Textiles Chapter One

TEXTILE PROCESSES

TEXTILE PROCESSES
QUALITY CONTROL AND DESIGN OF EXPERIMENTS
FIRST EDITION

Professor D.Sc. GEORGI BORISOV DAMYANOV Associate Professor PhD DIANA STOYANOVA GERMANOVA-KRASTEVA
Department of Textile Engineering Technical University of Soa

MOMENTUM PRESS, LLC, NEW YORK
Textile Processes: Quality Control and Design of Experiments Copyright Momentum Press, LLC, 2013. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any meanselectronic, mechanical, photocopy, recording, or any otherexcept for brief quotations, not to exceed 400 words, without the prior permission of the publisher. First published by Momentum Press, LLC 222 East 46th Street, New York, NY 10017 www.momentumpress.net ISBN-13: 978-1-60650-387-4 (hard back, case bound) ISBN-10: 1-60650-387-1 (hard back, case bound) ISBN-13: 978-1-60650-389-8 (e-book) ISBN-10: 1-60650-389-8 (e-book) DOI: 10.5643/9781606503898 Cover design by Jonathan Pennell Interior design by Exeter Premedia Services Private Ltd., Chennai, India 10 9 8 7 6 5 4 3 2 1 Printed in the United States of America
CONTENTS
PREFACE ABOUT THE AUTHORS LIST OF FIGURES LIST OF TABLES PART I: INTRODUCTION TO MATHEMATICAL STATISTICS I.1. GENERAL TERMS AND DEFINITIONS Parameters and numerical characteristics of the random variable Characteristics for location Dispersion characteristics Moments Properties of numerical characteristics I.2. LAWS OF RANDOM VARIABLES DISTRIBUTION Continuous distributions Discrete distributions I.3. STATISTICAL ESTIMATES Conducting the test Point estimates Interval estimates Condence intervals Condence intervals of the estimates in cases of normal distribution Condence interval of the l parameter for Poisson distribution Condence interval of the parameter p for binomial distribution I.4. STATISTICAL PROCESS CONTROL AND CONTROL CHARTS Statistical process control Control charts Design of statistical control charts Types of control charts xi xiii xv xvii 1 3 6 6 7 9 11 13 13 16 21 21 24 25 28 28 30 31 33 33 36 37 38
VI CONTENTS
I.5. CORRELATION ANALYSIS Coefcient of linear correlation Coefcient of determination, anticorrelation, and indeterminate coefcients Correlation in case of alternative indicatorsthe four-eld method Multiple and partial correlation I.6. ANALYSIS OF VARIANCE Single-factor analysis of variance Multifactor ANOVA PART II: DESIGN OF AN EXPERIMENT II.1. MAIN CONCEPTS IN MATHEMATICAL MODELING AND OPTIMIZATION II.2. CHOICE OF PARAMETERS OF OPTIMIZATION II.3. CHOICE OF INPUT FACTORS Methods of rank correlation Coefcient of rank correlation Coefcient of concordance Random balance method Design of the experiment Construction of a diagram of dispersion Separation of essential factors II.4. MAIN STAGES OF EXPERIMENTAL MODELING II.5. REGRESSION ANALYSIS II.6. FULL FACTORIAL EXPERIMENT Properties of the extended matrix of FFE Types of matrices Stages of derivation of the model Calculation of coefcients Verication of reproducibility of the process Calculation of the test variance Determination of the variance of regression coefcients Verication of the signicance of regression coefcients Registration of the derived model Calculation of the values of the output variable on the model Verication of the model adequacy II.7. FRACTIONAL FACTORIAL EXPERIMENT Stages of construction of a fractional factorial experiment Determination of the minimum number of tests for deriving a linear model Choice of main and additional factors
65 66 66 68 70 73 73 75 79 81 85 89 93 94 98 103 103 104 107 109 113 123 124 125 125 129 129 130 130 130 130 130 130 133 134 135 135
CONTENTS VII
Composing the design of the experiment Setting the determining contrasts Setting the generalized determining contrast Mixing the coefcients II.8. STATISTICAL METHODS FOR MOVEMENT TO AN OPTIMAL AREA BoxWilson method Condition for application of the method Method principle Method application conditions Disadvantages of the method Application Simplex method Simplex property Criterion for reaching the optimal area Specifying the optimal area Construction of the initial simplex Calculating the values of the coordinate points Calculating the coordinates of the mirrored point apex Determining the coordinates of the starting points Filling in the simplex table Determining the coordinates of an additional point Formation of a new simplex II.9. INVESTIGATION OF THE OPTIMUM AREA: COMPOSITE DESIGNS OF SECOND ORDER Orthogonal central composite experiment Number of tests Design of the experiment Determination of regression coefcients Determination of regression coefcients variances Signicance of coefcients of regression equation Recording of regression equation Verication of adequacy of the model Determination of the number of tests Determination of the size of the star arm, a, and the value, k Plan of the experiment Determination of the regression coefcients Determination of the variances of the regression coefcients Signicance of the coefcients of the regression equation Record of the regression equation Verication of the model adequacy Rotatable central composite experiment Number of tests Design of the experiment Determination of regression coefcients
135 136 136 137 139 139 140 140 141 141 141 147 147 148 148 148 150 150 151 151 152 152 153 154 154 154 155 156 156 157 157 158 158 158 158 160 161 162 162 163 163 164 165
VIII CONTENTS
Variance of regression coefcients Signicance of coefcients of regression equation Verication of adequacy of the model Determination of regression coefcients Determination of variance of reproducibility Determination of variances of regression coefcients Verication of signicance of regression coefcients Recording of the model Verication of model adequacy Optimal composite experiment Number of tests Design of the experiment Determination of regression coefcients Variance of regression coefcients Signicance of coefcients of regression equation Verication of adequacy Determination of the regression coefcients Determination of the variances of the regression coefcients Verication of the signicance of the coefcients of the regression equation Record of the model Verication of the model adequacy II.10. OPTIMIZATION OF TARGET FUNCTION Canonical analysis Algorithm for reduction to a canonical form Determination of the surface type Differentiation of target function Solving the system of linear equations Determination of the extreme value of output parameter Determination of the rotation angle Determination of regression coefcients in canonical equation Additional verication of calculation correctness Determination of surface type Determination of the optimal parameters of the cylinder drawing device II.11. TAGUCHI METHODS Innovations in the sphere of the designed experiment Off-line methods of control System design Parameter design Design of the parameter tolerances (tolerance plan) Loss function Application of the function
165 166 166 167 169 169 170 171 171 171 171 172 172 173 173 173 174 176 176 177 177 179 180 180 181 184 185 185 185 185 186 186 187 189 189 190 190 195 202 202 204
CONTENTS IX
APPENDIX 1: STUDENTS T-DISTRIBUTION APPENDIX 2: CHI-SQUARED 2-DISTRIBUTION APPENDIX 3: FISHERS F-DISTRIBUTION a = 0.05 APPENDIX 4: FISHERS F-DISTRIBUTION a = 0.01 APPENDIX 5: COCHRANS CRITICAL VALUES APPENDIX 6: DENSITY OF NORMAL DISTRIBUTION N(0, 1) APPENDIX 7: PROBABILITY P FOR Sd Sdo APPENDIX 8: RANDOM NUMBERS BIBLIOGRAPHY INDEX
209 211 213 217 221 225 227 229 231 235
PREFACE
This book has been written for textile experts, researchers, students, and PhD students in the eld who work on the control and optimization of textile processes and the quality of the manufactured products. The book presents a review of the methods for experiment design and for deriving and optimization of mathematical models. The individual models are illustrated by numerical examples, which allow for easier comprehension and implementation of the methods in practice. The rst part of the book is for those experts who are not familiar with the foundations of mathematical statistics. There, in a very accessible form, they will nd a review of the main notions of mathematical statistics used in textile processes modeling. The review covers the main properties of the laws on distribution of random variables, the types of statistical estimates, and the ways to determine them, which are used most often in the textile practice. There is a description of the algorithms for examination of various statistical hypotheses, as well as their graphic application for control of the technological processthe control charts. There is a brief presentation of the correlation analysis and the analysis of variance (ANOVA), their signicance, and mode of application. The Design of experiments section gives a description of the particulars of mathematical modeling and preparation of experiments. There is a review of the capabilities and the main stages of the regression analysis. The full and fractional factorial experiments for derivation of linear models are claried with examples. There is a description of the simplex method and the BoxWilson (experiment design) method for denition of the extreme area. There is a review of the three most wide-spread designs for derivation of second order mathematical models: the rotational central composite design, the orthogonal central composite design, and the optimal design. You will nd examples of the ways to present in a graphic manner the area, described by the model, and the ways to optimize the model. Explanations have also been provided on how to interpret the results of the designed experiment. Special attention has been given to the specics of the Taguchi method of experiment design. We have strived to create a book which claries, in an accessible manner, the methods for experiment design and for optimization of the derived models. We hope that with the examples provided we will be useful to the experts and researchers in their efforts to increase the quality in the area of textiles. Georgi Borisov Damyanov Diana Germanova-Krasteva
xi
ABOUT THE AUTHORS

Professor D.Sc Georgi Borisov Damyanov was born on 1927 in Soa. He obtained a degree in Textile Engineering (1951) and Doctor of Science (1975) from Moscow Textile University and has specialized in textile technology in Poland, Russia, England, and the USA. Professor Damyanov started his professional career as a manager and chief engineer in various textile factories in Soa (19511955). From 1985 to 1997 he worked as member and president of the board of directors of textile companies. For many years he has been president of Bulgarian Association of Textile Engineers. In 1956 he entered academia as a lecturer, became associate professor, and ultimately a full professor in 1969 in the Textile department of the Technical University of Soa. He has lectured on textile technology at the University of Economics, Academy for Social Sciences and also in Poland, USA, and North Korea. For many years Professor Damyanov has been head of the Textile department and also Dean and Prorector in the Technical University Soa. His scientic interests include investigation of textile machinery and technology, application of radioactive isotopes in textile technology, and mathematical optimization of textile processes. Professor Damyanov is the author of over 250 research works published in professional textile journals in ten countries. As author and co-author has also written 41 textile textbooks and monographs, four of which are in English and Russian. In 1966 Professor Damyanov was a visiting professor in the Textile department of the University of Manchester, England and in 19681969 he was a visiting professor at MIT in Boston, USA. Associate Professor. PhD Diana Stoyanova Germanova-Krasteva was born in 1967 in Soa. In 1991 she graduated from the Technical University of Soa as a Master of Science in Textile Engineering. She began working as an editor in the scientic journal Textile Industry issued by the Scientic and Technical Union of Textile, Clothing and Leather. In 1993 Diana Krasteva became professor in the Department of Textile Engineering. She leads classes in textile testing, textile materials, quality management, textile production, and others. Her interests are also in the eld of application of simulation modeling and statistical methods for modeling and optimization of textile processes.
xiii
XIV ABOUT THE AUTHORS
In 2005 she obtained a PhD degree and since 2007 is an Associate Professor in the Department of Textile Engineering at the Technical University of Soa. Professor Krasteva specializes in Belgium, Spain, and Germany. She is the author of over 50 publications in Bulgarian, German, Swiss, English, Polish, and Romanian scientic journals. She has published three textbooks in the eld of textiles, textile testing, and statistical analysis.
LIST OF FIGURES
Figure I.1.1. Figure I.1.2. Figure I.1.3. Figure I.1.4. Figure I.1.5. Figure I.1.6. Figure I.1.7. Figure I.1.8. Figure I.1.9. Figure I.2.1. Figure I.2.2. Figure I.2.3. Figure I.2.4. Figure I.2.5. Figure I.3.1. Figure I.3.2. Figure I.3.3. Figure I.3.4. Figure I.4.1. Figure I.4.2. Figure I.4.3. Figure I.4.4. Figure I.4.5. Figure I.4.6. Figure I.4.7. Figure I.4.8. Figure I.4.9. Figure I.4.10. Figure I.4.11. Figure I.4.12. Figure I.4.13. Figure I.4.14. Figure I.4.15. Figure I.4.16. Figure I.4.17. Figure I.4.18. Figure I.4.19. Figure I.4.20. Figure II.1.1. Function of distribution of a continuous variable Function of distribution of a discrete variable Probability distribution function Order of distribution Quantiles of the distribution function Probability distribution functions with different variances Distributions with positive asymmetry, symmetrical distribution, and negative asymmetry Distribution with positive, zero, and negative kurtosis Different cases of correlation dependencies Movement of the normal distribution in conformity with the mathematical expectation Characteristic values for the normal distribution Uniform distribution Binomial distributions Poisson distributions Histogram and polygon Cumulative curve Students t-distribution Chi-squared distribution Steps for quality improvement Principle of SPS Some typical cases of distribution of the controlled variable Graphical presentation of the control chart limits Algorithm for calculation of the control limits x / S control chart x / R chart with a moving average Control chart for practical limits Control charts for batches of similar products Control charts for small batches with different distribution characteristics Control charts for group values Control charts for extreme values x -chart during its calculation Final x / S chart for the example Final x / R control chart for the example p-chart p-chart in case of 100% control - and u-charts Final version of the p-chart for the example p-chart for 100% control for the example Scheme of the inuences on an object 4 5 5 5 6 8 10 10 11 14 15 16 17 19 23 24 29 30 34 34 35 36 37 41 42 44 46 48 49 49 51 52 53 55 56 58 60 61 82
xv
XVI LIST OF FIGURES Figure II.3.1. Figure II.3.2. Figure II.3.3. Figure II.3.4. Figure II.3.5. Figure II.5.1. Figure I.6.1. Figure II.8.1. Figure II.8.2. Figure II.8.3. Figure II.8.4. Figure II.8.5. Figure II.9.1. Figure .10.1. Figure .10.2. Figure .10.3. Figure .10.4. Figure .10.5. Figure .10.6. Figure I.11.1. Figure I.11.2. Figure I.11.3. Figure I.11.4. Figure I.11.5. Factors ranked by the extent of inuence on the output parameter Ranking of the factors according to the summarized opinion of the experts Diagram of dispersion Diagrams of dispersion of xi and xj Reconstructed diagram of dispersion for the remaining factors Graphical dependency between the measuring and simulation results Geometrical interpretation of experimental points Graphical presentation of the method of Box and Wilson Examples of simplices Rolling of the simplex Construction of an initial simplex Construction of an initial simplex Experimental point for composite designs Translation of the extremal point and rotation of the axes Elliptic paraboloid surface Equipotential lines of elliptic paraboloid surface Hyperbolic paraboloid surface Equipotential lines of hyperbolic paraboloid surface Surface described by the model The House of quality FMEA chart Classication of the acting factors Matrix form of the experiment design Taguchi loss function 90 102 105 106 107 120 124 140 147 147 148 149 154 180 182 183 183 184 186 193 194 196 197 203
LIST OF TABLES
Table I.3.1. Calculation of the data for drawing of the histogram, polygon, and cumulative curve Table I.3.2. Values of the coefcient an in accordance with the number of trials n Table I.3.3. Values of the coefcient A in accordance with the number of trials n Table .4.1. Percentage of defective production for a normally distributed random variable Table .4.2. Calculation of the control limits for x / S control chart Table .4.3. Auxiliary coefcients for calculation of the control limits Table .4.4. Standard deviation in accordance to the selected AQL Table .4.5. Calculation of the control limits for x / S control chart Table .4.6. Calculation of the control limits for x / R control chart Table .4.7. Calculation of the control limits for control chart for the individual and absolute values of the sequential differences Table .4.8. Data and statistical estimates for the example Table I.4.9. Calculation of the control limits for p-chart Table I.4.10. Calculation of control limits for np-chart Table I.4.11. Calculation of control limits for c-chart Table I.4.12. Calculation of control limits for u-chart Table I.4.13. Calculation of control limits for p-chart Table I.4.14. Number of established defects for the samples Table I.5.1. Data and calculation of the linear correlation coefcient Table I.5.2. Quadruple table for calculation of the degree of the causal dependency Table I.5.3. Input table for calculation of the degree of the causal dependency for the example Table I.6.1. Input table for single-factor ANOVA Table I.6.2. Formulae for single-factor ANOVA in tabular form Table I.6.3. Measured values of the yarn strength Table I.6.4. Calculation of the variance components for the example Table I.6.5. Input table for multifactor ANOVA Table I.6.6. Formulae for multifactor ANOVA in tabular form Table I.6.7. Measurement data Table I.6.8. Calculated mean values for the example Table I.6.9. Sums of the squares, degrees of freedom, and variance components for the example Table I.3.1. Ranking from the experts Table I.3.2. Computation of Spearmans coefcient of rank correlation Table I.3.3. Ranking with coinciding ranks Table I.3.4. Matrix of the ranks
23 27 28 35 40 40 40 41 43 45 50 55 56 57 58 59 62 67 69 69 74 74 75 75 76 76 77 78 78 94 95 96 98
xvii
XVIII LIST OF TABLES
Table I.3.5. Table I.3.6. Table I.3.7. Table I.3.8. Table II.5.1. Table II.5.2. Table II.5.3. Table II.5.4. Table II.6.1. Table II.6.2. Table II.6.3. Table II.6.4. Table I.7.1. Table II.8.1. Table II.8.2. Table II.8.3. Table II.8.4. Table II.8.5. Table II.8.6. Table II.8.7. Table II.8.8. Table II.8.9. Table II.8.10. Table II.9.1. Table II.9.2. Table II.9.3. Table .9.4. Table .9.5. Table .9.6. Table .9.7. Table .9.8. Table .9.9. Table .9.10. Table .9.11. Table .10.1. Table .11.1. Table .11.2.
Numbering of the factors Experts ranking of the factors Full factorial experiment for both groups Common matrix of the experiment Data arrangement for conduction of regression analysis Measuring and simulation results Measured ber and yarn characteristics Summary output from Excel Design of the experiment with two input factors Design of the experiment with three input factors Natural values of the levels of the factors and variance interval FFE table for the example Plan of the experiment for the example Main factor levels, interval of variation, upper and lower levels Experimental data and calculations for obtaining of the rst model Computational table for moving to the optimal area Main factor levels, interval of variation, upper and lower levels for the new FFE Experimental date for the FFE Computational table for moving to the optimal area according to the new direction Determination of the coordinates of the initial points Determination of the coordinates of the initial points Coordinates of the initial points Coordinates of the simplices Orthogonal central composite experiment for three factors Plan of the experiment for the example Number of tests for the rotatable central composite experiment Design of the rotatable central composite experiment for two factors Values of the auxiliary coefcients Levels of factors in natural and coded values Design of the experiment, measured values of the output parameter and supporting products for determination of regression coefcients Design of the optimal composite experiment for three factors Values of the auxiliary coefcients Experimental data for the example Optimal composite plan for the example Values of the semiaxes in the variation of yR Taguchis orthogonal table Taguchis dispersion table
100 101 104 104 114 120 121 121 124 124 127 128 136 141 142 144 145 146 146 149 150 151 152 155 159 164 164 165 167 168 172 173 174 175 187 199 200
PART I
INTRODUCTION TO MATHEMATICAL STATISTICS
The production of high quality textiles is based on the excellent knowledge of the properties of textile materials. The establishing of these properties is achieved by testing. The test results are data, the processing of which is performed according to the relationships of the theory of probability and mathematical statistics. The mathematical foundations of statistics were laid during the seventeenth century with the development of probability theory by Blaise Pascal and Pierre de Fermat. The extensive use of computers nowadays allows the mass application of statistical calculations on large data volumes and has lead to the development of new computational methods. In Part I, the main concepts and methods of processing and analyzing the results obtained, and applying them in the textile processes evaluation will be presented.
I.1. GENERAL TERMS AND DEFINITIONS

Mathematical statistics views the regularities and methods used for the study of mass phenomena that are characterized by individuals that are homogenous in a qualitative aspect. The combination of a large number of individuals (cases) is hereby referred to as statistical population. The statistical population can be: General, which includes all cases of the studied mass phenomenon, or Representational (sample), which includes part of the cases of the general population, the characteristics of which can be used to draw conclusions regarding the characteristics of the general population.
The volume of the general population (the batch) is usually assigned with N and the sample with n. Mathematical statistics is based on probability theory that uses a series of specic terms such as test, event, random variable, probability, and so forth. A test (observation) is the implementation of a specic sum of conditions. An event is any fact that has occurred as a result of holding the test. Probability of occurrence of a specic event is a number that expresses the possibility of occurrence of the specic event. A valid event is an event which always occurs as a result in the test. The probability of occurrence of this event is 1 or 100%. An impossible event is an event which never occurs as a result in the test. The probability of occurrence of this event is 0. A random event is an event which could or could not occur as a result of the test. The probability of its occurrence ranges between 0 and 1, or between 0 and 100%. A practically impossible event is the event for which the probability of occurrence is very close to 0, for example 0.1, 0.05, 0.01, 0.001. A practically valid event is an event for which the probability of occurrence is very close to 1, for example 0.9, 0.95, 0.99, 0.999. A signicant level is the probability of occurrence for a practically impossible event . The condence level is the probability = 1 for the occurrence of a practically valid event. In the textile production sector if no special instructions have been given, operations will be performed at the signicant level = 0.05. A random variable is a variable, which as a result of the performance of the test can adopt different values which could not be predicted in advance.
4 TEXTILE PROCESSES
A random variable can be discrete or continuous. A discrete random variable can adopt only separate values isolated from one another, for example, the number of defective articles in a single batch, the number of bers in a band, the number of stops of the machine. A continuous random variable can be adopted by any value in a given nite or innite interval, for example, yarn strength. A random variables distribution expresses the dependence between the possible values of the random variable and their respective probabilities. Random variables distribution can be described by means of the distribution function, or the density of probability distribution, or the order of distribution. Distribution function F(x), also referred to as integral distribution function or integral distribution principle, expresses the probability of the random variable to remain a smaller number than the number : F ( x) = P { X < x} . (.1.1)
Figure .1.1 presents the function of distribution of a continuous variable. The gure indicates that in this specic case, the probability P { X < x = 0} = 0.5.
F(x)
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 3 2 1 0 1 2 3 x
Figure I.1.1. Function of distribution of a continuous variable.
In case the variable is discrete, the distribution function F(x) has a terraced form (Figure .1.2). In case of Figure .1.2, the following probability has been determined: P { X < x = 4} = 0.265. The density of probability distribution f (x), also referred to as density of probabilities, differential principle of distribution, or differential distribution function, is determined as a derivative of F(x) along x: f ( x) = dF ( x) . dx (I.1.2)
Distribution density exists only for continuous variables and is of the type shown in Figure .1.3.
PART I: GENERAL TERMS AND DEFINITIONS 5

F(x)
0.4
0.3
0.2
0.1
0 0 1 2 3 4 5 x 6
Figure I.1.2. Function of distribution of a discrete variable.
f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 4 3 2 1 0 1 2 3 4
Figure I.1.3. Probability distribution function.
The order of distribution consists of the possible values of the discrete random variable and the probability for it to have the respective values. The order of distribution can be represented as a table or a graphic (Figure .1.4) and the separate discrete values are not connected.
f (x)
0.4 0.3 0.2 0.1 0
10
15
20
Figure I.1.4. Order of distribution.
6 TEXTILE PROCESSES
PARAMETERS AND NUMERICAL CHARACTERISTICS OF THE RANDOM VARIABLE

Very frequently, instead of the distribution rule for characterization of some essential aspects of the random variable, separate numerical parameters and characteristics are used. The distribution parameters are variables which are directly involved in the distribution function or the probability distribution function. Numerical characteristics are characteristics expressing certain properties of the random variable distribution rule. The parameters (characteristics) can be divided into several groups: those referring to the location, the dispersion, and the moments. CHARACTERISTICS FOR LOCATION 1. Quantile the value of the random variable , for which the probability P X < x p = p,
0 < p < 1.
(I.1.3)
For example, the quantiles 1/4, 1/2, and 3/4 separate the interval of the random variable to four subintervals with equal possibilities for being part of them (Figure .1.5).
F(x) 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 3 2 1 x1/4 0 x1/2 x3/4 1 2 x 3
Figure I.1.5. Quantiles of the distribution function.
2. Mathematical expectation {X}, x, mxthe average value of the random variable dened by the weight aspect according to the probability for its occurrence. For discrete random variables it is dened as M { X } = mx = mx = and for continuous variables as M { X } = mx = mx =
x p( x )
i i i
(I.1.4)
x f ( x)dx.
(I.1.5)
For certain types of distributions a mathematical expectation may not exist.
3. Median the point separating the interval of variation of the random variable in two parts so that the probability of occurring therein would be , that is, the median is a quantile at = . 4. Mode the value(s) of the random variable , which in case of a continuous random variable is compliant with the local maximum of the probability density, and in case of a discrete random variable is compliant with the local maximum of the series of probabilities. Depending on the number of modes, the distributions can be single-mode or multimode. It is possible that a mode does not exist for a specic type of distribution. A single continuous distribution is symmetrical if F ( M e x) = 1 F ( M e + x) , respectively, f ( M e x) = f ( M e + x) . (I.1.7) (I.1.6)
For those distributions, the mathematical expectation matches the median and the mode.
DISPERSION CHARACTERISTICS
2 1. Variance D{}, sx the characteristic of dispersion of the random variable around its mathematical expectation. For discrete random variables the following formula has been determined: 2 D { X } = sx =
(x m )
i x i
p ( xi ),
(I.1.8)
and for continuous ones it is

2 D { X } = sx =
(x m )
x
f ( x)dx.
(I.1.9)
The equivalent formula is frequently used as well:

2 D { X } = sx = M
{( X m ) } = M { X } ( M { X } ) .
2 2 2 x
(I.1.10)
Figure .1.6 represents the distribution density for three random variables with 2 matching mathematical expectations = 0 and variances, respectively, (a) s1 = 0.25, 2 2 (b) s2 = 1, and (c) s3 = 4. 2. Standard deviation s xcharacterizes the dispersion of the random variable around its mathematical expectation and has the same dimension as : sx = D { X } . (I.1.11)
8 TEXTILE PROCESSES
(a) f (x) 0.8 0.6 0.4 0.2 0 8 (b) f (x) 0.8 0.6 0.4 0.2 0 8 (c) f (x) 0.8 0.6 0.4 0.2 0 8
Figure I.1.6. Probability distribution functions with different variances.
3. Coefcient of variationa relative characteristic for evaluation of dispersion of the random variable. Percentage-wise it is determined according to the following formula: v= sx 100. mx (I.1.12)
4. Linear deviation dcharacterizes the average absolute deviation of the random variable: d { X } = M X mx .
(I.1.13)
5. Range Rthe difference between the maximum and the minimum values of X, for which f (x) 0. It is dened as R = X max X min . (I.1.14)
MOMENTS 1. Initial moments ak the initial moment from rank k is the average numerical characteristic: ak = M X k . The initial moment of rank 0 always equals 1: a0 = M X 0 = 1, and the initial rst rank moment always equals the mathematical expectation: a1 = M X 1 = mx .
{ }
(I.1.15)
{ }
(I.1.16)
{ }
(I.1.17)
2. Central moments mk the central moment from the k th rank is determined analogically to the initial rank and the random variable is centered according to the mathematical expectation: mk = M
{( X m ) } .
k x 1 x
(I.1.18)
The central moment from rst rank (k = 1) always equals 0: m1 = M
{( X m ) } = 0
2 x 2 x
(I.1.19)
and the central moment from second rank (k = 2) always equals the variance: m2 = M
{( X m ) } = s .
(I.1.20)
The central moments of uneven rank of random variables with symmetric distributions equals 0. 3. Asymmetry (skewness) Skcharacterizes the degree of asymmetry of a distribution around its mean. It is determined by the relation: Sk = m3
3 sx
(I.1.21)
10 TEXTILE PROCESSES
The skewness value is 0 with symmetrical distributions ( m3 = 0), its value being positive when the mode is situated before the median and negative when the mode is situated after the median (Figure .1.7).
f (x) Sk > 0
f (x) Sk = 0
f (x) Sk < 0
M0 Me (a)
M0 = Me (b)
x (c)
Me M0
Figure I.1.7. Distributions with positive asymmetry, symmetrical distribution, and negative asymmetry.
4. Kurtosis (declivity) characterizes the relative peakness or atness of a distribution in comparison with the normal distribution. It is determined according to the following formula: E= m4
4 sx
3.
(I.1.22)
Positive kurtosis indicates a relatively peaked distribution. Negative kurtosis indicates a relatively at distribution (Figure .1.8).
f(x)
E>0 E=0 E<0
Figure I.1.8. Distribution with positive, zero, and negative kurtosis.
The kurtosis is 0 when the form of the distribution curve is such that the following is true:
4 m4 = 3s x .
(I.1.23)
5. Correlation moment (co-variation factor) xythe central moment characterizing the degree of linear dependency of two random variables and Y: mxy = M
{ ( X m ) ( Y m )} .
x y
(I.1.24)
6. Linear correlation factor xythe standardized correlation moment, the values of which fall within the margins [1; 1]. Standardization is completed through the division of the correlation moment into the product of the standard deviations of the two random variables and : rxy = mxy sx s y . (I.1.25)
For | rxy | = 1 between the two random variables, a linear functional dependency exists. When 1 < rxy < 1 between the variables x and y, a statistical (stochastic) dependency exists, whereas the closer the value of | rxy | is to 0, the stronger the representation of this value is. If xy is positive, the increase in values of leads to increase in values of . If xy is negative, the increase of results in diminishing and a reverse dependency will be discussed. When rxy = 0, the random variables are uncorrelated. If the random variables x and y are independent, they are noncorrelated. The reverse is correct only for principles of normal distribution. All separate cases have been presented in Figure .1.9 [() reverse statistical dependency, (b) no correlation, () straight statistical dependency].
y y rxy = 0 0 < rxy < 1 (a) x (b) x (c) x y
1 < rxy < 0
Figure I.1.9. Different cases of correlation dependencies.
PROPERTIES OF NUMERICAL CHARACTERISTICS

Textile practice often requires working with variables that are a sum or a product of two or more random variables, as well as a product of a random variable and a constant. In such cases, the following properties of the numerical characteristics can be used. 1. The mathematical expectation of the constant ( = const ) is equal to the variable itself: M { C } = C = const. (I.1.26)
2. The mathematical expectation of the sum of random variables equals the sum of their mathematical expectations:
n M Xi = i =1
M { X }.
i i =1
(I.1.27)
The property is also valid for subtraction of random variables. In this case, a minus sign will be placed before the respective random variables.
3. The mathematical expectation of the product of two random variables equals the sum of the product of the mathematical expectations of the two components and the correlation moment: M { X Y } = M { X } M {Y } + mxy . If the random variables are noncorrelated: M { X Y } = M { X } M {Y } = mx + my . (I.1.29) (I.1.28)
4. The mathematical expectation of the product of a random variable with a constant follows: M {C X } = C M { X } . 5. The variance of a constant is zero: D { C } = 0. 6. The variance of the product of a random variable and a constant is D {C X } = C 2 D { X } . 7. The variance of a sum or a margin of two random variables equals D { X Y } = D { X } + D {Y } 2 mxy . Result: If the random variable X = X 1 X 2 variables are mutually uncorrelated, then D { X } = D { X1} + D { X 2 } + (I.1.33) (I.1.32) (I.1.31) (I.1.30)
X n and each pair of random D{ Xn}.
(I.1.34)
I.2. LAWS OF RANDOM VARIABLES DISTRIBUTION

This section presents those probability distributions that are used most often in the textile practice.
CONTINUOUS DISTRIBUTIONS
1. Normal (Gaussian) distribution The distribution density law is as follows:
1 P( x ) = e s 2p
( x mx ) 2
2s2
(.2.1)
The curve of the distribution is bell-shaped and with the increase in the absolute value of (| x | 0), it tends asymptotically to the x-axis. The distribution, respectively the form of its graph, is dened by two parameters. The mathematical expectation mx denes the location of the curve (Figure .2.1) and the standard deviation denes the variance around it. When the standard deviation is big the curve is at, and when it is small the curve is convex (Figure .1.6). The normal distribution is typical for many textile properties and is used when the property that is being studied is dispersed symmetrically in relation to the central value, for example, strength, count, elongation, linear dimensions (length, width, diameter), and so on. Very often, when studying random variables, including those distributed according to the normal distribution law, the values are centered and standardized. A centered random variable is a variable the values of which lie symmetrically around zero. The centering is done by extracting the respective mathematical expectation from the individual variable values:
o
X = X mx .
(.2.2)
13
The standardization of a random variable is done by dividing its centered value by the standard deviation: X mx X = = . s s
o
XN
(a) f (x) 0.8 0.6 0.4 0.2 0 8 Mu 2
(.2.3)
Sigma 1
8 x
(b) f (x) 0.8 0.6 0.4 0.2 0 8 Mu 0 Sigma 1
8 x
(c) f (x) Mu 0.8 0.6 0.4 0.2 0 8 2 Sigma 1
8 x
Figure I.2.1. Movement of the normal distribution in conformity with the mathematical expectation.
PART I: LAWS OF RANDOM VARIABLES DISTRIBUTION 15
A standardized random variable has a mathematical expectation of 0 and a variance of 1N(0; 1). For example, when determining the linear density of linear products, instead of putting down the value of the separate sections, it is better to give the deviations from the mean value in percentage. With the help of the standardized random function it is also very easy to determine the share of the values in some typical intervals (Figure .2.2): (mx s ) 68.26% (mx 2s ) 95.46% (mx 2.575s ) 99% (mx 3s ) 99.73%.
f (x) 0.6 0.4 0.2 s +s 0 8 6 4 2 0 2 68.26% 95.46% 99% 99.73% 4 6 8 x
Figure I.2.2. Characteristic values for the normal distribution.
2. Uniform distribution The uniform distribution can be used for both, continuous and discrete variables. It is applied in those cases when the probability of the occurrence of a certain value within a particular interval [a; b] is the same. The distribution density is described with the help of the following law: 1 , a xb P( x) = b a . 0, b x a The mathematical expectation is dened as follows: a+b , 2
(.2.4)
mx =
(.2.5)
and the formula for the variance is
(b a)2 . =
12
(.2.6)
The graph of the uniform distribution is shown in Figure .2.3.

f (x)
1/(b a)
Figure I.2.3. Uniform distribution.
0 a b
DISCRETE DISTRIBUTIONS
1. Binomial distribution The binomial distribution is the most widespread discrete distribution. It shows the probability of a certain event to occur times out of n cases (repetitions). The distribution law is dened as follows:
x P( x) = cn p x q n x =
n! p x qn x , x !(n x)!
(.2.7)
where is the probability of occurrence of a certain event and q = 1 p is the probability of nonoccurrence of the particular event. The mathematical expectation is mx = n p and the variance is s 2 = n p q. (.2.9) (.2.8)
Figure .2.4 presents different cases of binomial distribution when there is a change in the number of tests n and the probability of occurrence of the event . Cases (a) and (b) show the distribution when there is a change in the number of tests from 5 to 10, the probability of occurrence being = 0.1. The probability of nonoccurrence is q = 0.9.
The distribution is asymmetrical. In cases (c) and (d) the number of tests is changed again, but the probability of occurrence is now = 0.5, respectively, and the probability of nonoccurrence is q = 0.5. The distribution is symmetrical. It can be seen that in case of a symmetrical distribution and increase in the number of tests, the distributions tends to normal.
(a)
p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 x n=5 p = 0.1
(b)
p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 x n = 10 p = 0.1
(c)
p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 x n=5 p = 0.5
(d)
p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 x n = 10 p = 0.5
Figure I.2.4. Binomial distributions.
In order to illustrate the binomial distribution, the most common example used is the example of determining the probability when rolling a die. For example, let us try to determine the probability of rolling a three twice in 10 throws. The probability is determined when the following values are substituted in formula (.2.7): number of tries (throws) n = 10, probability of the event rolling a three p = 1 (the possible options 6 are 6, each has an equal chance of occurrence), and the desired number of occurrences of the event is = 2. After substituting, the probability of rolling a three twice in 10 throws can be determined: 10! 1 5 2!(10 2)! 6 6
2 10 2
P( x) =
= 0.29.
The binomial distribution is often used in the textile production when qualifying the production.
Example In the production of hemp technical fabrics for sacks it has been established that out of 100 samples of the fabric an average of 50 have defects. Determine the probability of having a maximum of two samples with defects out of a total of three samples. First determine the probability of having a fabric with a defect: p= 50 = 0.5. 100
In order to satisfy the condition of having a maximum of two defective samples, the possible options are no defective sample, one defective sample, and two defective samples. The probability of having no defective samples when the volume of the sample is n = 3 is P(0) = 3! ( 0.5) 0 ( 0.5)3 0 = 0.125. 0!(3 0)!
The probability to have one defective sample is P(1) = 3! ( 0.5)1 ( 0.5)31 = 0.375. 1!(3 1)!
The probability to have two defective samples is P(2) = 3! ( 0.5) 2 ( 0.5)3 2 = 0.375. 2!(3 2)!
Therefore, the probability to have a maximum of two defective samples is P( x 2) = P(0) + P(1) + P(2) = 0.875.
2. Poisson distribution The Poisson distribution is an instance of the binomial distribution in which with the increase in the number of tests n the probability for the occurrence of the event is decreased, so that p n = const = l. Therefore, it is often said that the Poisson distribution is a distribution of the rare events. The distribution law is the following: P( x ) = e ll x , x! x = 0, 1, 2, (.2.10)
What is typical about the Poisson distribution is that it is monoparametrical. The mathematical expectation and the variance equal the parameter l: mx = s 2 = l. (.2.11)
Figure .2.5 represents different cases of Poisson distribution when there is a change in the number of the parameter l.
(a) 0.6 0.5 0.4 0.3 0.2 0.1 0 (c) p(x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 x l =2 0 1 2 3 4 5 x l = 0.5 0.5 0.4 0.3 0.2 0.1 0 p(x) 0 1 2 3 4 5 x p(x) (b) 0.6 l =1 p(x)
(d) 0.6 0.5 0.4 0.3 0.2 0.1
l =6
9 10
Figure I.2.5. Poisson distributions.
What can be seen from the graphs in Figure .2.5 is that with the increase in the values of the parameter l, the distribution becomes more and more symmetrical and tends to the normal distribution. One of the most common applications of the Poisson distribution in the textile production is in determining the number of bers in a section of a textile mill product. The irregularity is determined with the help of the coefcient of variation according to the formula given (see formula (.1.12)). Taking into account that mx = n , where n is the mean number of bers in the product section and the property of the Poisson distribution for equality between the variance and the mathematical expectation (see formula .2.10), the coefcient of variation is the following: v= s n 100 100 = 100 = . mx n n (.2.12)
The resulting value of the irregularity (coefcient of variation) is theoretic as it has been obtained on the basis of the random distribution of bers without taking into account the irregularity of the bers and the impact of machine parts. The real irregularity of the products is much higher. The Poisson distribution is also used very often in assessing the breakage in processing.
Example What is studied is the breakage of a spinning machine with 96 spindles. It has been established that there are 80 breakages for 2 h. What has to be found out is the number of spindles for which the number of breakages is expected to be 0, 1, 2, 3, 4, and so on. The mean number of breakages for one spindle for a period of 2 h is x= 80 = 0.833. 96
The probability that there will be no breakage within a period of 2 h is P(0) = e 0.833 0.8330 = 0.435. 0! e 0.833 0.8331 = 0.363. 1! e 0.833 0.8332 = 0.151. 2!
The probability of one breakage is: P(1) =
The probability of two breakages is: P(2) = For three breakages: P(3) =
e 0.833 0.8333 = 0.047. 3! e 0.833 0.8334 = 0.0087. 4!
For four breakages: P(4) =
In order to determine the number of spindles that will have the respective number of breakages, the probabilities are multiplied by the total number of spindles: For a period of 2 h there will be no breakage on 96 0.435 = 42 spindles. One breakage is expected on 96 0.363 = 35 spindles. Two breakages on 96 0.151 = 14 spindles. Three breakages on 96 0.047 = 4 spindles. Four breakages on 96 0.0087 = 1 spindle. There are other types of distributions that are used in textile production such as the Weibull and Gamma distributions, but the ones presented above are the most widely used.
I.3. STATISTICAL ESTIMATES

The denition of the numerical characteristics of the law, or respectively of the distribution function of a population of observed data, is called statistical estimation and the resulting numerical values are statistical estimates. The statistical estimates dene the real values of the respective numerical characteristics only in an approximate manner. The reason is that they are calculated on the basis of a random sample with limited volume. Increasing the volume of the sample increases the accuracy of the estimates, which in turn is related to an increase in the test costs. One of the tasks of mathematical statistics is to nd the minimum sample volume in order to obtain such a degree of accuracy in estimation that satises the needs of the researcher. Mathematical statistics also makes use of different functions, called statistics, which depend solely on the observation data. Each statistic used for the estimation of a certain parameter or numerical characteristic of the distribution of the population is distributed according to some law around the respective parameter or characteristic. Under an analogy with the probability theory where the task is to nd the most accurate value of a parameter or a numerical characteristic, this estimation is called point estimation, and the estimatea point estimate. When with the help of the distribution of the statistic used is dened the interval in which there is a sufcient probability to nd the estimated parameter or numerical characteristic, the estimation is called an interval estimation and the estimatesinterval estimates.
CONDUCTING THE TEST

As a result of conducting the test, quantitative or qualitative estimates of the investigated indications are obtained, which are the respective realization of the variable when performing the test. The resulting data, recorded according to the order of their appearance, is called primary range. The initial form of results recording does not allow a denition of the rule in the change of the indication. In order to provide a possibility for scientic research, rst the data should be arranged according to value. What is obtained is the so-called arranged range. The arrangement can be in ascending 1 2 ... i xn or descending order 1 2 ... i xn. When the number of measurements is big (over 2530), it is more convenient to group the results in class intervals. In this case the data is entered into a primary table. When choosing the width of the class interval b, the following requirements should be met: 1. It should be the same for all intervals, which facilitates the determination of the statistical estimates. 2. The number of class intervals should fall within 6 k 25.
21
3. The approximate number of intervals can be determined according to the following formula: k = 1 + 3.2 lg n, where n is the number of measurements. x xmin . 4. The width of the interval is b = max k It is recommended to round off the value of the interval width in order to facilitate calculations, if they are done manually, or the perception thereof as a whole. For example, if the calculations result is b = 1.89 or b = 2.14, it is preferable to operate with a class interval width of b = 2. In this case the limits of the class intervals will be 02, 24, 46, and so on. For the rst case (b = 1.89) they would be 01.89, 1.893.78, 3.785.67, and so on, and for the second one (b = 2.14) 02.14, 2.144.28, and so on. The number of results in a given interval is called absolute frequency, hj, and the ratio of the absolute frequency to the volume of the sample n is called relative frequency: fj = hj n 100. (.3.2) (.3.1)
When grouping the results together in class intervals, the data is processed under the assumption that all values determined in the respective interval equal the mean of the interval. If we are interested not in the share participation of each interval but in the share of the results above or under a certain value, then what is used is the cumulative frequency, fj. It represents the sum of the relative frequencies from the initial to a given class interval. Example A measurement should be conducted in order to determine the strength of a certain yarn by performing 120 tests. The expected values fall within the (190300) cN interval. The approximate number of class intervals is determined according to (.3.1): k = 1 + 3.2 lg120 = 7.65. The resulting value is rounded off to 8. The width of the class interval is determined: b= 300 190 = 13.75 cN. 8
The class width that is selected for operational convenience is close to the calculated one. In order to allow for more precise calculations, the interval is reduced to 10 cN. Table .3.1 shows the processing order for the grouped data. The number of the interval is entered in column 1, the interval limits in column 2, the mean of the interval in column 3, the number of occurrences in each interval in column 4, the absolute frequency hj (the sum of the occurrences) in column 5, the relative frequency fj in column 6, and the cumulative frequency fj [as determined according to formula (.3.2)] in column 7.
PART I: STATISTICAL ESTIMATES 23
Table I.3.1. Calculation of the data for drawing of the histogram, polygon and cumulative curve Class Interval Mean of the interval limits interval (1) 1 2 3 4 5 6 7 8 9 10 11 (2) 190200 200210 210220 220230 230240 240250 250260 260270 270280 280290 290300 (3) 195 205 215 225 235 245 255 265 275 285 295 Stroke (4) Absolute Relative Cumulative frequency hj frequency fj frequency S fj (5) 2 2 7 12 18 28 23 14 10 3 1 120 (6) 1.67 1.67 5.83 10.00 15.00 23.33 19.17 11.67 8.33 2.50 0.83 100 (7) 1.67 3.33 9.17 19.17 34.17 57.50 76.67 88.33 96.67 99.17 100.00
The test results are graphically represented through: A histogram of the frequency distribution (Figure .3.1)a diagram consisting of rectangles with a width equal to the width of the class interval and a height equal to the absolute or relative frequencies for the respective interval. A distribution polygon (Figure .3.1)a curve that represents the relationship between the mean values of the intervals and their absolute or relative frequencies. It is an estimate of the density distribution law.
hj 30 25 20 15 10 5 0 195 205 215 225 235 245 255 265 275 285 295 xj
Figure I.3.1. Histogram and polygon.
fj (%) 100 90 80 70 60 50 40 30 20 10 0 190 200 210 220 230 240 250 260 270 280 290 300 xj + b/ 2
Figure I.3.2. Cumulative curve.
A cumulative curve (Figure .3.2)a curve that represents the probability of the occurrence of a value under the upper limit of the respective class interval. It is an estimate of the distribution function.
With the increase in the size of the sample and narrowing of the interval width, the frequency distribution polygon gets closer and closer to the law on random variable distribution and the cumulative characteristic closer to the distribution function.
POINT ESTIMATES
1. Arithmetic mean xan estimation of the mathematical expectation. It is calculated for: single values
x= where xi is ith consecutive value, n is the number of tests. grouped values
x
i =1
(.3.3)
x=
h x
j j =1
fx,
j j j =1
(.3.4)
where xj is the mean of the jth interval, hj is the absolute frequency of the jth interval, fj is the relative frequency of the jth interval, k is the number of class intervals. 2. Mode/Modal value xthe estimate of the mode/modal value of the population. It is determined as follows: for single valuesit corresponds to the value that has the highest frequency; for grouped values x = xm + b hm hm 1 , 2hm hm 1 hm +1 (.3.5)
where xm is the upper limit of the modal interval (the interval with the highest frequency), b is the width of the class intervals, hm is the absolute frequency of the modal interval, hm1 is the absolute frequency of the premodal interval, hm+1 is the absolute frequency of the postmodal interval. 3. Median xthe estimate of the mode/modal value of the population. The median divides the arranged range in twothe number of values before and after being equal. Geometrically, the median divides the area under the polygon in two equal parts. It is determined as follows: for single values x = x n +1 if n is an odd number,
2
(.3.6) (.3.7)
x= for grouped values
1 x n + x n +1 if n is an even number 2 2 2
jmed 1 j =1
0.5n x = xmed + b
hmed
(.3.8)
where xmed is the lower limit of the median interval (the rst interval in which the absolute sum frequency is higher than 0.5n), jmed1 is the consecutive number of the premedian interval, hmed is the absolute frequency of the median interval.
INTERVAL ESTIMATES
1. The range R is an estimate which is very easy to calculate but it serves only as a dispersion indication: R = xmax xmin. (.3.9)
2. Mean linear deviation d is an estimate that does not reect the deviating results to a sufcient degree. It is determined in the following manner: for single values
d= where x is the arithmetic mean; for grouped values
x x
i i =1
(.3.10)
d=
h
j =1
xj x n . (.3.11)
3. The variance S 2 is the estimate of the population variance D{x} and is determined as follows: for single values S2 = for grouped values S2 = 1 n 1 1 n 1
(x x) ;
2 i i =1 k
(.3.12)
h (x
j j =1
x )2 .
(.3.13)
It is a displacement estimate which, due to its adjusted value, is calculated by dividing the sum of the squares of the deviation by (n 1). The standard deviation S is determined on the basis of this estimate: S = S2 . (.3.14)
Just like the range and the mean linear deviation, the standard deviation has the dimension of the tested variable but eliminates their defects. It is used both as a variance estimate and as an estimate for the error that occurs due to the incomprehensive nature of the research. This deviation can be used to determine the sustainability of the manufacturing process, and so on. If the size of the sample is small and there is little dispersion in the results (e.g., when testing fabrics), the standard deviation can be calculated with the help of the range as follows: S= R , an (.3.15)
where an is a coefcient which can be taken from Table .3.2 in accordance with the size of the sample n.
Table I.3.2. Values of the coefcient an in accordance with the number of trials n n an 2 3 4 5 6 7 8 9 10 15 20 25 1.128 1.693 2.059 2.326 2.534 2.704 2.847 2.970 3.078 3.472 3.735 3.931
The described interval estimates have some disadvantages, namely: They bear the dimension of the investigated variable and therefore it is not possible to compare the degrees of dispersion of two different indicators, for example, whether the wool bers show greater dispersion in length or in diameter. They do not depend on the mean value. When the mean value is low even the smallest deviations are of importance. For example, if there are two yarns with mean strengths of 100 cN and 350 cN, respectively, and the same standard deviation S = 20 cN. At 100 cN, the standard deviation is 20% of the mean value and at 350 cN it is 5.7%.
These disadvantages are eliminated with the so-called irregularity estimates. In their nature there are interval estimates (for dispersion) but are calculated as a percentage of the mean value. 4. Linear irregularity coefcient H= d 100. x (.3.16)
It is used for the comparison of two empirical distributions in view of their dispersion around the mean. Since it is calculated from the mean linear deviation, it has certain drawbacks.
Example There are two different yarns, each having been measured 100 times. For the rst yarn there are 99 results of 100 cN and one of 1 cN. For the second yarn there are 50 measurements of 101 cN and 50 measurements of 97 cN. In both cases the average strength is 99 cN and the linear irregularity is 2%. In the rst case, due to the presence of a weak section the yarn will break, and in the second it will not.
5. Coefcient of variation v v= S 100. x (.3.17)
The coefcient of variation is preferable when analyzing the irregularity as it is calculated on the basis of the standard deviation in which the deviations from the mean value are at square degree. Thus the weak deviations die out and the strong ones intensify. In the example above, the coefcient of variation in the rst case is 10%, and in the second is 2%.
When the sample is big in size and the law of distribution is normal, the coefcient of variation can be dened by the linear irregularity coefcient using the following formula: v= p H 1.253H . 2 (.3.18)
When the size of the sample is small and the law of distribution is normal, the relationship is as follows: v = A H , (.3.19)
where is a coefcient, which can be taken from Table .3.3 in accordance with the size of the sample n.
Table I.3.3. Values of the coefcient A in accordance with the number of trials n n 4 1.447 6 1.373 8 1.340 10 1.321 15 1.291 20 1.286 50 1.266
CONFIDENCE INTERVALS
The determination of condence intervals is necessary due to the substitution of the unknown real values with estimates calculated on the basis of a sample. For the establishing of condence intervals such a distribution is used, the law of which does not depend on unknown values but only on the number of tests n and on the law of distribution of the random variable.
CONFIDENCE INTERVALS OF THE ESTIMATES IN CASES OF NORMAL DISTRIBUTION 1. Mean value If the random variable is normally distributed, the condence interval of its mathematical expectation is determined as follows: x q, (.3.20)
where q is the absolute condence error. It is determined with the help of the following formula: q = t S n , (.3.21)
where t is the value of Students t-distribution, which is determined according to the selected level of signicance a and the degrees of freedom f (Figure .3.3): t ( a, f = n 1). (.3.22)
0.4 0.3 0.2 0.1 0.05 0 8 6 /2 4 2 t( f, /2)
p(t)
/2 2 4 6 8 t
t( f, /2)
Figure I.3.3. Students t-distribution.
Students t-distribution is tabulated and therefore the notion of the table value of Students t-distribution (criteria) is used very often. The values of the distributions are given in Appendix 1. As can be seen from formula (.3.21), the absolute condence error has the dimension of the investigated value and depends on the standard deviation of the sample, on its size, and on the value of Students t-distribution. It increases with an increase in the dispersion and an increase in the condence probability g = 1 a and decreases with an increase in the number of tests. The condence probability is selected in accordance with the desired number of cases within the resulting condence interval. In the textile practice, people most often use condence probability of 0.95, which corresponds to a 0.05 level of signicance. In certain cases, higher values of the condence probability can be used. The condence interval can also be arrived at through the relative condence error. It is dened by relating the absolute condence error to the mean value, respectively, if instead of the standard deviation the variation coefcient is used: p = t S x n 100 = t v . n (.3.23)
With the help of these formulae one can establish the approximate number of tests nmin for acquiring the set maximum condence error percentage pmax: tv nmin = . pmax
2
(.3.24)
2. Variance When dening the condence interval of the variance, the Chi-squared (2) distribution is used (Figure .3.4).
f( c2) 0.4 2 a/2 0.05 0 g 0 5
2 2
a/2 10 15 c2
Figure I.3.4. Chi-squared distribution.
c (a/2, f ) c (1 a/2, f )
Chi-squared distribution is asymmetrical (it tends to symmetrical with the increase in the number of tests) and therefore the condence level of variance is determined by the following inequality: ( n 1) S 2 ( n 1) S 2 s2 2 , c 2 (a 2, n 1) c (1 a 2, n 1) where the degrees of freedom are f = n 1. (.3.25)
CONFIDENCE INTERVAL OF THE l PARAMETER FOR POISSON DISTRIBUTION When the value of the parameter l (l > 20) is big, the distribution tends to normal and therefore the condence interval of the mathematical expectation can be calculated in a manner similar to that of the normal distribution: x t s. As in Poisson distribution s = mx , then: x t x. (.3.27) (.3.26)
In the classical case (l < 20), the limits of the condence interval of the mathematical expectation are: x+ t2 t2 t2 t2 + x mx x + + + x, 2 4 2 4
(.3.28)
where the value of Students t-distribution is taken into consideration in the following case t2 << x , this member is very often neglected and the following equation is used t (a, ). Since 4 in order to determine the condence interval:
x+
t2 t2 x mx x + + x , 2 2
(.3.29)
which shows that the interval is symmetrical, however not in reference to the mean value but t2 in reference to x + . 2 CONFIDENCE INTERVAL OF THE PARAMETER p FOR BINOMIAL DISTRIBUTION Very often the parameter cannot be mathematically calculated like dice-throwing. So one should use tests to determine it. If n is the total number of tests, and m is the cases when the desired event occurred, then the ratio m/n in a sufciently big number of tests is the probability estimate . In this case the condence interval is p= m m m t 2 1 . n n n (.3.30)
If the probability can be calculated, the condence interval is m p ( p 1) m p ( p 1) t p +t . n n n n Example When calculating the mass of knit sweaters, the following values have been acquired (in g): 200, 210, 215, 210, 210, 195, 200, 190, 210, 205. Determine the statistical estimates. The arranged range is: 190, 195, 200, 200, 205, 210, 210, 210, 210, 215. The mean arithmetic value is: x= 190 + 195 + 200 + 200 + 205 + 210 + 210 + 210 + 210 + 215 = 204.5 g. 10 (.3.31)
The modal value is x = 20 g as it has the highest rate of occurrence (four times), and the median is determined as the mean arithmetic value of the fth and the sixth value 205 + 210 x= = 207.5 g. 2 The range is R = xmax xmin = 215 195 = 20 g. The mean linear deviation is d = 190 204.5 + 195 204.5 + 2 200 204.5 + 205 204.5 10 + 4 210 204.5 + 215 204.5 = 6.6 g. 10 (Continued )
(Continued ) The variance is S 2 = 1 2 2 2 190 204.5) + (195 204.5) + 2 ( 200 204.5) + (205 204.5) 2 ( 9 2 2 + 4 ( 210 204.5) + ( 215 204.5) = 63.61 g2
and the standard deviation S = S 2 = 7.98 g. 204.5 100 = 3.9%. The coefcient of variation is v = 7.98 In order to determine the condence errors from Appendix 1 at a signicance level of = 0.05 and degree of freedom f = 10 1 = 9, the tabular values of Students t-distribution are taken into account t = 2.26. 2.26 7.98 = 5.7 g. The absolute condence error is q = 10 The percentage condence error is p = = 2.8%. 10 The condence interval for the sweaters mass can be recorded, both, with the absolute and the percentage condence error: m = (204.5 5) g m = 204.5 g 2.8% It shows that when conducting new measurements, 95% ( = 1 0.05) of the measured sweaters will have a mass in the interval (195.5209.5) g. According to (.3.21), the condence interval of the variance can also be determined: 9 63.61 9 63.61 s2 , 16.92 3.33 that is, with a condence probability = 0.95, the variance lies within [33.84; 171.92] g2. 2.26 3.9
Example In 50 tests, 10 samples proved to be without defects. Find the approximate condence interval of pulling out a sample without defect. The probability of pulling out a sample without a defect can be determined approximately as the ratio between the number of samples without defects m and the total number of samples n: m 10 = = 0.2. n 50 The condence interval is calculated according to (.3.30), the value of Students t-distribution being determined from Appendix 1 at a level of signicance = 0.05 and degree of freedom f = . At t = 1.96, the condence interval is p = [0.0891, 0.3109].
I.4. STATISTICAL PROCESS CONTROL AND CONTROL CHARTS
STATISTICAL PROCESS CONTROL

The statistical process control (SP) is applied in established technological processes in order to observe, analyze, and regulate them. The use of SPC allows for timely warning in case of a deviation and, if necessary, to correct the parameters of the process. The application of SPC is a prerequisite for the timely detection of the reasons behind quality deterioration. It helps in decreasing the defects, maintaining the quality at a selected level, and improving the technological process (Figure .4.1). The control is realized through regular inspections carried out according to a certain schedule that uses the methods of mathematical statistics (Figure .4.2). The correct running of the technological processes is assessed with the coefcients of process capability. The coefcient of process stability, , represents the relationship between the tolerance of the controlled parameter and the area of dispersion V: Cp = T Tu Tl = , V 6 S (.4.1)
where S is the standard deviation of the parameter, u and l are the upper and lower limits of the parameter tolerance, respectively. When the dispersion area is lower than the tolerance, ( > 1), the technological process is stable and can be controlled by statistical methods. The control of this process can help in avoiding defective products. When < 1, the process is unstable and the occurrence of defects is inevitable. In order to make this process manageable it is necessary to make changes in the technology, to repair or replace the machines. A value over 1.33 is considered a good coefcient of process stability as it guarantees the stability of the process for a longer period. For a short period the stability is determined mainly by the machine, while for a longer period other factors may play a role, such as wearing out, changes in the parameters of the environment or in the parameters of the material. Apart from being stable, a process has to be set in the right manner. The setting of the technological processes is characterized by the so-called coefcient of adjustment, k, which
33
Information Process disturbance
Corrections
Disturbance decreasing
Figure I.4.1. Steps for quality improvement.
Quality improvement
Process
Impact on the process
Samples
Figure I.4.2. Principle of SPS.
Control charts
denes the displacement of the center of the grouped results toward the middle of the tolerance eld. Since the displacement can occur in both directions two coefcients are determined, k1 and k2, the coefcient of adjustment being the smaller of the two: C pk1 = Tl X , 3 S X Tu , 3 S (.4.2)
C pk 2 =
(.4.3)
PART I: STATISTICAL PROCESS CONTROL AND CONTROL CHARTS 35
C pk = min C pk1 , C pk 2 .
(.4.4)
The coefcient of adjustment k should have a value of more than 1. Otherwise, defects can occur in the production, and their percentage depends on the type of distribution. For a normally distributed random variable, the values of the defective production dependent on the coefcient of adjustment are given in Table .4.1.
Table .4.1. Percentage of defective production for a normally distributed random variable k Defective production (in %) 0 50 0.1 38 0.2 27 0.3 18 0.4 12 0.5 7 0.6 4 0.7 2 0.8 1 0.9 0.4 1 0
Figure .4.3 represents the typical cases of distribution of the controlled variable under the normal law of distribution [(a) set but unstable process, (b) set and stable process, (c) stable process with a borderline setting, (d) stable but unset process].
Tl
Tu
(a)
Cp < 1
Cp > 1; Cpk > 1 (b)
Figure I.4.3. Some typical cases of distribution of the controlled variable.
Cp > 1; Cpk = 1 (c) Cp > 1; Cpk < 1 (d)
CONTROL CHARTS
The operational control is implemented with the help of control chartsa piece of cardboard on the x-axis of which are recorded the discrete moments of time in which the control has been performed or the number of the sample (observation), and on the y-axis is the controlled indicator. The statistical hypothesis can be veried graphically by the control chart. The condence interval is built for a certain estimate q under the condition that the zero hypothesis is being met 0: q = q0, where q0 is the value desired by us or required by a certain standard or a document. The zero hypothesis is not rejected if the estimate of the specic parameter falls within the limits of the condence interval, and it is rejected in the opposite case. Most often, the numerical characteristics of the distribution of random variables (mathematical expectation, median, variance, range, and so on) are used for estimates. The creation of a control chart is based on the following prerequisites: the controlled property is a random value and, therefore, the individual values vary in a certain interval: m t s, where m is the mathematical expectation of the random value, s is the standard deviation, t is Students t-distribution value at a selected condence probability. Most often, the condence probability used is g = 0.99 which corresponds to t = 2.576 or g = 0.9973 which corresponds to t = 3. The second case is used more often since the value of t is a whole number and the percentage of case coverage is almost 100. The control chart consists of three linesone central that corresponds to the nominal or mean value and two parallel onesthe upper and lower control limits, which set the width of the condence interval at the selected condence probability (Figure .4.4). Sometimes on the control chart are also positioned two warning lines at a distance of 2S. Keeping a control chart means to record the value of the investigated property on the chart in selected moments. If there are values that fall outside the interval, dened on the basis of a preliminary sample, it means that the technological process is disturbed and an intervention is necessary in order to regulate it. If only one value falls outside the interval, this does not imply a correction in the process since the interval set does not cover 100% of the cases (when g = 0.99 it is permissible for 1 of a 100 values to fall outside the set limits). If there is a value outside the (.4.5)
Upper limit 3S Nominal 3S
Lower limit
Figure I.4.4. Graphical presentation of the control chart limits.
control limits dened by the standardization requirements (through the limits of the tolerance eld), this is already a signal for the appearance of defects. All products manufactured after the last regular sample should be subjected to overall control.
DESIGN OF STATISTICAL CONTROL CHARTS When designing a control chart it is necessary to comply with the following requirements: the controlled variable should be random; and the controlled technological process should be stable ( > 1).
Data from the preliminary measurement is used in chart design with the objective of nding the mean and the standard deviation. For the standard deviation to be close enough to that of the population, 100 to 200 primary data are necessary. The primary data should be taken from different packages or in different times which guarantees that the sample is representative.
Example When determining the limits of the control chart for yarn strength, it is preferable to make 20 measurements from ve spools than 100 measurements from one spool.
The next task is to establish whether the resulting values are from one and the same population. This could be checked with the help of Students criteria by comparing the mean values of the different packages. More often, the approach of excluding the strongly deviating results is used. The method includes calculating the mean and the standard deviation of the primary data and establishing the condence interval. All results that fall outside the limits of the condence interval are eliminated (Figure .4.5). New interval limits are calculated with the remaining values until all values fall within the newly calculated condence interval. Control charts can also be designed on the basis of standardization of requirements, including nominal value, tolerance, upper and lower tolerance limits, and acceptable quality level (AQL). AQL is the maximum medium-level of defectiveness which is considered to be
3 1
5 6 8
4 2
Figure I.4.5. Algorithm for calculation of the control limits.
acceptable. It is a matter of agreement, but as a reference it is selected within the following limits in dependence to the type of product and its responsibility: AQL = 0.040.4%for products with a high level of responsibility such as those used in the healthcare and defense sector. AQL = 0.651.5%for products with high level of responsibility used in technical facilities. AQL = 2.56.5%mass technical tasks (such as those in the textile production). AQL = 6.515%products with low level of responsibility.
TYPES OF CONTROL CHARTS for measurable (quantitative) indicators; and for countable (qualitative) indicators.
If possible, the control charts for quantitative indicators should always be preferred as they provide better accuracy.
Control Charts for Quantitative Indicators For this type of control charts, the time is recorded along the x-axis and the qualitatively measurable indicators about the quality of the production are along the y-axis. When setting the control limits it is necessary to know the type of distribution of the random variable. The most widely spread types of control charts are those for normally distributed random variables. In case the distribution of the random variable differs from the normal one, other suitable distributions or free of distribution criteria might be used for verication of hypotheses. In order to nd out the distribution type, usually a preliminary sample is made which consists of a large number of elements (no less than 150200 observations). These observations are used for the creation of a histogram and then visually it is estimated whether the distribution can be considered to be a normal one. If necessary, a statistical verication of the distribution type can be performed. If it is necessary to control two or more indicators, those indicators could be checked for independence with the help of the procedures for examination of hypothesis for coefcients of correlation. If two indicators are interrelated then only one of them should be used as it contains information about the other one. Thus the statistical control is greatly facilitated. On the basis of the preliminary sample the capability of the process is dened, such as its adjustment and stability. The charts for qualitative indicators with an established normal law of distribution have the biggest practical application. According to the Bulgarian State Standard BDS 11319:1990 Statistical quality regulation. Control charts the following combined charts are used: x / S chart for mean arithmetic value and standard deviation; x / R chart for mean arithmetic value and range; x / R chart for median and range; Control chart for the practical limits; Control chart for individual values and absolute values of the subsequent differences xi / | xi xi +1 |.
Initially the chart that characterizes the dispersion of the process is analyzed. After the standard deviation or the range is stabilized, the chart that characterizes the setting of the process is analyzed.
x / S CONTROL CHART The chart is effective if the size of the sample is bigger than 8. It is convenient for usage when there are tools for the quick calculation of x and S. What is necessary to calculate the charts are k = 2030 samples with a total number of single tests n = 1000200. The mean arithmetic value xi and the standard deviation Si are established for all samples: xi = 1 n
x ,
ij j =1 n ij
(.4.6)
Si =
1 n 1
(x
j =1
xi ) 2 .
(.4.7)
The resulting values are recorded on the control charts and the individual points are connected with a broken line that shows the trends in the change of certain characteristics. The mean values of xi and Si are determined by x= 1 k 1 k
x ,
i i =1 k
(.4.8)
S=
S .
i i =1
(.4.9)
The resulting values are recorded on the charts (Figure .4.6). The control limits Kl and Ku are determined according to the formulae given in Table .4.2, and the values of the coefcients 3, 6, 3, 4, 5, and 6 are given in Table .4.3. When the control chart is calculated according to the standardization requirements, the limits are calculated in relation to the tolerance eld and the standard deviation S0, as dened in big samples or according to the selected acceptable quality level (Table .4.4). The calculated control limits are recorded in the chart (Figure .4.6). On the variance chart only the upper limit is represented. The value of the lower limit has no practical signicance since the movement of the results in its direction is a positive sign that the process is stabilized.
x / R CONTROL CHART The chart is similar to the x / S chart but instead of the standard deviation the range is used to estimate the dispersion. The range for each sample is calculated: Ri = xi, max xi, min, (.4.10)
Table .4.2. Calculation of the control limits for x / S control chart Control limits Type of chart According to test data Ku = x + A3 S Kl = x A3 S Ku = B4 S Kl = B3 S According to standardization requirements Ku = Tu A6 (T / 2) Kl = Tl + A6 (T / 2) Ku = B6 S0 Kl = B5 S0 Comment In combination with S-chart When n 5 Kl = 0
Table .4.3. Auxiliary coefcients for calculation of the control limits Coefcients 2 3 5 6 3 4 5 6 D1 D2 D3 D4 d2 n 2 1.880 2.659 2.800 0.293 0 3.267 0 2.606 0 3.686 0 3.267 1.128 3 1.023 1.954 1.360 0.423 0 2.568 0 2.276 0 4.358 0 2.574 1.693 4 0.729 1.628 0.950 0.500 0 2.266 0 2.088 0 4.698 0 2.282 2.059 5 0.577 1.427 0.740 0.553 0 2.089 0 1.964 0 4.918 0 2.114 2.326 6 0.483 1.287 0.620 0.592 0.030 1.970 0.029 1.874 0 5.078 0 2.004 2.534 7 0.419 1.182 0.540 0.622 0.118 1.882 0.113 1.806 0.204 5.204 0.076 1.924 2.704 8 0.373 1.099 0.480 0.646 0.185 1.815 0.179 1.751 0.388 5.306 0.136 1.864 2.847 9 0.337 1.032 0.430 0.667 0.239 1.761 0.232 1.707 0.547 5.393 0.184 1.816 2.970 10 0.308 0.975 0.390 0.684 0.284 1.716 0.276 1.669 0.687 5.469 0.223 1.777 3.078
Table .4.4. Standard deviation in accordance to the selected AQL AQL (in %) S0 AQL (in %) S0 0.04 /7 1.0 / 5.16 0.065 / 6.8 1.5 / 4.86 0.10 / 6.6 2.5 / 4.46 0.15 / 6.4 4.0 / 4.12 0.25 /6 6.5 / 3.7 0.40 / 5.78 10.0 / 3.3 0.65 / 5.44 15.0 / 2.88
Ku x = x
Kl
Ku
10 11 12 13 14 15 16
Observation number
Figure I.4.6. x / S control chart.
and the individual values are used for the calculation of the mean range of all k number of samples: R= 1 k
R .
i i =1
(.4.11)
The range is a less effective estimate in comparison with the standard deviation but is easier in terms of calculation. The formulae for the calculation of the control limits are given in Table .4.5. Table .4.5. Calculation of the control limits for x / S control chart Control limits Type of card According to test data Ku = x + A2 R Kl = x A2 R Ku = D4 R Kl = D3 R According to standardization requirements Ku = Tu A6 (T / 2) Kl = Tl + A6 (T / 2) Ku = D2 S0 Kl = D1 S0 Comment In combination with R-chart When n 6 Kl = 0
Very often, with time the mean value changes under the inuence of a certain factor, for example, the wearing out of the card-clothing wire leads to an increase in the number of neps in the card sliver. In these cases an x / R chart with a moving average which allows the retrieval of information, both about the accidental disturbances and the effect of the dominant factor, is used. The chart of mean arithmetic value is developed on the basis of the sample data taken during the period between two changes of the card-clothing wire. After determining and recording the mean arithmetic values of the individual samples on the chart, the approximating line is drawn (Figure .4.7).
Process needs adjustment x M
Mu Ku
Ml Kl Specific disturbances Completed adjustment
Figure I.4.7. x / R chart with a moving average.
10
11
12
Observation number
Parallel to the line are drawn the limits l and u, located at a distance of A2 R from the line. If a value falls outside these limits, this is a signal for the occurrence of specic disturbances in the process. The horizontal lines Kl and Ku are also recorded on the chart and they are determined in accordance with the following: x K u ,l = x + A2 R , 2 (.4.12)
where x is the change in the mean values for the studied period. If the mean value falls outside these limits, this is a signal that it is necessary to adjust the settings or replace the tool. The range chart is the same as in the preceding case.
x / R CONTROL CHART The chart is similar to the x / R chart but the median is used for the estimation of the center of the grouped results and its determination is much easier.
When working with these types of charts it is recommended that the number of measurements is an odd number. If the measurements are an even number, the median is determined as the mean arithmetic values of the two central values. The median value xi is determined for every sample. The mean median is determined on the basis of the median values: x= 1 k
x .
i i =1
(.4.13)
The control limits are determined according to the formulae given in Table .4.6.
Table .4.6. Calculation of the control limits for x / R control chart Control limits Type of chart According to test data Ku = x + A5 R Kl = x A5 R Ku = D4 R Kl = D3 R According to standardization requirements Ku = Tu 0.8 A6 (T / 2) Kl = Tl + 0.8 A6 (T / 2) Ku = D2 S0 Kl = D1 S0 When n 6 Kl = 0 Comment
The mean value of the median and the control limits is recorded on the chart. The range chart is calculated in the same way as for the x / R chart. The standard deviation of the median is Sx = that is, with
p 2
S n
p , 2
(.4.14)
= 1.25 being bigger than the mean arithmetic value. Since the variance of the
median is 25% higher, the control limits are wider (5 > 2), which results in less effective control.
CONTROL CHART FOR PRACTICAL LIMITS This chart is designed for statistical regulation of parameters with normal distribution and a two-sided tolerance eld with limits u and l. Typically in this chart, the control limits change in a dynamic way because they are calculated on the basis of the practical limits of dispersion. They are determined for each sample in accordance with its size n and probability g for the
inclusion of a certain percentage of the values of the measured parameter in the control limits. The control limits u,i and l,i are calculated according to the following formulae: Ku ,i = xi + K Si Kl ,i = xi K Si , where K is a coefcient dependent on n, g, and . It can be taken from tables or determined with the help of Students t-distribution t(g, n 1), which reect the probability g for inclusion of a given percentage of the values of the measured parameter in certain limits: K= t n (.4.17) (.4.15)
(.4.16)
and g are most often assumed to equal 0.90, 0.95, and 0.99. For each sample, on the control chart are recorded the points xi , Ku ,i , and Kl ,i , the limits u and l, and the median of the tolerance eld (Figure .4.8).
Tu Ku,i xi TM Kl,i Tl 1 2 3 4 5 6 7 Observation number 8 9 10
Figure I.4.8. Control chart for practical limits.
The location of points Ku ,i and Kl ,i serves as a basis for the dispersion and the location of points xi as the basis for the adjustment of the process. The controlled process is considered to be statistically manageable and capable when, during a given period of time, corresponding to more than ve consecutive samples, the points Ku ,i and Kl ,i are located in the tolerance eld between u and l.
CONTROL CHART FOR THE INDIVIDUAL AND ABSOLUTE VALUES OF THE SEQUENTIAL DIFFERENCES x i / | x i x i +1 | This type of control chart is applicable in the production of small batches. The sequential differences are determined by |x1 x2|, |x2 x3|, and so on. The mean of the individual values and the range are determined according to the following formulae: x= and R= 1 k 1 k
x
i =1 i
(.4.18)
x x
i =1
i +1 ,
(.4.19)
where k is the number of individual values. The control limits are determined according to Table .4.7, and the coefcients according to Table .4.3 when n = 2. Table .4.7. Calculation of the control limits for control chart for the individual and absolute values of the sequential differences Control limits Type of chart xi According to test data K u = x + 3R / d 2 K l = x 3R / d 2 Ku = D4 R Kl = D3 R According to standardization requirements Ku = Tu 3S0 Kl = Tl + 3S0 Ku = D2 S0 Kl = D1 S0 Comment n=2
|xi xi+1|
n=2
For control of the manufacturing process for the production of small batches, the following types of control charts can be used: chart for small batches of similar products; and chart for small batches with different distribution characteristics.
CONTROL CHARTS FOR BATCHES OF SIMILAR PRODUCTS Very often a certain machine is used for the consecutive production of small batches of similar products. For example, a spinning machine could be used for the production of yarns with linear density of 20 and 25 tex or a knitting machine could make products of different sizes. In this case, it is more appropriate to use relative deviations instead of absolute values, that is, the control charts for mean value are designed according to a zero mean line.
The procedures for the design of these charts are identical with those for classical charts. The difference is that the limits of the different batches of the products are recorded on the control charts.
Batch 1
Batch 2
Batch 3
Batch 4
Ku = x x 0 0
Kl
Ku
0 Date
10 12 14 16 18 20 22 24 26 Sample
Figure I.4.9. Control charts for batches of similar products.
The sequence in the design of charts is the following: 1. The mean value x of the controlled parameter is determined for every sample. The mean line on the control chart is marked as zero (Figure .4.9). 2. The mean value is determined for every sample with size n and the differences are recorded on the chart as x = x x . The values for the sample range are recorded on the range chart. 3. The control limits are determined. For the mean values chart: Ku ,l = A2 R. For the range chart: Ku = D4 R. 4. The limits for the individual batches are recorded distinctly on the control limits (Figure .4.9).
CONTROL CHARTS FOR SMALL BATCHES WITH DIFFERENT DISTRIBUTION CHARACTERISTICS Very often, the batches differ not only in terms of the mean value but also in terms of the degree of variance. This is typical for such cases when the difference in the mean values is big or when one and the same product is made from different materials. For example, when there is a big difference in the linear density of the yarn, the deviations in the absolute values are also big. At the same time, yarns with the same linear density but made from cotton or polyester bers have a different variance, that is, the control limits of the range should be different. This requires a transformation of the regular x / R charts. For regular charts, the values of x should fall within the following limits x A2 R < x < x + A2 R. (.4.20)
If x is subtracted and the equation is divided by 3, the result is the following equivalent equation: A2 < xx < A2 . R (.4.21)
Thus, the permanent control limits 2 can be recorded on the control chart for the mean values. Such a transformation can be made for each batch. In a similar manner, the inequality for the range: D3 R < R < D4 R can be transformed into: D3 < R < D4 , R (.4.23) (.4.22)
which allows the determination of constant control limits for the range that are dened by the coefcients D3 and D4. The control charts of this type are designed according to the following algorithm: 1. The mean value x is determined for every batch. It can be determined on the basis of a preliminary investigation; when such information is not available or is insufcient it can be determined on the basis of the nominal value. 2. The mean value of the range R is determined. When there is no reliable information about the mean value of the range, it is determined by selecting the desired value of the coefcient of process capability : R= T d2 . 6C p (.4.24)
3. For each sample the value of 4. 5. 6. 7.
x x R
is determined and the resulting values are recorded on
the mean values chart according to the zero line. The control limits for the mean values 2 are determined. For each sample, the ratio R / R is determined and recorded on the range chart. The control limits for the range, D3 and D4, are determined. The limits between the separate batches are recorded.
In cases of small batches, the classical control charts can also be used but instead of the mean values of the individual samples, the individual values are recorded. The variance of singular values S is higher than the variance of the mean values Sn (Figure I.4.10). It is a special case when n = 1. Several other charts are applicable in the textile practice on the basis of the specic features of the production process.
t. t.s
S n
x, x
t.s
2 1
Figure I.4.10. Control charts for small batches with different distribution characteristics.
CONTROL CHARTS FOR GROUP VALUES (i-CHARTS) They are used when several measurements are made for one sample, for example, in the control of the linear density of the yarn. The equality of the mean values and the variances should be made in advance (according to Students and Fisher criteria). The resulting values are recorded one above the other and it should be observed whether some of the points do not fall outside the limits of the condence interval (Figure I.4.11). The condence interval is determined in the same way as for the chart for individual values. The i-chart is a chart for primary data and the chart for the individual values is a special case of this chart when n = 1.
CONTROL CHARTS FOR EXTREME VALUES (k-CHARTS) They are a special case of the group value charts. From the resulting n values, the only ones that are recorded are the two extreme valuesxmin and xmax. Since they fall within the control
110 108 106 104 102 100 98 96 94 92 90 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Figure I.4.11. Control charts for group values.
limits, it infers that the other values also fall within them. For better clarity the points are joined together with broken lines (Figure I.4.12). The control limits are calculated in the same way as for the group values.
110 108 105 104 102 100 98 96 94 92 90 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Figure I.4.12. Control charts for extreme values.
Example Determine the control limits for the main statistical estimates of the linear density of cotton sliver. Five measurements have been made in a period of 20 days and the linear density has been recorded in Nm. For facilitation, the values have been multiplied by 1000, that is. when the measured value has been Nm 0.254, the value recorded in the table is 254. The data from the measurements is represented in Table .4.8, columns (2)(6). (Continued)
Table .4.8. Data and statistical estimates for the example Observation day (1) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 (2) 254 258 256 252 244 248 252 252 248 260 254 251 250 258 254 250 252 250 254 250 (3) 252 258 250 246 250 246 254 256 256 258 256 247 248 254 258 252 250 250 260 250 Values (4) 250 256 252 252 246 258 252 252 258 260 250 240 250 263 248 244 252 252 252 276 (5) 252 256 254 258 254 254 250 256 252 250 246 244 244 254 252 250 246 252 270 276 (6) 256 252 252 254 248 254 252 250 252 252 252 244 248 250 252 252 248 250 272 276 x (7) 252.8 256 252.8 252.4 248.4 252 252 253.2 253.2 256 251.6 245.2 248 255.8 252.8 249.6 249.6 250.8 261.6 265.6 252.97 S (8) 2.28 2.45 2.28 4.34 3.85 4.90 1.41 2.68 3.90 4.69 3.85 4.09 2.45 4.92 3.63 3.29 2.61 1.10 9.10 14.24 4.10 R (9) 6 6 6 12 10 12 4 6 10 10 10 11 6 13 10 8 6 2 20 26 9.7 x (10) 252 256 252 252 248 254 252 252 252 258 252 244 248 254 252 250 250 250 260 276 253.2
Mean values
In columns (7), (8), (9), and (10) are calculated the mean value, the standard deviation, the range, and the median, respectively, of the results for different days. CALCULATION OF x / S CONTROL CHART The control limits for the mean values are determined according to the formulae given in Table .4.2: For the lower limit: Kl = x A3 S = 252.97 1.427 4.1 = 247.116; For the upper limit: Ku = x + A3 S = 252.97 + 1.427 4.1 = 258.824.
It should be checked if there are mean values that fall outside the set limits. Three cases fall outside the set limits: Day number 12, under the set lower limit, and Day numbers 19 and 20, above the set limit (Figure I.4.13).
Mean value 270 265 260 255 250 245 240 235 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Figure I.4.13. x -chart during its calculation.
The limits are calculated again and the quoted values are excluded. The new values of the mean of the mean values and the mean standard deviation are x = 252.176 and S = 3.21 and the new control limits are: For the lower limit: Kl = x A3 S = 252.176 1.427 3.21 = 247.592; For the upper limit: Ku = x + A3 S = 252.176 + 1.427 3.21 = 256.761. When the limits are calculated in this manner, there are no cases of values that fall outside them and therefore, it can be inferred that these are the nal values of the control limits for the mean value. In a similar manner the control limits for the standard deviation are calculated by excluding the cases that have already been excluded in the determination of the control limits for the mean value (Day numbers 12, 19, and 20). For the lower limit: Kl = B3 S = 0 3.21 = 0; For the upper limit: Ku = B4 S = 2.089 3.21 = 6.71. These limits for the standard deviation are nal since there are no cases that fall outside them. The x / S control chart is drawn with the help of the already calculated control limits (Figure I.4.14).
Mean value 270 265 260 255 250 245 240 235 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Standard deviation 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Figure I.4.14. Final x / S chart for the example.
CALCULATION OF THE x / R CONTROL CHART The calculation of the control limits for the mean value and the range is done according to Table I.4.5. For the mean value, the control limits are: For the lower limit: Kl = x A2 R = 252.97 0.577 9.7 = 247.37; For the upper limit: Ku = x + A2 R = 252.97 + 0.577 9.7 = 258.57. Again, several cases fall outside the calculated limitsDay numbers 12, 19, and 20. The adjusted limits are as follows: For the lower limit: Kl = x A2 R = 252.176 0.577 8.06 = 247.53; For the upper limit: Ku = x + A2 R = 252.176 + 0.577 8.06 = 256.83, and after the abovementioned cases are disregarded, R = 8.06. These are nal since there are no cases that fall outside them.
In regard to the range, the control limits are: For the lower limit: Kl = D3 R = 0 8.06 = 0; For the upper limit: Ku = D4 R = 2.114 8.06 = 17.04. These are also nal. They serve as a basis for drawing the control charts presented in Figure I.4.15.
Mean value 270 265 260 255 250 245 240 235 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Range 18 16 14 12 10 8 6 4 2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Figure I.4.15. Final x / R control chart for the example.
In a similar manner, the other types of control charts can be calculated and drawn.
Control Charts for Qualitative Indicators In numerous cases, the values of the controlled parameter can only be measured in a qualitative and not in a quantitative manner. For example, presence/absence of a defect, good for sale/ defective. In order to implement current statistical control, the so-called control charts for immeasurable (qualitative) indicators are used in these cases. In comparison with the charts for quantitative indicators, these are simpler and can be used to control several parameters and
the costs for control, design, and lling in are signicantly lower. Along with these advantages, they have several disadvantages: they are less sensitive to the changes in the technological processes and provide less information. There are four types of control charts for qualitative indicators: control chart for the relative number of defective products (-chart); control chart for the number of defective products (n-chart); control chart for the number of defects (-chart); and control chart for the relative number of defects (u-chart).
CHART FOR THE RELATIVE NUMBER OF DEFECTIVE PRODUCTS (p-CHART) The chart is based on the calculation of the relative number of defective products in the sample: pi = fi . ni (.4.25)
where fi is the number of defective products, ni is the number of the products in the sample. The difference in the size of the samples should not be higher than 25% and the minimum sample size is determined according to the following condition: nmin 4 p or (.4.26)
nmin
4 . AQL
(.4.27)
where p is the mean arithmetic value of the relative number of defective products and denes the location of the central line on the chart. It is determined as follows: 1 m
p= where m is the number of samples.
p,
i i =1
(.4.28)
The number of samples should not be less than 20, and in order to set the control charts correctly, the size of the samples should be big enoughapproximately 100200 products. The control limits are calculated according to the formulae given in Table I.4.9, the mean size of the sample being: n= n1 + n2 + m + nm . (.4.29)
When calculating the control limits according to the standardization requirements, the AQL is taken into consideration. The standardization requirements are compatible with the accuracy capabilities of the process, if AQL p and the statistical regulation of the technological process is easy to implement. The standardization requirements are difcult to achieve when AQL < p. In this case it is necessary to improve the processes. Table I.4.9. Calculation of the control limits for p-chart Control limits Type of chart According to test data Ku = p + 3 -chart Kl = p 3 p (1 p ) n p (1 p ) n According to standardization requirements Ku = AQL + 3 Kl = AQL 3 AQL(1 AQL) n AQL(1 AQL) n
The calculated mean line and the control limits are recorded on the control chart (Figure I.4.16).
Ku p
Kl
Sample n f p 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Figure I.4.16. p-chart.
If the control is 100% then the control limits are calculated everyday on the basis of the volume of the daily production ni: Ku = p + 3 p (1 p ) , ni (.4.30)
Kl = p 3
p (1 p ) . ni
(.4.31)
In this case the control limits are not straight but broken lines (Figure I.4.17).
6 p(%) 5 4 3 2 1
2 3
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Samples
Figure I.4.17. p-chart in the case of 100% control.
CHART FOR THE NUMBER OF DEFECTIVE PRODUCTS (np-CHART) The control n-charts resemble the -charts, but for them it is necessary that the number of samples, n, is the same as the actual number of defective products, fi, in the sample is recorded on the chart. The control limits are calculated according to the formulae given in Table I.4.10 where the mean defectiveness of the products in the samples is f = f1 + f 2 + m + fm (.4.32)
Table I.4.10. Calculation of control limits for np-chart Control limits Type of chart According to test data Ku = np + 3 np (1 p ) Kl = np 3 np (1 p ) According to standardization requirements Ku = nAQL + 3 nAQL(1 AQL) Kl = nAQL 3 nAQL(1 AQL)
n-chart
If the resulting value for the lower limit is negative then the lower limit is assumed to be zero.
CONTROL CHART FOR THE NUMBER OF DEFECTS (c-CHART) The p- and n-charts are used for the control of the percentage (part) of defective products. The c- and u-charts are used to control the number of defects in a group of products (a batch), in one product, or in a preselected notional unit (the number of defects in a roll of fabric, the number of neps in 1g or, the number of neps in 1000 m yarn, the number of defects in a batch of 50 suits). The c-chart is used when the number of products in the batch is one and the same. The control limits are calculated according to the formulae given in Table I.4.11, where c is the mean number of defects. Table I.4.11 Calculation of control limits for c-chart Control limits Type of chart According to test data Ku = c + 3 c Kl = c 3 c According to standardization requirements Ku = AQL + 3 AQL Kl = AQL 3 AQL
c-chart
The acceptable level of defectiveness is given in the number of defects, n, per unit of production or the number of defects per unit of length, area, or volume. The control limits of the -charts are rounded down to a whole number since the number of defects can only be a whole number.
CONTROL CHART FOR THE RELATIVE NUMBER OF DEFECTS (u-CHART) The u-chart is used when the number of products in the different batches is different and they should be reduced to a unit of production:
ui =
ci . ni
(.4.33)
The control limits are calculated according to the formulae given in Table I.4.12 where u is the mean number of defects in a unit of production/notional unit, and n is the mean size of the samples. The acceptable level of defectiveness is given in number of defects per unit of production. Figure I.4.18 gives examples of - and u-charts.
Table I.4.12. Calculation of control limits for u-chart Control limits Type of chart According to test data Ku = u + 3 u-chart Kl = u 3 u n u n According to standardization requirements Ku = AQL + 3 Kl = AQL 3 AQL n AQL n
c-chart 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Sample
u-chart 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Sample
Figure I.4.18. - and u-charts.
Example Design a control chart for controlling the defectiveness in a textile factory. A research has been carried out over a period of 25 days and the data about the products made, ni, and the defective products, xi, is given in Table I.4.13. (Continued )
In column 3 is the calculated percentage of defective products, pi, for each day. The mean percentage of defective products p = 3.79% and the mean number of products made n = 1573.2 are determined. Table I.4.13. Calculation of control limits for p-chart Observation day number (1) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 ni (2) 837 1540 1455 1230 1814 2151 2032 2094 1973 1823 2093 1536 1314 1756 1628 954 881 1136 1546 1338 1095 1959 1850 1813 1482 xi (3) 21 67 55 30 88 91 76 67 58 66 96 82 41 46 37 46 37 41 47 36 54 103 91 68 59 pi (4) 2.51 4.35 3.78 2.44 4.85 4.23 3.74 3.20 2.94 3.62 4.59 5.34 3.12 2.62 2.27 4.82 4.20 3.61 3.04 2.69 4.93 5.26 4.92 3.75 3.98 Ku (5) 5.77 5.25 5.29 5.42 5.14 5.03 5.06 5.04 5.08 5.13 5.04 5.25 5.37 5.16 5.21 5.64 5.72 5.49 5.25 5.36 5.52 5.08 5.12 5.14 5.28 Kl (6) 1.81 2.33 2.29 2.16 2.44 2.55 2.52 2.54 2.50 2.45 2.54 2.33 2.21 2.42 2.37 1.94 1.86 2.09 2.33 2.22 2.06 2.50 2.46 2.44 2.30 Ku (7) 5.68 5.17 5.21 5.34 5.05 4.94 4.98 4.96 5.00 5.05 4.96 5.29 5.07 5.56 5.63 5.40 5.16 5.27 5.44 5.04 5.05 5.19 Kl (8) 1.76 2.27 2.23 2.10 2.39 2.50 2.46 2.48 2.44 2.39 2.48 2.15 2.37 1.88 1.81 2.04 2.28 2.17 2.00 2.40 2.39 2.25
The upper and lower control limits for the p-chart are calculated according to the formulae given in Table I.4.9 (the np-chart is not applicable as the number of products made each day is different). (Continued )
(Continued ) Upper control limit: Ku = p + 3 p (1 p ) 0.0379(1 0.0379) = 0.0379 + 3 = 0.0524 = 5.24% n 1573.2
The limits could also be calculated directly in percentage. Lower control limit: Kl = p 3 p (100 p ) 3.79(100 3.79) = 3.79 3 = 2.35% n 1573.2
Cases 12, 15, and 22 fall outside the control limits. They are taken out and the new control limits are calculated on the basis of the new mean percentage of defective products p = 3.72% and the mean number of products made n = 1554.9. Upper control limit: p (100 p ) 3.72(100 3.72) = 3.72 + 3 = 5.17% n 1554.9
Ku = p + 3
Lower control limit: p (100 p ) 3.72(100 3.72) = 3.72 3 = 2.28% n 1554.9
Kl = p 3
These control limits are nal, since there are no cases that fall outside them. The control chart is represented in Figure I.4.19.
p-chart 6.00 5.00 4.00 3.00 2.00 1.00 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Figure I.4.19. Final version of the p-chart for the example.
If the control is not going to be a sample control but a 100% control, the control limits should be calculated according to formulae (.4.30) and (.4.31). The calculated values for the upper limits are entered in column 5 of Table I.4.13 and for the lower limits in column 6. As expected, cases 12, 15, and 22 fall outside the control limits. The values calculated after they have been taken out for the upper and lower limits are given, respectively, in columns 7 and 8. There is no case that falls outside the resulting values which allows for the drawing of the control chart, represented in Figure I.4.20.
p-chart (100% control) 6.00 5.00 4.00 3.00 2.00 1.00 0.00 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Figure I.4.20. p-chart for 100% control for the example.
Example On the basis of 15 samples, each consisting of 5 products, design the control charts for the number of defects and their relative number per unit of production. The number of established defects for every sample is given in column (2) in Table I.4.14. The number of defects per unit of production ui =
ci n
(n = 5) is calculated in column (3).
The formulae from Table I.4.11 are used to determine the control limits for the number of defects after the mean number of defects have already been established:
c= Upper control limit:
c
i =1
15
23 = 1.53 defects per sample. 15
Ku = c + 3 c = 1.53 + 3 1.53 = 5.25 It is assumed that Ku = 5. (Continued )
(Continued ) Table I.4.14. Number of established defects for the samples Sample number (1) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Lower control limit: Kl = c 3 c = 1.53 3 1.53 = 2.18. It is assumed that Kl = 0. The results of sample number 5 fall outside the control limits and therefore it is taken out and they are calculated anew. The mean number of defects for the other 14 samples is c = 1.21. The new control limits are: Upper control limit: Ku = c + 3 c = 1.21 + 3 1.21 = 4.52. It is assumed that Ku = 4. (Continued ) i (2) 3 2 1 2 6 1 3 2 0 0 0 0 0 1 2 ui (3) 0.6 0.4 0.2 0.4 1.2 0.2 0.6 0.4 0 0 0 0 0 0.2 0.4
Lower control limit: Kl = c 3 c = 1.21 3 1.21 = 2.09. It is assumed that Kl = 0. These limits are nal. The controlled limits for the u-chart are calculated in a similar way. The formulae from Table I.4.12 are used where:
u=
c
i =1 15 i =1
15
23 = 0.307 defects per product. 15 5
Upper control limit: Ku = u + 3 u 0.307 = 0.307 + 3 = 1.05 n 5
Lower control limit:
Kl = u 3
u 0.307 = 0.307 3 = 0.80. It is assumed that Kl = 0. n 5
After sample number 5 is dropped out, the new mean value is u = 0.243 defects per product and the new control limits are:
Upper control limit: Ku = u + 3 u 0.243 = 0.243 + 3 = 0.904. n 5
Lower control limit: Kl = u 3 u 0.243 = 0.243 3 = 0.75. It is assumed that Kl = 0. n 5
These control limits are nal.

Textiles Chapter One

Uploaded by

Copyright:

Available Formats

Textiles Chapter One

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Textiles Chapter One

Uploaded by

Copyright:

Available Formats

TEXTILE PROCESSES

MOMENTUM PRESS, LLC, NEW YORK

ABOUT THE AUTHORS

XIV ABOUT THE AUTHORS

XVIII LIST OF TABLES

INTRODUCTION TO MATHEMATICAL STATISTICS

I.1. GENERAL TERMS AND DEFINITIONS

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 3 2 1 0 1 2 3 x

Figure I.1.1. Function of distribution of a continuous variable.

PART I: GENERAL TERMS AND DEFINITIONS 5

Figure I.1.2. Function of distribution of a discrete variable.

f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 4 3 2 1 0 1 2 3 4

Figure I.1.3. Probability distribution function.

0.4 0.3 0.2 0.1 0

Figure I.1.4. Order of distribution.

PARAMETERS AND NUMERICAL CHARACTERISTICS OF THE RANDOM VARIABLE

Figure I.1.5. Quantiles of the distribution function.

For certain types of distributions a mathematical expectation may not exist.

PART I: GENERAL TERMS AND DEFINITIONS 7

and for continuous ones it is

The equivalent formula is frequently used as well:

Figure I.1.6. Probability distribution functions with different variances.

PART I: GENERAL TERMS AND DEFINITIONS 9

The central moment from rst rank (k = 1) always equals 0: m1 = M

E>0 E=0 E<0

Figure I.1.8. Distribution with positive, zero, and negative kurtosis.

PART I: GENERAL TERMS AND DEFINITIONS 11

1 < rxy < 0

Figure I.1.9. Different cases of correlation dependencies.

PROPERTIES OF NUMERICAL CHARACTERISTICS

X n and each pair of random D{ Xn}.

I.2. LAWS OF RANDOM VARIABLES DISTRIBUTION

(b) f (x) 0.8 0.6 0.4 0.2 0 8 Mu 0 Sigma 1

(c) f (x) Mu 0.8 0.6 0.4 0.2 0 8 2 Sigma 1

PART I: LAWS OF RANDOM VARIABLES DISTRIBUTION 15

f (x) 0.6 0.4 0.2 s +s 0 8 6 4 2 0 2 68.26% 95.46% 99% 99.73% 4 6 8 x

Figure I.2.2. Characteristic values for the normal distribution.

and the formula for the variance is

The graph of the uniform distribution is shown in Figure .2.3.

Figure I.2.3. Uniform distribution.

PART I: LAWS OF RANDOM VARIABLES DISTRIBUTION 17

p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 x n=5 p = 0.1

p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 x n = 10 p = 0.1

p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 x n=5 p = 0.5

p = f (x) 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 x n = 10 p = 0.5

Figure I.2.4. Binomial distributions.

PART I: LAWS OF RANDOM VARIABLES DISTRIBUTION 19

(d) 0.6 0.5 0.4 0.3 0.2 0.1

Figure I.2.5. Poisson distributions.

The probability of one breakage is: P(1) =

e 0.833 0.8333 = 0.047. 3! e 0.833 0.8334 = 0.0087. 4!

For four breakages: P(4) =

I.3. STATISTICAL ESTIMATES

CONDUCTING THE TEST

PART I: STATISTICAL ESTIMATES 23

Figure I.3.1. Histogram and polygon.

Figure I.3.2. Cumulative curve.

x= where xi is ith consecutive value, n is the number of tests. grouped values

PART I: STATISTICAL ESTIMATES 25

x= for grouped values