See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/266178907

Dead-time compensation in discrete time control

Article  in  Systems Science · January 2007

CITATIONS READS

0 2,247

2 authors:

P. Albertos Pedro García

Universitat Politècnica de València Universitat Politècnica de València
386 PUBLICATIONS   3,861 CITATIONS    94 PUBLICATIONS   1,659 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Profesion@l: Equilibrio de género en el espacio europeo View project

Artificial Pancreas - Spanish Consortium on Artificial Pancreas and Diabetes Technology View project

All content following this page was uploaded by Pedro García on 20 September 2016.

The user has requested enhancement of the downloaded file.

discrete dead-time compensator, unstable time-delay systems, digital implementation

Pedro ALBERTOS*, Pedro GARCÍA*

DEAD-TIME COMPENSATION IN DISCRETE TIME CONTROL#

Most industrial processes are characterized by the presence of time delays. These time delays may be intrinsic to
the process to be controlled or associated to the controller itself. The Smith Predictor scheme was the most
famous control method for controlling stable single-input-single-output linear processes showing a delay in their
input or output. Solution to the problems related to robustness as well as those involved in the control of
integrating and unstable delayed processes were presented. Most of these solutions do not cover all the situations
and, in any case, they lead to complex controllers. This paper reviews the most relevant problems and their
solutions in the literature, and a new methodology to design dead-time compensator for stable, integrating and
unstable processes is reported. Several illustrative examples show that the robustness and performance of the
proposed methodology are similar or better than the more recently proposed dead-time compensators, the design
approach being simple and straightforward.

1. INTRODUCTION

One of the main difficulties in control systems design is undoubtedly the presence of delays,
becoming more challenging if the system to be controlled is unstable in open loop and/or non
minimum phase. Time delays may be intrinsic to the process, such as in chemical and biological
processes, distillation columns, processes with thermal exchanges, and so on. Delays can also be
introduced in the control loop (computation time of the control algorithm, distributed systems,
remote control, communication networks, sensors and/or actuators induced delays, etc.).
In general, the control system performance is very sensitive to all these delays, even more
than to other model parameters. In fact, a closed-loop control system may become easily unstable as
a consequence of delays. Systems with delay are infinite-dimensional, and thus their transfer
function has an infinite number of poles. Thus, the tuning of the parameters of a conventional
regulator to adjust these poles is very difficult.
Conventional controllers, such as PID controllers [1], could be used when the dead-time is
small but they show poor performance when the process exhibits long dead-time since a significant
amount of detuning is required to maintain closed-loop stability. In these cases, it is convenient to
introduce a dead-time compensating structure [5]. The Smith Predictor (SP) [18], and its many
extensions, generically denoted as Dead-Time Compensator (DTC) [15], may be considered as the
first control design methods to cope with delays in single-input-single-output linear processes
[17,4].
The main advantage of the SP is that time-delay is eliminated from the characteristic equation
of the closed loop system. Thus, the analysis and control design problems for processes with delay
can be converted into one for systems without delay. However this common conclusion has been
*Department of Systems Engineering & Control, Technical, University of Valencia, E-46022 Valencia, SPAIN
{pedro,[email protected]}
#
This work has been partially granted by CICYT Project DPI2005-09327-C02-01
shown to be misleading because, even if there is small mismatch in the considered model, the SP or
DTC’s exhibits stability features that can be neither detected nor analysed through the seemingly
equivalent time-delay free system. On the other hand, [22], the SP cannot be used for processes
having an integral mode because a steady-state error will appear under constant load disturbances.
Also, linked to the internal instability of the prediction, they fail to stabilize unstable systems, as
explained in [15]. Different modifications have been proposed [10,6,7,19,21,9,8].
Although a computer is used to implement the control law in any practical application, [2],
most of the contributions concerning DTC only analyzes the continuous time case. The first work
dealing with the analysis of the stability of the sampled-data DTC was presented in [16]. In [20], a
complete analysis of the design procedure of the discrete version of the scheme proposed in [23] is
also presented.
Recently a new method to control stable, integrating and unstable systems with long delays
has been presented [3]. Two schemes are provided, in order to consider the option of minimum and
non-minimum-phase systems, being compared with the most recently published papers on control
of unstable processes with long delays, and showing the improvements obtained with respect to
previous structures. The scheme proposed for minimum-phase systems, called the New Smith
Predictor Controller, is also compared with those reported by Zhong et al. [23,24,25], and the
advantages are pointed out. In this case, the controllers may be designed according to the exact
process model, but if the model is a high-order model, it results in a high-order controller.
In this paper this new methodology to design DTCs for the control of stable and/or unstable
systems, minimum or non minimum phase, with delays is summarized and compared with
previously published solutions. Special attention is devoted to integrating and open loop unstable
systems.
Contrary to other proposals, this methodology allows the controller design without taking into
consideration the delay in the tuning of the characteristic equation for reference tracking as well as
for disturbance rejection. By this, a substantial improvement is achieved with respect to previous
approaches. Some conclusions are drafted at the end of the paper.

2. THE SMITH PREDICTOR: A DTC FOR STABLE TIME-DELAY SYSTEMS

In many cases, processes can be modelled by means of a transfer function such as

P( s ) = G ( s )e − Ls (1)

where L is the time-delay and G(s) is the undelayed dynamics. The dead-time part L not only
represents the actual delay but also the high order dynamics of the real process.
The basic SP scheme is represented in Figure 1. In the ideal case, i.e. when there are no
uncertainties ( G ( s) = G r ( s); L = Lr ), the following expressions can be obtained:

K ( s )G ( s )e − Ls  K ( s )G ( s )e − Ls  − Ls
=y(s) r ( s ) +  1 −  G ( s )e d ( s ) (2)
1 + K ( s )G ( s )e − Ls  1 + K ( s )G ( s )e
− Ls


From Equation (2) it appears clear that the controller K ( s ) can be designed as if the process
has no time delay but if the system has any unstable pole, the closed-loop system will be also
unstable. On the other hand, from easy algebraic manipulation it is possible to rewrite (2) as
G ( s )e − Ls K ( s )G ( s )e − Ls
G yd ( s ) = +
1 + K ( s )G ( s ) 1 + K ( s )G ( s )
( G(s) − G(s)e− Ls ) (3)

Fig. 1. Smith predictor scheme

Note that, the closed-loop system has zero steady-state output to step load disturbances only if
lim ( G ( s ) − G ( s )e − Ls ) =
0 . However, if a system with an integrating mode is considered,
s →0

G ( s ) = Gs ( s ) / s , the steady-state error is not zero:

lim ( G ( s ) − G ( s )e − Ls ) =
lim Gs ( s ) L (4)
s →0 s →0

3. DTCs FOR INTEGRATING AND UNSTABLE SYSTEMS

The SP-based control scheme has been improved by several authors to overcome these
drawbacks. Some proposals are reviewed in this section.

3.1. ROBUSTNESS IN THE DTCs

As already mentioned, the SP may be very sensitive to modelling errors. In [13] (Figure 2) a
low-pass filter F(s) in the prediction error (the difference between the real output and the estimated
delayed output) has been proposed. This modification allows improving the robustness against
model errors even if they are significantly large.

Fig. 2. Filtered Smith Predictor.

If multiplicative uncertainties are considered, i.e. =

Pr ( s ) P( s )(1 + W ( s )) , the robust stability
condition is given by
K ( s )G ( s ) W ( s )
<1 (5)
1 + K ( s )G ( s ) F ( s )

The filter F ( s ) may be tuned to cancel the uncertainities W ( s ) at the desired frequency ranges.

3.2. DTCs FOR INTEGRATING PROCESSES

To deal with this situation, several proposals have been reported. Matausek et al [12]
1 + Td s
proposed the control scheme depicted in Figure 3, where the filter is given by F ( s ) = K 0 .
1 + Tf s

Fig. 3. DTC proposed by Matausek et al.

In this case, the disturbance rejection design is rather involved due to the filter presence. To
overcome this limitation, a modification in the undelayed model of the plant to be used in the minor
feedback of the SP has been proposed by Normey-Rico et al [14], as depicted in Figure 4, where
G1 ( s ) = G ( s )(1 − Ls ) , and G ( s ) = kv / s . The output sensitivity to disturbances is given by

G ( s )e − Ls K ( s )G ( s )e − Ls
y(s)
= +
d ( s ) 1 + K ( s )G1 ( s ) 1 + K ( s )G1 ( s )
( G1 (s) − G(s)e− Ls ) (7)

Fig. 4. DTC proposed by Normey-Rico and Camacho

The term G1 ( s ) − G ( s )e − Ls leads to null steady-state output for constant disturbances.

Another strategy, proposed by Zhong et al [23], is depicted in Figure 5. Again, the output /
disturbance transfer function is given by
C ( s )G ( s ) − Ls
=
y(s) e + G ( s )e − Ls (1 − Q( s )e − Ls )d ( s ) (8)
1 + C ( s )G ( s )

Fig. 5. DTC proposed by Zhong et al.

It allows the design of the controller C (s) as for a delay free system. Nevertheless, the design
of the filter Q(s) is not straightforward, although it can be chosen to cancel the steady-state error,
and the last feedback loop presents implementation problems, as the causality is not at all ensured.

3.3 DTCs FOR UNSTABLE PROCESSES

Many different approaches to control time-delay unstable systems have been reported in the
literature. Among these solutions, in the scheme shown in Figure 6 proposed by Lu et al. [9], the
degrees of freedom of the scheme depicted in Figure 3 are augmented in such a way that it can be
used to control also unstable systems. Note, that if K=
1 K=2 K 4 , this structure is the same as the
one proposed by Matausek et al.
One general drawback of these control schemes is that the control design is not as direct as in
the SP.

Fig. 6. DTC proposed by Lu et al.

4. NEW PROPOSED DTC

Consider the continuous-time state-space representation of the plan model, Equation (1)

x (t ) = Ac x(t ) + bc u (t − L)
(9)
y (t ) = cx(t )

The discrete-time model of this plant is

x k +1 = Ax k + bu k − d N ( z ) −d
⇒ P= z − d c( zI − A) −1 bz − d =
( z) G( z)= z (10)
y k = cx k D( z )
where, in order to simplify the development, it has been assumed L = dT ( d ∈ Z + ), T being
the sampling period. Based on the previous inputs and the current state, the output can be predicted
d-steps ahead by using Equation (10), yielding
d
yˆ k ,d cAd xk + c ∑ Ad −i uk −i
= (11)
i =1

or, by using the Z-transformation representation: =

yˆ d ( z ) cAd x( z ) + cΦ d ( z )u ( z ) , where the
Φ d ( z ) vector is given by: Φ=
d ( z) Ad −1bz −1 +  + Abz −2 + bz −1 . It should be noted that this predictor
requires access to the process state xk . Thus, in practice, its applicability is doubtful.

4.1. NEW PREDICTOR CONTROL SCHEME

By using the previous idea, a new methodology to deal with any time delay system is
addressed, allowing the use of any classical control design approach, as for delay-free systems. It is
easy to verify that

( Ad −1 z − d +  + Az −2 + Iz −1 )( zI − A) =I − Az − d , and Φ ( z ) = ( I − A d z − d )( zI − A) −1 b (12)

Thus, the scalar cΦ ( z ) is numerically equivalent to the following equality:

cΦ ( z ) ≡ G ( z ) − G * ( z ) z − d (13)

−1 N * ( z)
where G ( z ) ≡ cA ( zI − A) b =
* d
D( z )

Then, the predicted signal yˆ d ( z ) , as defined in Equation (12), can be computed as

yˆ ( z ) = cΦ ( z )u ( z ) + F ( z ) y ( z ) (14)

where the filter F ( z ) is defined as

G * ( z) N * ( z)
F ( z) ≡ = (15)
G( z) N ( z)
The proof is straightforward. Note that

G* ( z )
(G ( z ) − G * ( z ) z − d )u ( z ) +
yˆ ( z ) = G ( z )u ( z ) =
G ( z )u ( z ) (16)
G( z)

This prediction scheme is depicted in Figure 7, and the following expression is obtained:

K ( z )G ( z ) z − d  G( z) z −d K ( z )cΦ ( z )(G ( z ) − G ( z ) z − d ) z − d 
y ( z=
) r ( z) +  +  d ( z) (17)
1 + K ( z )G ( z )  1 + K ( z )G ( z ) 1 + K ( z )G ( z ) 
Note that, similar to the SP setting, the design of the controller K ( z ) can be made as for delay
free systems. The closed-loop system being stable, it can be seen that the steady-state error is
similar to that of simple loops without delay, the second term on the disturbance transfer function
vanishing.

Fig. 7. Proposed prediction-control scheme

If multiplicative uncertainties are considered, i.e. = Pr ( z ) P( z )(1 + W ( z )) , it is well-known

that the robust stability condition could be obtained from the output-noise sensitivity function, that
is
K ( z )G ( z )
( F ( z ) z − d − F ( z ) − 1) W ( z) < 1 (18)
1 + K ( z )G ( z )

The definition of F ( z ) requires dealing with minimum-phase systems.

For unstable non-minimum-phase delayed systems the design procedure is split into two
parts. First, a modification on the prediction signal presented above is considered.

yψ ( z ) = Gψ ( z )u ( z )
cΨ d ( z )u ( z ) + y ( z ) = (19)

where cΨ d ( z ) = A− d Φ d ( z ) . The availability of this signal will allow stabilizing the plant by
means of K(z), leading to the transfer function M(z). Afterwards, an external control loop is
designed by using a classical SP. The complete control structure is shown in Figure 8.

Fig. 8 Proposed prediction-control scheme for non-minimum phase systems

4.2. COMPARASION WITH OTHERS DTCs

In this section the proposed methodology is compared with some of the most recently
reported methods conceived to improve the performances of stable, integrating and unstable time
delay systems. That is, the schemes proposed by [13], [23], [9].
e −10 s
Example 1: Let us consider an example taken from [5,13] where the plant model is P( s ) =
s +1
=
and there is only a delay error, i.e., W (s) ( e − ( Lr − L ) s
− 1) . Lr represent the actual delay. The control,
as proposed in [5], is a PI controller with parameters K = 1 and Ti = 1 . If the delay of the process is
Lr = 12 , both the SP and the proposed scheme (Figure 7) are unstable. But if the estimated error, ê ,
is filtered, for example by taking F ( s ) = 1 /(1 + 2s ) , the system is stabilized. The output responses and
the Nyquist plots are shown in Figure 9. Note that both, the robust stability condition and the output
responses are practically the same. As both plots are coincident, in order to differentiate both step
responses, an additional step disturbance has been applied in one of the schemes at time 120 s.

Fig. 9. Output response and Nyquist plot of both the proposed scheme and the filtered SP.

Thus, for stable systems, the SP design and the proposed control scheme behave similarly.
This is not the same for unstable or non-minimum phase systems, as detailed in the next examples.

Example 2: In this example, taken from [23], it is shown that the proposed NDTC can be also
tuned to reject ramp disturbances. Consider the following process:

1
P( s) = e −5 s
s ( s + 1)(0.5s + 1)(0.2 s + 1)(0.1s + 1)
In [22], in order to obtain a simpler controller, the proposed controller is designed according
−6.5 s 1 −6.5 s
to the reduced model:= Pm ( s ) G=
m ( s )e e . The filter in Figure 5 is then designed as:
s
193.62s 2 + 21.5s + 1 ,
Q( s) = C ( s ) = 0.2 being designed as a proportional controller.
(5s + 1)3
1
The reduced model used in NDTC is: Pi ( s ) = e −6.8 s . The K-controller is tuned to reject ramp
s
1
disturbances, K ( s) = 10 ( s + 0.1)
2

2
, and the pre-filter is C ( s) = 4s + 1 .The filter F2 ( s ) , applied to the
s
312.92 s 2 +27.8 s+1
estimated error, ê , is tuned in a similar way to Q( s ) .Then F2 ( s) = (7s+1) . With a

sampling time T=0.05 s, the results in Figure 10a are obtained. Note that the IAE for a disturbance
response in the NDTC is 5.76 whereas it is 5.85 in the Zhong's scheme. When the actual dead time
becomes L = 5.8s , the responses with the same controller are shown in Figure 10b. Note that the
Zhong's response is unstable.
Fig. 10. Nominal system response to reject ramp disturbances. NDTC (solid line), Zhong (dash line).

4
Example 3: Consider the unstable process P( s ) = e −5 s . Following [9], four controllers must
10 s − 1
be tuned, K1 = 1.5 , K 2 = 7.93s + 1.25 , K 3 = 0.35 , and K 4 = 1.6 . By using the proposed NSPC, the PI
0.793s + 1
s + 0.25 1
controller and pre-filter are, respectively: K ( s ) = 1.8 , F1 ( s ) = . By using a sampling
s 4s + 1
period T=0.1s, the delay is L = 50Ts . The result shown in Figure 11a, are obtained. But if the
4.4 −5.5 s
modeling errors are such that the real process is Pr ( s ) = e , the responses are shown in
11s + 1
Figure 11b. The Xiang method is unstable.

Fig. 11. Nominal system responses for a step load disturbance of -0.1 at t=50 s.

5. CONCLUSION

The SP-based control scheme presents robustness and stability problems when dealing with
uncertainties and integrative or unstable open-loop systems. Several proposals have been reported.
Most of them loose the attractive facility of designing the control as for delay-free systems. A new
predictor-based control scheme has been proposed. It can be applied, with minor modifications, to a
wider spectrum of systems: stable/unstable, minimum/non-minimum phase, disturbed/undisturbed
systems. It also presents improvements from the robustness viewpoint, as far as model uncertainties
are considered.
Several examples illustrate the possibilities of this new DTC predictor-based control structure.

REFERENCES

[1] ASTROM K. J., HAGGLUND T. : Advanced PID Control. ISA - The Instrumentation, Systems, and Automation
Society, Research Triangle Park, NC 27709, 2005.
[2] ASTROM K.J.,WITTENMARK B.: Computer-controlled systems - theory and design (3rd ed.). Englewood
Cliffs, NJ: Prentice-Hall. 1997.
[3] GARCÍA P., ALBERTOS, P. HAGGLUND, T.: Control of unstable non-minimum-phase delayed systems.
Journal of Process Control, Vol. 16, 2006, 1099—1111.
[4] GU K., NICULESCU S.I.: Survey on Recent Results in the Stability and Control of Time-Delays Systems.
Journal of Dynamic Systems, Measurement, and Control. Vol. 125, 2003, 158—165.
[5] HAGGLUND T.: An industrial dead-time compensating PI controller. Control Engineering Practice, Vol. 4,
1996, pp. 749--756.
[6] LEE Y., LEE J., PARK S.: PID controllers tuning for integrating and unstable process with time delay. Chem.
Eng. Sci. 55 (2000) 3481-3493.
[7] LEE Y., LEE J., PARK S.: IMC-based control system design for unstable process. Ind. Eng. Chem. Res. 41
(2002) 4288-4294.
[8] LIU T., ZHANG W., GU D.: Analytical design of two-degree-of-freedom control scheme for open-loop unstable
processes with delay. Journal of Process Control 15 (2005) 559-572
[9] LU X., YANG Y.-S., WANG Q.-G., ZHENG W.-X. A double two-degree-of-freedom control scheme for
improved control of unstable delay processes. Journal of Process Control 15 (2005) 605-614.
[10] MAJHI S., ATHERTON D.P.: Obtaining controller parameters for a new Smith Predictor using autotuning
predictor and its modifications. Automatica 36 (2000) 1651-1658.
[11] MATAUSEK M.R., MICIÉ A.D: A Modified Smith Predictor for Controlling a Process with an Integrator and
Long Dead-Time. IEEE Transactions on Automatic Control 41 (1996) 1199-1203.
[12] MATAUSEK M.R., MICIÉ A.D: On the modified Smith Predictor for Controlling a Process with an Integrator
and long Dead-Time. IEEE Transactions on Automatic Control 44 (1999) 1603-1606.
[13] NORMEY-RICO J.E. and BORDONS C. and CAMACHO E.F.: Improving the robustness of dead-time
compensating PI controllers, Control Eng. Practice 5, (1997) 801-810.
[14] NORMEY-RICO J.E. and CAMACHO E.F.: Robust Tuning of Dead-Time Compensators for Processes with an
Integrator and Long Dead-Time, IEEE Transactions on Automatic Control 44, (1999) 1597-1603.
[15] PALMOR Z.J.: Time delay Compensation- Smith predictor and its modifications. The Control Handbook, (W.S.
Levine, Eds), CRSC Press, 1996.
[16] PALMOR Z., HALEVI Y.: Robustness properties of sampled-data systems with dead time compensators.
Automatica 26 (1990) 637-640.
[17] RICHARD J.P.: Time-delay systems: an overview of some recent advances and open problems. Automatica,
Vol. 39, 2003, 1667—1694.
[18] SMITH O.J.M.: Closer Control of loops with dead time. Chemical Engineering Progress. Vol. 53, 1959, pp. 217-
-219.
[19] TAN W., MARQUEZ H.J., CHEN T.: IMC-based design for unstable processes with time delays. Journal
Process Control 13 (2003) 203-213
[20] TORRICO B.C., NORMEY-RICO J.E.: 2DOF discrete dead-time compensators for stable and integrative
processes with dead-time. Journal of Process Control 15 (2005) 341-352
[21] WANG, Q.-G. AND ZHOU, H.-Q. AND ZHANG, Y. AND ZHANG, Y.: A Comparative Study on Control of
Unstable Processes with Time Delay. 5th Asian Control Conference, Melbourne, Australia, 2004, pp. 2006—
2014.
[22] WATANABE K., ITO M.: A process model control for linear systems with delay. IEEE Transactions on
Automatic Control, Vol 26, 1981, 1261—1268.
[23] ZHONG Q.-C. , NORMEY-RICO J.: Control of integral processes with dead-time. Part 1: Disturbance observer-
based 2 DOF control scheme. Control Theory and Applications, IEEE Proceedings 149 (4) (2002) 285-290
[24] ZHONG Q.-C., LI H.-X.: 2-degree-of-freedom Proportional-Integral-Derivative-type controller incorporating the
Smith principle for processes with dead time. Ind. Eng. Chem. Res. 41 (2002) 2448-2454
[25] ZHONG Q.-C.: Control of integral processes with dead time-Part 3: deadbeat disturbance Response. IEEE Trans.
on Automatic Control 48 (1) (2003) 153-159

View publication stats