
but are dependent on the temperature and/or solid fraction of the mushy zone. These
physical parameters can be calculated according to Eqs. (11)–(14).

ρ = θρ1 + (1 − θ )ρ2 (10)

Fluctuant Solidification Behavior in the Centerline of Continuous … 101

1 ∂αm
Cp = [θρ1 C p,1 + (1 − θ )ρ2 C p,2 ] + L 1→2 (11)
ρ ∂t
1 (1 − θ )ρ2 − θρ1
αm = (12)
2 θρ1 + (1 − θ )ρ2

λ = θ λ1 + (1 − θ )λ2 (13)

where ρ 1 and ρ 2 are the density of solid and liquid steel, and the values are 7600 kg/m3
and 7020 kg/m3 , respectively; C p,1 and C p,2 are the heat capacity of solid and liquid
steel, and the values are 710 J/(kg °C) and 880 J/(kg °C), respectively; λ1 and λ2
are the thermal conductivity of solid and liquid steel, and the values are set to be
118.8 W/(m °C) and 29.7 W/(m °C). During the calculation process, the solidus
and liquidus temperatures of the selected medium-carbon steel were set to 1357 °C
and 1497 °C, respectively.

Boundary and Initial Conditions

The size of the numerical model is 160 mm × 160 mm, which corresponds to the
actual billets, and the length of the calculation domain is 15.9 m. The lengths, the
boundary conditions, and the relevant formulas of mold, secondary cooling zone,
and air cooling zone are shown in Table 2 [10–12], where qm , qk , and qa are the heat
flux of mold, secondary cooling zone, and air cooling zone, respectively, kW/m2 ;
δ m is the holding time in the mold, s; β is a coefficient about the shape of the mold
and the parameter of casting, kW/(m2 s1/2 ); hk is the heat transfer coefficient of the
secondary cooling zone, W/(m2 °C); w is water flow rate, L/(m2 min); η is the radia-
tion coefficient; σ is the Stefan-Boltzmann constant; t b , t w , and t a are the temperature
of the billets surface, the water of secondary cooling zone, and the environment of
air cooling zone, respectively, °C. The mold water flow rate is 114 m3 /h, according
to the actual production condition of selected billets. At the beginning of the calcu-
lation, it is assumed that the calculation domain is filled with molten steel and the
temperature is consistent with the casting temperature, which can be presented as:

tin = t1 + tsh (14)

where t in is the casting temperature of molten steel, °C; t l is the liquidus temperatures
of the selected medium-carbon steel, °C; t sh is the superheat of the molten steel, °C.
The straight bore type submerged entry nozzle is not considered in the established
numerical model, so the velocity of molten steel flowing into the mold is equal to
the casting speed, namely:

vin = vc (15)

where vin is the velocity of molten steel flowing into the mold.
102 D. Guo et al.

Table 2 Parameters and relevant computational formulas at different sections in calculation of

three-dimensional model
Section Length, m Boundary condition Computational formula
√
Mold 0.9 qm qm = (2.68 − β δm ) × 103
First section of secondary 0.5 qk = hk (t b − t w ) hk = 153.6(w/60)0.351
cooling zone
Second section of secondary 2.7 qk = hk (t b − t w ) hk = 160 + 8.35w0.851
cooling zone
Third section of secondary 2.9 qk = hk (t b − t w ) hk = 200 + 10.44w0.851
cooling zone
Fourth section of secondary 3.5 qk = hk (t b − t w ) hk = 200 + 10.44w0.851
cooling zone
Air cooling zone 5.4 qa = ησ (t b 4 − t a 4 ) η = 0.8

Acquisition of Solidification Structure in Billets

Billet samples produced by process parameters No. 1 and No. 2 were obtained during
the actual continuous casting process. According to the method shown in Fig. 2a,
two central longitudinal section samples of continuous casting billets under process
parameters No. 1 and No. 2 were obtained, and the solidification structure on the
central longitudinal section can be obtained by hot pickling experiment. During
the experiment, the samples were etched with 1:1 warm hydrochloric acid–water
solution for revealing the solidification structure more clearly. The temperature of
the hydrochloric acid–water solution was 60–80 °C, and the pickling time was 25 min
[13].

Fig. 2 Sampling method for central longitudinal section of selected continuous casting billets
(a) and measurement position of secondary dendrite arm spacing of fine dendrites inside central
segregation spot (b)
Fluctuant Solidification Behavior in the Centerline of Continuous … 103

Results and Discussion

Fluctuation of Solid Fraction and Local Cooling Rate

in the Centerline of Billets

Based on the three-dimensional numerical model presented, the solid fraction in the
centerline of billets No. 1 and No. 2 can be obtained. In this work, the effect of molten
steel solidification in the centerline of billets on centerline segregation defect was
investigated, and the centerline segregation was formed at the end of solidification.
Therefore, to reflect the solidification behavior at the end of solidification more
intuitively, the change data of the centerline solid fraction of billets No. 1 and No.
2 from 90 to 100% are extracted, and the results are shown in Fig. 3. It can be
seen that the centerline solid fraction gradually increases, but the increase process
is fluctuating. Assuming that the actual continuous casting process is in an ideal
state, that is, the position of the solidification end point and the shape of the liquid
core are constant, the centerline solid fraction of billets should increase smoothly.
Meantime, at the solidification end point, due to the stable centerline solid fraction,
the centerline local cooling rate of the billets should also be constant under specific
cooling conditions. At this time, the solute element distribution and solidification
structure in the centerline will be relatively uniform. However, the fluctuation of
the centerline solid fraction of the actual billets means that the cooling process and
solidification behavior in the centerline are not uniform, which will aggravate the
fluctuation of the centerline segregation along the casting direction.
To verify the above analysis, the centerline local cooling rate was calculated
through the actual solidification structure of billets No. 1 and No. 2. Figure 4 shows

Fig. 3 Trend of solid fraction from 90 to 100% in the centerline of billets No. 1 and No. 2
104 D. Guo et al.

the solidification structure in the central area on the longitudinal section of billets No.
1 and No. 2, and the spot segregation defect solidified by the enriched solute liquid
phase between dendrites at the end of solidification [14, 15]. Due to the solidification
characteristic of continuous billets from outside to inside, spot segregation in the
central area can reflect the solidification behavior of molten steel in the final stage.
Then, the central spot segregation defect on the longitudinal section of billets No.
1 and No. 2 was analyzed, and it was found that there were fine dendrites, different
from the conventional solidification structure, in the spot segregation, as shown in
Fig. 5. The secondary dendrite arm spacing (SDAS) can reflect the local cooling rate
during the formation of dendrites. As the local cooling rate increases, the dendrites
will be finer. Therefore, the fine dendritic structure inside the spot segregation can
be used to intuitively reflect the local cooling rate at the final solidification of the
central area of the billets.
According to the method shown in Fig. 1b, the SDAS of fine dendrites inside the
spot segregation on the longitudinal section of billets No. 1 and No. 2 was measured.
Starting from the initial position in Fig. 1b, the SDAS of the fine dendrites inside
each spot segregation was measured along the solidification direction. When there

Fig. 4 Solidification structure in central area on the longitudinal section of billets No. 1 (a) and
No. 2 (b)

Fig. 5 Fine dendrites inside spot segregation in central area on longitudinal section of billets No.
1 (a) and No. 2 (b)
Fluctuant Solidification Behavior in the Centerline of Continuous … 105

Fig. 6 Local cooling rate along the solidification direction in the centerline of billets No. 1 (a) and
No. 2 (b)

were multiple fine dendrites inside the spot segregation, the measurement results
were represented by the average SDAS of multiple fine dendrites. Then, the local
cooling rate can be calculated by Eq. (16).

d2 = β · R −α (16)

where is the SDAS of fine dendrites inside the centerline spot segregation of billets,
µm; R is the local cooling rate at the fine dendrites, °C/s; α and β are coefficients.
For the billet selected in this paper, α is 0.38 and β is usually 148 [16].
The calculated centerline local cooling rate along the solidification direction is
shown in Fig. 6. It can be seen that during the final stage of the solidification process in
billets, the local cooling rate along the solidification direction is fluctuating, which
corresponds to the calculation result of the centerline solid fraction. In the actual
continuous casting process, the fluctuation of the centerline solid fraction represents
the volume fluctuation of the solute-enriched liquid phase, and the solidification of
the solute-enriched liquid phase with different volumes will lead to the fluctuation
of the centerline local cooling rate. The non-uniformity of the solidification process
and cooling rate in the central area will finally lead to the fluctuation of centerline
segregation and properties of the billets.

Fluctuation of Solidification End Point in Continuous Casting

Process

According to the centerline solid fraction of billets No. 1 and No. 2 shown in Fig. 3,
a relatively typical section is taken for analysis, and the results are shown in Fig. 7. It
can be seen that the centerline solid fraction of the billets can be regarded as periodic
fluctuation. In a single fluctuation period, the centerline solid fraction decreases first
and then increases. Thus, the fluctuation period of the centerline solid fraction marked
by the dotted line in Fig. 7 can be obtained. Since the centerline solid fraction is
106 D. Guo et al.

Fig. 7 Period of local solid fraction in centerline of billets No. 1 (a) and No. 2 (b)

related to the local cooling rate, the centerline local cooling rate should also fluctuate
periodically.
Subsequently, the centerline local cooling rate of billets shown in Fig. 6 was
analyzed, and it was found that the local cooling rate was also periodic along the
solidification direction. According to the calculation results, the polynomial fittings
of the different cooling rate periods along the solidification direction were carried
out, and the results are shown as the blue curves in Fig. 8. It can be seen that the local
cooling rate decreases first and then increases in one period along the solidification
direction, where T 1-1 and T 1-2 are the periods of centerline local cooling rate of billet
No. 1, T 2-1 and T 2-2 are the periods of centerline local cooling rate of billet No. 2.
The common periodic fluctuation characteristic of the solid fraction and cooling rate
in centerline represent that periodicity is the inherent characteristic of solidification
and cooling in billets.
The centerline solid fraction and local cooling rate are the reflections of the final
stage of billets solidification. When the solute-enriched liquid phase in the center-
line is solidified under certain cooling rate, the solidification end point is formed.
Combined with the periodic fluctuation of the solid fraction and local cooling rate
in the centerline, it is speculated that the actual solidification end point of contin-
uous casting also fluctuates periodically in a certain range. The periodic fluctuation
process of centerline solid fraction, local cooling rate, and solidification end point
of continuous casting billets is shown in Fig. 9. Firstly, the fluctuation characteristic
of the macroscopic or microscopic flow of molten steel in the billets [17, 18] will

Fig. 8 Fitting period of local cooling rate in the centerline of billets No. 1 (a) and No. 2 (b)
Fluctuant Solidification Behavior in the Centerline of Continuous … 107

affect the temperature distribution of molten steel in the liquid core, resulting in a
periodic distribution of centerline solid fraction. With the movement of the billets,
the fluctuating centerline solid fraction will complete the final solidification. For the
cooling process ➀ in Fig. 9, the volume of the solute-enriched liquid phase increases
with the decrease of the centerline solid fraction. Under certain cooling conditions,
the cooling rate will decrease and the cooling time will increase. When the casting
speed is constant, the increase of cooling time at the end of solidification makes the
solidification end point move away from the meniscus, that is, the position marked
by the dotted line e2 . As for the cooling process ➁, the volume of the solute-enriched
liquid phase decreases with the increase of the centerline solid fraction. Thus, the
cooling rate will increase and the cooling time will decrease under certain cooling
conditions, which will make the solidification end point move to the meniscus, that
is, the position marked by the dotted line e1 . The cooling process ➂ is roughly the
same as the cooling process ➀, and the solidification end point moves again to the
position marked by the dotted line e1 .
In summary, the fluctuation characteristic of the centerline solid fraction in billets
was found through the three-dimensional numerical model of medium-carbon steel
billets. Subsequently, according to the solidification structure on the longitudinal
section of selected billets, the fluctuation of the centerline local cooling rate was
analyzed. During the continuous casting process, the fluctuation of the centerline
solid fraction under certain cooling conditions will cause the fluctuation of the
centerline local cooling rate. The common fluctuation characteristic of these two
parameters reflects the uneven solidification behavior and cooling process along the
casting direction at the end of solidification, which eventually leads to the uneven
solidification structure and centerline segregation defect of the billets. On this basis,
the fluctuation periodicity of the centerline solid fraction and local cooling rate was

Fig. 9 Schematic diagram of periodic fluctuation process of solid fraction, local cooling rate in
centerline, and solidification end point in continuous casting billets
108 D. Guo et al.

analyzed, and the periodic fluctuation mechanism of the solidification end point of
continuous casting was proposed based on this periodicity.

Conclusions

Based on the three-dimensional numerical model of medium-carbon steel billets, the

fluctuation characteristic of the centerline solid fraction of billets was assessed. At
the same time, the centerline local cooling rate was calculated through the actual
solidification structure characteristics of billets, and the fluctuation characteristic
corresponding to the solid fraction was shown. The fluctuation characteristic of the
centerline solid fraction and local cooling rate results in the uneven cooling and
solidification in the centerline of billets and aggravate the fluctuation of the centerline
segregation along the casting direction.
The fluctuation periodicity of the centerline solid fraction and local cooling rate
was analyzed. Based on this periodic characteristic, the periodic fluctuation mecha-
nism of the solidification end point of continuous casting was proposed. This paper
can provide a theoretical reference for the cooling process and solidification behavior
of molten steel in the central area of continuous casting billets and the fluctuation
mechanism of the solidification end point of continuous casting.

Acknowledgements The authors are very grateful for COMSOL Multiphysics 5.6 software and
the support from National Natural Science Foundation of China (52274318).

References

1. Ai XG, Han D, Li SL, Zeng HB, Li HY (2020) Optimization of flow uniformity control device
for six-stream continuous casting tundish. J Iron Steel Res Int 27(9):1035–1044
2. Zhou YJ, Xu K, He F, Zhang ZY (2022) Application of time series data anomaly detection
based on deep learning in continuous casting process. ISIJ Int 62(4):689–698
3. Choudhary SK, Ganguly S (2007) Morphology and segregation in continuously cast high
carbon steel billets. ISIJ Int 47(12):1759–1766
4. Chen HB, Long MJ, Chen DF, Liu T, Duan HM (2018) Numerical study on the characteristics
of solute distribution and the formation of centerline segregation in continuous casting (CC)
slab. Int J Heat Mass Transf 126:843–853
5. Jiang DB, Wang WL, Luo S, Ji C, Zhu MY (2018) Numerical simulation of slab centerline
segregation with mechanical reduction during continuous casting process. Int J Heat Mass
Transf 122:315–323
6. Wan Y, Li MH, Chen LJ, Wu YC, Li J, Pan HB (2019) Effect of final electromagnetic stirring
parameters on central cross-sectional carbon concentration distribution of high-carbon square
billet. Metals 9(6):665–681
7. Ji C, Luo S, Zhu MY (2014) Analysis and application of soft reduction amount for bloom
continuous casting process. ISIJ Int 54(3):504–510
Fluctuant Solidification Behavior in the Centerline of Continuous … 109

8. Zeng J, Chen WQ, Wang QX, Wang GS (2016) Improving inner quality in continuous casting
rectangular billets: comparison between mechanical soft reduction and final electromagnetic
stirring. Trans Indian Inst Met 69(8):1623–1632
9. Xiao C, Zhang JM, Luo YZ, Wei XD, Wu L, Wang SX (2013) Control of macrosegregation
behavior by applying final electromagnetic stirring for continuously cast high carbon steel
billet. J Iron Steel Res Int 20(11):13–20
10. Wang W, Hou ZB, Chang Y, Cao JH (2018) Effect of superheat on quality of central equiaxed
grain zone of continuously cast bearing steel billet based on two-dimensional segregation ratio.
J Iron Steel Res Int 25(1):9–18
11. Choudhary SK, Mazumdar D (1995) Mathematical modelling of fluid flow, heat-transfer and
solidification phenomena in continuous-casting of steel. Res Int 66(5):199–205
12. Cai K, Yang J (1989) Investigation of heat transfer in the spray cooling in continuous casting.
J Univ Sci Technol Beijing 11(6):510–515
13. Hou ZB, Guo ZA, Guo DW, Cao JH, Chang Y, Wen GH (2019) A new method for carbon
content distribution based on grayscale image of casting blank macrostructure in carbon steel.
J Iron Steel Res 31(7):620–627
14. Tsuchida Y, Nakada M, Sugawara I, Miyahara S, Murakami K, Tokushige S (1984) Behavior
of semimacroscopic segregation in continuously cast slabs and technique for reducing the
segregation. Trans Iron Steel Inst Jpn 24(11):899–906
15. Ji Y, Lan P, Geng H, He Q, Shang CJ, Zhang JQ (2018) Behavior of spot segregation in
continuously cast blooms and the resulting segregated band in oil pipe steels. Steel Res Int
89(3):700331
16. ElBealy M, Thomas BG (1996) Prediction of dendrite arm spacing for low alloy steel casting
processes. Metall Mater Trans B 27(4):689–693
17. Liu ZQ, Li BK, Jiang MF, Tsukihashi F (2013) Modeling of transient two-phase flow in a
continuous casting mold using euler-euler large eddy simulation scheme. ISIJ Int 53(3):484–
492
18. Li XL, Li BK, Liu ZQ, Niu R, Liu YQ, Zhao CL (2019) Large eddy simulation of multi-phase
flow and slag entrapment in a continuous casting mold. Metals 9(1):7–20
Part III
Slag and Ladle Treatment
Toward Meso-scale Modelling of Slag
Foaming Phenomena in Pyrometallurgy

Quinn G. Reynolds and Oliver F. Oxtoby

Abstract Dispersed multiphase fluid flows, in which one phase is distributed as

small inclusions in another, occur in a wide range of chemical and process indus-
tries. In pyrometallurgical smelting, these may manifest in the form of decoupling
of gases from one of the molten phases. This can cause slag foaming, which occurs
when gas bubbles are unable to escape from the viscous slag rapidly enough and a
low-density foam layer builds up at the surface of the slag pool. Although uncon-
trolled slag foaming can cause hazardous equipment failures, controlled foaming has
the potential to significantly reduce energy consumption of many smelting processes.
Improvements in the understanding of slag foaming are therefore of value both
from health and safety as well as economic and environmental aspects. This paper
presents the evaluation and application of the dynamic multi-marker (DMM) method,
a novel meso-scale computational fluid dynamics algorithm for efficient modelling
of dispersed-phase systems, for slag foaming problems. Foaming behaviour and gas–
liquid decoupling are studied using numerical simulations of test systems, and the
results are compared to established empirical relationships.

Keywords Slag foaming · Computational modelling · Pyrometallurgy

Introduction

The pyrometallurgical production of commodities such as iron and steel, ferroal-

loys, precious metals, and numerous others underpins much of modern industri-
alised society [1]. Pyrometallurgical smelting processes are frequently energy- and
carbon-intense, and the drive to reduce direct and indirect emissions from heavy
industry worldwide is placing considerable pressure on such plants to optimize their

Q. G. Reynolds (B)
Pyrometallurgy Division, Mintek, Randburg, South Africa
e-mail: quinnr@[Link]
Department of Process Engineering, University of Stellenbosch, Stellenbosch, South Africa
O. F. Oxtoby
ENGYS, London, UK

© The Minerals, Metals & Materials Society 2023 113

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
114 Q. G. Reynolds and O. F. Oxtoby

existing operations while shifting toward cleaner and more efficient technologies for
the future. Combined with the rapid growth in processing of complex secondary ores
from recycling sources such as e-wastes and batteries, there is increasing demand for
cross-cutting energy saving techniques which can be applied to pyrometallurgical
furnaces of many types.
Most high-temperature reductive smelting and recycling processes operate with
multiple immiscible molten phases inside the furnace vessel. Presence of a slag phase,
typically a mixture of metal oxides and other refractory materials, is common. Due to
the unique properties of molten slag such as high viscosity and surface tension, slag
foaming is a phenomenon that occurs under certain conditions in many pyrometallur-
gical processes. As the name suggests such foams consist of many small bubbles of
gas trapped together in a low-density layer, separated by thin films of liquid slag. The
bubbles may originate either from external gas injection lances, or more commonly
from solid carbonaceous reductants reacting with the process material to produce
CO gas.
Slag foaming is frequently viewed as undesirable as it can lead to poor control
of the process energy and metallurgical balances, and in extreme cases rapid uncon-
trolled slag foaming events may cause eruptions or explosions on the furnace vessel
which are hazardous for operators working in the area. As shown in Fig. 1, controlled
slag foaming does, however, have great potential for protecting the furnace walls,
reducing consumption of (typically carbon-based) electrodes in electric furnaces,
and improving process energy efficiency; the steel industry in particular has applied
slag foaming to good effect in this regard [2].
The existing body of academic literature on slag foaming focuses predominantly
on experimental tests conducted in laboratory and industrial environments, and the
formulation of empirical correlations based on these data. The pioneering research
of Fruehan and colleagues [3–5] dominates the field, and data from much of their
work is frequently collated and re-analysed in reviews such as [6].
Empirical correlations developed for slag foaming are usually expressed in terms
of dimensionless groups. Per the Buckingham π theorem any suitable set can be
chosen, but in these applications it is preferred to use the Morton, Archimedes, and

Fig. 1 Arc furnace without (l) and with (r) an insulating slag foam layer, showing reduction in
energy losses
Toward Meso-scale Modelling of Slag Foaming Phenomena … 115

Table 1 Dimensionless gμ

N σ
groups in slag foaming
gμ4
NMo ρσ 3
gρ 2 db3
NAr μ2
μu g
NCa σ

Capillary numbers as independent values, with the foam index number N as the
dependent variable. Definitions of these numbers in terms of dimensioned parameters
are given in Table 1, with g the normal acceleration due to gravity, μ the viscosity of
the slag, σ the surface tension of the slag, ρ the density of the slag, db a representative
diameter of the bubbles in the foam, and u g a superficial gas velocity based on the
volumetric gas flow rate and the surface area over which bubbles are being emitted
into the slag. The foam index is defined as the rate of change of the stable foam
height with the superficial gas velocity and has the units of time. It should be noted that
the expressions in Table 1 apply only in the simplified case of negligible minimum
foaming velocity, and slag density much greater than gas density; interested readers
are referred to Lotun and Pilon [6] for a more detailed treatment including these
additional phenomena.
Using these quantities, researchers have developed many different empirical corre-
lations to describe the available slag foaming data, several of which are shown below
[3, 4, 6].

N = 359NMo
0.5
(1)

0.4 −0.3
N = 900NMo NAr (2)

0.067 −0.867 −0.2

N = 2617NMo NAr NCa (3)

Some authors have questioned the inclusion of the bubble diameter db as an inde-
pendent variable. Stadler et al. [7] conducted additional slag foaming experiments
and concluded that there was no statistical justification for choosing between Eqs. (1)
and (2) given the errors involved in such work. Lotun and Pilon [6] also raised doubts
from a more theoretical standpoint, observing that well-established relationships of
foam index to viscosity, density, and surface tension are only recovered from corre-
lations such as Eq. (3) when the bubble diameter is assumed to be a function of the
physical properties of the slag. Although bubble diameter has been retained as an
independent variable in the current study, it is clear that this will require investigation
in future work.
The application of numerical and computational models to dispersed-phase flows
such as slag foaming is desirable as an alternative to the challenges of conducting
experimental investigations at high temperatures but presents a number of challenges
of its own. The primary problem is the very wide range of length scales involved; a
116 Q. G. Reynolds and O. F. Oxtoby

typical industrial reactor or furnace may be tens of metres in size while the individual
particles of the dispersed phase are of the order of centimetres or millimetres, and the
thin films between adjacent particles may be down to micrometres or nanometres in
thickness. A hierarchical approach is therefore advisable, with computational mate-
rials science models describing the properties of the film at atomistic level, meso-scale
computational fluid dynamics (CFD) models describing the resolved flow patterns
at the particle scale, and CFD models using simplified empirical closures to capture
the bulk statistical effects of the dispersed phase at industrial scale. Many examples
of industrial-scale models are already available in the literature, e.g. Sattar et al. [8],
however, information required for the empirical closures needed to complete these
models generally comes from limited experimental data or simplified theoretical
treatments and is a significant source of error.
Computational modelling methods to address the micro- and meso-scales remain
in their infancy but have the potential to serve as an extensive data source supporting
the development of more robust engineering simulation tools via bottom-up numer-
ical experimentation. With this in mind, this paper aims to evaluate the applicability of
a novel algorithm for CFD modelling of dispersed multiphase flows at the meso-scale,
the dynamic multiple marker (DMM) method, to the problem of slag foaming.

Model Description

The DMM method is based on the volume-of-fluids (VOF) technique for multiphase
flows and makes use of dummy or marker phase fields to prevent any adjacent bubbles
in a simulation from merging prematurely even when the thickness of the film of slag
between them drops below the resolution limit of the mesh on which the problem
is being solved. This is combined with a dynamic graph-based contact detection
and phase reassignment algorithm which minimises the number of marker phases
required at each time step in the simulation. In addition, a sub-grid model based on
the Reynolds equation for film thickness evolution is activated when bubbles are in
contact with one another and may be used to trigger film rupture and coalescence
of bubbles. Detailed descriptions of the various components of the DMM algorithm
may be found in [9, 10].
In the present work, the DMM solver and specialised inlet–outlet boundary condi-
tions for the dispersed phase were implemented using the v2206 release build of
the OpenFOAM® open-source computational mechanics platform [11]. Software
automation tools written in Python 3.9 facilitated with pre- and post-processing of the
data from the simulations [12]. Field visualisations were performed using ParaView
5.10 [13].
Toward Meso-scale Modelling of Slag Foaming Phenomena … 117

Results and Discussion

For the purposes of this paper, the choice was made to study the foaming problem
in 2D planar geometries. Although the DMM method and OpenFOAM® software
implementation extend trivially to three dimensions, the computational cost asso-
ciated with such simulations is much higher. 3D studies are therefore best left
to production simulations after the method has been more thoroughly tested and
validated.
Based on the literature summaries of previous experimental work on slag foaming
[6], a representative reference set of material property and numerical parameters was
selected as shown in Table 2. These are used for all simulations presented here unless
otherwise mentioned. h crit indicates the thickness of the thin films between bubbles
at which point rupture is assumed to occur—this has been observed to be of the order
of 500 nm for silicic materials similar to metallurgical slags [14].
Readers will note that in real slag foaming problems the density of the gases
involved is typically very low (< 1 kg/m3 ). The elevated gas density used in this
work is an approximation which greatly improves numerical stability of standard
VOF multiphase flow solution algorithms in cases of high surface tension and low
fluid density, by damping the effect of parasitic currents at the phase interfaces. The
suitability of this choice will be examined in the following section.

Numerical Considerations

In this section, a single gas bubble was simulated as it rises from 0.05 m below the
surface of a slag layer and subsequently ruptures into the bulk gas phase above. An
example sequence showing the system behaviour over time is given in Fig. 2. The
bubble deforms considerably as it rises, and the surface of the slag is pushed upward
by the bubble before it settles. After a short period of time the slag film surrounding
the bubble thins and ruptures.

Table 2 Reference
Parameter Value
parameter set for slag
foaming simulations Slag phase density, ρ 2750 kg/m3
Slag phase viscosity, μ 0.5 Pa.s
Gas phase density, ρg 30 kg/m3
Gas phase viscosity, μg 5 × 10–5 Pa.s
Interfacial tension, σ 0.35 N/m
Bubble diameter, db 0.01 m
Mesh resolution, δl 5 × 10–4 m
Critical film thickness, h crit 5 × 10–7 m
118 Q. G. Reynolds and O. F. Oxtoby

(a) 0 s (b) 0.2 s (c) 0.4 s (d) 0.46 s (e) 0.52 s (f) 0.56 s

Fig. 2 Phase fields at different times for bubble rise reference case (ρ g = 30 kg/m3 , δ l = 500 um).
Red is slag phase, blue is gas

Fig. 3 Model dependence on mesh resolution (l) and gas phase density (r)

This simple test case was used to test the effect of both mesh resolution and
gas density on the results. The position of the slag (or bubble) phase interface on
the bubble centreline was monitored over time to give a quantitative indication of
the dynamics. The results are shown in Fig. 3. Relatively low sensitivity to mesh
resolution was observed, with all simulations showing similar behaviour in both
the mesh-level (bubble rising and slag surface deformation) and sub-grid-level (film
thinning and time to rupture) solutions. The choice of gas density was also seen to
have relatively little impact on the bubble rising and rupturing behaviour provided
the value was kept at 30 kg/m3 or less.

Slag Foaming Case Study

Using the reference set of parameters in Table 2, slag foaming was simulated in a
region 20 × 20 cm in size. The region was initialised with a layer of slag 0.05 m
thick. The left and right boundaries of the region are cyclic and allow material to
Toward Meso-scale Modelling of Slag Foaming Phenomena … 119

pass freely between them. The upper boundary is open to atmosphere, and the lower
boundary is assumed to be a solid wall with 20 equally spaced inlets located along
it. Bubbles are emitted continuously from the inlets, on which gas injection is timed
to reproduce both the specified bubble diameter and superficial gas velocity. 20 s
of model time was simulated, with the period of stable foam height (typically the
final 12.5 s) being used for further analysis in each case. At each time step in the
simulation the distribution of phase fields was sampled along vertical lines at five
different positions in the region and used to measure both the upper and lower extent
of the low-density foam layer. The results were then averaged to obtain an indicative
foam height as a function of time.
Figures 4 and 5 show the evolution of the simulation at a superficial gas velocity
of 0.02 m/s. During slag foaming new bubbles enter the base of the foam layer and
travel slowly upward, occasionally merging with adjacent bubbles as they do so,
before reaching the top of the layer and rupturing into the bulk gas phase. This leads
to a distinct stratification of bubble sizes in the foam, with smaller bubbles at the
bottom of the layer and the largest near the top surface.
The foam thickness in the simulation is shown as a function of time in Fig. 6.
The general behaviour can be roughly divided into three periods. During the start-up
phase, from time zero to approximately 0.5 s, the injected bubbles rise through the
expanding slag layer. The growth phase, between 1.5 and 7.5 s, is characterised by
the formation and increase in height of the foam layer. The stable foaming phase,
from 7.5 s to the end of the simulation, shows an approach to a steady average foam

(a) 1 s (b) 2 s (c) 5 s

(d) 10 s (e) 15 s (f) 20 s

Fig. 4 Phase fields at different times for ug = 0.02 m/s case. Red is slag phase, blue is gas
120 Q. G. Reynolds and O. F. Oxtoby

(a) Marker phases, 15 s (b) Velocities, 15 s

Fig. 5 Dispersed-phase marker phases and velocities for ug = 0.02 m/s case

height as the rates of bubble addition and rupture in the layer begin to come into
equilibrium.
Dividing the foam thickness during the final 12.5 s of the simulation by ug =
0.02 m/s gives a crude estimate of the slag foaming index. The distribution of values
so obtained has a mean of = 2.08 s and a standard deviation of 0.48 s. 95% of
the values therefore lie between 1.12 and 3.04 s, indicating considerable variability;
longer simulation times are recommended to narrow this range in future work. The
values obtained do nonetheless compare reasonably well with the predictions of
correlation Eqs. (1), (2), and (3) which give 1.85 s, 1.42 s, and 1.84 s, respectively,
using the reference parameters in Table 2.
A common test procedure in experimental slag foaming studies is the variation of
ug over a range. Additional simulations were therefore performed using superficial
velocities of 0.01 and 0.03 m/s. The foam thickness evolution for all three simulations
is shown in Fig. 7. It is immediately obvious that at low superficial velocities the

Fig. 6 Foam levels in reference case simulation for ug = 0.02 m/s, (l) position of upper and lower
boundaries of foam layer, (r) estimated foam thickness
Toward Meso-scale Modelling of Slag Foaming Phenomena … 121

Fig. 7 Foam thickness results with increasing ug

(a) 15 s, = 0.01 m/s (b) 15 s, = 0.02 m/s (c) 15 s, = 0.03 m/s

Fig. 8 Phase fields during stable foaming for different superficial gas velocities

bubbles are not replenished at the surface of the slag faster than they burst, and no
foam layer develops. This effect has been noted and discussed at length in [6] and
others in terms of a minimum foaming velocity u g0 ; in such work increasing the
superficial velocity past the minimum value results in the formation of a foam layer
roughly proportional to u g − u g0 . A similar relationship is observed qualitatively in
the present simulation results, although further work will be required to confirm this
rigorously. Comparative visualisations of the phase fields during the stable foaming
period at different superficial gas velocities are shown in Fig. 8, from which the
differences in foam layer thickness are obvious.

Conclusions

Preliminary testing of the dynamic multiple marker method for applications in slag
foaming studies was successfully conducted and indicates that the method shows
122 Q. G. Reynolds and O. F. Oxtoby

promise for studying slag foaming problems at the meso-scale. Comparisons against
empirical relationships from the literature showed reasonable agreement in the quan-
titative and qualitative trends, with similar foam indices obtained from the simulations
and foam heights increasing at higher superficial gas velocities.
Although the model shows good mesh independence and low dependence on gas
densities, more work is required to assess its sensitivity to the various numerical
settings available in the DMM algorithm. In particular, a detailed investigation into
the numerical performance of the sub-grid model for the evolution of the thin films
between bubbles is currently underway. Additional challenges in the case of slag
foaming include very high interfacial tension, a high-viscosity continuum phase, and
a low-density dispersed phase, all of which exacerbate various numerical issues and
instabilities in the underlying VOF algorithm on which the method is based. These
will need to be addressed before numerical experimentation using this approach can
be considered production-ready. Computational costs also remain a challenge with
such high-resolution meso-scale models, despite the good efficiency and scalability
obtained with the DMM approach. With the transition to 3D, which will be necessary
to accurately capture foam topology and bubble size statistics, this will become even
more significant and will need to be investigated further in future work.
In addition to the purely numerical issues, more work on revising and opti-
mising the solver’s implementation in OpenFOAM® is planned and can easily
be combined with extensions to include heat transfer and other effects. Enhance-
ment of the sub-grid film model to include additional physics related to ionic and
metallic fluids will also be considered. Most importantly, a rigorous testing and vali-
dation programme is planned in which DMM simulations will be compared against
cold-model experiments to more thoroughly demarcate its strengths and weaknesses.

Acknowledgements This paper is published by permission of Mintek. The authors acknowledge

the Centre for High Performance Computing (CHPC), South Africa, for providing computational
resources to this research project.

References

1. Pariser HH, Backeberg NR, Masson OCM, Bedder JCM (2018) Changing nickel and chromium
stainless steel markets—a review. J S Afr Inst Min Metall 118(6):563–568. [Link]
17159/2411-9717/2018/v118n6a1
2. Luz AP, Tomba Martinez AG, López F, Bonadia P, Pandolfelli VC (2018) Slag foaming practice
in the steelmaking process. Ceram Int 44(8):8727–8741. [Link]
2018.02.186
3. Jiang R, Fruehan RJ (1991) Slag foaming in bath smelting. MTB 22(4):481–489. [Link]
org/10.1007/BF02654286
4. Zhang Y, Fruehan RJ (1995) Effect of the bubble size and chemical reactions on slag foaming.
MMTB 26(4):803–812. [Link]
5. Matsuura H, Fruehan RJ (2009) Slag foaming in an electric arc furnace. ISIJ Int 49(10):1530–
1535. [Link]
Toward Meso-scale Modelling of Slag Foaming Phenomena … 123

6. Lotun D, Pilon L (2005) Physical modeling of slag foaming for various operating conditions
and slag compositions. ISIJ Int 45(6):835–840. [Link]
7. Stadler SAC, Eksteen JJ, Aldrich C (2007) An experimental investigation of foaming in acidic,
high FexO slags. Miner Eng 20(12):1121–1128. [Link]
8. Sattar MA, Naser J, Brooks G (2013) Numerical simulation of slag foaming in high temperature
molten metal with population balance modeling. Procedia Eng 56:421–428. [Link]
1016/[Link].2013.03.142
9. Musehane NM, Oxtoby OF, Reddy BD (2018) Multi-scale simulation of droplet–droplet
interaction and coalescence. J Comp Phys 373:924–939. [Link]
07.027
10. Reynolds QG, Oxtoby OF, Erwee MW, Bezuidenhout PJA (2021) An extension of the multiple
marker algorithm for study of phase separation problems at the mesoscale. MATEC Web Conf
347:00025. [Link]
11. ESI (2022) OpenFOAM v2206. [Link]
v2206. Accessed 09 Dec 2022
12. Python Software Foundation, “Python,” (2022). [Link] Accessed 09 Dec
2022
13. Kitware, “ParaView,” (2022). [Link] Accessed 09 Dec 2022
14. Castro JM, Burgisser A, Schipper CI, Mancini S (2012) Mechanisms of bubble coalescence in
silicic magmas. Bull Volcanol 74(10):2339–2352. [Link]
Simulation on the Slag Desulfurization
During the LF Refining in a Gas-Blowing
Ladle

Jujin Wang, Yuexin Zhang, Binyu Lyu, and Lifeng Zhang

Abstract A three-dimensional numerical model coupled the fluid-flow simulation

and the reaction kinetics was established to study the evolution of the sulfur content
in the molten steel. The VOF, k-ε, and DPM models were employed to simulate the
fluid flow of three phases in the ladle furnace, including the steel, the slag, and the
air. Reactions between the steel and the slag were considered in the simulation using
a series of user defined functions. The steel-slag reactions had a significant effect on
the sulfur content in the molten steel, resulting the continuous decrease of the sulfur
content. The [S] content reduced from 0.077 wt% into 0.01 wt% in 50 min. The
distribution of [S] in the molten steel was uneven, with the lowest near the steel-slag
interface and the highest in the middle zone between the argon blowing point and
the ladle wall.

Keywords Desulfurization · VOF-DPM · Gas-stirred ladle · Numerical model

Introduction

Sulfur is one of the detrimental impurities in the steel, which decreases dramatically
the impact resistance [1] and weldability [2] of the steel products. The removal of the
sulfur in the molten steel is usually produced in the KR process. For the steelmaking
without the KR process, the ladle metallurgy plays an important role in the sulfur
removal. During the secondary refining, the desulfurization reaction occurs between
the steel and the slag, as expressed in Eq. (1). At the steel-slag interface, the sulfur
reallocates according to the thermodynamic equilibrium [3, 4].

(CaO) + [S] = (CaS)+[O] (1)

J. Wang · L. Zhang (B)

School of Mechanical and Materials Engineering, North China University of Technology,
Beijing 100144, China
e-mail: zhanglifeng@[Link]
Y. Zhang · B. Lyu
School of Metallurgical and Ecological Engineering, University of Science and Technology
Beijing (USTB), Beijing 100083, China
© The Minerals, Metals & Materials Society 2023 125
S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
126 J. Wang et al.

The desulfurization mechanism of the synthetic slag is complex and affected by

several factors including the composition and weight of the slag [5–7], the injected
gas flow rate [8], the temperature [9], and the dissolved element in the steel. Sulfur
capacity was often used to evaluate the desulfurization efficiency of the slag in
thermodynamics. For convenience, the sulfur partition ratio, which expresses the ratio
of sulfur in the slag to sulfur in the metal at the interface, was employed by mostly
researchers to calculate the desulfurization thermodynamic instead of calculating the
equilibrium directly [10].
The desulfurization efficiency is also determined by the flow field of the steel and
slag, as the fluid flow at the steel-slag interface has a big effect on the mass transfer
kinetic of the sulfur. In recent years, many researchers have employed computational
fluid dynamics (CFD) to provide the flow field for the mass transfer. Wang et al. [10]
established a transient three-dimensional coupled model and analyzed the desulfu-
rization behavior during the VE-ESR process. Antonio et al. [11] coupled the fluid
dynamics, thermal, and kinetic behaviors to study the steel desulfurization process in
a gas-stirred ladle furnace. Jonsson et al. [12] used an axisymmetric-two-dimensional
CFD model with a thermodynamic solver to analyze the desulfurization during the
refining process. Andersson et al. developed a similar two-dimensional model as the
Jonsson did for evaluating the desulfurization and reoxidation of the slag. Margareta
et al. [13] proposed a two-dimensional fluid-flow model covering three phases (steel,
slag, and gas), coupling the thermodynamic equation to study the desulfurization.
Lou et al. [14–16] proposed a CFD-SRM coupled model to predict desulfurization
efficiency in an 80 tons ladle.
However, though the numerical simulation has been widely studied, most of
kinetic models simplified the solution for the interface equilibrium, such as using a
constant distribution ratio at the interface instead of solving equilibrium equations
[10, 14–16]. Therefore, in the current study, an accurate and real-time method for the
kinetic process was proposed, where the equilibrium at the interface was solved by
coupling the mass transfer and the thermodynamic based on the coupled reactions
model [17–19]. The ion and molecular coexistence theory (IMCT) was employed to
calculate the activity of CaO and CaS in the slag, and the associate model was used
to calculate the activity of [Ca], [O], and [S] in the steel.

Model Description and Numerical Scheme

Model Schematic

Figure 1 shows the schematic of the current established model. The model consisted
of two parts. One was the solver of the flow of the steel-slag-air three phases, which
was solved using the Fluent 22.0. And another was the species transport, including
the [Ca], [O], [S] in the steel phase, and the (CaO), (CaS) in the slag phase. The
momentum source and the turbulent source were considered using user defined
Simulation on the Slag Desulfurization During the LF Refining … 127

Fig. 1 Model schematic for the fluid flow and the desulfurization

functions (UDFs). Besides, the desulfurization at the steel-slag interface was realized
using the coupled reaction model, which was also attached to the Fluent using UDFs.

Three-Dimensional Multiphase Fluid Flow

A coupled discrete particle model (DPM) and volume of fluid (VOF) model was
applied to simulate the motion of injected argon bubbles and the fluid flow during
the ladle furnace (LF) process. Besides, the turbulent flow was solved using the k-ε
two-equation model. The detailed description and validation of these models could
be find elsewhere [20–24], which would not be discussed here.
Continuity equation (VOF model):

∂αq ρq
+ ∇ · αq ρq uq = 0 (2)
∂t

Momentum conservation equation (VOF model):

∂
(ρM uM ) + ∇ · (ρM u M uM ) = −∇ p
∂t

+ ∇ · μM ∇ uM + ∇ uM T
+ ρM g + F (3)

Turbulence equations (k-ε two-equation model):

∂ρM k μt,M
+ ∇ · (ρM uM k) = ∇ · μM + ∇k + G k − ρM ε + Sk (4)
∂t σk
128 J. Wang et al.

∂ρM ε μt,M
+ ∇ · (ρM uM ε) = ∇ · μM + ∇ε
∂t σε
ε
+ (C1ε G k − C2ε ρM ε) + Sε (5)
k

The motion of argon bubbles (DPM):

d
uB
= FD + FG + FB + FVM + FP + FL (6)
dt

Species Transport for Components in the Steel and Slag

The concentration field of components in the steel and slag was simulated by solving
the transport equation, as shown in Eqs. (7)–(8), where [Ca], [O], and [S] were
considered to only transport in the steel phase and (CaO) and (CaS) only transport in
the slag phase. Source terms in Eqs. (7)–(8) were indicated the composition change
due to the desulfurization.

∂ μeff
ρst c[i] + ∇ · ρst uc[i] = ∇ · ∇c[i] +Sc[i] (7)
∂t Sct

∂ μeff
ρst c(i) + ∇ · ρst uc(i) = ∇ · ∇c(i) +Sc(i) (8)
∂t Sct

Kinetic Model for the Desulfurization at the Steel-Slag

Interface

A kinetic model based on the coupled reaction model was established to calculate
the desulfurization at the steel-slag interface. In practice, Fluent called the kinetic
model using UDFs in cells with volume fraction of the steel phase larger than 0.1
and that of the slag phase larger than 0.1. The detail of the kinetic could be find in
Refs. [17–19]. The variation of the sulfur in the steel and components in the slag
were calculated using Eqs. (9)–(10). Source terms in Eqs. (7)–(8) were equal to the
value of Eqs. (9)–(10) multiplied by the density of the steel and slag, respectively,
which was carried out using UDFs.

d[%M] Ainterface · kM
=− [% M]b − [%M]∗ (9)
dt Vsteel
Simulation on the Slag Desulfurization During the LF Refining … 129

d(%MOn ) Ainterface · kMOn

=− (%MOn )b − (%MOn )∗ (10)
dt Vslag

where [%M] and (%MOn ) represent the content of components in the steel and slag,
respectively, wt%; A is the area of the reaction interface, m2 ; V is the volume of
the phase, m3 ; k is the mass transfer coefficient, m/s; the superscript “b” and “*”
represent the bulk and equilibrium of the phase.
The IMCT was used to calculate the activity of the slag phase, while the asso-
ciate model coupled with the UIPF model was employed to calculate the activity of
the steel phase. The initial composition of the slag was 61.64%CaO-19.39%Al2 O3 -
2.43%SiO2 -8.42%MgO-2.16%CaS-5.96%CaF2 , and that of the steel was 0.108%C-
0.013%Si-0.764%Mn-0.077%S-0.003%O-0.028%Al-0.0004%Ca-Fe.

Simulation Results

Three-Dimensional Flow Field

The three-dimensional flow field of the steel-slag-air was solved employing the Fluent
22.0 on a working station equipped with two Intel Xeon Gold 6226R CPUs and
96 GB RAM. Figure 2 shows the stable flow field in a 230 tons ladle with bottom-
blowing argon of 1200 NL/min. There were two circulating flows in the molten
steel. A relatively obvious circulating flow was formed on the side far from the argon
blowing point. Another was a less obvious circulating flow which was formed in the
middle region between the argon blowing point and the ladle wall.

Fig. 2 Distribution of the

velocity vector on the
longitudinal section passing
the argon blowing point of
the molten steel in the ladle
130 J. Wang et al.

0.08
Lines: calculated

Composition of the steel (wt%)

0.07
Points: measured
0.06

0.05

0.04

0.03

0.02

0.01

0
0 600 1200 1800 2400 3000 3600
Time (s)

Fig. 3 Evaluation of the average content of [S] and [O] in the molten steel

Variation of the Sulfur Content in the Steel

Figure 3 shows the calculated variation of the average content of the dissolved [S] and
[O] in the molten steel in several hundred seconds. The [S] content has a continued
decrease during the refining process. At the beginning of the refining, the [S] content
was about 0.077 wt%. Through the desulfurization reaction between the steel and
the slag, the [S] content reduced to about 0.01% in 5 min. At the same time, the
dissolved [O] had a little bit increase followed by a decrease. As the followed two
reactions were considered in the current model, a positive direction of Eq. (11) and
a negative direction of Eq. (12) could be figured out in the first 600 s.

[Ca] + [S] = (CaS) (11)

[Ca] + [O] = (CaO) (12)

Three-Dimensional Distribution of the Sulfur

Due to the desulfurization reaction occurred exclusively at the steel-slag interface,

the [S] content near the steel-slag interface must be the lowest, as shown in Fig. 4.
What’s more, the steel and the slag kept flowing during the refining, so the [S]
transferred from the interior to the interface. Therefore, the distribution of the [S]
content showed an uneven state. In the zone in the middle between the argon blowing
Simulation on the Slag Desulfurization During the LF Refining … 131

point and the ladle wall, the [S] content was the highest. This was owing to the low
speed flow in this zone. The distribution tendency of the [S] was the same at different
refining periods. The difference lied on that the overall [S] content kept decreasing
when the refining was going on.

Conclusion and Summary

In the current study, a numerical model coupled with a kinetic model was proposed to
evaluate the desulfurization of the slag during the refining process. The VOF model,
k-ε model, and DPM model were employed to solve the three-dimensional flow
of the steel-slag-air system. The thermodynamic equilibrium and the mass transfer
at the steel-slag interface was coupled to solve the kinetic process. The following
conclusion was obtained.
1. Two circulating flows in the molten steel were formed under the influence of
argon blowing. A relatively obvious circulating flow was formed on the side far
from the argon blowing point. Another was a less obvious circulating flow which
was formed in the middle between the argon blowing point and the ladle wall.
2. On the condition of high initial [S] content, the slag with high content of CaO
had an apparent desulfurization ability. The [S] content reduced from 0.077 wt%
into 0.01 wt% in fifty minutes.
3. The distribution of the [S] content in the molten steel showed an uneven state
with the lowest near the steel-slag interface and the highest in the middle zone
between the argon blowing point and the ladle wall. The distribution tendency
of the [S] was the same at different refining periods. The difference lied on that
the overall [S] content kept decreasing when the refining was going on.
132 J. Wang et al.

Fig. 4 Three-dimensional
distribution of the sulfur in
the molten steel at different
times

(a) 50 s

(b) 100 s

(c) 300 s

(d) 600 s
Simulation on the Slag Desulfurization During the LF Refining … 133

Acknowledgements The authors are grateful for support from the National Natural Science Foun-
dation of China (Grant No. U1860206, No. 51725402, 51874031, 51874032), the S&T Program
of Hebei (Grant No.20311004D), the High Steel Center (HSC) at North China University of Tech-
nology and the High Quality Steel Consortium (HQSC) at University of Science and Technology
Beijing, China.

References

1. Maciejewski J (2015) The effects of sulfide inclusions on mechanical properties and failures
of steel components. J Fail Anal Prev 15(2):169–178
2. Weider C, Pichard C (1991) Effects of sulphur and inclusion morphology in friction welding
of structural steels. Weld Int 5(5):369–377
3. Shi C, Huang Y, Zhang J, Li J, Zheng X (2021) Review on desulfurization in electroslag
remelting. Int J Miner Metall Mater 28(1):18–29
4. Visuri VV, Vuolio T, Haas T, Fabritius T (2020) A review of modeling hot metal desulfurization.
Steel Res Int 91(4):1900454
5. Kang JG, Shin JH, Chung Y, Park JH (2017) Effect of slag chemistry on the desulfurization
kinetics in secondary refining processes. Metall Mater Trans B 48(4):2123–2135
6. Takahashi K, Utagawa K, Shibata H, Kitamura S, Kikuchi N, Kishimoto Y (2012) Influence
of solid CaO and liquid slag on hot metal desulfurization. ISIJ Int 52(1):10–17
7. Yang AF, Karasev A, Jönsson PG (2015) Characterization of metal droplets in slag after
desulfurization of hot metal. ISIJ Int 55(3):570–577
8. Cao Q, Pitts A, Nastac L (2018) Numerical modelling of fluid flow and desulphurisation kinetics
in an argon-stirred ladle furnace. Ironmaking Steelmaking 45(3):280–287
9. Fan W, Zhang J, Ao W, Huang J (2015) Effect of Al2 O3 and CaF2 on melting temperature of
high calcium ladle desulfurization slag. Adv Mater Res 1094:325–328
10. Wang F, Tan J, Huang X, Wang Q, Liu Z, Baleta J, Li B (2022) Mathematical and numer-
ical predictions of desulfurization behavior in the electromagnetically controlled vibrating-
electrode electroslag remelting furnace. Metall Mater Trans B 53(3):1792–1805
11. Uriostegui Hernandez A, Garnica Gonzalez P, Ramos Banderas JA, Solorio Diaz G, Hernandez
Bocanegra CA (2022) Desulphurization kinetic prediction into a steel ladle by coupling
thermodynamic correlations, fluidynamics and heat transfer. ISIJ Int 62(6):1189–1198
12. Jonsson L, Sichen D, Jönsson P (1998) A new approach to model sulphur refining in a gas-stirred
Ladle—a coupled CFD and thermodynamic model. ISIJ Int 38(3):260–267
13. Andersson AMT, Jonsson LTI, Josson PG (2003) A model of reoxidation from the top slag and
the effect on sulphur refining during vacuum degassing. Scand J Metall 32(3):123–136
14. Lou W, Wang X, Liu Z, Luo S, Zhu M (2018) Numerical simulation of desulfurization behavior
in ladle with bottom powder injection. ISIJ Int 58(11):2042–2051
15. Lou W, Zhu M (2015) Numerical simulation of slag-metal reactions and desulfuriza-
tion efficiency in gas-stirred ladles with different thermodynamics and kinetics. ISIJ Int
55(5):961–969
16. Lou W, Zhu M (2014) numerical simulation of desulfurization behavior in gas-stirred systems
based on computation fluid dynamics—simultaneous reaction model (CFDSRM) coupled
model. Metall Mater Trans B 45(5):1706–1722
17. Zhang Y, Ren Y, Zhang L (2017) Modeling transient evolution of inclusion in Si-Mn-killed
steels during the ladle mixing process. Metall Res Technol 114(3):308
18. Zhang Y, Ren Y, Zhang L (2018) Kinetic study on compositional variations of inclusions, steel
and slag during refining process. Metall Res Technol 115(4):1–7
19. Wang J, Zhang L, Cheng G, Ren Q, Ren Y (2021) Dynamic mass variation and multiphase inter-
action among steel, slag, lining refractory and nonmetallic inclusions: laboratory experiments
and mathematical prediction. Int J Miner Metall Mater 28(8):1298–1308
134 J. Wang et al.

20. Luo Y, Liu C, Ren Y, Zhang L (2018) Modeling on the fluid flow and mixing phenomena in a
RH steel degasser with oval down-leg snorkel. Steel Res Int 89(12):1800048
21. Liu C, Li S, Zhang L (2018) Simulation of gas-liquid two-phase flow and mixing phenomena
during RH refining process. Acta Metall Sin 54(2):347–356
22. Duan H (2016) Simulation on multiphase flow and mixing phenomena during ladle metal-
lurgy process. In: The eighth China-Korea joint symposium on advanced steel technology,
Chongqing, China, pp 55–62
23. Duan H, Ren Y, Zhang L (2019) Fluid flow, thermal stratification, and inclusion motion during
holding period in steel ladles. Metall Mater Trans B 50(3):1476–1489
24. Duan H, Zhang L, Thomas B, Conejo A (2018) Fluid flow, dissolution, and mixing phenomena
in argon-stirred steel ladles. Metall Mater Trans B 49(5):2722–2743
A Modified Thermodynamic Software
to Control the Composition of Inclusions
During Calcium Treatment Process

Weijian Wang and Lifeng Zhang

Abstract Based on the minimum Gibbs free energy method, a modified thermo-
dynamic model was established to predict the composition of inclusions during the
calcium treatment process. Three types of phase were contained in the steel-inclusion
system, including the molten steel phase, the liquid inclusions phase, and the solid
inclusions phase. The Wanger model was applied to calculated activities of elements
in the molten steel. Reported interaction parameters were evaluated and used to
improve the accuracy of thermodynamic calculation results. The coexistence theory
was applied to calculate activities of Al2 O3 and CaO in liquid inclusions. According
to the thermodynamic calculations, the effect of [Link] content in the molten steel on
the transformation of inclusions was discussed. Combining the established thermo-
dynamic model and a calcium yield prediction model, a software for precise control
of calcium treatment process was also developed.

Keywords Calcium treatment · Software · Calcium yield · Thermodynamic model

Introduction

Aluminum is widely used in steelmaking process due to its strong deoxidation ability.
However, a large amount of Al2 O3 inclusions were generated after the addition of
aluminum in the molten steel, leading to nozzle clogging, and some product defects
[1–6]. Calcium treatment technology was applied to modify solid Al2 O3 inclusions
into liquid calcium aluminates [7–10]. The optimal calcium addition to the molten
steel was relative to the composition and temperature of the steel. Solid inclusions
were hardly modified to liquid state if the calcium content was too low, while solid
CaO and CaS were generated with the excessive calcium addition, which can also
cause nozzle clogging easily [9, 11–13]. Moreover, it was difficult to control the
yield of calcium stably due to its low solubility in the molten steel and low boiling

W. Wang · L. Zhang (B)

School of Mechanical and Materials Engineering, North China University of Technology,
Beijing 100144, China
e-mail: zhanglifeng@[Link]

© The Minerals, Metals & Materials Society 2023 135

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
136 W. Wang and L. Zhang

temperature. Therefore, it was an important issue to realize the precise control of

calcium treatment.
In the current study, a thermodynamic model was established to predict the compo-
sition of inclusions based on the minimum Gibbs free energy method during calcium
treatment process. Next, a calcium yield prediction model was established through the
neural network technology. Finally, a software was developed to achieve the online
control of precise calcium treatment with the objective of precisely controlling of the
inclusions composition. Combined with the calcium yield model, the precise length
of the calcium wire required in the liquid steel could be reported online.

Prediction of Calcium Yield

The calcium yield, in other words, the efficiency of calcium recovery was calcu-
lated through Eq. (1). It was difficult to control the calcium yield precisely during
calcium treatment process. The factors having influence on calcium yield during the
calcium treatment process were numerous and complex including temperature, steel
composition and steel grade, and so on. The solubility of calcium in the molten steel
was affected by the temperature. The time for the calcium wire melting completely
in the molten steel was shorten with the increase of temperature so that the posi-
tion where the calcium wire completely melted became shallower at the same wire
feeding speed, which had an adverse effect on the improvement of calcium yield.
After entering the molten steel, calcium reacted with the dissolved oxygen to generate
CaO, meaning that the high dissolved oxygen content in the molten steel was bene-
ficial to the increase of calcium yield. The effect of the temperature and dissolved
oxygen content in the molten steel on calcium yield is shown in Figs. 1 and 2.

% CaFinal − % CaInitial
Calcium yield = (1)
% CaAddition

40
Fig. 1 Relationship between
the temperature and calcium
yield [15]
30
Calcium yield (%)

20

10

0
1530~1540 1540~1550 1550~1560 1560~1570 >1570
Temperature (oC)
A Modified Thermodynamic Software to Control the Composition … 137

40
Fig. 2 Relationship between
the dissolved oxygen content
and calcium yield [16]
30

Ca yield (%)
20

10

0
<2 2-3 3-4 4-5 5-6 >6
Dissolved oxygen at the start of RH (ppm)

where %CaFinal means the [Link] content after calcium addition, %; %CaInitial means
the initial [Link] content before calcium addition, %; %CaAddition means the amount
of calcium addition during calcium treatment process, %.
A calcium yield prediction model [14] that had been introduced elsewhere was
established based on the neural network technology to predict and improve the
calcium yield during steel refining process so that the required feeding length of
the calcium wire could be calculated precisely and timely once the steel composi-
tion and temperature were determined. Three steps were contained for the calcium
yield prediction by neural network technology. Firstly, the initial data was collected,
analyzed, and preprocessed and the appropriate parameters were selected to construct
the neural network model. Then, the constructed neural network was trained until the
accurate calculation results were obtained. Finally, the trained neural network model
could be applied to predict the calcium yield during the calcium treatment process.
The comparison between predicted calcium yields using neural network model and
the detected yields in industrial trials shows a good agreement with each other, as
shown in Fig. 3. The composition, temperature of the molten steel, the amount of
slag addition, and some operation parameters that had a great effect on the calcium
yield were included in the dataset. There are totally 511 sets of data collected from a
steel plant, while 461 sets of data were selected as teaching data to train the model.
The input layer, middle layer, and output layer were included in the neural network
model. The unit number of input layer was 22, and the unit number of output layer
was 1. The maximum training epoch of the training process was set as 1500, and the
learning rate was 0.2. The unit number in each middle layer was determined by an
empirical formula which was set as 6 in the current model.
138 W. Wang and L. Zhang

50
Fig. 3 Comparison between Fitted

20
Predicted Ca yield by DNN (%)
predicted calcium yields

%
Data

10
%
using neural network model 40
and the detected yields in
industrial trials [14] 30

20

10

0
0 10 20 30 40 50
Measured Ca yield (%)

Development of the Software for the Online Control

of Precise Calcium Treatment

For thermodynamic equilibrium calculation of multicomponent and multiphase

systems, two methods were contained, one was the minimizing Gibbs free energy
method and the other was the classical thermodynamic calculations based on chem-
ical reaction equilibrium between different elements in the molten steel. In the current
thermodynamic model, the minimizing Gibbs free energy method was adopted to
calculate the thermodynamic equilibrium in the steel-inclusions system. Three types
of phase were contained in the steel-inclusion system, including the molten steel
phase, the liquid inclusions phase, and the solid inclusions phase. The Wanger model
was applied to calculated activities of elements in the molten steel [17, 18]. Reported
interaction parameters were evaluated and used to improve the accuracy of thermody-
namic calculation results. The coexistence theory was applied to calculate activities
of Al2 O3 and CaO in liquid inclusions. The calculated effects of [Link] content on the
composition of inclusions in the molten steel is shown in Fig. 4. Inclusions could be
controlled in liquid state when the [Link] content is 12–18 ppm.
Combined the established thermodynamic model and calcium yield prediction
model, the software for the online control of precise calcium treatment was developed.

0.008
Fig. 4 Effect of [Link]
Composition of inclusions (wt.%)

T.S=20 ppm
T.O=20 ppm
content on the composition [Link]=400 ppm
Al2O3 Liquid inclusions T=1600 oC
of inclusions 0.006
CaO·6Al2O3
CaO·2Al2O3

0.004 CaS
CaO(l)

0.002 Al2O3(l)

CaO
0.000
0.000 0.001 0.002 0.003 0.004 0.005
[Link] content (wt.%)
A Modified Thermodynamic Software to Control the Composition … 139

Fig. 5 Operation interface of the software

The operation interface of which is shown in Fig. 5. After connecting the industrial
database, the information on the composition and temperature of the molten can
be acquired so that the appropriate [Link] content could be calculated through the
established thermodynamic model. Then, the calcium yield could be predicted based
on the reported deep neural network model according to the information of steel
composition, temperature, and other operation parameters during calcium treatment
process. Then, the optimal additional length of the calcium wire can be calculated
based on the prediction of calcium yield on the basis of Eq. (2).

W × (n[Ca]T − n[Ca]O )
L= (2)
η×β ×μ

where L is the appropriate addition length of calcium wire, m; W is the mass of

molten steel, t; n[Ca]T is the target calcium content in the molten steel calculated
by the established model, ppm; n[Ca]O is the initial calcium content before calcium
treatment process, ppm; η is the calcium yield calculated by the prediction model,
%; β is the calcium content of calcium wire, %; μ is the mass of calcium wire per
meter, g/m.

Conclusions

A calcium yield prediction model was established based on the neural network tech-
nology. After the composition, temperature of the molten steel and other opera-
tion parameters was read, the calcium yield can be predicted to guide the industrial
calcium treatment process. Based on the minimizing Gibbs free energy method, a
thermodynamic model was established to predict the composition of inclusions after
calcium addition during the refining process. Combing the calcium yield prediction
140 W. Wang and L. Zhang

model, a software for the online control of precise calcium treatment was developed,
in which the optimal calcium wire feeding length can be reported so that the precisely
and timely guidance of calcium treatment operation can be realized.

Acknowledgements The authors are grateful for support from the National Science Foundation
China (Grant No. 52004025 and No. 51704018, No. 51725402), Fundamental Research Funds for
the Central Universities (Grant [Link]-TP-17-001C2 and FRF-TP-19-037A2Z), the High Steel
Center (HSC) at North China University of Technology, Yanshan University and University of
Science and Technology Beijing, China.

References

1. Choudhary SK, Ghosh A (2008) Thermodynamic evaluation of formation of oxide—sulfide

duplex inclusions in steel. ISIJ Int 48(11):1552–1559
2. Zhang L, Thomas BG, Associate R (2002) Alumina inclusion behavior during steel deoxidation.
In: 7th European electric steelmaking conference, Venice, Italy, pp 77–86
3. Zhang L, Thomas BG (2006) State of the art in the control of inclusions during steel ingot
casting. Metall and Mater Trans B 37(5):733–761
4. Wang QY, Bathias C, Kawagoishi N, Chen Q (2002) Effect of inclusion on subsurface crack
initiation and gigacycle fatigue strength. Int J Fatigue 24(12):1269–1274
5. Larsson M, Melander A, Nordgren A (1993) Effect of inclusions on fatigue behaviour of
hardened spring steel. Mater Sci Technol 9(3):235–245
6. Lamut J, Falkus J, Jurjevec B, Knap M (2012) Influence of inclusions modification on nozzle
clogging. Arch Metall Mater 57(1):319–324
7. Ren Q, Yang W, Cheng L, Hu Z, Zhang L (2020) Effect of calcium treatment on magnetic
properties of non-oriented electrical steels. J Magnetism and Magnetic Mater 494
8. Li M, Liu Y, Zhang L (2019) Effect of reoxidation on inclusions in steel during calcium
treatment. Metallurgical Res Technol 116(2):206–214
9. Zhang L, Liu Y, Zhang Y, Yang W, Chen W (2018) Transient evolution of nonmetallic inclusions
during calcium treatment of molten steel. Metall and Mater Trans B 49(4):1841–1859
10. Liu Y, Zhang L (2018) Relationship between dissolved calcium and total calcium in Al-killed
steels after calcium treatment. Metall and Mater Trans B 49(4):1624–1631
11. Yang W, Zhang L, Wang X, Ren Y, Liu X, Shan Q (2013) Characteristics of inclusions in low
carbon al-killed steel during ladle furnace refining and calcium treatment. ISIJ Int 53(8):1401–
1410
12. Zhang L (2019) Non-metallic inclusions in steels: fundamentals (in Chinese). Metallurgical
Industry Press, Beijing
13. Zhang L (2019) Non-metallic inclusions in steels: industrial practice (in Chinese). Metallurgical
Industry Press, Beijing
14. Wang W, Zhang L, Ren Y, Luo Y, Sun X, Yang W (2021) Prediction of calcium yield during
calcium treatment process performed in steelmaking using neural network. Metall and Mater
Trans B 53(1):1–7
15. Wang W, Luo Y, Sun X, Zhang L (2019) Study on effect of calcium treatment on inclu-
sions evolution mechanism and Ca yield. In: Ladle refining 2019 annual meeting proceedings.
Suzhou, China, pp 416−420
16. Sun X, Wang W, Luo Y, Zhang J, Ren Y, Zhang L (2019) Evolution of inclusions in high-sulfur
and low-oxygen steels after calcium treatment. In: AISTech 2019-proceedings of the iron and
steel technology conference Pittsburgh, Pa., USA, pp 1233–1247
A Modified Thermodynamic Software to Control the Composition … 141

17. Suito H, Inoue R (1996) Thermodynamics on control of inclusions composition in ultra-clean

steels. ISIJ Int 36(3):528–536
18. Wan Y, Chen W, Zhang S (2014) Thermodynamic analysis and experimental research of
50W600 non-oriented silicon steel after calcium treatment. Shanghai Metals 36(1):37–41
Part IV
New Processes and Insights
Reductant Formation Enthalpy in DC
Ferrochrome Smelting: Merely Academic
or Fundamental to Operation?

H. J. Oterdoom, M. A. Reuter, and J. H. Zietsman

Abstract The start-up and operation of DC furnaces for ferrochrome and titania slag
but also ferronickel and ferrocobalt have proven again and again to be more chal-
lenging than anticipated. A combination of few furnaces in operation, a limitation on
shared operational experience, quick response times between operational decisions
and consequences, and a limited fundamental understanding of process mechanisms
is a part of the causes for the results. Understanding the process chemistry, energy
balance, and interactions between them is fundamental to investigating the mech-
anisms taking place in a DC furnace. As part of a PhD investigation of process
mechanisms and behaviour of DC ferrochrome furnaces, this paper will describe the
potential operational consequences of ignoring something as basic as the formation
enthalpy of reductants. Only one of many pitfalls, reductant is used as an example
to demonstrate that fundamental understanding of process chemistry is required for
designing, operating, and studying DC furnaces. Only with at least an understanding
of process chemistry it can be possible to connect operational experiences to process
mechanisms. The paper will conclude with a brief discussion of the presented results,
and how these relate to the PhD study.

Keywords DC furnace · Operation · Feed-to-power · Ferrochrome · Dynamic

simulation · Digital twin · Reductant

H. J. Oterdoom (B)
University of Pretoria, Dusseldorf, Germany
e-mail: u18195530@[Link]
M. A. Reuter
SMS Group, Dusseldorf, Germany
e-mail: [Link]@[Link]
J. H. Zietsman
University of Pretoria, Pretoria, South Africa
e-mail: [Link]@[Link]
Ex Mente Technologies, Pretoria, South Africa

© The Minerals, Metals & Materials Society 2023 145

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
146 H. J. Oterdoom et al.

Introduction

In multiple projects, the start-up and operation of DC furnaces has been problem-
atic. There is no clear singular cause for the experienced delays in reaching design
capacity. The main issues -in hindsight- appear again and again to be unrealistic
expectations, underestimation of challenges, and a lack of preparation. Merrow and
McNulty are only two of many that have written about project failure [1, 2]. There are
at least the following two topics that reoccur in analyses of project disappointments:
• Equipment design: the equipment initially installed was not fit for the task.
• Raw materials to be processed: something unexpected happened in the processing
or handling.
Both are closely related. Only when the process is fully understood, then the
optimal design can be developed, but only when the equipment has been installed,
then the process can be observed and optimized. Especially new processes in larger-
than-known equipment have proven to be a risky business.
On a global level, the climate change is pushing the metallurgical industry into
looking at ways to reduce greenhouse gas emissions. Power to these furnaces can be
green, but reductants will be needed, also in a green future. Replacing reductants from
fossil origin by alternative reductants can contribute to greener metal production. A
better fundamental understanding of the behaviour of reductants in production and
their role in the energy balance is important to support a transition towards alternative
reductants.
Two papers had a significant impact on this paper: one by Barcza et al. [3] and
the other by Theron [4]. These papers and personal industrial experiences, as well
as an interest in digital twinning, environment, risk reduction, process fundamentals
and mechanisms, and the availability of increased computer capabilities led to this
investigation.

Physics and Furnaces

The bigger furnaces get, the more physics affect the operation. This is not only true for
the mechanical aspects like refractory expansion, but also in heat and mass transfer
mechanisms. Thermodynamics remain the same, but distances and volumes change,
and that means time and temperature (differences) become increasingly important,
as these affect kinetics and therefore the operation.
Small-scale testwork may therefore be more misleading than thought, especially
when such work is done to tick off boxes for management instead of having actual
targets relevant to the large-scale operation. Risks can remain high if an interme-
diate step between small-scale and large-scale industrial furnace is ignored, and an
upscaling factor of 10 should be seen as a maximum [5, 6].
Reductant Formation Enthalpy in DC Ferrochrome Smelting: Merely … 147

Because physics becomes more important, it is up to the metallurgists to better

connect thermochemistry and physics and investigate what happens over time and
space. Single black-box calculations will probably not be capable to describe those
phenomena that are relevant for an optimal mechanical design.
Feed materials will not all follow the same pathway through the reactor until the
products leave the furnace via gas duct or tapholes. It is therefore that a dynamic
multizone model is being developed to be able to investigate heat and mass transfer
as well as process pathways and kinetics in a DC furnace. As a test for a basic
version of this model, it was decided to investigate the possible effect of (ignoring)
the reductant formation enthalpy (RFE) in the production of ferrochrome (FeCr) in
an open bath DC furnace.

DC Furnace and Ferrochrome Production

An open bath furnace, like a DC furnace for FeCr, relies heavily on a strictly main-
tained balance between power input, feedrate, feed composition, reactions, and heat
losses. A change in any of these factors will require other factors to be adjusted.
Reactions can be considered constant if both feed composition and the furnace power
input are constant. Changes in heat losses are relatively slow for most furnaces and
considered constant in this investigation.
Ferrochrome is made by reducing chromite ore with a reductant. The required
amount of reductant is calculated to achieve a certain recovery of chrome, and fluxes
are added to achieve a targeted slag composition. The mass and energy balances are
calculated, and then a furnace is designed including the compensation for heat losses.
The premixed feed is charged continuously into the open bath and smelted to
form alloy, slag, and process gas. The balance between power input and feedrate,
known as feed-to-power ratio (FtP), is in modern furnaces carefully controlled by
using an automated loss-in-weight system. Alloy and slag are tapped intermittently.
An incorrect FtP will lead to overheating of the bath, unwanted side reactions, or
freezing of the bath. The FtP usually has to be adjusted if there is a change in
composition, temperature or behaviour of raw materials. This can be caused by a
change in, for example,
• used raw material,
• preheating,
• prereduction,
• particle size distribution,
• weather,
• supplier, or
• source of raw material.
148 H. J. Oterdoom et al.

Feed Materials for Ferrochrome

Ore and reductant are the main raw materials of interest for mass and energy balances.
In published calculations, and probably in many industrial situations too, compounds
are used instead of minerals for ores, fluxes, and reductants. For example, in
ferrochrome calculations, the ore is commonly described by a combination of Cr2 O3 ,
MgO, SiO2 , Al2 O3 , CaO and, when lucky, both FeO and Fe2 O3 as main components
[7–9]. Mineralogical compositions like FeCr2 O4 , MgAl2 O4, or MgSiO3 are hardly
used, let alone an accurate composition like Mg0.56 Ti0.01 Cr1.33 Fe0.51 Al0.59 O4 [10].
As demonstrated by for example Sweeten [9], mineralogy can affect the mass and
energy balances of a smelting operation because of the role of the formation enthalpy.
Barcza [3] showed that a 0.5% deviation between feedrate and power input, already
creates problems:
The control over the feed rate and power level to within about 0.5 percent accuracy appears to
be necessary, particularly at power levels greater than 20 MW, so that the unwanted increase
in energy consumption caused by increased bath temperature and the favouring of undesir-
able side reactions can be avoided. Having a model that can investigate effect of formation
enthalpy on actual DC ferrochrome operation therefore seemed a useful investigation.

An accuracy of 0.5% equals 200 kW on a 40 MW furnace. With such a furnace

having heat losses in the range of 4−8 MW, the 200 kW is 2.5−5.0% on the
heat losses, and only 0.56−0.63% on the metallurgical energy consumption of
32−36 MW. Both percentage ranges are at best within the range where heat loss
measurements or thermochemical calculations are accurate.

Reductant

Reductants are important in metal production. Nowadays even more so, because
fossil-fuel-based reductants contribute to climate change, and their use should be
minimised. A better understanding of the behaviour of reductants in smelting oper-
ations can support the change to, for example, alternative reductants like biochar or
end-of-life plastics.
For reductants often the proximate analysis is used:
• moisture,
• C-fix,
• volatile matter, and
• ash.
A detailed ash analysis is usually known and applied in the mass and energy
balances. The exact composition of the volatile matter, typically given by VM or
volatiles, is sometimes either assumed, or recalculated based on the ultimate analyses
(C, H, N, O, S) in combination with assumptions [4].
Reductant Formation Enthalpy in DC Ferrochrome Smelting: Merely … 149

One factor not often used in calculations with reductants is the gross calorific
value (GCV). This can give information on the RFE of complex substances like coal,
anthracite, or coke.
This paper will focus on the effect of ignoring the RFE on DC FeCr operation.

Challenges with Using Reductants in Calculations

A reductant is a highly complex “mineral” of mainly C-H–O connections with other

elements connected (for example S, N, Na), and locked in materials, like pyrite or
clay.
It is important to understand that volatile matter in the cold reductant is not present
in the gaseous state, but only released as gas from the solid during heating. The
relative quantities of fixed carbon and gases and the composition of the volatiles
depend strongly on variables like particle size distribution and process conditions.
Especially, heating rate and final temperature are important [4, 11–13].
Reductants therefore are a tricky substance to do calculations with. Some issues
are:
• When and where how many volatile products are released remains uncertain.
• There is a wide range of non-organic and organic compounds possible (clay, pyrite,
sulfates, alkalis loose or as organic compound, etc.) that can affect the behaviour
of the reductant, catalyse reactions, or obstruct them.
• Getting reliable values for proximate and ultimate analyses, and determination of
the GCV are a problem because accurate sampling and analysing of reductants is
extremely hard, and even with utmost care results may not be consistent [14].
To make things harder, one should be extremely careful to make sure if the values
are based on analyses as received (AR), air dried (AD), dry basis (DB), or dry ash free
(DAF). Even with support literature [15, 16] errors are easily made. For example,
there is a difference between the GCV values when given based on analyses “as
received” or “dry ash free”.

Reductant Enthalpy of Formation and GCV

Reductants have an enthalpy of formation, just like chromite ore, bauxite, or methane
(CH4 ), but the RFE is harder to determine. Reductants with identical chemical compo-
sitions can be structured completely different on an atomic level, and thus have
very different enthalpies of formation and GCV values. Theron showed the effect
on formation enthalpy by using different compositions for the volatiles based on
the same ultimate analyses [4]. Reductants as a bulk commodity do have a (quite)
constant GCV.
150 H. J. Oterdoom et al.

Table 1 Example based on methane to show the effect of enthalpy of formation on an energy
balance, calculated with HSC
1 kg CH4 1 kg (C + 2H2 )
Comment All gas Solid graphite + gas
Enthalpy at 25° before combustion kWh/kg −1.3 0
Enthalpy at 25° after combustion kWh/kg −16.7 −16.7
Theoretical GCV as calculated kWh/kg 15.4 16.7
Measured GCV of CH4 [26] kWh/kg 15.4 15,4
Difference theoretical to measured GCV kWh/kg 0,0 −1,3
Stoichiometric combustion temperature °C 1677 1804

To explain the effect of the formation enthalpy, the GCV determination of methane
is given in Table 1. Pure methane has no issues with solids, clay minerals, or sulphur.
Methane has a GCV of 15.4 kWh/kg, measured by calorimeter [26]. The calculated
GCV can be based on combustion of CH4 or C+2H2 . The difference is the enthalpy
of formation: the energy that was required to form the CH4 from the elements. That
cost causes a lower final combustion temperature because less energy is available
from CH4 in reality than from C+2H2 in theory.
A similar approach is applicable to reductants like anthracite or coke. These
reductants have measured GCV values, but only the elemental composition can be
used in thermodynamic calculations. It is possible to calculate a theoretical GCV
based on the ultimate analysis and consider the difference to the actual GCV as a
correction factor [4] or RFE [27]. Note that enthalpy is given as a negative value,
whereas GCV is a positive value.

RFE = GCV − theoretical GCV

If RFE > 0, then more chemical energy is available than calculated based on the
ultimate analysis. In a DC furnace this means that more feed should be smelted per
hour than calculated or else the bath heats up.
If RFE < 0, then less chemical energy is available than calculated based on the
ultimate analysis. In a DC furnace this means that less feed should be smelted per
hour or else the bath will freeze in.
Kleynhans [18] provided an excellent dataset that is used in this paper. Table
2 shows a selection from that dataset with the theoretical GCV and RFE added.
Anthracites and cokes are given by An and Co, respectively.
Note that these anthracites have a positive deviation, while all cokes have a nega-
tive one. Secondly, most anthracites are in the same range, meaning that changing
from one anthracite to another mostly has less impact than changing from anthracite
to coke or vice versa, but not in all cases.
Reductant Formation Enthalpy in DC Ferrochrome Smelting: Merely … 151

Table 2 Selection of data from Kleynhans [18] and theoretical GCV values based on Zietsman
[27]. The ultimate analysis is “air dried”
An1 An2 An3 An4 An5 An10 Co1 Co2 Co5
Inh. mois % 3.8 1.9 2.9 2.1 1.4 3.2 1.2 0.2 0.4
Ash % 15.7 20.3 11.8 12.1 15.5 16.3 20.4 7.7 12.4
VM-volatiles % 5.0 9.7 4.9 4.5 6.5 5.0 1.6 0.6 0.9
Fixed C % 75.5 68.1 80.4 81.3 67.6 75.5 76.8 91.5 86.3
C % 74.1 70.6 78.2 79.8 76.0 73.6 75.5 89.1 84.7
H % 1.62 2.60 2.12 2.01 2.49 1.69 0.20 0.03 0.01
N % 1.63 1.69 1.77 1.44 1.55 1.64 0.87 1.33 1.19
O % 2.01 2.14 2.05 1.80 2.15 2.12 1.04 0.80 0.48
S % 1.14 0.74 1.07 0.70 0.84 1.42 0.78 0.80 0.78
GCV kWh/t 7558 7567 8192 8197 8153 7606 6481 8125 7508
Th. GCV1 kWh/t 7411 7468 7979 8,072 7919 7400 6970 8141 7733
RFE kWh/t 147 98 212 125 234 205 −90 −16 −224
1 Theoretical GCV determined by calculation in HSC of combustion of the ultimate analysis of the
air dried reductant at 25 °C with H2 O as liquid product

On GCV Formulas

Several GCV formulas exist to approximate the real GCV as determined by official
guidelines. Reviews are included in [14, 17]. Care must be taken to use a consistent
calculation method with the correct analyses as explained. Be consistent with the
units kWh, MJ, btu per kg, t, lb.
Considering the assumptions in theoretical GCV calculations, it may be more
appropriate to talk about an enthalpy correction based on a certain method, than
claiming to have “the” method to determine the RFE.
Applying the method for calculating the RFE as described by Zietsman on data
from Kleynhans gave results ranging from around +250 to −1000 kWh/t of reductant.
Using Theron’s slightly different calculation for a correction value gave values in
the same order of magnitude.

Calculation in HSC

A dynamic calculation was set up using the software HSC [19]. This software allows
for dynamic changes in operation, parameters, and compositions. The effect of such
a change can then be visualised.
The HSC dynamic simulation software allows, i.e.,
• Inclusion of multiple reaction zones within a reactor.
152 H. J. Oterdoom et al.

• Changes in operating conditions while the simulation is running.

• Varying temperatures of output streams.
• Simulate events like tapping or process disruptions.
• Monitoring the involved energy flows between the different zones.
The aim of this investigation is limited to demonstrating how ignoring the RFE
affects the bath temperature and chemical reactions.
If these effects lead to operational problems and how fast what will happen depends
partially on the process itself. In open bath smelting, this can be very different for
FeCr, titania slag, or smelting of direct reduced iron. Secondly, the furnace design,
for example bath volumes, plays a role: large slag volumes will respond slower than
small slag volumes.

Simplifications−Conditions

When changing from one reductant to another, the main calculation parameter is the
carbon to ore ratio. This is because most carbon is used for the reduction of Cr2 O3
and formation of chromium carbides. Assuming the ore does not change, then the
same amount of C-fix will be required to achieve the same reduction of the ore. If
the reductant has a different fixed carbon content, then the quantity of reductant per
ton of ore is adjusted accordingly.
Slag, alloy, and off-gas should remain the same. However, the difference in mois-
ture, ash quantity, and ash composition may affect the required quantity of fluxes.
A change in the quantity of fluxes may then affect the required C-fix if the quantity
of especially iron oxides changes. Moisture and reactions between carbon and water
will affect the required reductant quantity too. An interpolation loop is thus formed
where a new balance must be found. The overall effect may be a different energy
consumption, requiring a change in FtP setpoint, even without considering the RFE.
Such loops, and they appear when changing for example ore or reductant compo-
sition, moisture content, furnace load, and type or source of fluxes, are the main
challenge for metallurgists in the highly sensible open bath operations.
Because too many variables would cloud the purpose of this paper, i.e., to
show how the use of RFE and GCV by themselves alone are significant, some
simplifications have been made.
• The ore was taken as compounds and included a selection of only the most
important ones.
• The same ore was used in all calculations.
• S from the reductant is neglected in the mass balance.
• No dust losses are in the simulation because these are not relevant to show the
effect of the RFE.
• Formed process gas leaves the furnace instantly.
• Temperatures for all exit streams are fixed at 1,800 °C in the calculations. This is
to prevent confusion by having internal energy streams.
Reductant Formation Enthalpy in DC Ferrochrome Smelting: Merely … 153

• In this model, raw materials enter the calculations at the same point of energy
input: the electrode tip

Input Data

Table 3 shows the analyses for the raw materials that were used in the calculations.
Data were taken from the literature and simplified to not unnecessarily complicate the
calculations. An anthracite and a coke were deliberately selected to create a bigger
difference.

Table 3 Raw materials used in the calculations

Ore1 Bauxite Quartz Co12 An12
Feedrate tph 20 As required by the calculations
Moisture % 2.0 2.0 2.0 1.2 3.8
GCV kWh/t 6,480 7,560
Theoretical GCV kWh/t 6,9703 7,4113
RFE kWh/t −4904 + 1474
Solid part (Dry
Basis)
Cr2 O3 % 51.00
FeO % 12.50
Fe2 O3 % 0.00 2.82 4.08
SiO2 % 7.00 10.50 100 13.30 7.86
MgO % 20.50 0.00 0.45 0.64
Al2 O3 % 8.50 0.00 4.03 4.40
CaO % 0.50 0.00 0.39 0.24
C % 0.00 79.01 82.79
Al(OH)3 % 76.00
Fe(OH)3 % 12.00
TiO2 % 1.50
Sum % 100 100 100 100 100
1 Based on [8, 20, 21] and simplified
2 From Kleynhans unless specified differently; ash adjusted to have less components, normalised,
and dry basis; moisture is calculated in separately
3 Based on [27] as described in Table 2
4 RFE = GCV—theoretical GCV
154 H. J. Oterdoom et al.

Results

Step 1 Model Test Base Case Ignoring RFE for Both

Reductants

The dynamic calculation was tested by first doing a “batch” calculation of one hour
to verify if the results compare to industrial data.
Two calculations were done with different reductants. From Kleynhans, an
anthracite and a coke, An1 and Co1, were selected. The target slag composition
was based on literature [22]. The calculated required energy, quantities of reductant
and fluxes, and some other results are given in Table 4.
Some values in Table 4 are not the same as industrial values. The calculated alloy
is not high carbon ferrochrome as produced industrially: that has C in the range
8−9.5% [24, 25]. As in other simulations, for example with FactSage in [23], C in
FeCr and FeO in slag are too low. These are indications the approach to add more
reaction zones could be meaningful: by allowing more reaction zones and pathways,
a better approximation of industrial values could probably be achieved.
The chrome recovery is very high. This should be corrected for mechanical furnace
dust losses. Taking these in the order of 2−4% leaves 91−94% recovery, which is
in line with values as found in the literature for high carbon FeCr [24].
Changing reductant from coke-1 to anthracite-1 has multiple effects. For one,
the moisture content already affects carbon consumption and energy requirement.
Secondly, the different ash requires a change in charged fluxes if the slag is to be

Table 4 Results from HSC

Coke-1 Anthracite-1
calculations with 20 tph of
chromite and the fluxes and Cr2 O3 % 4.5 4.5
reductants needed to achieve Al2 O3 % 27.7 27.8
the given target slag
SiO2 % 22.9 22.9
composition at 1800 °C
MgO % 43.4 43.3
FeO % 0.1 0.1
Cr % 71.2 71.0
C % 5.4 5.3
Si % 0.6 0.6
Load MW 40.26 41.99
Ore tph 20 20
Reductant tph 5.14 5.24
Bauxite tph 1.08 1.04
Quartz tph 0.02 0.32
Feed-to-power kg/MW 652 634
Chrome recovery % 96.1 96.1
Reductant Formation Enthalpy in DC Ferrochrome Smelting: Merely … 155

kept the same. As a result, a different furnace power is required to smelt the same
quantity of ore.
Thirdly, the FtP has changed because of the different reductant. The change of
18 kg/MW may seem small but is about 3%. A similar change can be seen in the MW
required: 1.7 MW is about 4%. This is well above the 0.5% accuracy limit given by
Barcza [3]. This change requires that the previously applied FtP must be adjusted
to prevent overheating when changing from coke-1 to anthracite-1. To conclude,
already a change in reductant without consideration of the RFE can be a reason to
adjust any preprogrammed FtP relationship in the control software.

Step 2 Application of Different RFE’s on the Previous

Calculation

The aim was to see if the ignoring the RFE could be a cause for problems in DC FeCr
smelting. Changing the reductant without considering the RFE is already a reason to
adjust the FtP ratio as shown under step 1. If no calculations are done to investigate
the effects of a reductant change, then any possible effects of the RFE could simply
be taken as part of the adjustment, without understanding where part of the required
change came from.
A different approach was needed to isolate the RFE effects.
Based on the RFE’s, changing from anthracite-1 to coke-1 during production
would lead to an energy shortage and thus freezing of the bath. By changing from
coke-1 to anthracite-1, the overheating of the bath could be investigated.
Assume the operation with coke-1 has been running stable and in equilibrium for
days at 40.26 MW without consideration for the RFE. That means feedrate, energy
balance, and assumptions for heat losses are all in equilibrium, irrespective of how
accurate these are individually. Assuming the feedrate and analyses of raw material
analyses are stable and highly accurate, then more detailed calculations on the RFE
become possible.
In step 1 with coke-1, the total required load was 40.26 MW, of which 88%
(35.43 MW) was assumed to cover metallurgy excluding RFE. The rest, 4.83 MW,
are considered to be heat losses, but actually cover both real heat losses and all
metallurgical issues ignored or not known, like the RFE.

Assumed energy balance = 40.26 MW = 35.43 MW + 4.83 MW

The calculated RFE for coke-1 = -490 kWh/t. This means that 490 kWh/t reductant
have been used in the calculation to produce FeCr but were not available in reality.
This means that the initially calculated heat loss of 4.83 MW was smaller, because
part of this energy was needed to compensate for the ignored RFE.
The energy balance including RFE is thus
156 H. J. Oterdoom et al.

40.26 MW = (35.43 + 5.14 ∗ 0.490) + (4.83−5.14 ∗ 0.490) = 37.95 + 2.31

The new heat loss is 2.31 MW and assumed to be also the actual heat loss for the
case with the new reductant, anthracite-1.
In the case with anthracite-1 as a reductant, Table 4 shows the required anthracite,
fluxes, and energy to have a similar process. The furnace load of 41.99 MW is,
however, calculated without considering the RFE correction for anthracite-1. The
41.99 MW included 12% of heat losses, 5.04 MW, “Because with 12% heat losses
the furnace operated stable with coke-1 and the given feedrate”.
The real furnace heat losses are only 2.31 MW. The difference of 2.73 MW must
be kept in mind, because this by itself would already upset the process and raise the
bath temperature.
The anthracite-1 calculation without RFE or heat loss correction has 36.95 MW
for the metallurgical reactions (41.99−5.04). The RFE correction is 5.24t
anthracite*(147) kWh/t = 0.77 MW.
The actual required energy with anthracite-1 is therefore: 36.95 + 2.31−0.77 =
38.49 MW. Relative to the actually required energy of 38.49 MW, the 41.99 MW
is 9.1 percent too much energy input. These 3.50 MW surplus are available in the
reaction zone and should not be written away as thermal losses: it is at the electrode
tip where the feed is processed and where the surplus energy will overheat the bath
and make side reactions happen.
A solution is to set the 2.31 MW as heat losses, being 5.5% of the furnace load. This
directs 2.73 MW from heat losses to the reaction zone. Then to compensate for the
RFE, 0.77 MW is added as a stream of energy directly into the reaction zone operation.
This method seems appropriate as it compensates for the real thermal losses whilst
having identical bath conditions and compensate for the energy “released” from the
coke.
Note that the difference in RFE between anthracite-1 and coke-1 is relatively
big. Changing from anthracite-1 to most other anthracites would not be so drastic
regarding the RFE.
Secondly, in industry, the feedrate would be adjusted to the power, whereas in
this calculation the power has been adjusted to a fixed rate of charging ore and the
required additions. This was done to show clearer the effect raw materials have on
the energy consumption.

Step 3 Run the Continuous Model with the New Values

and See What Happens

The continuous model was tested once the batch mass and energy balances were
validated. The principle of the continuous model is as shown in Fig. 1. Feed materials
and power enter the reaction zone. Reactions happen based on temperature and the
equilibrium calculation that HSC provides. The model allows for collecting products
Reductant Formation Enthalpy in DC Ferrochrome Smelting: Merely … 157

Fig. 1 Representation of the

model with input of raw
materials and energy input at
a single point used as
reaction zone. From the
reaction zone, alloy and slag
collect in the alloy and slag
bath, respectively, to be
tapped intermittently

in the furnace in a slag and alloy bath (called “tanks” in HSC) and tapping from these
tanks once certain bath levels have been reached. By assigning volumes to the tanks,
it is possible to simulate how a furnace acts over time. Following bath volumes were
used:
• Slag bath volume approximately 55 m3
• Alloy bath volume approximately 70 m3

First, the operation was run with coke-1 to verify the energy balance was stable.
Next, anthracite-1 was used together with the values for power, reductant and fluxes
as given in Table 4 without considering the RFE. To then simulate over time the
effect of the corrected heat loss and the RFE, the following values were used:
• Power input: 41.99 MW.
• Heat loss: 2.31 MW = 5.5%this already adds 2.73 MW to the reaction zone.
• Energy input into the reaction zone due to RFE: 0.77 MW.
The results of a 24-h simulation run are shown in Fig. 2. Reaction zone and slag
and alloy baths all start at 1800 °C. Slag and alloy go 100% to the slag and alloy
baths, and both are tapped when certain levels are exceeded. This is visible in the
upper left graph of Fig. 2. Process gas leaves the furnace continuously and instantly.
The first five hours are with the original settings with heat losses of 12% and no RFE
correction.
At t = 5 h, the RFE and heat loss correction as given above are activated. The
excess of available energy causes the temperature of the reaction zone to instantly
increase as can be seen in the lower left image of Fig. 2. Because the reaction zone
has not been assigned a volume, it instantly jumps to the new equilibrium value
of 1.885 °C. The formed slag and alloy mix with the slag and alloy baths and the
temperatures of both slowly rise. Compositions of alloy bath and slag bath are shown
in the graphs on the right.
With the higher temperature, the carbon content in the alloy drops while the Cr2 O3
and Al2 O3 contents in slag increase. At the same time, both MgO and SiO2 in slag
decrease. Note that Barcza et al. [3] came to similar graphs, where C and Cr from alloy
were consumed and the MgO content in the slag was reduced due to the formation
of Mg-gas.
158 H. J. Oterdoom et al.

Fig. 2 Screenshots from HSC SIM results. Tapping (top left) ensures the bath levels remain within
certain limits. Applying the corrections based on ERF’s show a strong increase in reaction zone
temperature (bottom left). As the temperature increases, the alloy bath (AB) is losing C (top right),
while the slag bath (SB) gets richer in Cr2 O3 (middle right) but is losing MgO and SiO2 due to Mg
and SiO gas formation

Summary and Conclusions

Aiming for an accuracy of 0.5% for the feed-to-power is an admirable target but may
be very hard in reality due to the lack in accuracy in heat loss measurements and
limitations in thermochemical calculations.
By isolating the effects of the enthalpy of formation of reductants on a FeCr
DC smelting operation in a dynamic simulation, it could be shown this can have a
significant impact on the operation. A check of the formation enthalpy of available
reductants on site by comparing the actual GCV with the theoretical GCV should be
done to see what the impact of a reductant change can be. The information can be
used as an extra warning and to calculate a pre-emptive FtP correction.
Single step black-box models have worked well for many years in many processes.
The modern open bath furnaces may have been proof that this approach does not
sufficiently address the importance of heat and mass transfer. The presented model
got close to industrial values but requires more accuracy.
A dynamic model to investigate and visualize the effects of changes in raw
materials or operational parameters has been demonstrated to lead to increased
understanding and insight.
Better understanding of the behaviour of carbonaceous reductants in a furnace
will support a transition to using recycled reductants, like (pyrolyzed) plastics or
biowaste.
The advice remains to stick to one reductant and if change is needed, to do so
gradually. Not only because of the RFE but also because in reality multiple variables
can change simultaneously. Therefore, paying attention carefully to the operation,
Reductant Formation Enthalpy in DC Ferrochrome Smelting: Merely … 159

like tapping temperatures and compositions, remains an important tool to keep the
process controlled.
The basic functionality of the dynamic model has been demonstrated, so next
is to include more zones for more accurate process simulation. With the improve-
ments made, ideas about the mechanisms inside a DC FeCr furnace will be further
investigated as part of the PhD.

References

1. Merrow EW, Phillips KE, Myers CW (1981) Understanding cost growth and performance
shortfalls in pioneer process plants. Rand Corporation for Department of Energy
2. McNulty T (1998) Developing innovative technology. Min Eng 50(19):50–55
3. Barcza NA, Curr TR, Jones RT (1990) Metallurgy of open-bath processes. Pure Appl Chem
62(9):1761–1772
4. Theron JA, Le Roux E (2015) Representation of coal and coal derivatives in modelling. J South
Afr Inst Min Metall 115(5):339–348
5. Barnes A, Mackey P, Alvear G (2009) Process development and growth of non-ferrous metals
production: the role of pilot plants. Plenary address, European Metallurgical Conference,
Innsbruck, Austria, 28 June−1 July 2009
6. Barnes RA, Jones RT (2011) Cobalt from slag—lessons in transition from laboratory to industry.
In: Paper presented at Conference of Metallurgists, Montreal, Canada, 2–5 October 2011
7. Pan X (2013) Effect of South Africa reductants on ferrochrome production. Paper presented
at the 16th IFAC Symposium on Automation in Mining, Mineral and Metal Processing, San
Diego, California, USA, 25–28 August 2013
8. Geldenhuys IJ (2013) Aspects of DC chromite smelting at Mintek—an overview. In: Paper
presented at Infacon XIII, Almaty, Kazakhstan, 9–13 June 2013
9. Sweeten NJ et al. (2018) Chrome ore mineralogy and the furnace mass and energy balance. In:
Paper presented at Infacon XV, Cape Town, South Africa, 25–28 February 2018
10. Sommerfeld M, Friedrich B (2022) Proposition of a thermogravimetric method to measure the
ferrous iron content in metallurgical grade chromite. Minerals. [Link]
20109
11. Merrick D (1983) Mathematical models of the thermal decomposition of coal. Fuel 62:534–539
12. Sampaio RS, Fruehan RJ, Ozturk B (1991) Rate of coal devolatilization in iron and steelmaking
processes. In: Paper presented at 1991 Ironmaking conference proceedings, Washington, USA
13. Wei Q, Qiang X, Yuyi H, Jiatao D, Kaidi S, Qian Y, Jincao W (2012) Combustion characteristics
of semicokes derived from pyrolysis of low rank bituminous coal. Int J Min Sci Technol
22:645–650. [Link]
14. Neavel RC, Hippo EJ, Smith SE, Miller RN (1986) Coal characterization research: sample
selection, preparation, and analyses. Preprints of papers. American Chemical Society. Div
Fuel Chem 25(3):246–257
15. WCI World Coal Institute (2009) Coal conversion facts: basis of analysis. [Link]
com/wp-content/uploads/coalconversionfacts200704_06_2009.pdf. Accessed 6 September
2022
16. SGS (2022) Coal calculations. [Link]
Accessed 6 September 2022
17. Mason DM, Gandhi KN (1983) Formulas for calculating the calorific value of coal and coal
chars: development, tests, and uses. Fuel Process Technol 7:11–22
18. Kleynhans ELJ, Beukes JP, Van Zyl PG, Bunt JR, Nkosi NSB, Venter M (2017) The effect of
carbonaceous reductant selection on chromite pre-reduction. Metall Mater Trans B 48b:827–
840
160 H. J. Oterdoom et al.

19. HSC chemistry 10, HSC SIM Flowsheet Module, version [Link]
20. Strakhov VM, Medyanik VS, Malinovskaya LI, Tsernikel AA, Veselovsky IA (2016)
Production of high-carbon ferrochrome from dry-quenched semicoke. Coke and chemistry
59(7):254–259
21. McCullough S, Hockaday S, Johnson C, Barcza NA (2010) Pre-reduction and smelting
characteristics of Kazakhstan ore samples. In: Paper presented at Infacon XII, Helsinki, Finland
22. Oterdoom HJ et al (2018) (2018) High carbon ferrochrome smelting in a DC furnace: smelt
small stuff and stop sintering. Paper presented at Infacon XV, Cape Town, South Africa
23. Lu J (unknown) Ferrochrome production. [Link]/wp-
content/uploads/[Link] Accessed 28 July 2022
24. Yessenzhulov AB (2020) New and reconstructed JSC “TNC “Kazchrome” facilities. In: Paper
presented at 2nd conference on The best available technologies for steel and ferroalloys
production, Moscow, Russia
25. Jones RT, Erwee MW (2016) Simulation of ferro-alloy smelting in DC arc furnaces using
Pyrosim and Factsage. CALPHAD: Comput Coupling of Phase Diagrams and Thermochem
55:20–25
26. Dale A, Lythall C, Aucott J, Sayer C (2002) High precision calorimetry to determine the
enthalpy of combustion of methane. Thermochim Acta 382:47–54
27. Zietsman J (2022) Essential theory of thermochemistry—mass and energy balances: organic
feed materials. Lecture notes on mass and energy balances
Measuring and Processing of Electrical
Parameters in a Submerged Arc Furnace

Hákon Valur Haraldsson, Halldór Traustason, Yonatan A. Tesfahunegn,

Merete Tangstad, and Gurún Sævarsdóttir

Abstract Electric arcs are necessary for high Si yield in submerged arc furnaces
(SAFs) for Si/FeSi production, and a certain fraction of heat dissipation in the arc
enables optimal operating conditions. Direct measurement of the arc characteristics is
impossible due to hostile conditions inside the SAF, so controlling the heat dissipation
is both a science and an art. The arcs exhibit non-linear electrical characteristics and
behave in a complex manner. Hence, implementing a data acquisition (DAQ) system
to collect current and voltage waveforms typically on the electrodes or transformer
connections combined with appropriate signal processing offers an estimate of the
actual arc parameters, enabling improved understanding of the arcing in the furnace,
and improving furnace operation. In this paper, a DAQ system gathering data from a
FeSi SAF will be discussed, and the data is processed and used to determine various
furnace conditions including arc and charge current as well as harmonics.

Keywords Submerged arc furnace · Arc footprint · Arc harmonics · Electric arc

H. V. Haraldsson · H. Traustason · Y. A. Tesfahunegn · G. Sævarsdóttir

Department of Engineering, Reykjavík University, Reykjavík, Iceland
e-mail: [Link]@[Link]
Y. A. Tesfahunegn
e-mail: yonatant@[Link]
G. Sævarsdóttir
e-mail: gudrunsa@[Link]
H. V. Haraldsson (B) · M. Tangstad · G. Sævarsdóttir
Department of Materials Science and Engineering, Norwegian University of Science and
Technology, Trondheim, Norway
e-mail: hakonh12@[Link]
M. Tangstad
e-mail: [Link]@[Link]

© The Minerals, Metals & Materials Society 2023 161

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
162 H. V. Haraldsson et al.

Introduction

The furnace used to produce silicon is called a submerged arc furnaces (SAF) because
the electrodes that carry the electric current are submerged in the raw materials used
in the production process. In some processes operated in submerged arc furnaces
such as in ferromanganese production, there is actually no arc in the furnace and
all the heat is dissipated as the current passes through the materials in the furnace,
such as the coke bed below the electrodes. For silicon and ferrosilicon production,
however, the presence of an arc is essential. In this case, much of the current will
pass through an electric arc, which burns in a cavity or crater which is filled with gas
not solid materials, this is thought to be due to the high viscosity of the SiO2 and
the formation of a SiC crust around the cavity. Liquid silicon forms in a pool below
the arcs and is tapped into ladles using tap holes. The extremely hostile environment
inside SAFs makes any direct observation of the arc impossible.
A submerged arc furnace is a fairly large machine usually spanning a few floors
in the building it is situated in. In simple terms, the furnace converts electricity
and internal energy of carbon materials to heat which melts the raw materials, thus
creating usable metals. The furnace itself is circular and varies from approximately
5–11 m in diameter and from 3–5 m in height. Raw materials are added from the top
and will descend slowly in the furnace while they are heated. A diagram of a SAF
can be seen in Fig. 1.
Since the arc cannot be measured directly, estimating its parameters can be dif-
ficult. In an operating furnace, we can only measure the voltage and current going
into each electrode, but by creating a complete electric circuit model of the furnace
and knowing its parameters, either by simulation or measurements, we can use the
measured voltage and current waveforms in each electrode to indirectly estimate the
arc. There are not many published studies on measuring arc parameters but previous
work on this has been done in [1] and [2]; in [3], measurements are used to model the

Fig. 1 A schematic drawing

of a SAF in operation
Measuring and Processing of Electrical Parameters in a Submerged Arc Furnace 163

Fig. 2 A SAF in knapsack

configuration

effects of electrode height on arc currents. All measurements on any type of furnace
require some sort of data acquisition (DAQ) system to collect and process the data.
For industrial applications, reliable and accurate equipment is essential, so selecting
the correct equipment can be a challenge. Lorenzo et al. [4] detail this for a Spanish
ferroalloy company, and Gerritsen et al. [5] describe some of the methods used and
impact of error in the measurements. These should not be confused with direct arc
modeling, which is a different approach [6].
In this paper, we will describe the DAQ system that was implemented on a FeSi
SAF in Elkem Iceland. We will discuss the SAF properties required for data pro-
cessing to estimate actual SAF parameters. Finally, we will describe how the data
is processed from raw measurements to actual useful parameters that are used to
estimate current distribution, arc footprint, and arc harmonics in the furnace.
Inside the furnace, the electrodes are submerged in the raw materials, below each
electrode there is a cavity where the arc is burning. This cavity or crater is filled with
gas not solid materials, and this is thought to be due to the high viscosity of the SiO2
and the formation of a SiC crust around the cavity. Liquid silicon forms in a pool
below the arcs and is tapped into ladles using tap holes.

Electric Circuit Description of a SAF

A 3-phase SAF generally consists of 3 transformers placed concentrically around

the furnace with 120◦ between them, and this is done to minimize mutual inductance
between them. The transformers are connected to 3 electrodes that conduct electric
power to the furnace through bus bars and flexibles. This setup is called a knapsack
configuration and is shown on Fig. 2.
When discussing a SAF from an electrical standpoint, the main focus is on the
currents present in the electrode, and what electrical parameters affect them. The
electrode current is the total current that travels down the electrode from the delta
connection connecting the transformers, before it diverges to different paths. To
understand what is happening, we must see what the total electrical characteristics
of the SAF are. Industrial high current arcs are generally present within the electric
164 H. V. Haraldsson et al.

Fig. 3 The electric circuit

used to describe a SAF [7]

arc furnaces, and to determine the behaviour of the furnace and the arc itself, a
complete electric circuit must be implemented.
The circuit should include most of the actual components present, transformer
resistances RTi , impedances LTi , and mutual inductance between them (MTi,Ti+1 and
MTi,Ti+2 ), as well as the resistances of the electrodes (Re ) and the phase inductance
including the mutual inductance (L e and Mei,ei+1 and Mei,ei+2 ). Mutual inductance
due to currents in the bus bars and flexibles are normally lumped into the phase
inductance, which is accurate if the furnace is reasonably symmetrical.
We consider the electrode current to split up into two parts, the current going into
the arc and then the current in the charge material. Rc represents the charge resistance
and Va the arc voltage drop. This is a simplification but gives a reasonable estimate
since other components are rather small in a well-operating furnace. The physical
properties of the charge materials differ with location in the furnace, as intermediate
reactions change their composition and temperature changes with depth both of Fig. 3
shows the electric circuit of a three-phase SAF and where each component is located
in it.

Estimating SAF Parameters

The DAQ System

The data acquisition system for SAF samples both the voltage and current in each
electrode as well as a few of the higher harmonics that are present in the signal. Here,
Measuring and Processing of Electrical Parameters in a Submerged Arc Furnace 165

Fig. 4 The simplified circuit

model used to estimate arc
characteristics from the
measurements

the measurement frequency f S is 3000 Hz which gives about 60 samples per period.
This sampling rate allows for the estimation of harmonics up to 1500 Hz which is
the 30th harmonic 50 Hz.
The DAQ system consists of three voltage sensors that measure the electrode
voltage through step-down modules since it is too high to measure directly, the
current sensors measure the current in each electrode using the hall effect. A detailed
description can be found in [8] where the DAQ system design and implementation
are described in detail.

Data Processing

A method has been developed to estimate the current that passes through the arc and
charge material. The simplified circuit presented in Fig. 4 shows the model used to
process the measured current and voltage waveforms. The model is similar to the
section surrounding the arc in Fig. 3 but most circuit elements have been combined
and the furnace resistance is included. Current and voltage measurements for each
electrode are defined as i phi and Vphi . Here, the subscript i denotes the electrode
numbering. The total phase voltage Vphi is a combination of the voltage induced
in the furnace VLi , the voltage over the arc Vpi (which is over the time varying arc
resistance R ai and the assumed constant charge resistance Rci connected in parallel),
and finally, the voltage over the furnace short-circuit resistance Rsi , the liquid metal
resistance, and the base resistance in the furnace. Vpi can be considered as the power
producing voltage or load voltage that contributes to the actual silicon production.
Now, the equivalent phase resistance Rph , the arc resistance Ra , and the charge
resistance Rc can be defined using Ohms law as:
1
Rph = + Rs (1)
1
Ra
+ 1
Rc
166 H. V. Haraldsson et al.

Rc (Rph − Rs )
Ra = (2)
Rs + Rc − Rph

Ra (Rph − Rs )
Rc = (3)
Rs + Ra − Rph

The induced voltage present in the phase voltage causes a phase shift in the current,
and this results in the current lagging behind the voltage by 30◦ to 45◦ . The magnitude
of the phase shift varies with the size of the furnace but also with electrode movement
withing the furnace. The induced voltage VLi must therefore be removed from the
total phase voltage Vphi before the resistances can be calculated. This is done by a
simple time-dependent differentiation of the current signal to estimate the induced
voltage:
di phi
Vpi = Vphi − Li (4)
dt
Here, L i is the total inductance present in the electrode, and this value is found
by iteration until the phase shift is negligible. With the phase shift removed, the total
phase resistance for each electrode Rphi can be determined. The phase resistance or
zero crossing resistance is the resistance when both the phase current and the power
producing voltage cross the zero point simultaneously. Rphi is found by plotting the
Lissajous curve for i phi and Vpi , and then, the slope is estimated as the curve passes
zero. To find Rc , we need both Rs and Ra , and the short-circuit resistance is either
known from the modeling, such as in [9], or can be assumed to be around 5% of
the phase resistance similar to what is done in [8]. Determining the arc resistance
Ra can be difficult since you ideally need measurements from when the furnace
being measured when only an arc is present without any charge materials present,
this is important since different furnaces will have different parameters. Previous
measurements carried out in the same furnace when a pure arc was present were
used and determined the arc resistance to be around 2.5–4 m [2]. This will give an
upper and lower limit for the charge resistance Rc , and consequently, both the arc
resistance Ra and the arc current i a will have similar limits; here, we will show the
upper limit only for simplicity of the graphs. Now, all that is needed to determine
the range for the arc current i a which is the voltage over the arc Va given as:
Vai
Vai = Vpi − i phi Rs ⇒ i ai = (5)
Rai

Results

The ability to estimate furnace parameters both after or preferably during operation
will allow for better control of the overall operation of a furnace. The data used in this
Measuring and Processing of Electrical Parameters in a Submerged Arc Furnace 167

Total Current Data Total Voltage Data

118 230

116
Average RMS current per second (kA)

Average RMS voltage per second (V)

220
114

112 210

110
200
108

106 190

104 Phase
1 Phase
1
Phase
2
180
Phase
102 2
Phase 3
Phase
3

100 170
0 50 100 150 200 250 0 50 100 150 200 250
Time (sec) Time (sec)

Fig. 5 Total RMS electrode current and voltage data averaged over one second periods

Fig. 6 To find the zero 200

Finding the zero crossing resistance

crossing phase resistance, we

do a linear fit around the 150
center of the voltage versus
current curve 100

50
Voltage [V]

-50

Total Electrode values

-100
Without phase shift
Fit data
Fit
-150

-200
-200 -150 -100 -50 0 50 100 150 200
Current [kA]

paper for all three electrodes can be seen in Fig. 5 where the RMS values for current
and voltage have been averaged over one second periods. From these figures, we can
easily observe the intensely variable nature of the furnace’s electric properties.
Now, we can pick a piece of the data to investigate in more detail. First, we need
to remove the induced voltage in the electrodes; Fig. 6 shows the voltage vs. current
profile before and after the induced voltage is removed and the fit used to estimate
the zero passage resistance using Eq. 4.
Now, we can find out the charge resistance Rc using Eq. 3, and since it is assumed
constant over each period, the change in the phase resistance must be due to the arc
resistance Ra . We can now divide the electrode current into charge and arc currents,
and in Fig. 7, we can see all three current for two periods. Here, a large part is going
into the arc so we estimate that it is probably rather short and that this will increase
the harmonic distortions in the furnace as a result.
168 H. V. Haraldsson et al.

Electrode 1 Electrode 2 Electrode 3

200 200 200
I Electrode I I Electrode
Electrode
150 I Charge 150 I Charge 150 I Charge
I I I
Arc Arc Arc
100 100 100

50 50 50
Current [kA]

0 0 0

-50 -50 -50

-100 -100 -100

-150 -150 -150

-200 -200 -200

0 0.01 0.02 0.03 0.04 0 0.01 0.02 0.03 0.04 0 0.01 0.02 0.03 0.04
Time [s] Time [s] Time [s]

Fig. 7 Current distribution for all three electrodes. The arc and charge currents vary somewhat
between them

Fig. 8 The arc footprints for all three arcs and the footprint area for arc 2

Another factor of interest is the arc footprint or the voltage and current graph area.
This area has the units kVA so it tells us how much power is present in the arc over a
whole period but the shape of the curve can also be connected to furnace operations.
Theses can be tapping, how the charge material is distributed and so forth. Figure 8
shows all three arc footprints, and the area for arc 2 is 5566.2 kVA.
More data during various furnace operations is needed to connect these foot-
prints to what is actually happening in the furnace, and ideas would be to properly
observe during tapping, charge refilling, and electrode operation and to catalog the
arc footprints to get a better understanding of the electric behaviour during these
events.
The final thing looked into in this paper are the harmonics. These are very impor-
tant to consider since they can be used along with the rest of the parameters help give
a better image of what is going on as well as being important on their own. Minimiz-
ing the total electrode harmonics is important so that the total harmonic distortion
(THD) does not go over the limit allowed by the electric utility supplying power to
the furnace. Seeing what frequency components are present in which current can also
Measuring and Processing of Electrical Parameters in a Submerged Arc Furnace 169

Frequency components in electrode 2 Frequency components in the arcs

90 70
I electrode I arc 1
80 I charge I arc 2
60
I arc I arc 3
70
% of total electrode current

% of total electrode current

50
60

50 40

40 30

30
20
20
10
10

0 0
0
5
0
5
0
5
0
5
0
5
0
5
0
5
0
5
0

0
5
0
5
0
5
0
5
0
5
0
5
0
5
0
5
0
25
50
75

25
50
75
0

0
10
12
15
17
20
22
25
27
30
32
35
37
40
42
45
47
50

10
12
15
17
20
22
25
27
30
32
35
37
40
42
45
47
50
Frequency [Hz] Frequency [Hz]

Fig. 9 The frequency components present in electrode 2 and in all three arcs as a percentage of
total current in each electrode

give an idea of the situation at each electrode and in the charge material. This can be
arc length and material properties in the charge and metal bath below the electrodes.
Figure 9 shows the current harmonics in the furnace for electrode 2 and all the arcs,
and we can see which frequency components are present in the furnace currents in
the given period.

Conclusion

We have used data gathered using a DAQ system implemented in Elkems FeSi plant
in Iceland to estimate various furnace parameters. The data is initially a set of voltage
measurements that must be converted to actual voltage and current values based on
the selected furnace, and measurement equipment. Using this data and previous
measurements done on the furnace, we have devised a method of estimating the
furnace current distribution profile for arc and charge currents.
The method described allows us to estimate all electrode current and voltage
parameters for each time period. Using reference data for a pure arc, we can find the
constant charge resistance based on each electrode’s slope of the Lissajous curve.
This, in turn, allows us to estimate the arc resistance, which is very variable during
each period.
By having a better idea of what is going on inside the furnace in relation to
the arc and charge currents should allow for smother furnace operation, which
should improve silicon yield and lower the specific energy consumption. More mea-
surements are however needed during operations to collect more data on current
behaviour, preferably during tapping, filling, and general electrode operation.
170 H. V. Haraldsson et al.

Acknowledgements The Icelandic Research Fund is greatly acknowledged for their funding of
this work.

References

1. Saevarsdottir G, Magnusson T, Bakken JA (2007) Electric arc on a coke bed in a submerged arc
furnace
2. Saevarsdottir G (2012) Bakken: Current distribution in submerged arc furnaces for silicon
metal/ferrosilicon production, international ferroalloys congress; 12th, INFACON 12. In: INFA-
CON 12, INFACON, International Ferroalloys Congress; 12th, INFACON 12. Norwegian Fer-
roalloy Research Organization;. Backup Publisher: Norwegian Ferroalloy Industry, pp 717–728.
[Link]
3. Hauksdottir AS, Soderstrom T, Thorfinnsson YP, Gestsson A (1995) System identification of a
three-phase submerged-arc ferrosilicon furnace. IEEE Trans Control Syst Technol 3(4):377–387
4. Lorenzo A, Lage M, Bullon J, Rivas J, Fondado A, Torres A, Farina J, Rodriguez-Andina
JJ (2007) Measurement of electrical parameters in high-current arc furnaces. In: 2007 IEEE
international symposium on industrial electronics, pp 1565–1568 (2007). [Link]
1109/ISIE.2007.4374836
5. Gerritsen T, Tracy PE, Saber FNM (2015) Electrode voltage measurement in electric furnaces:
analysis of error in measurement and calculation
6. Haraldsson H, Tesfahunegn YA, Tangstad M, Sævarsdottir G (2021) Modelling of electric arcs
for industrial applications, a Review. SSRN J
7. Saevarsdottir G, Larsen H, Bakken J. Modelling of industrial AC-arcs 3:1–15. [Link]
10.1615/HighTempMatProc.v3.i1.10
8. Traustason HG. Data acquisition of electrical parameters and interpretation of SAF data
9. Tesfahunegn YA, Magnusson T, Tangstad M, Saevarsdottir G (2018) Effect of carbide configu-
ration on the current distribution in submerged arc furnaces for silicon production-a modelling
approach. In: Nastac L, Pericleous K, Sabau AS, Zhang L, Thomas BG (eds) CFD modeling
and simulation in materials processing 2018. The Minerals, Metals & Materials Series. Springer,
Heidelberg, pp 175–185 [Link]
Volatilization Behavior of Arsenic
from a Hematite Ore During
Non-isothermal Heating in Argon
Atmosphere: An Overview

E. K. Chiwandika and S.-M. Jung

Abstract The high grade iron ore has depleted necessitating the need to inves-
tigate the use of low grade ores that may include arsenic-containing iron ores.
Arsenic (As) is volatile increasing the possibility of its removal from the ore by
roasting or sintering. As volatilization from iron ore is limited by many chemical
interactions with other metallic elements and oxides within the ore. The possi-
bility of the volatilization of As from a hematite ore was investigated using the
thermogravimetric-differential scanning calorimetry (TG-DSC) and the vertical tube
furnace by raising the temperature at a rate of 10 K/min in the temperature range of
298 to 1623 K. Predictions of the most stable As compound formed in the sintering
temperature range were done first using basic thermodynamics, then X-ray diffrac-
tion (XRD) measurements were done on the cooled and crushed samples to identify
the change in phases after heating to temperatures above 1273 K. Electron probe
microanalysis (EPMA) verified the possibility of the existence of some of the phases
identified by XRD. Results showed that the percentage weight loss of the pellet and
As loss from the pellet increased with temperature. Metallic element distribution
above 1273 K showed that As was highly concentrated in the same area as Ca, Si,
and Al in agreement with the XRD results and the thermodynamic predictions that
showed that 3CaO·2As2 O5 (s) and the AlAsO4 (s) as the most stable As-containing
phases even after heating to temperatures of about 1623 K.

Keywords Arsenic · Volatilization · Percentage weight loss · Volatilization rate ·

Sintering · Non-isothermal heating

E. K. Chiwandika (B)
Department of Materials Technology and Engineering, Harare Institute of Technology, P. O. Box
BE 277, Belvedere, Harare, Zimbabwe
e-mail: chiwandikae@[Link]
S.-M. Jung
Graduate Institute of Ferrous Technology, Pohang University of Science and Technology,
Pohang 37673, Korea

© The Minerals, Metals & Materials Society 2023 171

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
172 E. K. Chiwandika and S.-M. Jung

Introduction

The global annual steel production is growing, increasing the demand for raw mate-
rial. The available high grade iron ore has depleted necessitating the utilization of
low grade ores. The use of such ores has some economic advantages since prices of
such resources are still low due to little or no competition in demand of such ores.
Arsenic (As) containing iron ores are an example of such underutilized resource.
Arsenic exists in a wide range of arsenates and associated minerals and also shows
some polymorphic behavior. As2 O3 (s) that can exist as arsenolite, claudetite or a
glassy form [1]. This makes it difficult to understand the mode of existence of As in
the ores. The oxidation state of As in some of the compounds of As may form ranges
from + 2, + 3, and + 5. Examples of such compounds include but not limited to
arsenopyrite (Fex Asy Sz (s)), oxides (As2 O5 (s) or As2 O3 (s)) or sulfides (As2 S3 (s)) in
ores [2, 3]. Iron ores may contain As as high as 0.5 weight percent [4] and when
such ores are directly charged into the blast furnace, As readily dissolves into the
molten iron rendering its removal very difficult. The As in steel negatively impacts
the mechanical properties of steel products such that engineers demands that steels
for special purposes like those used in petroleum bars, large dynamo rotor, and in
nuke industries contain no As [5]. As removal methods prior to the blast furnace
process should be developed for the effective utilization of this resource.
Some researchers have tried to remove As by roasting or sintering, but these
methods are not effective enough as 50–75% of the As was retained inside the ore
[6]. Contreras et al. [7] used the HSC chemistry 5.0 software to predict the distribution
of trace element between gas, liquid, and solid phases as a function of temperature
in coal combustion. Their results showed that 3CaO·As2 O5 (s) was the most stable
phase, and remains non-volatile even at temperatures above 1573 K in addition, the
water solubility of this compound was found to be low [7]. As volatilization was
strongly related to the chemical bonds holding As in coal at 873 to 1773 K [4].
The behavior of As in iron ores during sintering is yet to be cleared due to the
complexity of As existence. The standard gas for calibrating state-of-the-art equip-
ment like the quadrupole mass spectrometer (QMS) is not easily available for online
gas analysis during As volatilization. Furthermore, development of the pollution
prevention method caused by the treatment of such an ore demands great attention.
This investigation is aimed at providing the basic understanding of some of the phases
that might develop during sintering of the hematite ore containing As and finding the
most suitable temperature to remove As which might also help in the development
of the pollution prevention methods caused by As.

Thermodynamic of As Volatilization

The generation of calcium ferrite is of great importance in the sintering of iron ores
due to better strength and reducibility caused by the presence of derivatives of calcium
Volatilization Behavior of Arsenic from a Hematite Ore During … 173

ferrites in sinter [8]. Volatilization of As from iron ore during sintering or roasting
can be complicated by interactions with other metal components in the ore. The
possible compounds formed by these interactions includes FeAsO4 (s), AlAsO4 (s),
KAs3 O4 (s), NaAs3 O8 (s), 3MgO·As2 O5 (s), and CaO·As2 O5 (s) [2, 4, 7, 9]. A ther-
modynamic analysis of some the compounds that may be formed in the sinter in
the presence of As was done by the plot of Go vs temperature, Fig. 1, using the
thermodynamic data from the equations below [10]

CaO(s) + Fe2 O3 (s) = CaO · Fe2 O3 (s) (1)

G ◦ = −30, 000 − 4.8T (973 − 1489 K)J/mol (2)

2CaO(s) + Fe2 O3 (s) = 2CaO · Fe2 O3 (s) (3)

◦
G = −53, 100 − 2.5T (973 − 1723K)J/mol (4)

2CaO(s) + SiO2 (s) = 2CaO · SiO2 (s) (5)

◦
G = −120, 000 − 11.3T (298 − 2403 K)J/mol (6)

Al2 O3 (s) + As2 O3 (g) + O2 (g) = 2AlAsO4 (s) (7)

◦
G = −536, 068 + 93.49T (1373 − 1673K) J/mol (8)

3CaO(s) + As2 O3 (g) + O2 (g) = 3CaO · 2As2 O5 (s) (9)

◦
G = −775, 454 + 268.8T (1373 − 1678 K) J/mol (10)

The Go for Eqs. (7) and (9) was calculated using FactSage 6.2 software.
It can be concluded that the 3CaO·2As2 O5 (s) and 2AlAsO4 (s) are the more
stable compounds at the selected temperature range than the calcium ferrite and
the 2CaO·SiO2 (s).
174 E. K. Chiwandika and S.-M. Jung

Fig. 1 Plot of Go versus temperature for some of the possible reactions that can occur during
sintering

Experimental

Materials

To a hematite ore with As content of 0.035 wt%, loss on ignition (LOI) of 7.10 wt%
and an average particle size of under 1 mm, about 1 wt% reagent grade As2 O3 (s) was
added targeting an As content above 0.5 wt% in the blend. The mixture was blended
at 34 revolutions per minute (rpm) for 1 h using the Turbula Mixer (T2 F Nr. 120,942
Switzerland) to make a blend of chemical composition shown in Table 1.
The As content in the blend was then measured using the inductively coupled
plasma atomic emission spectroscopy (ICP-AES) and found to be 0.74 wt%. X-ray
fluorescence (XRF) analysis was done on the ore to have a general understanding
of the other components in the ore. The other components in the ore were found
to be TiO2 (s), Cr2 O3 (s), NiO(s), P2 O5 (s), and Na2 O(s). The phases in the hematite
ore-As2 O3 blend were identified by X-ray diffraction (XRD) (Bruker AXS) with a
Cu tube at a scan angle of 20° to 80°, scanning rate of 2°/min, a voltage of 40 kV,
and a current of 40 mA. The results were as shown in Fig. 2.
The Fe-containing phase identified in the hematite ore-As2 O3 blend was the
Fe2 O3 (s). As existed as a As2 O3 dimer, As4 O6 (s) in the blend. The gangue existed
in the ore as the SiO2 (s), MgO·SiO2 (s), and the Al2 O3 ·SiO2 (s).
Table 1 Chemical composition of hematite ore-As2 O3 blend (wt%)
Analyte Total Fe FeO Fe2 O3 SiO2 Al2 O3 CaO S K MgO MnO As LOI Others
Fe2 O3 ore- As2 O3 blend 54.36 0.25 77.45 6.08 1.59 0.36 0.07 0.01 0.09 0.42 0.74 9.49 12.94
Volatilization Behavior of Arsenic from a Hematite Ore During …
175
176 E. K. Chiwandika and S.-M. Jung

Fig. 2 XRD pattern of the

hematite ore-As2 O3 blend

Experimental Procedure

Loss of Mass Profile Based on the TG-DSC

The loss of mass profile in the ores was determined by the TG-DSC (TGA/DSC 1
Star system Mettler Toledo). About 100 mg of the hematite ore only and hematite
ore-As2 O3 blend were placed in an aluminium oxide crucible and placed inside the
TG-DSC furnace, and the temperature was raised from room temperature to 1573 K at
a rate of 10 K/min in argon atmosphere. An Ar atmosphere at a flow rate of 70 mL/min
was used during this procedure. The loss of mass profile was also investigated using
the vertical tube furnace where both the ore and the ore-As2 O3 blend were pressed
at 20 MPa into cylindrical disc pellets weighing 1 g. Two pellets with total mass
of 2 g were placed into a platinum crucible (basket) and suspended in the vertical
tube furnace using a platinum wire, and temperature was raised at 10 K/min to target
temperature in an Ar atmosphere at 1 L/min. When the target temperature ranging
from 298 to 1623 K was reached, the crucible was pulled to the cold end of the
furnace and cooled to room temperature in Ar atmosphere. After cooling the final
mass of the pellets was measured. Pellets heated to temperatures above 1273 K were
crushed to fine powder and supplied for ICP analyses to determine the amount of
As remaining. XRD analyses were also performed on these heated samples, and
some of the pellets were prepared for electron probe microanalysis (EPMA; (Joel
(Japan)/JXA-8530F) analyses for verification of the existence of phases identified by
XRD. Elemental mapping by EPMA analyses were done using a voltage of 20 keV
and a beam current of 50 nA at a magnification of 400X. Arsenic volatilized was the
evaluated using data obtained from ICP using Eq. (11)
Volatilization Behavior of Arsenic from a Hematite Ore During … 177

As f
Asi
Wi
− Wf
% As volatilized = × 100 (11)
Asi
Wi

where Asi is the initial As content in the sample, Asf is the final As content in the
sample, W i is the initial weight of the sample, and W f is the final sample weight.

Results and Discussion

Mass Loss as a Function of Temperature

The mass loss profile from the hematite ore and from the hematite-As2 O3 blend were
investigated using the TG-DSC where the weight loss form the hematite ore only
was used as the reference. The results were as shown in Fig. 3a and b.
The mass loss profiles from both the ore and from the blend showed a similar
trend. This mass loss was from the volatilization of all the volatile matter in the ores.
Comparison between Fig. 3a and b showed that the mass loss from the ore-As2 O3
blend was higher than that from the ore alone. This led to the conclusion that part
of the difference in mass loss curve between that shown in Fig. 3a and b might be
attributed to the presence of the As2 O3 since the mass loss curve of the hematite ore
alone was used as the baseline. In addition, the loss on ignition (LOI) of the hematite
ore was 7.10 wt% while that of hematite-As2 O3 was 9.49 wt%. The difference was
about 2.39 wt% and part of this might be from the volatilized As2 O3 . However,
more research should be done after calibrating the QMS by nebulizing a solution
of known As concentration for online analyses. This will enable the quantification
of the exact amount of As volatilized since a natural ore with other diverse volatile
materials was used. The mass loss profiles based on the vertical tube furnace shown

Fig. 3 Plot of mass loss and volatilization rate with increasing temperature-using data from TG-
DSC of: a hematite ore only and b hematite ore-As2 O3 blend [11]
178 E. K. Chiwandika and S.-M. Jung

Fig. 4 Plot of mass loss with

increasing temperature-using
data based on the vertical
tube furnace [11]

in Fig. 4 showed some similar trend with that obtained from the TG-DSC data in
Fig. 3. Thus, the change in phases were therefore investigated using samples heated
non-isothermally in the vertical tube furnace.
The mass loss profile shown in Fig. 3a and b must be related to the volatiliza-
tion rate and the bonding strength holding the volatile material in the pellet. The
volatilization rate curve was plotted by taking the derivative of the mass loss profile
shown in Fig. 3a and b. A closer inspection of the volatilization rate curves shown
in Fig. 3a and b showed some specific peaks at 473 to 673 K, 700 to 800 K, 1000 to
1100 K, and above 1400 K on the volatilization rate curves of both the hematite ore
and the hematite ore-As2 O3 blend. This led to the conclusion that the volatilization
of the volatile matter from the ore depends on the chemical bonds holding them in
the ore [4].
The peak at 473 to 673 K in Fig. 3a might be due to dehydration of the ore while
that in Fig. 3b might be due to dehydration along with the As lost from the phase
transformation of As2 O3 (s). The phase transformation of As2 O3 (s) from the arseno-
lite to the claudetite was predicted to occur at around 503 K thermodynamically. This
phase transformation temperature is expected to occur at 453 K in the presence of
water vapor [1]. Furthermore, the sublimation of As2 O3 (s) occurs at around 588 K
[12]. The TG-DSC analysis of the As2 O3 (s) reagent grade was done to provide more
evidence as shown in Fig. 5.
Figure 5 shows that the mass loss started at around 503 K corresponding to the
phase change of the As2 O3 (s) and increased rapidly up to temperatures of just above
600 K. This acceleration might have been caused by the sublimation of As2 O3 (s)
that was reported to occur at around 588 K [12].
Part of the mass loss of the hematite ore-As2 O3 blend at 700 to 800 K in Fig. 3b
was from the evaporation of the remaining As2 O3 (s) among other reactions, the
boiling temperature of As2 O3 (s) was reported to be around 703 K [12] Fig. 5 shows
Volatilization Behavior of Arsenic from a Hematite Ore During … 179

Fig. 5 Plot of mass loss and volatilization rate with increasing temperature-using data from TG-
DSC of As2 O3 [11]

a characteristic peak between 700 to 800 K corresponding to the evaporation of

the As2 O3 (s). The increase in volatilization rate observed above 1400 K shown in
Fig. 3a and b might be due to the loss of O2 (g) from the thermal decomposition of
hematite that occurs above 1473 K [13]. Volatilization of As from the decomposition
of arsenates other than 3CaO·As2 O5 (s), 2CaO·As2 O5 (s), and AlAsO4 (s) above 1
373 K is also expected. It is believed that these were the cause for the increase in
mass loss observed above 1400 K in Fig. 3b.
Temperature range of 1373–1673 K is of great importance in iron ore sintering,
sample non-isothermally heated to this temperature range were crushed and the As
content analyzed using the ICP-AES to determine the amount of As lost from the
samples. Figure 6 shows the As volatilization trend from the ore.

Fig. 6 Arsenic volatilized 80

As volatilized (%)

from the sample with 70

increasing temperature
60
50
40
30
20
10
0
1200 1300 1400 1500 1600
Temperature (K)
180 E. K. Chiwandika and S.-M. Jung

The loss of As from the blend increased with increasing temperature and complete
removal of As from the ore was not accomplished even at 1623 K, Fig. 6. Part of
the As might have remained trapped inside the ore due to the interactions with other
metal oxide in the ore.

Phase Development with Increasing Temperature

In the sintering of iron ores, phases developed at temperatures higher than 1373 K
are of great importance. Figure 7a−d shows the XRD results of the hematite blend
after heating at 1373 K and above.
The phases identified at 1373 K and 1473 K were the Fe2 O3 (s), SiO2 (s),
MnAsO4 (s), 2CaO·As2 O5 (s), and AlAsO4 (s), at 1573 K, MnAsO4 (s) decomposed
and the Fe2 O3 (s), Fe3 O4 (s), SiO2 (s), 3CaO·As2 O5 (s), and AlAsO4 (s) were also
phases identified while the 3CaO·As2 O5 (s), 2CaO·As2 O5 (s), and AlAsO4 (s) phases
were the most stable As-containing phases at 1623 K. Figure 7c−d provided the
evidence of the thermal decomposition of hematite shown by the presence of Fe3 O4 (s)
peaks at 1573 K and 1623 K. The complete escape of the As from the ore was
complicated by the presence of other metallic oxides in the ore in agreement with
the thermodynamic prediction in Fig. 1.

Fig. 7 XRD analysis of the hematite ore-As2 O3 pellet after non-isothermal heating to: a 1373 K,
b 1473 K, c 1573 K, and d 1623 K [11]
Volatilization Behavior of Arsenic from a Hematite Ore During … 181

Fig. 8 Elemental mapping of the pellet non-isothermally heated to: a 1373 K, b 1473 K, c 1573 K,
and d 1623 K [11]

Elements Distribution in the Sintered Pellets with Increasing

Temperature

The EPMA (Joel JXA-8530F) mappings were performed to provide more evidence
that As remained trapped inside the sintered ore even after heating to temperatures
above 1273 K, Fig. 8a–d
It can be clearly shown in Fig. 8a–d that As was highly concentrated in the same
area as Ca, and Al shown by the points marked in red on the EPMA mappings in agree-
ment with the thermodynamic predictions in Fig. 1. It can be safely concluded that
2CaO·As2 O5 (s), 3CaO·As2 O5 (s), and the AlAsO4 (s) were likely to be the most stable
As-containing phase basing on the XRD, EPMA, and thermodynamic predictions.
The results of this research only provide the basic understanding of the behavior
of As during non-isothermal heating in an Ar atmosphere but in actual iron ore
sintering process, CaO(s) is intentionally added to produce fluxed and semi-fluxed
sinters in a gaseous atmosphere that varies from reducing during the burning of
the coal to oxidizing during cooling. Further, research should be done in reducing
and oxidizing conditions and on the kinetics of As volatilization to fully understand
arsenic volatilization during sintering. The production of acidic sinters should be
investigated since lower amount of As is expected to be trapped inside the sinter
in the form of 2CaO·As2 O5 (s) or 3CaO·As2 O5 (s). Some preliminary results on the
182 E. K. Chiwandika and S.-M. Jung

Fig. 9 Plot of
ln[(1 − (1 − α)1/3 )2 /T 2 ]
versus 1/T based on
non-isothermal As
volatilization

kinetics of non-isothermal As volatilization [14] based on results shown in Fig. 7 are

shown in Fig. 9.
The activation energy, E A , and the pre-exponential factor, A, were calculated to
be 17.0 kJ/mol and 246.6 min−1 , respectively. The evaluated activation energy value
falls within the range of the diffusion-controlled mechanism, 8.37–20.93 kJ/mol,
reported in the literature [15]. However, more results are needed were the heating
rate is changed to fully validate the kinetics results. Further investigation, especially
in gold production on the use of CaO(s) or Al2 O3 (s) to hold As from the off gas
in the form of 2CaO·As2 O5 (s), 3CaO·As2 O5 (s), or AlAsO4 (s) to prevent pollution,
must be done since these arsenates can thermodynamically form and also appears to
be stable even at 1623 K.

Conclusions

(1) The volatilization of As and other volatile components in the ore increased
with temperature. The rate of volatilization may depend on the chemical bonds
holding the volatile materials in the ore.
(2) The presence of metal oxide like CaO or Al2 O3 have potential of preventing
the complete volatilization of As by the formation of the stable 2CaO·As2 O5 (s),
3CaO·As2 O5 (s), and AlAsO4 (s). These interactions with different metal oxide
components in the ore limited the complete escape of As from the pellet during
the non-isothermal heating process even at temperatures as high as 1623 K.
Volatilization Behavior of Arsenic from a Hematite Ore During … 183

(3) Further, investigations on the formation of the 2CaO·As2 O5 (s) or

3CaO·As2 O5 (s) in developing technologies for As pollution prevention by using
CaO(s) to trap gaseous arsenic compounds in the off gas are highly encouraged.
(4) Preliminary kinetic results showed the evaluated activation energy, E A , and
the pre-exponential factor, A, were 17.0 kJ/mol and 246.6 min−1 , respectively.
The evaluated activation energy value falls within the range of the diffusion-
controlled mechanism, however, more results are required for validation.

References

1. Safarzadeh MS, Miller JD, Haung HH (2014) The behavior of arsenic trioxide in non-ferrous
extractive metallurgical processing. Metallurgical Res Technol 111:95–105
2. Cheng R, Ni H, Zhang H, Zhang X, Bai S (2017) Mechanism research on arsenic removal from
arsenopyrite ore during a sintering process. Int J Miner Metall Mater 24(4):353–359
3. Grund SC, Hanusch K, Wolf HU (2005) Arsenic and arsenic compounds. Wiley-VCH Verlag
GmbH & Co. 4:1–43
4. Lui H, Pan W, Wang C, Zhang Y (2016) Volatilization of As during coal combustion based on
isothermal thermogravimetric analysis at 600–1500°C. Energy Fuels 30:6790–6798
5. Lu Q, Zhang S, Hu X (2011) Study on removal arsenic from iron ore with arsenic in sintering
process. Trans Tech Publication 284:238–241
6. Chakraborti N, Lynch DC (1983) Thermodynamics of roasting arsenopyrite. Metall Trans B
14:239–251
7. Contreras ML, Arostegui JM, Armesto L (2009) Arsenic interactions during co-combustion
process based on thermodynamic and equilibrium calculations. Fuel 88:539–549
8. Harvey T, Pownceby MI, Chen J, Webster NAS, Nguyen TBT, Matthews L, O’Dea D,
Honeyands T (2021) Effects of temperature, time, and cooling rate on the mineralogy,
morphology and reduction of iron ore sinter analogues. JOM 78:345–355
9. Mikhail SA, Turcotte AM (1992) Thermal decomposition of arsenopyrite in the presence of
calcium oxide. Thermochim Acta 212:27–37
10. Lee H (1999) Chemical thermodynamics for metals and materials. Imperial College Press,
London, p 287
11. Chiwandika EK, Jung SM (2022) Volatilization behavior of arsenic from a hematite ore during
non-isothermal heating in argon atmosphere. J Sustain Metallurgy 8:458–467
12. Helsen L, Bulck EV, Bael MKV, Vanhoyland G, Mullens J (2004) The behavior of arsenic
oxide (AS2 O5 and As2 O3 ) and the influence of reducing agents (glucose and activated carbon).
Thermochim Acta 414:145–153
13. Kim BS, Xing L, Qu Y, Wang C, Shao L, Zou Z, Song W (2020) Kinetic study on thermal
decomposition behavior of hematite ore fines at high temperature. Metall and Mater Trans B
51:395–406
14. Kim BS, Kim EY, Kim CK, Lee HI, Sohn JS (2008) Kinetics of oxidative roasting of complex
copper concentrate. Minerals Trans 49(5):192–1198
15. Biswas AK (1981) Principles of blast furnace ironmaking. Cootha Publishing House, Brisbane,
Australia, pp 34–39
Comprehensive Recovery of Elemental
Sulfur and Sulfide Minerals
from Pressure Acid Leaching Residue
of Zinc Sulfide Concentrate
with an Integrated Flocculation
Flotation-Hot Filtration Process

Guiqing Liu, Bangsheng Zhang, Zhonglin Dong, Fan Zhang, Fang Wang,
Tao Jiang, and Bin Xu

Abstract An efficient flocculation flotation-hot filtration process was developed to

recover elemental sulfur and sulfide minerals from a pressure acid leaching residue of
zinc sulfide concentrate. The particle size distribution of the residue indicated that the
portion of thin particle with the size of -37 um reached 67.65 and 36.27% of sulfur
in the residue was distributed in this size fraction, necessitating the flocculation
flotation for recovering the elemental sulfur and sulfide minerals. 94.59% of the
sulfur was recovered with the flotation process of one-time blank rougher, two-time
agent-added roughers, and two-time blank cleaners with Z-200 as the collector and
polyacrylamide as the flocculant. After hot filtration for the flotation concentrate at
145 °C for 2 h, 85.32% of the elemental sulfur in the concentrate was recovered, and
its product purity reached 98.64%. The filter cake of sulfide minerals can be returned
to the leaching stage for zinc recovery.

Keywords Pressure acid leaching residue of zinc sulfide concentrate · Elemental

sulfur · Sulfide minerals · Flocculation flotation · Hot filtration

G. Liu
School of Metallurgy, Northeastern University, Shenyang 110819, Liaoning, China
G. Liu · B. Zhang · F. Zhang · F. Wang
Jiangsu BGRIMM Metal Recycling Science and Technology Co. Ltd, Xuzhou 221121, Jiangsu,
China
Z. Dong · T. Jiang · B. Xu (B)
School of Minerals Processing and Bioengineering, Central South University, Changsha 410000,
Hunan, China
e-mail: xubincsu@[Link]
B. Xu
Peace Building, School of Minerals Processing and Bioengineering, Central South University,
Changsha 410083, Hunan, China

© The Minerals, Metals & Materials Society 2023 185

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
186 G. Liu et al.

Introduction

Zinc is an important non-ferrous metal and has been widely applied in many fields
such as automobile, battery, and construction [1]. Traditionally, zinc is extracted by
pyro-hydrometallurgical process from sphalerite (i.e., zinc sulfide) [2, 3]. However,
high operation cost is required because the process is conducted at high temperatures
[4–7]. In addition, hazardous gas containing SO2 and some heavy metals (e.g., Hg
and Pb) will be generated, which causes a great threat to the environment and human
beings [8–12].
In order to avoid the disadvantages of pyro-hydrometallurgical processes,
hydrometallurgical processes have been developed. Among the alternatives, oxygen
pressure acid leaching process has attracted the most attention [13], and the related
reaction is shown in Eq. (1). It can be seen that apart from ZnSO4 , part of sulfur in zinc
sulfide is converted into elemental sulfur [14–16]. Thus, the residue from oxygen
pressure acid leaching of sphalerite can be regarded as a resource for recovering
elemental sulfur [17].

2ZnS + 2H2 SO4 + O2 = 2ZnSO4 + 2H2 O + 2S (1)

It should be noted that some undissolved sulfides in the sphalerite during leaching
also enter into the residue. Thus, the separation and recovery of elemental sulfur and
sulfides have become a vital issue. Froth flotation is a predominant beneficiation tech-
nique that can selectively separate target minerals from gangue minerals by making
use of their difference in natural hydrophobicity [18]. However, two problems needed
to be solved when froth flotation was adopted. First, oxygen pressure acid leaching
residue generally possesses a very small particle size, and thus, it is difficult to recover
it by conventional froth flotation. Second, the elemental sulfur and sulfides have very
close hydrophobicity, and thus, their selective flotation separation is difficult.
In this study, an efficient flocculation flotation-hot filtration process was devel-
oped to recover elemental sulfur and sulfide minerals from a pressure acid leaching
residue of zinc sulfide concentrate. One-time agent-added rougher flotation was
first conducted to ascertain the optimal floatation condition. Then, open-circuit
and closed-circuit flocculation flotation experiments were successively performed
to ascertain flotation flow sheet and corresponding test indexes. Finally, hot filtra-
tion was adopted to separate elemental sulfur from sulfide minerals. High-purity
elemental sulfur product was recovered, and sulfide minerals can be returned to the
leaching stage for zinc recovery, and thus, the comprehensive recovery of the pressure
acid leaching residue was realized.
Comprehensive Recovery of Elemental Sulfur and Sulfide Minerals … 187

Materials and Methods

Materials and Reagents

The raw material in this study was oxygen pressure leaching residue of zinc sulfide
concentrate from a metallurgical plant in Qinghai Province. The residue was mixed
adequately with wet sampler for the next tests and experiments. The chemical compo-
sition of leaching residue in Table 1 showed that sulfur was the most abundant element
whose content achieved 44.41%. The phase analysis of sulfur in Table 2 indicated
that 37.73% of the sulfur in the residue existed in the form of elemental sulfur. The
mineral composition of the residue was detected by X-ray diffraction (XRD), and
the result is shown in Fig. 1. The main minerals are elemental sulfur, jarosite, pyrite,
and quartz.
The particle size distribution of the residue is shown in Table 3. The sulfur distri-
bution ratio in the size fraction of -37 µm was the highest (36.27%), and the sulfur
grade in the size fraction of −74 + 44 µm was the highest (89.38%), and the sulfur
distribution ratio in this fraction also achieved 32.71%. Thus, sulfur was mainly
distributed in the two fractions. The result indicated that the particle size of leaching
residue was very fine, and flocculation flotation may be a feasible method to recover
the sulfur in the residue.
All the reagents used in this study such as O-isopropyl-N-ethyl thionocarbamate
(Z-200), ammonium dibutyl dithiophosphate (ADDTP), and ethyl xanthate (EX)
were of analytical grade, and tap water was used in all experiments.

Table 1 Chemical composition of the oxygen pressure leaching residue (%)

Element Cu Pb Zn S Fe SiO2 MgO
Content 0.055 2.24 2.28 44.41 21.23 14.34 0.33
Element Al2 O3 CaO As Mn Cd Ag*
Content 2.01 3.40 0.12 0.11 0.016 96.5
Note “*” unit is g/t

Table 2 Sulfur phase distribution of the oxygen pressure leaching residue (%)
Sulfur phase Elemental sulfur Sulfide Sulfate Total
Content 37.73 1.94 2.91 42.58
Distribution 81.97 12.03 6.00 100
188 G. Liu et al.

900
★ ●
★ ★ —Elemental sulfur
800
●— jarosite
◆ —pyrite
700
▼—quartz
●
◆

★
Intensity (a.u.)

600

★
500 ● ★
★ ★
★
400 ●
● ▼★ ●
★ ●
★ ● ●
▼ ★ ★ ●
◆

●
★ ★ ★★
◆

300 ★
●●
◆

★★ ★ ★
◆

◆
★ ★

200

0 10 20 30 40 50 60 70

Two theta (°)

Fig. 1 XRD spectrogram of the oxygen pressure leaching residue

Table 3 Particle size distribution of the oxygen pressure leaching residue

Size fraction (mesh) Weight (g) Yield (%) Grade (%) Distribution ratio (%)
+100 1.8 0.35 67.94 0.53
−100+200 47.2 9.07 85.79 17.39
−200+325 85.2 16.37 89.38 32.71
−325+400 34.2 6.57 89.19 13.10
−400 352.2 67.65 23.97 36.27
Total 520.6 100.00 44.72 100.00

Flotation Experiment

The flotation experiments of oxygen pressure acid leaching residue were carried
out in an XFD-63 flotation machine. First, the tap water and residue were put into
the cell, and the pH value of formed pulp was regulated to around 8.0 with HCl
or NaOH solution. After that, the collector and flocculant polyacrylamide (if used)
were successively added into the pulp which was separately agitated at 1650 rpm
for 2 min. The aeration flotation was performed for 5 min, and the concentrate and
tailing were dried for sulfur content analysis.
Comprehensive Recovery of Elemental Sulfur and Sulfide Minerals … 189

Fig. 2 Flow sheet of Feed

one-time rougher flotation 3' Z-200 40 g/t
3' Polyacrylamide
Rougher flotation 5'

Concentrate Tailing

Analytical Methods

The elemental composition of the oxygen pressure leaching residue was analyzed
with inductively coupled plasma-atomic emission spectrometer (ICP-AES). The
sulfur phase composition was detected with a chemical selective dissolution method
on the basis of the different dissolution behavior of different sulfur phase in various
solvents. The particle size distribution was ascertained by a wet screen analysis. The
mineralogical composition was determined by X-ray diffractometer.

Results and Discussion

One-Time Rougher Flocculation Flotation

Effect of Flocculant Dosage

Polyacrylamide was used as the flocculant, and the effect of its dosage on the grade
and recovery of sulfur in flotation concentrate was first investigated at pulp concen-
tration 35%. The used flow sheet is shown in Fig. 2, and the results are indicated in
Fig. 3.
With the increase of polyacrylamide dosage, the sulfur grade of the concen-
trate first increased and achieved its maximum 73.93% with 50 g/t polyacrylamide.
However, when polyacrylamide dosage was over this value, the sulfur started to
decline. The sulfur recovery rose with increased polyacrylamide dosage, but as the
dosage was beyond 50 g/t, no obvious increase of the sulfur recovery was obtained.
Therefore, the optimal polyacrylamide dosage was 50 g/t.

Effect of Type of Collector

The effect of collector type on the grade and recovery of sulfur in flotation concentrate
was also studied at a pulp concentration 35%. The used flow sheet is presented in
Fig. 4, and the results are shown in Fig. 5.
From the figure, when EX was adopted, the sulfur grade arrived at 75.26%, but
its recovery was only 80.02%. As for ADDTP, the sulfur recovery was enhanced to
190 G. Liu et al.

86

74
84

72 82

Sulfur recovery (%)

Sulfur grade (%)

80
70
78

68
76

66 74

72
64
30 40 50 60 70 80

Polyacrylamide dosage (g/t)

Fig. 3 Effect of polyacrylamide dosage on the sulfur grade and recovery of flotation concentrate

Fig. 4 Flow sheet of Feed

one-time rougher flotation 3' Collector 40 g/t
3' Polyacrylamide 50 g/t
Rougher flotation 5'

Concentrate Tailing

90
Sulfur grade
Sulfur recovery 83.24
81.82
80.02
80

75.26
74.02 73.93

70
%

60

50
EX ADDTP Z-200

Collector type

Fig. 5 Effect of collector type on the sulfur grade and recovery of flotation concentrate
Comprehensive Recovery of Elemental Sulfur and Sulfide Minerals … 191

80 90

88

Sulfur recovery (%)

78
Sulfur grade (%)

86

76
84

82
74

80
40 60 80 100

Z-200 dosage (g/t)

Fig. 6 Effect of Z-200 dosage on the sulfur grade and recovery of flotation concentrate

81.82%, but its sulfur grade was lower than that of EX. In comparison, 83.24% of
the sulfur was recovered, and the sulfur grade of the concentrate achieved 73.93%
when Z-200 was used. Thus, Z-200 exhibited the optimal flotation performance on
the sulfur in the oxygen pressure leaching residue. Therefore, Z-200 was selected as
the flotation collector.

Effect of Collector Dosage

The effect of Z-200 dosage on the grade and recovery of sulfur in flotation concentrate
was investigated at a pulp concentration 35% and polyacrylamide dosage of 50 g/t,
and the results are indicated in Fig. 6.
With the increase of Z-200 dosage, the sulfur grade and recovery of the concentrate
first increased and achieved their maximums 77.68 and 87.49% at 80 g/t. However,
when Z-200 dosage was beyond this value, both the sulfur grade and recovery began
to decline. This is because Z-200 also adsorbed on the surfaces of gangue minerals
when its dosage was too high. As a result of this, these gangue minerals were also
floated with elemental sulfur and sulfide minerals, leading to a decrease of sulfur
grade and recovery. Thus, the optimal Z-200 dosage was 80 g/t.

Effect of Pulp Concentration

The effect of pulp concentration on the grade and recovery of sulfur in flotation
concentrate was investigated at Z-200 dosage 80 g/t and polyacrylamide dosage of
50 g/t, and the results are displayed in Fig. 7.
192 G. Liu et al.

85 100

80

Sulfur recovery (%)

95
Sulfur grade (%)

75

90

70

65 85
20 25 30 35

Pulp concentration (%)

Fig. 7 Effect of pulp concentration on the sulfur grade and recovery of flotation concentrate

Clearly, pulp concentration had a significant influence on the grade and recovery
of sulfur. With the increase of pulp concentration from 20 to 25%, both the grade and
recovery of sulfur increased. Nevertheless, when the pulp concentration was over
25%, the sulfur recovery started to decrease although the sulfur grade continued to
increase. Thus, the optimal pulp concentration was 25%.

Open-Circuit Flocculation Flotation

Based on the above conditional tests and further flotation flowchart experiments, the
flow sheet of open-circuit flocculation flotation was ascertained, as is shown in Fig. 8,
and the results are shown in Table 4. Here, the purpose of adding one-time blank
rougher before agent-added rougher was to allow the easily floatable elemental sulfur
and sulfide to adhere to the surface of the foam by virtue of their hydrophobicity.
As a result, part of the elemental sulfur and sulfide could be recovered, which was
beneficial to save the reagent dosage in the subsequent flotation as well as improve
the product purity. It can be seen that the open-circuit flocculation flotation process
consisted of one-time blank rougher, two-time agent-added roughers, and two-time
blank cleaners. Under the optimal conditions, the sulfur grade of the concentrate
reached 90.16%. However, the sulfur recovery was only 58.01%.
Comprehensive Recovery of Elemental Sulfur and Sulfide Minerals … 193

Feed

Blank flotation

Blank rougher flotation 5'

3' Z-200 80 g/t

3' Polyacrylamide 50 g/t

Rougher flotation Ⅰ 3'

3' Z-200 48 g/t

3' Polyacrylamide 30 g/t

Rougher flotation Ⅱ 3'

Tailing

Cleaner flotation Ⅰ 5'

Cleaner flotation Ⅱ 3' Middling 1

Concentrate Middling 2

Fig. 8 Flow sheet of open-circuit flocculation flotation

Table 4 Test results of

Product Yield (%) Grade (%) Recovery (%)
open-circuit flocculation
flotation of the leaching Concentrate 35.60 90.16 58.01
residue Middling 1 13.81 59.69 14.90
Middling 2 24.97 51.82 23.38
Tailing 25.62 8.02 3.71
Feed residue 100.00 55.33 100.00

Closed-Circuit Flotation Experiment

To investigate the effect of middling return on sulfur flotation, a closed-circuit flota-

tion experiment was carried out. The flowchart is shown in Fig. 9, where the tailing of
the 1st cleaner (middling 1) was returned to the blank rougher, and the tailing of the
2nd cleaner (middling 2) went back to the 1st cleaner. The results of closed-circuit
flotation in Table 5 indicated that the sulfur recovery was increased to 94.59%, and
its grade also arrived at 87.92%. Therefore, after closed-circuit flocculation flotation,
most of the sulfur in the leaching residue was recovered, and the sulfur grade was
also high.
194 G. Liu et al.

Feed

Blank flotation

Blank rougher flotation 5'

3' Z-200 80 g/t

3' Polyacrylamide 50 g/t

Rougher flotation Ⅰ 3'

3' Z-200 48 g/t

3' Polyacrylamide 30 g/t

Rougher flotation Ⅱ 3'

Tailing

Cleaner flotation Ⅰ 5'

Cleaner flotation Ⅱ 3'

Concentrate

Fig. 9 Flow sheet of closed-circuit flocculation flotation of the leaching residue

Table 5 Test results of

Product Yield (%) Grade (%) Recovery (%)
closed-circuit flocculation
flotation of the leaching Concentrate 60.98 87.92 94.59
residue Tailing 39.02 7.86 5.41
Feed residue 100.00 56.68 100.00

Hot Filtration of Flotation Concentrate

The result of sulfur phase analysis for the concentrate is displayed in Table 6.
81.16% of the sulfur in the concentrate existed as elemental sulfur, i.e., the purity of
elemental sulfur concentrate product achieved 81.16%. However, the sulfide content
also arrived at 6.37%, which decreased the elemental sulfur grade. In order to obtain
the elemental sulfur with higher purity, hot filtration process was conducted for the
concentrate in a home-made equipment.

Table 6 Sulfur phase distribution of the elemental sulfur concentrate product (%)
Sulfur phase Elemental sulfur Sulfide Sulfate Total
Content 81.16 6.37 0.39 87.92
Distribution 92.31 7.25 0.44 100
Comprehensive Recovery of Elemental Sulfur and Sulfide Minerals … 195

Table 7 Sulfur phase distribution of the elemental sulfur product from the hot filtration of flotation
concentrate (%)
Sulfur phase Elemental sulfur Sulfide Sulfate Total
Content 98.64 0.76 0.32 99.72
Distribution 98.92 0.76 0.32 100

After repeated conditional test, the optimal hot filtration temperature and time
were separately 145 °C and 2 h, under which 85.32% of the elemental sulfur in
the concentrate was recovered in the form of hot filtrate. The result of sulfur phase
analysis for the elemental sulfur product in Table 7 indicated that the elemental sulfur
purity reached 98.64%. Therefore, after hot filtration, a high-purity elemental sulfur
product could be obtained. The filter cake contained abundant sulfide minerals that
can be returned to the leaching stage for zinc recovery.

Conclusions

The present work put forward an efficient flocculation flotation-hot filtration process
for recovering elemental sulfur and sulfide minerals from a pressure acid leaching
residue of zinc sulfide concentrate. The chemical composition and phase analyses
showed that the sulfur content achieved 44.41 and 37.73% of the sulfur existed as
elemental sulfur. The mineral composition analysis showed that the main minerals
are elemental sulfur, jarosite, pyrite, and quartz. The particle size distribution analysis
indicated that the portion of thin particle with the size of -37 um reached 67.65%.
When Z-200 was used as the collector and polyacrylamide as the flocculant, after
closed-circuit flocculation flotation consisting of one-time blank rougher, two-time
agent-added roughers, and two-time blank cleaners, 94.59% of the sulfur in the
residue was recovered hot filtrate, and the sulfur grade arrived at 87.92%. The flotation
concentrate was subjected to hot filtration; 85.32% of the elemental sulfur in the
concentrate was recovered at temperature 145 °C and time 2 h, and the purity of
obtained elemental sulfur product reached 98.64%. The filter cake containing sulfide
minerals was returned to the leaching stage for zinc recovery. Thus, after the process
of flocculation flotation-hot filtration of flotation concentrate, the elemental sulfur
and sulfide minerals in the pressure acid leaching residue of zinc sulfide concentrate
were comprehensively recovered.

Acknowledgements This research was funded by National Key Research and Development
Program of China (Nos. 2018YFC1902005 and 2018YFC1902006).
196 G. Liu et al.

References

1. Halfyard JE, Hawboldt K (2011) Separation of elemental sulfur from hydrometallurgical

residue: a review. Hydrometallurgy 109:80–89
2. Wang ZY, Cai XL, Zhang ZB, Zhang LB, Wang SX, Peng JH (2015) Separation and enrichment
of elemental sulfur and mercury from hydrometallurgical zinc residue using sodium sulfide. T.
Nonferr Metal Soc 25:640–646
3. Pacyna EG, Pacyna JM, Sundseth K, Munthe J, Kindbom K, Wilson S, Steenhuisen F, Maxson
P (2010) Global emission of mercury to the atmosphere from anthropogenic sources in 2005
and projections to 2020. Atmos Environ 44:2487–2499
4. Nriagu JO, Pacyna JM (1988) Quantitative assessment of worldwide contamination of air,
water and soil by trace metals. Nature 333:134–139
5. Li GH, Feng XB, Li ZG, Qiu GL, Shang LH, Liang P, Wang DY, Yang YK (2010) Mercury
emission to atmosphere from primary Zn production in China. Sci Tot Environ 408:4607–4612
6. Li GH, Feng XB, Qiu GL (2008) Environmental mercury contamination of an artisanal zinc
smelting area in Weining County, Guizhou, China. Environ Pollut 154:0–31
7. Xu ZF, Qiu DF, Lu HM, Wang HB (2005) Review on research of oxidic-acidic pressure of zinc
concentrates. Nonferr. Metal. 57:101–105 (in Chinese)
8. Liu SP, Wang HB, Jiang KX, Zhang BS (2009) New development of zinc hydrometallurgical
in China. Min Metall 18:25–279
9. Dong ZL, Jiang T, Xu B, Yang YB, Li Q (2019) An eco-friendly and efficient process of low
potential thiosulfate leaching-resin adsorption recovery for extracting gold from a roasted gold
concentrate. J Clean Prod 229:387–398
10. Li HL, Wu XY, Wang MX, Wang J, Wu SK, Yao XL, Li LQ (2014) Separation of elemental
sulfur from zinc concentrate direct leaching residue by vacuum distillation. Sep Purif Technol
138:41–46
11. Gu Y, Zhang TA, Liu Y, Mu WZ, Zhang WG, Dou ZH, Jiang XL (2010) Pressure acid leaching
of zinc sulfide concentrate. Trans Nonferrous Met Soc China 20:s136–s140
12. Dong ZL, Jiang T, Xu B, Yang J.K, Chen Y.Z, Yang YB, Li Q (2020) Comprehensive recoveries
of selenium, copper, gold, silver and lead from a copper anode slime with a clean and economical
hydrometallurgical process. Chem Eng J 393
13. Padilla R, Vega D, Ruiz MC (2010) Pressure leaching of sulfidized chalcopyrite in sulfuric
acid–oxygen media. Hydrometallurgy 86:80–88
14. Xu B, Chen YZ, Dong ZL, Jiang T, Zhang BS, Liu GQ, Yang JK, Li Q, Yang YB (2021)
Eco-friendly and efficient extraction of valuable elements from copper anode mud using an
integrated pyro-hydrometallurgical process. Resour Conserv Recycl 164
15. Liu FP, Wang JL, Peng C, Liu ZH, Wilson BP, Lundström M (2019) Recovery and separation
of silver and mercury from hazardous zinc refinery residues produced by zinc oxygen pressure
leaching. Hydrometallurgy 185:38–45
16. Fan YY, Liu Y, Niu LP, Jing TL, Zhang TA (2019) Separation and purification of elemental
sulfur from sphalerite concentrate direct leaching residue by liquid paraffin. Hydrometallurgy
186:162–169
17. Rao S, Wang DX, Liu ZQ, Zhang KF, Cao HY, Tao JZ (2019) Selective extraction of zinc,
gallium, and germanium from zinc refinery residue using two stage acid and alkaline leaching.
Hydrometallurgy 183:38–44
18. Qin SC, Jiang KX, Wang HB, Zhang BS, Wang YF, Zhang XD (2020) Research on behavior
of iron in the zinc sulfide pressure leaching process. Minerals 10:224–239
Part V
Additive Manufacturing and Materials
First Principles
Automatic Process Mapping for Ti64
Single Tracks in Laser Powder Bed
Fusion

Toby Wilkinson, Massimiliano Casata, and Daniel Barba

Abstract Using an unsupervised convolutional neural network classifier, an auto-

mated workflow to generate the process map for printing Ti-6Al-4 V with laser
powder bed fusion with minimal human supervision is proposed. Single scan vectors
using a range of laser powers and scan speeds were printed on a bare Ti-6Al-4 V
baseplate, which were then imaged using optical microscopy without further material
preparation steps. After resizing and thresholding, the resulting dataset was used to
train the neural network into automatically differentiating the tracks into categories.
Post-analysis reveals that the model can differentiate between commonly observed
track morphologies and map out the viable processing window automatically for the
alloy.

Keywords Additive manufacturing · Titanium alloy · Machine learning · Laser

powder bed fusion (LPBF)

Introduction

Metal additive manufacturing (AM) has become a subject of immense academic and
commercial interest in recent years due to its ability to process a wide range of alloys
for use in many different sectors and with high geometric complexity. The material
feedstock is in powder form and is melted by a laser in a layer-by-layer fashion,
gradually forming a 3D object. While conceptually simple, the process involves
many different physical phenomena including the multi-phase flow of the melt pool,
residual stresses from high thermal gradients, and the resulting microstructure of the
material. All these factors, plus many more uncited, play a key role in determining
the quality and mechanical properties of the final part.
Of all the possible variables involved in the laser powder bed fusion (LPBF)
process, the laser power (P) and scan speed (v) are two of the most important, having
a significant impact on the resulting part quality. LPBF is a hierarchical process

T. Wilkinson (B) · M. Casata · D. Barba

E.T.S de Ingeniería Aeronáutica y del Espacio, Universidad Politécnica de Madrid, Madrid, Spain
e-mail: [Link]@[Link]

© The Minerals, Metals & Materials Society 2023 199

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
200 T. Wilkinson et al.

where individual scan vectors are the fundamental unit of construction, and defects
caused by inappropriately chosen processing parameters will propagate through the
hierarchy and affect the 3D part performance. Such defects include keyhole and
lack of fusion porosity and spatter [1–3]. As a result, optimisation of the processing
parameters should begin at the fundamental scan vector level.
The traditional optimisation process is expensive in terms of time, labour, and
materials costs, requiring the printing of many 3D samples and their subsequent
metallurgical analysis to determine reasonable parameters. These density values of
these sample cubes are usually used to then select the “optimal” printing parameters.
Other methods that focus on the single-track morphology using experimental [4] or
computational [5] results use a P–v plot to investigate the effect of these parameters
on the melt pool dimensions or defects. These results are then used to determine
the printing parameters. This work will also use a P–v plot to investigate the track
morphologies but using machine learning (ML) instead of manual methods.
ML is a type of artificial intelligence (AI) that uses large quantities of data to
train a computer model to accomplish a complex task that it may not have been
explicitly written for. Depending on the problem to be solved and the availability of
training data, algorithms can learn in three different ways: supervised, unsupervised,
or with reinforcement. AM is a manufacturing method that lends itself well for use
in combination with ML, given the digital nature of many processing steps within
the AM manufacturing chain. In fact, there is a rich history of ML being used to
solve various problems within AM, including process parameter optimisation [6–8]
and part design [9].

Methods

Computational Methodology

Invariant Information Clustering

The chosen paradigm for the unsupervised clustering of the images was taken from
[10] and is called Invariant Information Clustering (IIC). This method aims to train
a neural network (NN) to learn representations of the input dataset that are invariant
under image transformations. The full mathematical description of the method can
be found in the corresponding article but will be summarised briefly here.
IIC clustering uses contrastive learning, which means a pair of images forms
the input, and the loss is calculated by comparing the outputs of the NN for each
image in the pair in some way. This version uses positive image pairs as training
data, i.e., images that should be classified as the same object by the neural network.
Since we are performing unsupervised clustering, this kind of image pairing is rarely
available, so the authors propose the use of synthetic augmentation to modify the
original image in such a way that it remains recognisable, but to the neural network
Automatic Process Mapping for Ti64 Single Tracks in Laser Powder … 201

is a different image. Acceptable transformations include rotation, translation, and

cropping. Colour corrections, such as changing the brightness and contrast, are also
acceptable.
The output layer of the NN creates a discrete probability distribution across the
number of chosen categories for the input image using a SoftMax activation function.
This is the traditional activation function used by classification NNs. The loss func-
tion used by this network uses the concept of mutual information. In simple terms,
the mutual information between two variables is a way of measuring the dependence
of these variables on each other. What this means in this case, is that the mutual infor-
mation of two completely different images should be 0. As stated in [10], maximising
mutual information has the effect of making the representations of each image in a
pair the same, hence assigning them the same cluster label.

Neural Network Architecture

As described above, the chosen ML paradigm was an unsupervised convolutional

neural network (CNN). This is widely regarded as the industry standard network
type for machine vision applications, due to its efficient sharing of weights within
the convolutional layers. Famous convolutional architectures include AlexNet [11]
and the VGG configurations [12] which have shown impressive accuracies in large-
scale image classification tasks and continue to inspire the construction of models
today.
For this work, a custom CNN model was implemented using TensorFlow [13]
and Python 3.7, with various design choices inspired by models in literature and
common practice. The model architecture is shown in Fig. 1. As is standard practice
when using classifying images, the architecture has two main parts: the convolutional
network and a fully connected prediction head. Between the two, a global average
pooling layer [14] is used to vectorise the resulting feature maps before the fully
connected layers. This results in fewer parameters in the network and is considered
more “convolutional” in nature than the traditional flatten layers, since each feature
map then resembles a categorical variable.
The network begins with a rescaling layer, which simply normalises all pixel
values to between 0 and 1. This is standard practice when preparing data for various
ML methods since it ensures all inputs are distributed across the same range. For
neural networks, this may improve training performance, but it is important if using
pretrained networks whose training data was in this standard range.
In total, there are 12 convolution layers, each with a ReLU activation function [11]
and followed by a batch normalisation (BatchNorm) [15]. The convolution layers all
have a filter size of 3 × 3, with a stride of 1 and are padded, such that the output
shape matches the input shape. There are a total of three max pooling layers that
reduce the height and width of the feature maps by a factor of 2, and after each max
pooling layer, the number of filters in each convolution layer is increased by a factor
of two, starting at 32 and finishing at 256.
202 T. Wilkinson et al.

Fig. 1 Custom CNN architecture. Layer types are identified by the colours shown in the key. For
conciseness, the BatchNorm layers are not shown

For the prediction head, this has two hidden, fully connected layers, each with 512
neurons and ReLU nonlinearity, followed by BatchNorm layers. The final, output
layer has the same number of neurons as the desired number of categories, in this
case, 5, and has a SoftMax activation function to provide a probability distribution.
The following augmentations were chosen:
• Random rotation
• Random horizontal translation
• Random horizontal mirroring
• Random brightness
• Random contrast
To utilise the power of GPU acceleration available within TensorFlow [13],
batches of 512 images were passed through the network simultaneously. With each
new epoch, a new augmentation is randomly generated for each input image. With
each training step, the original and augmented image passes through the network in
the same state, so with the same weights and biases for each layer. Then, the value
of the loss is calculated, and the parameters updated.
The model was trained for 500 epochs using the Adam optimiser [16] on a PC
with an Nvidia RTX A4000 GPU with 8 GB dedicated memory and an Intel i9 CPU.
Automatic Process Mapping for Ti64 Single Tracks in Laser Powder … 203

Fig. 2 The chosen

combinations of laser power
and scan speed plotted in a
scatter graph with the track
ID annotated

Experimental Methodology

LPBF Equipment

For these experiments, the LBPF equipment was an Aconity Mini (Aconity GmBH,
Germany) system equipped with a 400 W IR single-mode laser with a spot size of
80 µm. No substrate heating was used. Argon gas was used to create the inert gas
atmosphere, and the oxygen concentration was kept below 200 ppm for the duration
of the laser irradiation. The substrate plates used were solid Ti-6Al-4 V with equal
diameters of 25 mm and thickness of 15 mm. Before laser irradiation, the upper
surface of the substrates was ground using 180-grit sandpaper.
50 combinations of P and v were chosen using a Latin hypercube sampling scheme
within a restricted range, based on commonly observed values for Ti-6Al-4 V in
literature [4, 8, 17]. The data points with very low calculated energy densities were
removed from the study to ensure visible melt tracks were printed.
The final parameter set is shown in Fig. 2.

Track Printing and Processing

A custom Python script was used to generate the build file for the LPBF machine,
using the chosen parameter set. A total of 50 single scan vectors 10 mm in length,
spaced 800 µm apart, were printed on two identical substrate plates across two builds.
The lines were not sorted in any particular order on the substrates. A diagram of the
substrate plate with the placement of the single tracks is shown in Fig. 3. The upper
surface of each substrate plate was aligned with the machine bed level to ensure
accurate spot size.
204 T. Wilkinson et al.

Fig. 3 Geometry of the

substrate plates with the
position of the single tracks
indicated. The red dashed
line shows where the sample
plate is cut post-print to
measure the track widths and
depths

Data Collection and Image Processing

The printed samples were directly imaged using an optical microscope. Between 15
and 20, 8-bit grayscale images with pixel dimensions of 2168 × 2168 were captured
for each track. The pixel resolution used was 2.0290 pixels/µm.
A similar image processing method to [6] was followed to generate a significantly
large dataset for training the NN. A custom Python script was written to scale down
each image by a factor of 4, and then slice them into smaller samples with dimensions
of 136 × 27 pixels. These images were then segmented using the WEKA segmenta-
tion tool [18] available within Fiji [19], removing the background. An overview of
the process is given in Fig. 4. Images that did not contain a track, or only, a small
fraction of a track were removed from the dataset, leaving a total of 16,231 images. A
training split of 0.15 was used to separate the training and testing datasets, so 13,797
images were used to train the network, with the remaining 2434 used to test.
To view the cross-sections of the tracks, the substrate plates were cut perpendicular
to the track direction, embedded in resin, polished, and then etched. An optical
microscope was then used to capture images of the cross-sections, which were then

Fig. 4 An overview of the image preparation workflow. a shows the original image, which is then
sliced as shown in (b), resulting in slices as shown in (c). WEKA segmentation is used to create a
mask d which finally creates the image for training (e)
Automatic Process Mapping for Ti64 Single Tracks in Laser Powder … 205

measured using image processing software. Only, a single cross-section per track
was measured. Note, the cross-sectional images are not used to train the CNN. These
are only used for measurements of the track dimensions.

Results

The results of the clustering are summarised in Fig. 5, with different colours
separating the cluster assignments.
The 5 categories are identified as categories 0−4. After visually assessing the
different images within each category, a representative track for each was selected
and is displayed in Fig. 6. The tracks exhibit different morphologies, depending on
where they lie on the process map.

Fig. 5 Parameter set studied

in this work, clustered
according to the output of the
CNN

Fig. 6 Representative tracks

from each of the 5
categories. Each image is
labelled with the track ID
and then the category
206 T. Wilkinson et al.

For each of the identified groups, the number of members, as well as the mean
laser power and scan speed with standard deviation, is shown in Table 1.
Within the range of parameters selected, category 0 covers the largest number of
tracks in this study and has the highest mean scan speed compared to the other groups.
In fact, the entire group of tracks have scan speeds exceeding 1450 mm/s, suggesting
that there is a well-defined threshold at which this track morphology exists. Given
that the mean track width of the group is higher than the mean depth, this group
is unlikely to result in keyhole mode melt pools. Track 15 is an example track in
category 0 and lies at the top right of the process map, with high power and scan
speed. It has a much darker colour in general with a light central section, evidence
of a convex upper surface. There are no visible discontinuities or spatter particles in
the image.
The category 1 tracks have the highest mean laser power and a small standard
deviation, meaning this morphology only covers a small range of powers. Since the
mean track depth is greater than the mean track width for this group, it is likely that
printing in this parameter range would result in keyhole mode melt pools and cause
porosity in a 3D part. Track 35 is labelled as category 1 and lies at the top of the
process map, having a higher power compared to track 32. The upper surface of track
35 is more convex compared to that of track 32, which can be determined by the
difference in colour between the centre and external edges of the track.
Category 2 covers the central region of the process map, and, looking at the group
statistics, does not appear to produce keyhole mode melt pools on average. Track 32
is category 2 and lies centrally on the process map. The image shows a consistent
track, with minor variation of the width and flat top. There are no obvious defects
visible seen for this track in Fig. 6.
Category 3 is located to the left of the process map, with a low mean laser power.
On average, this group produces tracks with the lowest depth and joint lowest width.
If the layer thickness chosen for printing a part was 60 µm or greater (common in
LPBF), lack of fusion defects may appear between the layers. Track 3 is an example
of a category 3 track. The upper surface is flat, and the width appears consistent along
the length. Compared to the other 4 tracks, it is lighter in colour on average.
Category 4 lies in the upper left part of the process map, with the lowest mean
scan speed of all the categories while covering a large range of scanning speeds. The

Table 1 Calculated mean and standard deviation of the laser power, scan speed width and depth
for the tracks assigned to each of the 5 groups by the ML model
Category Track Mean laser Mean scan speed Mean track Mean track
number count power (W) (mm/s) width (µm) depth (µm)
0 13 164.00 ± 18.43 1703.15 ± 157.80 92.38 ± 11.59 78.05 ± 23.38
1 8 179.00 ± 13.57 1161.38 ± 173.17 114.20 ± 8.01 125.78 ± 18.02
2 9 117.00 ± 19.82 912.44 ± 239.94 121.08 ± 8.41 92.89 ± 15.86
3 10 91.10 ± 21.47 1013.10 ± 251.72 92.88 ± 9.97 54.99 ± 12.03
4 10 148.20 ± 28.59 665.50 ± 101.14 157.21 ± 16.21 180.77 ± 33.79
Automatic Process Mapping for Ti64 Single Tracks in Laser Powder … 207

mean depth and width of these tracks are greater than those in the other categories,
and on average, the tracks are in keyhole mode since the depth is greater than the
width. Track 39 is an example of category 4. This track shows more inconsistency
compared to the other 4, with spatter and high variation in the width along the length.
In summary, given the images and data presented, the tracks in categories 0 and
2 are consistent and the least likely to result in keyhole or lack of fusion defects, or
to suffer from excessive spattering.

Discussion

The results presented in the previous section will now be compared to examples from
the literature to determine the success of the proposed method and place the results
in context.
Several studies present a processing window for Ti-6Al-4 V based on computa-
tional or physical experimentation [3, 4, 20]. They provide thresholds for the different
melting regimes of single tracks in terms of the applied energy density. Comparing
the process map in that work (Fig. 13 of [20]) with Fig. 5 in this work, there are
numerous similarities. In their paper, Promoppatum et al. identify 4 key regions: over
melting, incomplete melting, balling effects, and continuous melting. The regions
are represented clearly in the process map given in this work:
• Category 0 matches with “Balling Effects”. While no balling of the tracks is
observed in the data presented here, there is a distinct morphology change, and
with higher laser speeds or powers, this may be seen to occur.
• Category 2 exhibits no visible defects and is unlikely to exhibit keyhole defects.
This matches the “continuous melting” zone highlighted in [20].
• Category 3 closely matches the “incomplete melting” zone since the track depths
are the lowest.
• Category 4 corresponds to the “over melting” zone, and this is clearly observed in
the data presented in this paper, since there are defects visible in the upper surface
photo in Fig. 6, and the dimensions given in Table 1 are evidence of keyhole mode
melt tracks.
• Category 1 is not represented in [20], but could be thought of a transition region
between “complete melting” and “over melting”, since the upper surface appears
consistent, but the tracks measurements suggest keyhole mode melt tracks.
Note, the exact numerical representations of these regions do not exactly match
with those from Promoppatum et al. but the general location is clearly visible. The
same arguments can also be made if the work presented here is compared to the
process map seen in [4], and the same trends and general locations of the groups are
seen. This variation is likely explained by the use of powder in that paper, whereas
in this work, bare substrate plates are used. This variation can be observed in 316L
stainless steel single tracks [21], where the varying powder thickness varies where
these regions sit on the process map. Interestingly however, similar groups are also
208 T. Wilkinson et al.

observed in the 316L samples, suggesting that the general group layout may be
consistent across multiple materials. Hence, this CNN could be used to characterise
other alloys, besides Ti-6Al-4 V, for automatically generating process maps.

Conclusion

In this study, we have demonstrated how unsupervised image clustering can be used
to automatically generate a process map for the printing of LPBF single tracks. A
CNN was used to create 5 distinct groups of single tracks, based on the upper surface
morphology. These five groups showed key differences in their measured widths and
depths. The results from this clustering could be used to better refine a processing
window for the printing of 3D test samples, such as cubes.
The data presented in this work alone is not sufficient to fully determine the
processing parameters for high-quality 3D objects, since there are many other vari-
ables at play, such as the hatch spacing and layer thickness. However, it has been
shown that this technique can offer a good starting point for the development of
optimal printing parameters without the high material and time cost of traditional
methods.

Acknowledgements This project has received funding from the European Union’s Horizon 2020
Marie Skłodowska-Curie Actions—Innovative European Training Networks under grant agreement
No 956401.

References

1. Yang T et al. (2020) Laser powder bed fusion of AlSi10Mg: influence of energy intensities
on spatter and porosity evolution, microstructure and mechanical properties. J Alloys Compd
849. [Link]
2. Li Z et al. (2022) A review of spatter in laser powder bed fusion additive manufacturing:
in situ detection, generation, effects, and countermeasures. Micromachines (Basel) 13(8):1366.
[Link]
3. Luo Q, Yin L, Simpson TW, Beese AM (2022) Effect of processing parameters on pore struc-
tures, grain features, and mechanical properties in Ti-6Al-4V by laser powder bed fusion. Addit
Manuf 56. [Link]
4. Soylemez E (2018) Modelling the melt pool of the laser sintered Ti6Al4V layers with Goldak’s
double-ellipsoidal heat source process mapping and transient melt pool geometry analysis of
Ti-6Al-4V in direct metal laser sintering view project. 2018. [Online]. Available: [Link]
[Link]/publication/328792215
5. Tapia G, Khairallah S, Matthews M, King WE, Elwany A (2018) Gaussian process-based
surrogate modeling framework for process planning in laser powder-bed fusion additive manu-
facturing of 316L stainless steel. Int J Adv Manuf Technol 94(9–12):3591–3603. [Link]
org/10.1007/s00170-017-1045-z
Automatic Process Mapping for Ti64 Single Tracks in Laser Powder … 209

6. Silbernagel C, Aremu A, Ashcroft I (2019) Using machine learning to aid in the parameter
optimisation process for metal-based additive manufacturing. Rapid Prototyp J 26(4):625–637.
[Link]
7. Liu Q et al (2020) Machine-learning assisted laser powder bed fusion process optimization
for AlSi10Mg: new microstructure description indices and fracture mechanisms. Acta Mater
201:316–328. [Link]
8. Liu S, Stebner AP, Kappes BB, Zhang X (2021) Machine learning for knowledge transfer
across multiple metals additive manufacturing printers. Addit Manuf 39. [Link]
1016/[Link].2021.101877
9. Yao X, Moon SK, Bi G (2017) A hybrid machine learning approach for additive manufacturing
design feature recommendation. Rapid Prototyp J 23(6):983–997. [Link]
03-2016-0041
10. Ji X, Henriques JF, Vedaldi A (2018) Invariant information clustering for unsupervised image
classification and segmentation, July 2018. [Online]. Available: [Link]
06653
11. Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional
neural networks. In: Proceedings of the 25th international conference on neural information
processing systems, vol 1. pp 1097–1105
12. Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image
recognition, Sep. 2014, [Online]. Available: [Link]
13. Abadi M et al. (2016) TensorFlow: a system for large-scale machine learning. May 2016, pp
44. [Online]. Available: [Link]
14. Lin M, Chen Q, Yan S (2013) Network in network, Dec. 2013. [Online]. Available: [Link]
org/abs/1312.4400
15. Ioffe S, Szegedy C (2015) Batch normalization: accelerating deep network training by reducing
internal covariate shift, Feb. [Online]. Available: [Link]
16. Kingma DP, Ba J (2014) Adam: a method for stochastic optimization. Dec. 2014, [Online].
Available: [Link]
17. Barba D, Alabort C, Tang YT, Viscasillas MJ, Reed RC, Alabort E (2020) On the size and
orientation effect in additive manufactured Ti-6Al-4V. Mater Des 186. [Link]
[Link].2019.108235
18. Arganda-Carreras I et al (2017) Trainable Weka segmentation: a machine learning tool for
microscopy pixel classification. Bioinformatics 33(15):2424–2426. [Link]
informatics/btx180
19. Schindelin J et al (2012) Fiji: an open-source platform for biological-image analysis. Nat
Methods 9(7):676–682. [Link]
20. Promoppatum P, Onler R, Yao SC (2017) Numerical and experimental investigations of micro
and macro characteristics of direct metal laser sintered Ti-6Al-4V products. J Mater Process
Technol 240:262–273. [Link]
21. Tran HC, Lo YL (2019) Systematic approach for determining optimal processing parameters
to produce parts with high density in selective laser melting process. Int J Adv Manuf Technol
105(10):4443–4460. [Link]
Part VI
Poster Session
Simulation of Ferroalloy Casting
in Copper Moulds

Haifei An, Weijian Tian, Hao Chen, Shaojun Chu, Lihong Li, and Hao Bai

Abstract In China, for the production of ferroalloys such as ferrosilicon and ferroa-
luminium, the mould casting is usually used, and the material of casting mould is
usually cast iron. However, in the process of periodic heating and cooling, cracks
appear at the edge of mould. This severely affects the service life, thus increasing
the casting cost, which has become a bottleneck of the development of ferroalloy
industry. In the casting process, the lower thermal conductivity of the cast iron causes
greater alternating thermal stresses, which are the reason for damages of cast iron
moulds. In this study, the copper which has excellent heat transfer properties was
used as the mould material to develop a new type of ferroalloy mould. In order to
study the service performance of copper mould, the casting process of ferroalloy
in copper mould and the temperature field of copper mould under different water
cooling intensities were simulated by using ANSYS Workbench 17.0 software. The
results show that under water cooling conditions, the maximum temperature of the
mould cannot reach 1083°C in the casting process, so the molten ferroalloy would
not destroy the copper mould, which has low melting point, proofing the feasibility
of copper as the mould material. In addition, the water cooling intensity of the mould
has great influence on the cooling rate of the mould. And the optimized water cooling
intensity is important for proper operation of the copper moulds.

Keywords Ferroalloy · Mould · Copper · Temperature field · Cooling

H. An · W. Tian · H. Chen · S. Chu · H. Bai (B)

State Key Laboratory of Advanced Metallurgy, University of Science and Technology Beijing,
Beijing 100083, China
e-mail: baihao@[Link]
School of Metallurgical and Ecological Engineering, University of Science and Technology
Beijing, Beijing 100083, China
L. Li
Shantou Huaxing Metallurgical Equipment Co., Ltd., Shantou 515063, China

© The Minerals, Metals & Materials Society 2023 213

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
214 H. An et al.

Introduction

Ferroalloy is one of the important basic raw materials in the iron and steel industry,
which plays a key role in improving the quality of steel [1, 2]. With the rapid develop-
ment of the modern steel industry, the demand for ferroalloys is increasing. At present,
China’s ferroalloy production accounts for about 40% of the world’s output, and it has
become the world’s largest ferroalloy producer and consumer [3]. According to the
database of China Commercial Industry Research Institute, the national ferroalloy
output in 2018 was increased by 4.5% to 31.23 million tons [4]. In the ferroalloy
production, casting is the key process to ensure the production efficiency and quality
of ferroalloy. In general, there are three casting methods for the solidification of
ferroalloy: ingot mould casting, pit-layered casting, and machine casting [5–8]. The
technological process of all the methods is generally composed of pouring, slag
removal, primary crushing, and secondary processing. In this process, the cast needs
to be crushed further manually or mechanically in order to meet the requirement of
the users for product size. After the crushing process, the powdering loss of products
reaches 8–10%, and especially, the powdering loss of high carbon ferromanganese
even reaches 16–17%, resulting in a declination in yield and product quality [9].
In recent years, some technologies of continuous mould casting granulation for
ferroalloy have been proposed in order to decrease powdering loss [10] by eliminating
the crushing process. In this technology, the molten ferroalloy is continuously poured
into the drainage device, and then dispersed and injected into the circulating mould.
The ferroalloy solidifies in the cavities, demoulds, and the granulated ferroalloy
products are obtained without further processing. This process reduces the product
loss and improves the environment. For this technology, the mould plays a key role
in solidification and forming of the ferroalloy. At present, cast iron is widely used as
the mould material [11]. Although the cast iron has the advantages of high strength,
easy processing, and so on, its disadvantages are also obvious. For example, after a
period of thermal shocks, the mould will crack locally, thus reducing the service life.
Many researches have been carried out to improve the quality and service life of the
moulds [5]. In terms of moulds materials, several kinds of cast iron, such as normal
gray iron, medium manganese gray iron, ductile iron, compacted graphite cast iron,
rare earth gray iron, and vanadium titanium cast iron, have been tried. However, the
service life of the mould is still not ideal.
The casting mould is not only the shaping equipment, but it is also a heat exchanger
[12]. In the casting process, due to the repeated scouring at the mould cavities surface
by the high temperature molten ferroalloy, the periodic temperature changes cause
expansion and contraction alternately, so that the mould experiences non-isothermal
temperature field and nonuniform stress field. The maximum thermal stress occurs
on the surface of the mould cavities, which exceeds the yield strength of the mould
material, causing plastic deformation. In addition, the heat dissipation capacity of
the mould is insufficient during the process of cyclic use, which lead to demould of
casts if the cooling intensity is not enough. Therefore, the alternating thermal stress
and thermal load are the main factors affecting the use of mould.
Simulation of Ferroalloy Casting in Copper Moulds 215

Consider that copper has excellent heat transfer performance, it was used as the
mould material to develop a new type of ferroalloy mould in this study. As is known,
the thermal conductivity of pure copper (CW004A) can reach 400 W/(m K), and it is
10 times that of gray iron, which will make the temperature distribution more uniform.
Therefore, pure copper has lower alternating thermal stress and lower overall temper-
ature than cast iron in the same environmental conditions. In order to study the service
performance of the copper mould, the casting process of ferroalloy in copper mould
and the temperature field of copper mould under different cooling intensity were
simulated by using ANSYS Workbench 17.0 software. This study would provide
a theoretical reference for further research and development of ferroalloy casting
system with copper mould.

Model Description

The development ferroalloy granulation casting system developed, as shown in Fig. 1,

is composed of the pouring device, moulds, and circulating rotating device. The
motors drive the moulds to move counter-clockwise at a constant speed while the
molten ferroalloy is poured into the moulds by the pouring equipment. Then, the
ferroalloy is solidified in the moulds and demoulded. The moulds enter the next
cycle after water cooling.
In order to study the service performance of copper moulds, the temperature field
change of a single mould in a cycle was simulated numerically, which involves the
analysis on solidification of the ferroalloy and transient heat transfer of the mould.
The complete casting cycle is divided into two stages according to the different initial
and boundary conditions before and after demoulding, and the demoulding time is
ignored. The temperature field of the mould from pouring to demoulding is simulated
at stage 1, and the mould temperature field of natural cooling or forced water cooling
after demoulding is simulated at stage 2.

Molten Pouring
Move ferroalloy device
direction
Stage 1 Mould

Demould Motor Motor

Stage 2

Fig. 1 Schematic diagram of ferroalloy granulation casting system

216 H. An et al.

Fig. 2 Three-dimensional
model of copper mould

Fig. 3 Single cavity size

Physical Model

The mould can be set as a copper cuboid of 650 mm × 349 mm × 97 mm, with
108 cavities on its upper surface. The three-dimensional model of copper mould is
built by SolidWorks software, as shown in Fig. 2. The size of each cavity is shown in
Fig. 3. The mass of a single copper mould is 183 kg, and the cavity can hold 4.8-kg
ferrosilicon alloy, so the copper-metal ratio of mould is 38.1.

Mathematical Model

Governing Equations

The release of latent heat during the solidification of ferroalloys is a major feature of
the casting forming process. The latent heat of metal solidification accounts for about
50% of the heat transferred from the casting to the mould, so it has a great influence
on the temperature field of the casting and must be considered. The relationship
between temperature, time, and space can be described by Fourier partial differential
equations for heat conduction, which actually becomes an unsteady-state conduction
process with internal heat source [13]:
Simulation of Ferroalloy Casting in Copper Moulds 217

∂T λ ∂2T ∂2T ∂2T Q
= + + + (1)
∂t Cρ ∂ x 2 ∂ y2 ∂ z2 Cρ

where T is the temperature (°C); C is the specific heat capacity (J/(kg K)); ρ is the
average density of the liquid and solid phases (kg/m3 ); Q is the amount of latent
heat released, which is the internal heat source (J); λ is the thermal conductivity
(W/(m K)).
How to deal with the latent heat of crystallization is key to the success of casting
solidification simulation. The enthalpy method is usually used for alloys with a
wide crystallization temperature range [14]. Enthalpy method introduces an enthalpy
function into Fourier partial differential equation, which specifies the relationship
between enthalpy and temperature. When solving the equation, the enthalpy value
of a certain temperature node is obtained as a known value, and then, the node
temperature is obtained through the solution of the equation.
When the solidification phase transition is considered, the enthalpy function of
the metal is as follows:

T
H = H0 + C P dT + (1 − f s )L (2)
T0

where T0 is the reference temperature or reference temperature (°C); H0 is the

enthalpy at the reference temperature (J); f s is the solid fraction.
Derive the time from the above formula:

∂H ∂ fs ∂ T
= cP − L (3)
∂t ∂ T ∂t

Substitute formula (3) into formula (1):

2
∂H ∂ T ∂2T ∂2T
ρ =λ + + (4)
∂t ∂x2 ∂ y2 ∂z 2

The enthalpy and temperature of the next node can be obtained according to the
initial enthalpy and the relationship of enthalpy and temperature.

Thermophysical Parameters

According to Neumann Kopp law [15], when the temperature of composition is above
Debye temperature, the heat capacity can be obtained by the following formula:

Cm = xi cim (5)
i
218 H. An et al.

Table 1 Specific heat capacity of 75 ferrosilicon

Temperature (°C) 300 600 900 1100 1207 1340 1500
Specific heat capacity J/(kg K) 796 946 1000 985 980 975 965

Table 2 Thermophysical parameters of pure copper (CW004A)

Temperature (°C) Density (kg/m3 ) Thermal conductivity Specific heat capacity
(W/(m K)) (J/(kg K))
20 8920 405.5 385
100 8920 404.1 388
300 8920 390.5 396
500 8920 367.7 404
700 8920 342.4 414
900 8920 322.9 427
1000 8920 312.7 445
1600 8920 290.0 470

where x is the mole fraction of the component. The difference between the specific
heat capacity calculated by the law and the experimental results is not more than 4%.
The specific heat capacity of 75 ferrosilicon varying with temperature was calculated,
as shown in Table 1. The copper mould is made of pure copper (CW004A), and the
thermophysical parameters are shown in Table 2.
The phase diagram of ferrosilicon shows that Ts = 1207 °C and Tl = 1340 ◦ C,
the phase transition interval between fluid and solid is defined as 133°C. The thermal
conductivity of ferrosilicon is set to 25 W/(m K). The relationship between enthalpy
and temperature of 75 ferrosilicon is calculated by enthalpy method, as shown in
formula (6).
⎧
⎨ 2.12 × 106 T T ≤ 1207 ◦ C
H = 3.42 × 107 T − 3.87 × 1010 1207 ◦ C < T ≤ 1340 ◦ C (6)
⎩
2.12 × 106 T − 2.13 × 1010 1340 ◦ C < T ≤ 1600 ◦ C

Initial and Boundary Conditions

At stage 1, the mould temperature field from pouring to demoulding was simulated.
The initial temperature of molten iron is 1550 °C; the initial temperature of mould
is 40 °C, and the ambient temperature is 30 °C.
According to the literature [12, 16], the interfacial heat transfer coefficient
increases linearly with the increase of temperature, which means that the higher
Simulation of Ferroalloy Casting in Copper Moulds 219

the temperature of the casting itself, the greater the interfacial heat transfer coeffi-
cient between the casting and air. In order to simplify the calculation, the convective
heat transfer coefficient between casting and air was set as 100 W/(m2 K), and the
convective heat transfer coefficient between mould and air was set as free convection
heat transfer coefficient of 10 W/(m2 K).
At Stage 2, the mould temperature field of natural cooling or forced water cooling
after demoulding of castings is simulated. The temperature load at the end of the
previous stage was taken as the initial condition of the cooling process in the next
stage. The convective heat transfer coefficient of natural cooling was set to be free
convection heat transfer coefficient of 10 W/(m2 K), and the convective heat transfer
coefficient of forced water cooling was set to 500 W/(m2 K), 1000 W/ W/(m2 K),
2000 W/(m2 K), 5000 W/(m2 K), and 8000 W/(m2 K), respectively, corresponding
to different water cooling intensity.

Mesh Generation

The pre-processing software MESH was used to generate structured hexahedral

meshes for the mould and the casting, respectively, and the boundary layer encryption
was performed on the contact area between the casting and the mould. Considering
the accuracy and the speed of calculation, it is determined after the mesh indepen-
dence test that the minimum size of the mesh was 2 mm, divided into 1.43 million
grids, and the average quality of the grid was 0.8495. The physical model meshing
was generated as shown in Fig. 4.

Fig. 4 Mesh generation of

physical model
220 H. An et al.

Results and Discussion

Solidification Process of Ferroalloy

The boundary conditions of the mould change before and after demoulding, and the
demoulding time affects the temperature field of the mould, so the demoulding time
should be determined first. Therefore, the solidification process of the ferroalloy was
studied. After pouring molten iron, the temperature changes of the mould and molten
iron under the cooling intensity of 10 W/(m2 K) (natural cooling) were simulated
and analyzed to determine the appropriate demoulding time.
Seven temperature monitoring probes were set at the central axis of the mould
cavity to monitor the temperature change during the casting process. The probe 1
is in the molten iron, and the probes 2–7 are in the mould, as shown in Fig. 5. The
temperature changes at seven probes were collected for further studies. Also, the
maximum and minimum temperature in the solidification process were monitored,
respectively. The results are shown in Fig. 6.

Fig. 5 Distribution of
temperature monitoring
probes

Fig. 6 Temperature change

diagram of mould and
molten iron under different
water cooling intensity
Simulation of Ferroalloy Casting in Copper Moulds 221

Due to the excellent heat transfer performance of the copper mould, the temper-
ature of molten iron decreases rapidly after pouring, and the corresponding temper-
ature of copper mould increases rapidly. The mould temperature rose to the highest
at 362.2 °C in 7.5 s, and then, the temperature dropped slowly, indicating that the
copper mould can fully bear the thermal shock of molten iron under this copper-iron
ratio and will not have destructive influence on the mould. Ignoring the minimum
temperature curve, it was found that the temperature of all probes of the mould is
about 220°C in 50 s, and the overall temperature difference is within 50°C, which
is very close to the temperature of hot metal, indicating that the mould has fully
absorbed the heat of molten iron, and extending the time is not conducive to the
cooling of the mould. Therefore, it was suggested that the longest demoulding time
is 50 s after pouring.

Temperature Field Before Demoulding

The casting demoulding time was determined to be 50 s, and the data of the first 50 s
were analyzed, as shown in Fig. 7. It can be found that the maximum temperature
curve of molten iron has a plateau area, which is due to the release of a large amount
of solidification latent heat during the cooling process of ferrosilicon that slows down
the temperature drop. The temperature plateau area of the probe near the mould was
not obvious. This was because the excellent heat transfer performance of the copper
mould offsets the effect of latent heat release, which makes the cooling rate of the
molten iron faster. The latent heat of solidification is basically released at 22.5 s, and
the molten iron is completely solidified without any liquid core. Regardless of other
factors, it can be used as the minimum time point for demoulding. When the time
exceeds 22.5 s, the maximum temperature of the casting decreased rapidly. The heat
flux of molten iron between the starting point and point B was much greater than
the heat flux of the copper mould, so the mould temperature rose sharply and can
reach 350.4 °C in 2.7 s. Between point B and point C, that is, within 2.7–7.5 s, the
output heat flux of molten iron was slightly larger than the heat dissipation flux in
the copper mould, so the mould temperature slowly rose to a maximum of 362.2 °C.
The highest temperature was within the safe use range of copper metal, which shows
that the copper mould meets the production requirements. After 7.5 s, the molten
iron output heat flux was less than the heat dissipation flux in the copper mould,
so the mould temperature was slowly reduced. Among the common ferroalloys, the
sum of sensible heat and latent heat of ferrosilicon is the highest, which verifies the
feasibility of copper as the mould material for common ferroalloys (Fig. 8).
222 H. An et al.

Fig. 7 Temperature change

curve of mould and molten
iron before demoulding

˄a˅Section view ˄b˅Oblique view

˄c˅Front view ˄d˅Bottom view

Fig. 8 Temperature field of the mould at 50 s before demoulding

Temperature Change of the Mould After Demoulding Under

Different Cooling Conditions

At stage 2, the mould temperature changes due to natural convection cooling or forced
water cooling after demoulding of castings was simulated. The temperature load at the
end of the previous stage was taken as the initial condition of the cooling process in the
next stage. The convective heat transfer coefficient of natural convection cooling in
Simulation of Ferroalloy Casting in Copper Moulds 223

air was 10 W/(m2 K), and the intensity of forced water cooling was set to 500, 1000,
2000, 5000, and 8000 W/(m2 K), respectively. The temperature changes at seven
temperature probes were collected according to Fig. 5, and the result was shown in
Fig. 9. The cooling process was divided into two areas. Before demoulding, the heat
of molten iron was mainly absorbed by the copper mould, and the heat lost by natural
convection with air accounted for a small part, which was referred to as the copper
cooling area. After demoulding, forced water cooling was used for copper mould,
which was referred to as the water cooling area.
In order to keep the stability of the copper mould and reduce the heat accumu-
lation, the mould temperature was cooled to below 100 °C. The greater the water

(a) (b)

(c) (d)

(e) (f)

Fig. 9 Temperature change diagram of mould and molten iron under a natural convection and
different water cooling of: b 500, c 1000, d 2000, e 4000, and f 8000 W/(m2 K), respectively
224 H. An et al.

cooling intensity, the shorter time the mould cooling to 100 °C. Figures 10 and 11,
respectively, show the variation curves of the highest temperature and the lowest
temperature of copper mould under different water cooling intensities. The results
show that when the water cooling intensity is doubled, the temperature of copper
mould can be reduced by about 30 °C after 50 s of cooling; when the water cooling
intensity exceeds 4000 W/(m2 K), the cooling enhancement effect becomes slower. In
order to ensure the safe use of copper mould, the cooling time is calculated according
to the maximum temperature change. At 4000 W/(m2 K) water cooling intensity, it
takes 49.2 s to cool to 100 °C and 111.0 s to cool to 40 °C. At 8000 W/(m2 K)
water cooling intensity, it takes 26.3 s to cool to 100 °C and 76.8 s to cool to 40 °C.
The water cooling intensity of the mould has great influence on the cooling rate of
the mould. Considering the pouring cycle and cooling effect comprehensively, the
cooling intensity should be set at 8000 W/(m2 K), and the corresponding cooling time
is only 26.3 s, which can ensure the continuous and stabilized work of the copper
mould.

Fig. 10 Maximum
temperature change of
copper mould under different
water cooling strength

Fig. 11 Minimum
temperature change of
copper mould under different
water cooling intensity
Simulation of Ferroalloy Casting in Copper Moulds 225

Conclusions

When the copper-iron ratio of mould is 37.8, the highest temperature of the copper
mould during the casting process of the ferrosilicon rise to 362.2 °C, and the residence
time of the peak temperature was short, so the molten ferroalloy would not destroy
the copper mould which has low melting point. Among the common ferroalloys, the
sum of sensible heat and latent heat of ferrosilicon is the highest, which proofs the
feasibility of copper as the mould material for common ferroalloys.
At 50 s after pouring, the mould has fully absorbed the heat of molten ferroalloy,
indicating that an appropriate demoulding time can be 50 s.
The water cooling intensity of the mould has great influence on the cooling rate
of the mould. The results show that when the water cooling intensity is doubled, the
temperature of copper mould can be reduced by about 30 °C after 50 s of cooling.
On the other hand, when the water cooling intensity exceeds 4000 W/(m2 K), the
cooling enhancement effect becomes slower.
Finally, the optimized water cooling intensity is important for proper operation
of the copper moulds. Considering the pouring cycle and cooling effect comprehen-
sively, the cooling intensity should be set at 8000 W/(m2 K) for a corresponding
cooling water flow of 5 m/s. It takes 26.3 s to cool to 100 °C at its intensity, which
can ensure the continuous and stable condition of the copper mould.

References

1. Xiwang M, Bai Zhitao L, Guanghua. (2020) Review of comprehensive utilization of typical

ferroalloy slags. Chin J Eng 42(6):663–679
2. Surup GR, Trubetskaya A, Tangstad M (2020) Charcoal as an alternative reductant in ferroalloy
production: a review. Processes 11(8):1432
3. Xiao L, Jinyan L, Yuanyuan Z (2020) Prospects of resource utilization techniques of solid
waste in China’s ferroalloy industry. Ferro-Alloys 51(01):44–48
4. Website: [Link]
5. Zhao LJ, Le J, Dejian Z (2014) Current situation and trends of research on the ingot mould
design. Mech Res Appl 27(06):189–191
6. Hongyi Y (2020) SA brief discussion on application of combined ingot mould in ferroalloy
production. J Jisco Technol 02:44–46
7. Dongsheng D, Yanlin X, Yanyou Z (2009) Application of water cool ingot mould in ferroalloy
production. Ferro-Alloys 40(05):24–26
8. Curcio P, Kerezsi B, Brown P (2004) Failure investigation of lime coated grey cast iron moulds
at TEMCO ferroalloy smelter. Eng Fail Anal 11(6):925–933
9. Jalkanen H, Gasik M (2013) Chapter 3—Theory of ferroalloys processing. In: Handbook of
Ferroalloys. Elsevier Ltd
10. Wenchao H, Xuewei L, Feifei P et al (2019) Granulation of ferrosilicon alloy by rotary
multi-nozzles cup atomizer: granulation behavior and model formation. Adv Powder Technol
30(5):895–902
11. Kaila VN, Dave IB (2020) The influence of coating sand materials on shell mould properties
of investment casting process. Mater Today Proc
226 H. An et al.

12. Stieven GDM, Soares DDR, Oliveira EP et al (2021) Interfacial heat transfer coefficient in
unidirectional permanent mould casting: modeling and inverse estimation. Int J Heat Mass
Transf 166(1):120765
13. Cuesta FJ, Lamúa M (2009) Fourier series solution to the heat conduction equation with an
internal heat source linearly dependent on temperature. J Food Eng 90(2):291–299
14. Knig-Haagen A, Franquet E, Faden M et al (2020) Influence of the convective energy
formulation for melting problems with enthalpy methods. Int J Therm Sci 2020:106477
15. Leitner J, Voňka P, Sedmidubský D et al (2010) Application of Neumann–Kopp rule for the
estimation of heat capacity of mixed oxides. Thermochimica Acta 497(1):7–13
16. Liu X, Fakir OE, Cai Z et al (2019) Development of an interfacial heat transfer coefficient model
for the hot and warm aluminium stamping processes under different initial blank temperature
conditions. J Mater Process Technol 273
Effects of Temperature and Density
on Transition Slab Length During Steel
Grade Transition

Sicheng Song, Yanhui Sun, Yaoguang Li, and Chao Zhuo

Abstract The length of the transition slab may be decreased by manipulating differ-
ences in temperatures or densities of new and old grade steels. The molten steel
liquidus temperature and density are related to the composition. For most elements
in steel, both the liquidus temperature and density of the molten steel decrease with
an increase in their mass fraction. Therefore, the temperature and density of molten
steel cannot simultaneously shorten the transition slab. This study compared the
effect of temperature and density on the length of the transition slab using numerical
simulation methods based on flow, heat transfer, and mass transfer. The model was
verified using a physical simulation and an industrial experiment. Finally, the effects
of temperature and density on the transition slab length were elucidated.

Keywords Steel grade transition · Transition slab · Intermix · Tundish ·

Simulation · Continuous casting

Introduction

Compared with die casting, continuous casting offers higher yield and higher quality
of slabs; for this reason, continuous casting has become a common method for molten
steel solidification. As steel products become more refined, steel mills are often
required to produce steel in small batches. To further improve the production effi-
ciency, steel grade transition is increasingly becoming a common method in steel
production. However, steel grade transition inevitably results in transition slabs.
Therefore, studies on continuous casting have long been focused on reducing the
length of the transition slab [1–5].
Thomas [2] developed a series of measures to shorten the length of the transition
slab by considering the tundish furniture, minimum tundish weight at the opened
ladle, cross-sectional area of the mold, size of the tundish, and casting order of steel
grades. Kant et al. [6] mainly considered the effect of the location of the dam in the

S. Song · Y. Sun (B) · Y. Li · C. Zhuo

School of Collaborative Innovation Center of Steel Technology, University of Science and
Technology Beijing (USTB), Beijing 100083, China
e-mail: sunyanhui_ustb@[Link]

© The Minerals, Metals & Materials Society 2023 227

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
228 S. Song et al.

six-strand caster on the length of the transition billet, and their study concluded that
placing the dam at the outlet close to the inlet can significantly reduce the length
of the transition billet. Siddiqui and Jha [7] used the volume of fluid method for
numerical simulations, which considered the effect of the turbulence inhibitor on the
length of the transition slab. They concluded that the turbulence inhibitor increases
the length of the transition slab. Amorim et al. [4] established a steel grade transition
model for a six-strand billet continuous caster. Huang and Thomas [8] suggested that
during the steel grade transition, argon injection has a slight effect on the surface
composition of the slab but has no effect on the mixing inside the slab. Li and
Tsukihashi [9] developed a mathematical model to analyze the mass transfer during
steel grade transition using a static magnetic field. The model reduces the length
of the transition slab by approximately 50% by applying a 0.5 T magnetic field.
Ahn et al. [10] found that an electromagnetic field suppressed the lower vortex in
the mold and formed a plug flow zone, which reduced the mixing of two different
grades of steel and shortened the length of the transition slab. Chattopadhyay et al.
[11] considered the effect of a turbulence inhibitor on steel grade transition under
non-isothermal conditions. Alizadeh et al. and Damle and Sahal [3, 12] found that
the weight of mixed-grade steels is influenced by the new molten steel volume, slab
size, casting rate, and non-isothermal effects.
To ensure the quality of the slab during actual production, low-superheat casting
is generally required. The liquidus temperature and density of the molten steel are
both related to its composition. To determine a more suitable casting sequence,
we calculated the steel grade transition at different temperatures and densities via
numerical simulation. This helped to further shorten the length of the transition slab.

Simulation Model

Model Assumptions

(1) The molten steel in the model is considered to be an incompressible Newtonian

fluid under turbulent conditions.
(2) The model ignores chemical reactions in molten steel.
(3) The variation in the molten steel element mass fraction in the model is mainly
caused by the flow, and diffusion has little effect on it.
(4) The model ignores the rise and fall in the molten steel level in the tundish during
ladle replacement.
(5) The shear stress at the steel–slag interface in the model was set to 0.
(6) The model considers the area below 2 m of the meniscus of the mold to have
laminar flow, and it is assumed that the composition of the molten steel in the
laminar flow area does not change.
Effects of Temperature and Density on Transition Slab Length During … 229

Governing Equations

The continuity and Navier–Stokes equations can be written as follows:

∂ρ
+ ∇ · (ρ v) = 0 (1)
∂t
∂(ρ v)
+ ∇ · (ρ vv) = −∇ p + μ∇ · ∇ v + ∇ vT + ρ g (2)
∂t

where ρ is the density, t is the time, v is the velocity, p is the static pressure, μ is the
laminar viscosity, and g is the gravity.
The turbulence kinetic energy k, and its rate of dissipation ε, are employed as
follows:

∂(ρk) ∂(vi ) ∂ μt ∂k ∂vi ∂v j ∂vi
+ ρk = μ+ + μt + − ρε (3)
∂t ∂ xi ∂x j σk ∂ x j ∂x j ∂ xi ∂ x j

∂ ∂ ∂ μt ∂ε
(ρε) + (ρεvi ) = μ+
∂t ∂ xi ∂x j σε ∂ x j

ε ∂vi ∂v j ∂vi ε2
+ C1ε μt + − C2ε ρ (4)
k ∂x j ∂ xi ∂ x j k

The turbulent viscosity, μt , is expressed as:

k2
μt = ρCμ (5)
ε
where C1ε (1.44), C2ε (1.92), and Cμ (0.09) are constants, σk (1.00) and σε (1.30)
are the turbulent Prandtl numbers for k and ε, respectively.
The conservation equation for species transfer is expressed as follows:

∂ μt
ρCspe + ∇ · ρ vCspe = −∇ · − ρ Dspe,liq + ∇Cspe (6)
∂t Sct

where Cspe is the local mass fraction of the speth species, Dspe,liq is the liquid-phase
diffusion coefficient, and Sct (0.70) is the turbulent Schmidt number.
The energy conservation equation is expressed as follows:

∂ ρC p T ∂ ρC p T u i ∂ c p μt ∂ T
+ = λ+ (7)
∂t ∂ xi ∂ xi Prt ∂ xi

where C p is the specific heat capacity, T is the temperature, λ is the thermal

conductivity, and Prt (0.85) is the turbulent Prandtl number.
230 S. Song et al.

Boundary Conditions and Properties

The computational domain of this model includes the tundish, mold, and part of the
secondary cooling area. The three-dimensional model and the mesh are shown in
Fig. 1. The remaining calculation parameters are listed in Table 1. Owing to the long
mixing time of molten steel, we imported the data on the submerged entry nozzle
(SEN) in the tundish area into the SEN in the mold area and calculated the tundish
and mold parts separately, thereby improving the calculation efficiency.
To bring the model closer to the actual situation, the boundary conditions of this
model were set as follows:
(1) The ladle nozzle in the tundish area and SEN in the mold area were both set as
velocity inlets, (the inlet conditions are listed in Table 1).
(2) The outlet at the bottom of the mold and the SEN in the tundish area were both
set as outflows.

Fig. 1 Graph and grid of the computational domain

Table 1 Calculated parameters in the model

Parameter Value Parameter Value
Ladle nozzle inner 75 mm SEN inner diameter 56 mm
diameter
Ladle nozzle outer 140 mm SEN outer diameter 116 mm
diameter
Tundish length 4214 mm Billet cross-section 1260 mm × 220 mm
Tundish width 1181 mm Casting speed 1.5 m/min
Ladle nozzle velocity 1.569 m/s SEN velocity 2.802
Ladle nozzle turbulence 3.668% SEN turbulence intensity 3.531%
intensity
Effects of Temperature and Density on Transition Slab Length During … 231

Table 2 Temperature parameters

Parameter Value Parameter Value
Tundish wide wall 3200 W/m2 Tundish narrow wall 3800 W/m2
Tundish free surface 15,000 W/m2 Tundish bottom wall 1400 W/m2
Mold wide face 2,420,000 W/m2 Mold narrow face 2,230,000 W/m2
Loop 1 wide face 1520 W/ (m2 K) Loop 1 narrow face 1320 W/ (m2 K)
Loop 2 wide face 847 W/ (m2 K) Loop 2 narrow face 685 W/ (m2 K)
Mold free surface 15,000 W/m2

Table 3 Material parameters

Parameter Value Parameter Value
Specific Heat 650 J/(kg K) Viscosity 0.0062 kg/ (m
s)
Thermal 45 W/(m K) Liquid-phase 2.0e−09 m2 /s
Conductivity diffusivity [13]

(3) The rest of the boundaries were set as walls. The shear forces at the steel–
slag interface in the tundish part and the steel–slag interface in the mold part
were both set to zero. The heat-dissipation parameters of the walls are listed in
Table 2.
The molten steel density was calculated in real time using the volume-weighted
mixing law. In addition, because the same density or the same liquidus temperature
can correspond to multiple steel grades, the molten steel parameters were all set as
constants (Table 3).

Experimental and Numerical Procedure

In this study, the temperature difference was obtained by subtracting the temperature
of the new grade from that of the old grade. The density difference was obtained by
subtracting the density of the new grade from that of the old grade. Four temperature
and four density differences were used to study the effects of temperature and density
on the length of the transition slab. The 16 numerical simulation schemes are listed
in Table 4.
In this study, the flow field and temperature field of the tundish and mold areas
during normal production were obtained first. The mixing process in the tundish was
then calculated, and the residence time distribution (RTD) data at the outlet of the
tundish were obtained. Finally, the profile method was used to import the mold, and
the molten steel compositions on the surface and center of the slab at the exit of the
mold were calculated.
232 S. Song et al.

Table 4 Numerical simulation experiment scheme

No New grade Old grade Temperature New grade Old grade Density
temperature temperature difference (K) density density Difference
(K) (K) (kg/m3 ) (kg/m3 ) (kg/m3 )
1 1670.0 1820.0 −150.0 7100 6800 300
2 1770.0 1820.0 −50.0 7100 6800 300
3 1820.0 1770.0 50.0 7100 6800 300
4 1820.0 1670.0 150.0 7100 6800 300
5 1670.0 1820.0 −150.0 7100 7000 100
6 1770.0 1820.0 −50.0 7100 7000 100
7 1820.0 1770.0 50.0 7100 7000 100
8 1820.0 1670.0 150.0 7100 7000 100
9 1670.0 1820.0 −150.0 7000 7100 −100
10 1770.0 1820.0 −50.0 7000 7100 −100
11 1820.0 1770.0 50.0 7000 7100 −100
12 1820.0 1670.0 150.0 7000 7100 −100
13 1670.0 1820.0 −150.0 6800 7100 −300
14 1770.0 1820.0 −50.0 6800 7100 −300
15 1820.0 1770.0 50.0 6800 7100 −300
16 1820.0 1670.0 150.0 6800 7100 −300

Results and Discussion

Validation Experiments

We used physical simulations and industrial experiments to verify the model. The
ratio of the tundish model to the actual tundish used in the physical simulation was
1:3. In the physical simulation, water and brine represent different steel grades,
and a schematic of the water model experiment is shown in Fig. 2a. The physically
simulated ink distribution is shown in Fig. 3. During the industrial test, samples were
taken every 1 min at two positions: the center and the surface, as shown in Fig. 2b.
The steel grades used in the industrial test are presented in Table 5.

Fig. 2 Schematic of verification experiments: a physical simulation, b industrial test

Effects of Temperature and Density on Transition Slab Length During … 233

Fig. 3 Black ink dispersion photo in the water model

Table 5 Composition and physical parameters of the experimental steels

Steel grade Weight (%) Density Liquidus
C Si Mn P S (kg/m3 ) temperature (K)

DC01 (New) 0.011 0.024 0.174 0.012 0.006 7134.95 1806.47

SPHC (Old) 0.058 0.032 0.188 0.015 0.008 7124.49 1802.49

We transformed the RTD data from water model experiments using the following
equation.

Ct − Co
F= (8)
Cn − Co

where Ct is the fraction of a given element at time t; Co and Cn are the fractions
of that element measured in the old and new grades, respectively. Additionally, we
compared the results of the field experiments and model predictions. The results
showed good agreement, as shown in Fig. 4.

Flow Field

This study found that the mixing process in the tundish may be roughly divided into
two types. We chose schemes 6 and 11 to represent these two cases, as shown in
Fig. 5. The temperature difference in scheme 6 is − 50 K, and the density difference
is 100 kg/m3 . In scheme 11, the temperature difference is 50 K, and the density
difference is − 100 kg/m3 .
The new grade in scheme 6 enters the turbostop from the ladle nozzle and sinks
to the bottom of the tundish along the outside of the turbostop after the speed is
234 S. Song et al.

Fig. 4 Comparison of
predicted and measured slab
compositions

reduced, and soon reaches the SEN. The old grade residing in the upper part of the
tundish needs to flow and mix for a long time before it can gradually flow out of the
tundish to complete the steel grade transition. At the same time, the new grade in
scheme 11 enters the turbostop from the ladle nozzle and returns to the upper part
of the tundish after deceleration. The new grade then flows to the stopper rod along
the steel–slag interface. In the early stage of mixing, it does not affect the flow of
the old grade out of the tundish, causing a rapid decrease in the proportion of the old
grade in the tundish, thereby effectively shortening the length of the transition slab.
The dam in the tundish prevents the last part of the old grade from flowing out of the
tundish to a certain extent, and this part of the old grade can only flow out by mixing
with the new grade or through two holes in the dam.
In scheme 6, the new grade reaches the SEN earlier, and the old grade completely
flows out of the tundish later than those in scheme 11. Therefore, we adopted a
scheme similar to scheme 11 rather than scheme 6 to shorten the transition slab.
After the molten steel is mixed in the tundish, the composition of the molten steel
flowing from the SEN to the mold is relatively uniform; therefore, the variation in the
molten steel composition in the mold is smaller than that in the tundish. As shown
in Fig. 6, after the molten steel flows from the SEN into the mold, it first reaches
the narrow face of the mold. Subsequently, part of the molten steel flows down the
mold, so that the new grade mass fraction on the surface of the slab is slightly higher
than that in the center of the slab. Similarly, the new grade in scheme 6 enters the
mold earlier than in scheme 11, and the old grade in scheme 6 completes the mixed
casting process later than in scheme 11.
As shown in Fig. 7, although the center and surface compositions of the slab
differ, the difference is small. Compared with the RTD curve at the SEN, the center
and surface components of the slab have a certain hysteresis, but the change trends
in the three are consistent. In scheme 6, the new grade reaches the SEN earlier than
in scheme 11, and the old grade in scheme 6 flows out of the tundish completely
later than in scheme 11. The RTD curves of schemes 6 and 11 are consistent with
the molten steel flow laws shown in Figs. 5 and 6. The research results show that the
Effects of Temperature and Density on Transition Slab Length During … 235

Fig. 5 Numerical simulation results for the mass fraction of new grade in the tundish as a function
of time (a–g) steel grade transition in scheme 6, (h–n) steel grade transition in scheme 11
236 S. Song et al.

Fig. 6 Numerical simulation results for the mass fraction of new grade in the mold as a function
of time: (a–f) steel grade transition in scheme 6, (g–l) steel grade transition in scheme 11

Fig. 7 Mass fraction of the new grade at the tundish outlet and the steel stream outlet in the model

molten steel flow in the mold has little effect on the composition distribution in the
slab; therefore, only, the mixing process in the tundish needs to be investigated to
obtain the change rule for the length of the transition slab.

RTD Curves

The RTD curves of the 16 groups of tundish experiments at SEN are shown in Fig. 8.
We assumed that the molten steel with the mass fraction of the new grade below 20%
met the requirements of the old grade, above 80%—the new grade, and between 20
and 80%—the transition slab.
Effects of Temperature and Density on Transition Slab Length During … 237

Fig. 8 RTD curves of 16 experiments (corresponding to Table 4)

By analyzing the RTD curves of the 16 groups of experiments, we determined the

time at which the mass fraction of the new grade reached 20%, reached 80%, and
the time for producing the transition slab (Table 6). A contour map was drawn based
on the data in Table 6, as shown in Fig. 9.

Table 6 Critical moment at SEN during the steel grade transition

No New grade New grade Transition No New grade New grade Transition
20% (s) 80% (s) slab 20% (s) 80% (s) slab
duration (s) duration (s)
1 1070.1 1892.3 822.2 9 1114.5 1976.8 862.3
2 1069.9 1902.0 832.1 10 1244.5 1859.2 614.7
3 1069.4 1914.6 845.2 11 1300.6 1836.9 536.3
4 1063.3 1964.0 900.7 12 1310.1 1830.0 519.9
5 1068.6 1918.9 850.3 13 1303.2 1833.1 529.9
6 1063.5 1937.7 874.2 14 1310.6 1831.4 520.8
7 1082.4 2005.6 923.2 15 1317.7 1833.1 515.4
8 1255.2 1862.3 607.1 16 1324.0 1835.6 511.6
238 S. Song et al.

Fig. 9 Contour plot drawn from critical moments at SEN during steel grade transition: a new grade
20%, b new grade 80%, c transition slab duration, d the duration of the transition slab for the usual
range of temperature and density differences during steel grade transition

As shown in Fig. 9a, the smaller the density difference and the greater the tempera-
ture difference, the later the mass fraction of the new grade reaches 20%. At the same
time, the greater the density difference and the smaller the temperature difference,
the earlier the mass fraction of the new grade reaches 20%.
As shown in Fig. 9b, the lower the density difference and the greater the tempera-
ture difference, the earlier the mass fraction of the new grade reaches 80%. However,
with the increase in density difference and decrease in temperature difference, the
time when the mass fraction of the new grade reached 80% did not monotonously
increase.
As shown in Fig. 9c, a decrease in density difference and an increase in temper-
ature difference result in shorter transition slab production time. In this study, the
transition slab took the longest to be produced at a temperature difference of 50 K and
density difference of 100 kg/m3 . Notably, at greater density differences and lower
temperature differences, the time for producing the transition slab decreases. This
result is attributed to the combined flow and temperature fields of the molten steel in
the tundish.
At present, in most steel companies, the old and new grades involved in the steel
grade transition process have small differences in composition, meaning that the
temperature and density differences are also small. Thus, the casting sequence of
Effects of Temperature and Density on Transition Slab Length During … 239

steel grades needs to be arranged in such a way that the mixed casting is conducted
under the conditions in the lower half of Fig. 9d, and conditions in the upper half of
Fig. 9d are avoided. The length of the transition slab could be changed by changing the
casting sequence of the old and new grades. At the same time, the effect of density on
the length of the transition plate cannot be ignored. When the composition difference
between the old and new grades is small, the effect of a 1 kg/m3 change in density
on the length of the transition slab is equivalent to that of a temperature change of
3 K.

Conclusions

In this study, the effects of the density and temperature of the old and new grades
on the length of the transition slab were studied using numerical simulation. We
verified the model using physical simulations and industrial experiments to ensure
its accuracy. From the above analysis, the following conclusions can be drawn.
(1) The density of the new grade is lower than that of the old grade; therefore, the
tundish can preferentially discharge the old grade accumulated at the bottom of
the tundish, quickly reducing the proportion of the old grade in the tundish and
shortening the transition slab.
(2) During the steel grade transition, after the new and old grades are mixed in
the tundish, the composition of the molten steel entering the mold does not
change drastically. The simulation results show that the difference between the
compositions at the surface and center of the slab is small, indicating that the
flow field has little effect on the compositions at the surface and center of the
slab.
(3) With the decrease in density difference and increase in the temperature differ-
ence, the time required to produce the transition slab decreases. However, after
the generation time of the transition slab reaches the maximum, further increase
in the density difference and decrease in the temperature difference reduce the
time to produce the transition slab. This was attributed to the combined action
of the molten steel flow and temperature fields.
(4) If the composition difference between the old and new grades is small, the effect
of a 1 kg/m3 change in density on the length of the transition slab is equivalent
to that of a 3 K temperature change. Therefore, the effect of molten steel density
on the length of the transition slab cannot be ignored.
240 S. Song et al.

References

1. Huang X, Thomas BG (1993) Modeling of steel grade transition in continuous slab casting
processes. Metall Mater Trans B 24:379–393. [Link]
2. Thomas BG (1997) Modelling study of intermixing in Tundish and strand during a continuous-
casting grade transition. Iron and Steelmaker 24:83–96
3. Alizadeh M, Edris H, Pishevar AR (2008) Behavior of mixed grade during the grade transition
for different conditions in the slab continuous casting. ISIJ Int 48:28–37. [Link]
2355/isijinternational.48.28
4. Amorim LLC, Silva CA, Resende AD, Silva IA, Oliveira MJM (2018) A study of intermix in a
six-strand billet caster. Metall Mater Trans A 49:6308–6324. [Link]
018-1177-z
5. Siddiqui MIH, Kim M (2019) Two-phase numerical modeling of grade intermixing in a
steelmaking Tundish. Metals 9:40. [Link]
6. Kant S, Jha PK, Kumar P (2011) Investigation of effect of dam on intermixing during ladle
changeover in six strand billet caster tundish. Ironmaking Steelmaking 38:391–397. https://
[Link]/10.1179/1743281211Y.0000000007
7. Siddiqui MIH, Jha PK (2014) Assessment of turbulence models for prediction of intermixed
amount with free surface variation using coupled level-set volume of fluid method. ISIJ Int
54:2578–2587. [Link]
8. Thomas BG, Huang X, Sussman RC (1994) Simulation of argon gas flow effects in a continuous
slab caster. Metall Mater Trans B 25:527–548. [Link]
9. Li B, Tsukihashi F (2001) Effect of static magnetic field application on the mass transfer in
sequence slab continuous casting process. ISIJ Int 41:844–849. [Link]
ternational.41.844
10. Ahn JH, Yoon JK, lee JE (2002) Analysis of mixed grade transition in continuous thin slab
casting with EMBR. Met Mater Int 8:271.[Link]
11. Chattopadhyay K, Isac M, Guthrie RIL (2012) Modelling of non-isothermal melt flows in a
four strand delta shaped billet caster tundish validated by water model experiments. ISIJ Int
52:2026–2035. [Link]
12. Damle C, Sahal Y (1996) A Criterion for Water Modeling of Non-isothermal Melt Flows in
Continuous Casting Tundishes. ISIJ Int 36:681–689. [Link]
36.681
13. Li YG, Sun YH, Bai XS (2012) Numerical simulation of the flow, solidification, and solute
transport in a billet mold under electromagnetic stirring. Metall Res Technol 118:221. https://
[Link]/10.1051/metal/2021015
Mathematical Simulation Study
on the Effect of Nozzle Side Hole
Structure Parameters on the Behavior
of Molten Steel in Stainless Steel Mold

Si-kun Peng, Ming-mei Zhu, Kun-chi Jiang, and Cheng-hong Li

Abstract In this paper, the influence of structural parameters such as the area and
shape of the side hole of the submerged nozzle on the flow behavior of steel and the
distribution of liquid slag layer in the mold of large section stainless steel contin-
uous casting slab at high casting speed is studied by combining physical model
with mathematical simulation. The results show that with the increase of side hole
area, the fluctuation of molten steel level in the mold decreases; the impact depth
increases; the flow rate of molten steel decreases; the distribution of liquid slag layer
is gradually uniform, and the number of slag entrainment decreases. Compared with
the rectangular side hole shape, when the side hole shape is oval, the fluctuation of
molten steel level in the mold is reduced; the distribution of liquid slag layer is more
uniform, and the number of slag entrainment is reduced. The research results can
provide a theoretical basis for designing a reasonable submerged nozzle.

Keywords Stainless steel · Mold · Submerged nozzle · High efficiency continuous

casting

Introduction

Stainless steel has the characteristics of low solidification temperature, small thermal
conductivity, large thermal expansion coefficient, large shell shrinkage, long solidifi-
cation time, etc. Therefore, unlike other steels, the side holes of the submerged nozzle
of the mold are usually inclined upward to generate an upward reflux to promote
slagging. However, this nozzle design is also easy to cause excessive upflow in the
mold and large disturbance to the steel slag interface, leading to slag entrapment and
secondary oxidation of molten steel [1]. This is also often an important reason for
defects such as warping and sagging on the surface of stainless steel billet and leads
to surface defects of subsequent plates and strips. Therefore, it is very important to
design a reasonable nozzle structure to optimize the flow field of the mold to reduce

S. Peng · M. Zhu (B) · K. Jiang · C. Li

College of Materials Science and Engineering, Chongqing University, Chongqing 400044, China
e-mail: zhumingmei@[Link]

© The Minerals, Metals & Materials Society 2023 241

S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
242 S. Peng et al.

the occurrence rate of slag or inclusion defects in the hot rolling plate of a specific
section slab [2–4].
Using the methods of physical simulation and mathematical simulation, many
metallurgical workers have studied the influence of submerged nozzle structure on
the flow behavior of molten steel in slab mold. Deng Xiaoxuan [5] and other scholars
have conducted corresponding experiments on different shapes of the side holes of
the nozzle. It is concluded that the velocity of square nozzle is the highest. The
flow stream of elliptical nozzle and rectangular nozzle is more dispersed, and the
velocity distribution at the outlet is uniform, while the flow stream of square nozzle
is concentrated at the lower part of the side hole. Zhang et al. [6] studied the fluid
flow in the mold in combination with industrial test, water model, and numerical
simulation, and found that increasing the inner diameter of the nozzle and the area
of the side hole can slow down the liquid level fluctuation. Bai et al. [7] studied the
influence of nozzle structure on turbulence, and proposed that the vertical injection
angle and the area of the return flow area of the molten steel stream at the outlet of
the nozzle increase with the length of the side hole. Wu et al. [8] used numerical
simulation to study the influence of nozzle structural parameters on the liquid level
fluctuation, and found that increasing the height and width of the side hole of the
nozzle has little effect on reducing the jet exit velocity, but will reduce the effective
area fraction of the side hole. Najjar et al. [9] studied the influence of nozzle structure
parameters on the jet by means of finite element model, and found that reducing the
height of the side hole can increase the average jet velocity and turbulence intensity,
improve the effective area fraction of the side hole, which is conducive to reducing
the accumulation of alumina inclusions at the bottom of the nozzle and easing the
blockage of the nozzle.
In this paper, the large section stainless steel continuous casting rectangular slab
mold under high drawing speed is taken as the research object. The numerical simula-
tion and physical simulation experiments are carried out, and the influence of nozzle
structure parameters on the flow of molten steel is obtained. Finally, the theoretical
basis is provided for the design of reasonable mold nozzle.

Numerical Simulation and Physical Simulation

Physical Simulation and Research Methods

Physical Model

The hydraulic model is established based on the similarity principle. The geometric
similarity ratio of 0.6:1 was selected. The mold and submerged nozzle model are
made of transparent plexiglass, and the normal temperature water is used to simulate
the molten steel. The water model test device is shown in Fig. 1.
Mathematical Simulation Study on the Effect of Nozzle Side Hole … 243

Fig. 1 Schematic diagram

of experimental device. 1.
Tundish, 2. Nozzle, 3.
Stopper, 4. Mold, 5. Wave
height meter, 6. Computer, 7.
Water pump, 8. Flowmeter

Research Method

(1) The wave height meter is used to measure the fluctuation of each point in Fig. 2.
The X measuring point is at the narrow side; the Y measuring point is at 1/4 of
the mold, and the Z measuring point is near the water inlet. The data acquisition
time of each measuring point is 600 s, and the average wave height is obtained
by arithmetic mean. At the same time, blue ink is used as a tracer to find the
impact point position of the stream and record its impact depth through flow
field observation.
(2) In the experiment, the liquid slag layer distribution at different times under each
scheme was photographed with a camera. The maximum and minimum liquid
slag thickness on both sides of the crystallizer in each picture was obtained
through picture processing, and the maximum liquid slag thickness at different
times was averaged. Finally, the average value of liquid slag thickness difference
at different times is used to characterize the uniformity of liquid slag distribution.
At the same time, observe the liquid level of the crystallizer within 5 min, and
record the average number of slag inclusions per minute.

Fig. 2 Wave height

measurement points
244 S. Peng et al.

Numerical Simulation

Basic Assumptions

In order to facilitate the mathematical simulation, the assumptions for the calculated
model are as follows:
(1) The solid, liquid phase metal and liquid slag in the mold are all regarded as
Newtonian fluids, that is, incompressible viscous fluids, and they are all stable,
that is, the fluid density is constant;
(2) The influence of metal phase transformation on the temperature distribution in
the mold and the influence of solidified shell on the flow of molten steel in the
mold are ignored;
(3) The influence of billet shrinkage and mold vibration on the flow field of molten
steel during solidification is not considered;
(4) The liquid level of the crystallizer is free.

Basic Equation

The flow of molten steel in the mold is a turbulent three-dimensional unsteady flow.
The governing equations describing the flow of molten steel in the mold include conti-
nuity equation, momentum equation, energy equation, and standard k-ε equation; the
governing equation is as follows:

∂(ρu i )
=0 (1)
∂ xi

∂ ui u j ∂ 2ui ∂P
ρ = μeff − + ρgi (2)
∂x f ∂x j2 ∂ xi

∂ ρμ j H ∂ ∂T
= keff (3)
xj ∂x j ∂x j

∂k ∂ μeff ∂k ∂vi ∂vi ∂v j
ρvi = + μt + − ρε (4)
∂ xi ∂ x j σk ∂ xi ∂x j ∂x j ∂ xi

∂ε ∂ μeff ∂ε ε ∂v j ∂vi ∂v j ε
ρv j = + C1 μt + − C2 ρε (5)
∂x j ∂ x j σε ∂ x j K ∂ xi ∂ x j ∂ xi K

wherein ρ is the fluid density, kg m−3 ; μi is the velocity component, m s−1 ; xi is

directional component; p is pressure, Pa; gi is gravitational acceleration, m s−2 ; T
is temperature, K; keff is the effective thermal conductivity W m−1 K−1 ; k is the
turbulent kinetic energy, m2 s2 ; μeff is the effective viscosity coefficient, Pa s; ε is
the energy dissipation rate of turbulent flow, m2 s3 ; σk , σε , C 1 , and C 2 are empirical
constants with values of 1.0, 1.3, 1.44, and 1.92, respectively.
Mathematical Simulation Study on the Effect of Nozzle Side Hole … 245

Model Parameters and Boundary Conditions

Based on the consideration of symmetry, this paper takes 1/2 of the mold area as
the research object, and the calculation area of the mold is divided by hexahedral
non-uniform grid. The inlet of the model nozzle is defined as the velocity inlet, whose
velocity can be calculated by the billet drawing speed according to the flow balance
principle, and its temperature is the sum of the liquidus temperature and superheat
of the molten steel. The outlet of the calculation domain is defined as the velocity
outlet. The non-slip boundary condition is used for the wall surface, and the standard
wall function is used for the calculation of the flow field near the wall surface. The
normal derivatives of the boundary velocity and other variables on the symmetry
plane are zero. Table 1 shows process parameters required for numerical simulation.

Simulation Scheme

The cross-section size of the prototype mold studied in this paper is 200 mm ×
1530 mm, and the immersion depth of the long nozzle in the simulation process is
120 mm. In this paper, four types of nozzles with different geometries are considered.
As shown in Table 2 for structural parameters.

Table 1 Numerical simulation parameters of mold

Parameters Density of Viscosity of Inlet Inlet turbulent Turbulent
molten steel molten steel vertical kinetic energy dissipation rate at
speed inlet
Data 6908 0.00807 1.16 m/s 0.01 μ2in m2 /s2 1.5 /(0.5d m2 /s3
kin 0)
kg m−3 kg m−1 s−1

Table 2 Structural parameters of water inlet

Nozzle number Side hole Mesopore Side hole shape Side hole
inclination/° diameter/mm size/mm2
1 15 60 Rectangle 50 × 56
2 15 60 Rectangle 52 × 58
3 15 60 Rectangle 56 × 60
4 15 60 Oval like 56 × 16
246 S. Peng et al.

Table 3 Liquid level fluctuation

Nozzle number Wave height at narrow Wave height at 1/4 of Wave height at 1/2 of
edge (mm) mold (mm) mold (mm)
1 4.74 4.12 3.76
2 4.53 3.93 3.55
3 3.51 3.40 3.26
4 4.16 3.83 3.68

Table 4 Impact depth

Nozzle number Impact depth (mm)
1 138.3
2 141.0
3 148.0
4 145.3

Results and Analysis

Physical Simulation Experiment Results and Analysis

Liquid Level Fluctuation and Impact Depth

Table 3 shows the experimental results of the liquid level fluctuation of the crystallizer
under nozzles 1–4 and Table 4 shows the experimental results of the impact depth.
It can be seen from the analysis of the experimental data that with the increase of
the side hole area of the nozzle, the fluctuation of the liquid steel level in the mold
decreases, and the impact depth of the liquid steel increases. Under the same side
hole area, changing the shape of the side hole from rectangular to oval will increase
the liquid level fluctuation of molten steel in the mold and reduce the impact depth.

Uniformity of Liquid Slag and Number of Slag Curling

As shown in Table 5 for the experimental results of the uniformity of liquid slag
distribution in the crystallizer under No. 1 to No. 4 nozzles, and as shown in Table
6 for the experimental results of the number of slag entrapment. According to the
analysis of the experimental data, when No. 1 and No. 2 nozzles are used, there is
no liquid slag layer at the narrow side of the mold, and the exposed length is 7.3 mm
and 2.3 mm, respectively, and there is slag entrapment, which will lead to no flux
inflow at the meniscus, which will easily lead to cohesive breakout, thus affecting
the smooth progress of continuous casting. However, when No. 3 and No. 4 nozzles
are used, the liquid slag layer is evenly distributed, and there is no slag entrapment.
With the increase of the side hole area of the nozzle, the uniformity of the liquid
Mathematical Simulation Study on the Effect of Nozzle Side Hole … 247

Table 5 Uniformity of liquid slag distribution

Nozzle number Maximum Minimum Thickness Length of
thickness of liquid thickness of liquid difference/mm exposed part/mm
slag layer/mm slag layer/mm
1 19.3 0 19.3 7.3
2 18.7 0 18.7 2.3
3 18.3 3.3 15.0 0
4 16.7 6.3 10.3 0

Table 6 Times of slag

Nozzle number Slag curling times
curling per minute
1 0.7
2 0.3
3 0
4 0

slag layer becomes better. When the area of side holes is the same, the shape of side
holes changes from rectangular to oval, which will further improve the uniformity
of liquid slag layer.

Numerical Simulation Results and Analysis

The velocity distribution in the central section of the mold, the velocity distribution at
the outlet of the two side holes and the liquid surface velocity distribution are shown
in Figs. 3 and 4. Table 7 shows the maximum, minimum, and average velocities of
the crystallizer liquid level. It can be seen from Figs. 3 and 4 that the flow field of
the crystallizer using these four kinds of nozzles has formed an obvious upper loop,
and the upward stream forms a small reflux area in the upper part of the crystallizer,
which has a certain impact on the removal of inclusions. In this area, if the flow rate
of molten steel moving upward along the mold wall is too large, turbulence will be
generated, and the mold flux will be drawn into the molten steel. It can be seen from
Table 7 that with the increase of the side hole area, the maximum speed, minimum
speed, and average speed of the liquid steel decrease, while changing the shape of the
side hole from rectangular to oval like will reduce the maximum speed and minimum
speed of the liquid surface of the mold, while the average speed of the liquid surface
increases.
248 S. Peng et al.

(a)1# (b)2# (c)3# (d)4#

Fig. 3 Velocity distribution in central section

(a)1# (b)2#

(c)3# (d)4#

Fig. 4 Liquid surface velocity distribution

Table 7 Velocity value of

Nozzle Maximum Minimum Average
liquid level
number velocity (m/s) velocity (m/s) velocity (m/s)
1 0.365387 0.000803 0.150020
2 0.354292 0.001503 0.149764
3 0.351240 0.001046 0.140803
4 0.341723 0.000162 0.145382
Mathematical Simulation Study on the Effect of Nozzle Side Hole … 249

Conclusion

The effects of the structural parameters such as the area and shape of the side hole
of the submerged nozzle on the steel flow behavior and the distribution of liquid
slag layer in the mold of large section stainless steel continuous casting slab at
high drawing speed were studied by combining physical model with mathematical
simulation. The results show that:
(1) The liquid level fluctuation of molten steel in the mold decreases with the
increase of the side hole area, and the liquid level wave heights at the narrow
side 1/4 and 1/2 of the mold decrease from 4.74 mm, 4.12 mm, and 3.76 mm
to 3.51 mm, 3.40 mm, and 3.26 mm, respectively; the liquid level fluctuation
increases as the shape of the side hole changes from rectangular to oval like, and
the liquid level wave heights at the narrow side 1/4 and 1/2 of the mold increase
from 3.51 mm, 3.40 mm, and 3.26 mm to 4.16 mm, 3.83 mm, and 3.68 mm,
respectively.
(2) The impact depth of molten steel in the mold increases with the increase of
the side hole area, from 138.3 mm to 148.0 mm; as the shape of the side hole
changes from rectangular to oval like, it decreases from 148.0 mm to 145.3 mm.
(3) The uniformity of liquid slag in the mold becomes better with the increase
of the area of the side hole, and when the area of the side hole is too small,
there will be no liquid slag layer at the narrow edge of the molten steel, and
there will be slag curling, which will lead to no inflow of protective slag at the
meniscus, resulting in cohesive steel leakage. Changing the shape of side hole
from rectangle to ellipse will further improve the uniformity of liquid slag.
(4) With the increase of the area of the side hole of the nozzle, the flow pattern of
molten steel in the mold has no obvious change, and the flow rate has decreased.
The maximum velocity of the free surface has decreased from 0.36539 to
0.35124 m/s, and the average velocity of the free surface has decreased from
0.15002 to 0.14080 m/s. The shape of the side hole changed from rectangle
to ellipse; the flow pattern of molten steel in the mold did not change signif-
icantly; the maximum velocity of free liquid surface decreased from 0.35124
to 0.34172 m/s, and the average velocity of free liquid surface increased from
0.14080 to 0.14538 m/s.

Acknowledgements This research was funded by the Natural Science Foundation of Chongqing,
China, grant number cstc2019jcyj-msxmX0526.

References

1. Youlin J, Yanping B, Jianhua L et al (2009) Simulation of steel/slag interface behavior

and analysis of slag entrapment in stainless steel slab continuous casting mold. J Eng Sci
31(005):618–624
250 S. Peng et al.

2. Jingfeng C Control and optimization of wide width stainless steel continuous casting process.
Wuhan University of Science and Technology
3. Yingdong W, Zhongqiu L, Mingjie Q et al (2020) Experimental study on steel/slag interface
fluctuation and slag entrainment behavior in continuous casting mold. J Mater Metall 19(1):8
4. Gang L, Yong G, Chong L et al (2020) Research on the improvement and control of edge warping
defects of ultra low carbon if steel. Continuous Casting 45(2):6
5. Xiaoxuan D, Xiao X, Xinhua W et al (2014) Effect of nozzle outlet shape on flow field
characteristics in slab continuous casting mold with high drawing speed. J Iron Steel Res
26(7):35–40
6. Zhang LF, Yang SB, Cai KK et al (2007) Investigation of fluid flow and steel cleanliness in the
continuous casting strand. Metall Mater Trans B 38(1):63–83
7. Bai H, Thomas BG (2001) Turbulent flow of liquid steel and argon bubbles in slide-gate tundish
nozzles: part 2 effect of operation conditions and nozzle design. Metall Mater Trans B 32(2):269–
284
8. Wu DF, Cheng SS (2008) Effect of SEN design on surface fluctuation and solidifying shell in
slab mold and its optimization. Acta Metall 21(5):341–350
9. Najjar FM, Thomas BG, Hershey DE (1995) Numerical study of steady turbulent flow through
bifurcated nozzles in continuous casting. Metall Mater Trans B 26(4):749–765
Author Index

A Higley, Emily A., 3

Anderson, Alexandra, 3 Hou, Zibing, 65, 97
Andersson, Stefan, 25
An, Haifei, 213
J
Jiang, Kun-chi, 241
B Jiang, Tao, 185
Bai, Hao, 213 Jung, S.-M., 171
Barba, Daniel, 199

L
C Li, Cheng-hong, 241
Casata, Massimiliano, 199 Li, Lihong, 213
Chen, Hao, 213 Liu, Guiqing, 185
Chen, Wei, 15 Liu, Qian, 65
Chiwandika, E. K., 171 Li, Yaoguang, 227
Chu, Shaojun, 213 Lyu, Binyu, 125

D O
Dalaker, Halvor, 25 Oterdoom, H. J., 145
Dong, Zhonglin, 185 Oxtoby, Oliver F., 113
Dugic, Izudin, 83

P
E Pan, Wen, 35
Einarsrud, Kristian Etienne, 25 Peng, Si-kun, 241
Peng, Zhiqiang, 65

G
Grogan, Joseph, 3 R
Guo, Dongwei, 97 Reuter, M. A., 145
Guo, Kunhui, 97 Reynolds, Quinn G., 113

H S
Haraldsson, Hákon Valur, 161 Sævarsdóttir, Gurún, 161
© The Minerals, Metals & Materials Society 2023 251
S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
252 Author Index

Shen, Huiqiang, 47 Wen, Guanghua, 65

Silaen, Armin K., 3 Wilkinson, Toby, 199
Song, Sicheng, 227
Sun, Yanhui, 227
X
Xiao, Pengcheng, 47
T Xu, Bin, 185
Tangstad, Merete, 161
Tesfahunegn, Yonatan A., 161
Tian, Weijian, 213
Traustason, Halldór, 161 Z
Zeng, Zihang, 97
Zhang, Bangsheng, 185
V Zhang, Fan, 185
Vachaparambil, Kurian J., 25 Zhang, Haihui, 47
Zhang, Lifeng, 15, 75, 125, 135
Zhang, Yapeng, 35
W Zhang, Yuexin, 125
Walla, Nicholas J., 3 Zhao, Yanyu, 15
Wang, Dongqing, 35 Zhao, Zhixing, 35
Wang, Fang, 185 Zhou, Chenn Q., 3
Wang, Jujin, 125 Zhu, Ming-mei, 241
Wang, Weijian, 135 Zhuo, Chao, 227
Wang, Yadong, 75 Zietsman, J. H., 145
Subject Index

A F
Additive manufacturing, 199 Feed-to-power, 147, 154, 158
Arsenic, 171, 172, 176, 179, 181, 183 Ferroalloy, 113, 163, 213–216, 220, 221,
225
Ferrochrome, 145, 147, 148, 154
Flocculation flotation, 185–187, 189,
C
192–195
Calcium treatment, 135–140
Fluctuation of centerline solid fraction, 98
Calcium yield, 135–139
Fluctuation of local cooling rate, 97–99,
Casting simulation, 85
103–108
Coke breeze combustion, 40, 42
Fluctuation of solidification end point, 105
Computational Fluid Dynamics (CFD),
116, 126
Computational modelling, 116
Continuous casting, 47–50, 57, 65–69, 71, G
75, 97–99, 102, 103, 105–108, 227, Gas porosity, 83, 84, 89
242, 246 Gas-stirred ladle, 126
Continuous casting slab, 65, 241, 249
Cooling, 11, 50, 59, 65, 67, 71, 72, 75–78,
80, 81, 97–99, 101–108, 176, 181, H
213–215, 219–225, 230 Heat transfer, 3–5, 16, 27, 48–50, 56, 59,
Copper, 50, 56, 57, 213, 215, 216, 218, 60, 70–73, 75, 76, 101, 122, 213,
221, 223–225 215, 218, 219, 221, 222, 227
High efficiency continuous casting, 242,
246, 249
D Hot filtration, 185, 186, 194, 195
Data mining, 65, 66 Hot rolled coils, 65–70
DC furnace, 145–147, 150
Desulfurization, 15, 16, 19, 20, 22, 36,
125–128, 130, 131 I
Dynamic simulation, 151, 158 Intermix, 227
Iron and steel, 35, 36, 113, 214

E
Elemental sulfur, 185–187, 191, 192, 194, K
195 KR process, 15, 125
© The Minerals, Metals & Materials Society 2023 253
S. Wagstaff et al. (eds.), Materials Processing Fundamentals 2023, The Minerals, Metals
& Materials Series, [Link]
254 Subject Index

L S
Laser Powder Bed Fusion (LPBF), 199, Segregation, 75, 97, 98, 102–105, 107, 108
203, 206, 208 Simulation, 3–12, 16, 26, 29–31, 36, 86, 91,
Lead recycling, 4 99, 113, 116–122, 125, 126, 129,
Low oxygen gas, 35, 36 152, 154, 157, 159, 162, 217, 227,
228, 231, 232, 235, 236, 239, 241,
242, 244, 245, 247, 249
M Sinter, 36, 37, 39, 173, 181
Machine learning, 200 Sintering, 31, 32, 35–40, 42, 43, 171, 172,
Microsilica, 25–32 174, 179–181
Modeling and simulation, 228 Slag foaming, 113–122
Mold, 16, 47–51, 56–61, 65, 67, 69–73, 76, Software, 35–38, 85, 91, 116, 117, 135,
77, 84, 88, 89, 101, 102, 227, 228, 136, 138–140, 151, 155, 172, 173,
230, 231, 234, 236, 239, 241–247, 205, 213, 215, 216, 219
249 Solidification, 48, 59, 67, 75, 76, 83, 85,
Mould, 213–216, 218–225 97–100, 102–108, 214–217, 220,
Multiphase flow, 15, 16, 18–20, 116, 117 221, 227, 241, 244
Stainless steel, 207, 241, 242, 249
Steel casting, 49, 83
N Steel grade transition, 227, 228, 234–239
Non-isothermal heating, 180–182 Submerged nozzle, 241, 242, 248
Numerical Model, 5, 16, 97, 98, 101, 103, Sulfide minerals, 185, 186, 191, 195
107, 108, 125, 131 Surface defects, 48, 66, 70, 241
Numerical Modeling, 4

T
O Temperature field, 26, 32, 36, 213–216,
OpenFOAM, 25, 26, 31, 116, 117, 122 218–222, 231, 238, 239
Operation, 4, 5, 7, 9, 10, 37, 114, 137, 139,
Thermodynamic model, 135, 136, 138, 139
140, 145, 146, 148, 149, 151, 152,
Thermodynamics and kinetics, 36
155–158, 161, 162, 166, 168, 169,
Titanium alloy, 199
186, 213, 225
Transition slab, 227, 228, 231, 234,
236–239
Tundish, 67, 70, 227, 228, 230–236, 238,
P
239
Parameters optimization, 66
Percentage weight loss, 171
Pressure acid leaching residue of zinc
sulfide concentrate, 185, 186, 195 V
Process technology, 47 VOF-DPM, 15, 20, 116, 125, 127, 131
Pyrometallurgy, 4, 113, 114 Volatilization, 171–173, 177–179, 181, 182
Volatilization rate, 177–179

R
Reductant, 114, 145–152, 154–158 W
Reverberatory Furnace, 3–6, 8, 9, 12 Wearing impellers, 15