- Email: [email protected]
Benefits from multivariable control in the PRP-industry
Copyright
© IFAC
Session 15
PRP 4 Automation, Ghent, Belgium 1980
BENEFITS FROM MULTIVARIABLE CONTROL IN THE PRP-INDUSTRY Chairman:
C. Foulard (Grenoble, France)
Multivariable control theory was developed since about 30 years. But it takes a long time to be able to define robust control structures and also to implement such controllers. Both constraints were solved since 10 years until today. We can define robust multivariable decoupled control structures and we can use computers or microcomputers for realization.
Multivariable control gives important benefits and is a necessity to solve specific problems (for instance decouplin9 or quasidecoupling, stability, high level performances ... ). Some unsolved problems have to be taken into account and we can estimate to have solutions (for instance for failures). The chairman can observe that such a division is classical in the history of control theory and its applications: use of digital computers 10 - 15 yea~ ago, use of state variables, use of optimization techniques.
Having now some experience, we can ask if it is interesting or not to use multivariable control instead of single variable loops. The question is not trivial because we can observe that very often single variable control is used and proposed by computer and controller manufacturers.
CONTRIBUTION by R.M.C. DE KEYSER (Automatic Control Laboratory, University of Ghent, Belgium)
So the chairman asked to 7 authors of papers in the PRP Congress concerned with control applications to give their opinion on that problem.
Abstract The benefits and the realization possibilities of multivariable control in industrial practice are considered. It is stressed that a pure multivariable control approach of coupled MIMO processes is theoretically attractive but leads to control strategies that are far more complex than a combination of single loop controllers.
These authors were successively: R. DE KEYSER, Ghent University, Belgium S. HEM, Reed International Consultants Ltd., UK S. SUH, St. Regis Paper Co., USA B. LEBEAU, Centre Technique du Papier, France Th. OLSEN, Sintef, Norway T. TAKAt,1ATSU, Kyoto Uni vers i ty, Japan P. URONEN, Oulu University, Finland
Some aspects of the control of coupled MIMO Processes
After these presentations a discussion during 20 minutes took place, based on 4 questions and for each several answers. We can summarize the discussion dividing participants into two main groups: - multivariable control techniques are 07 too high level abstraction to be acceptable in industry; there are also problems linked to their applications: reliability, identification, failures of a part of instrumentation consequently this g\oup prefer to use single variable technlques.
471
We will bring out these aspects using a specific example of a coupled r~I~~O system in the PRP environment: the control of a pressurized paper head box. In its most general configuration the paper head box can be considered as a system with three inputs (air valve, pump speed, thick stock valve) and three outputs (total head at the slice, liquid level, consistency). The thick stock valve-consistency subsystem is not always present but it might be of great use in a slave control loop of the basis weight controller on machines with variable speed (Lebeau et al., 1980). Most of the inputs and outputs of the above system are cross-coupled. This cross-coupling has
472
c.
Foulard
two drawbacks as far as control performance is concerned: - the stability problem of the global system is by far more complex to analyse than that of three non-interacting single loops even if the coupled MIMO process is controlled by three single loop controllers. However this difficulty should not be overestimated in control applications: most coupled MIMO processes are until now controlled by single loop (PI) controllers with reasonable although suboptimal control performance. It might however be suspected (and it was experienced during simulation studies) that the use of more powerful control strategies (e.g. time optimal strategies, model following strategies, minimum variance strategies) will also emphasize the stability problem. - the interaction may lead to a degradation of control performance especially on machines with speed or grade changes. Speed variation e.g. requires an adaptation of the total head pressure at the slice. This can be done by controlling the pump speed but as a by-effect the consistency varies leading to a disturbance in the basis weight. When the basis weight falls outside the specification range this means a production loss. However to evaluate the economical loss we should further investigate the characteristics of the disturbance. The amount of production loss is dependent on ~ the magnitude of the disturbance ~ the duration of the disturbance. The magnitude of the disturbance can be influenced by the control structure: multivariable or pseudo-mult1var;able control. The amplitude of the disturbance can even be brought back to zero when using decoupl;ng control. However due to some drawbacks pure dynamic decoupling does not seem to be a robust method for practical control. Strategies based on multivariable control theory are powerful but rather complex. Every output and every input of the MI~10 process ; s used in a coup 1ed t'lIt10 regul ator even if some subsystems of the physical process are noninteracting (e.g. air valve and consistency) or even if some process outputs are less important (e.g. liquid level). The e~eldo multivariable strategy is more flex1b e : it 1S a logical extension of the single loop structure for which the coupling in the process might be compensated by what is called "feedforward from the other input/output signals" (De Keyser; 1980) (see figure 1 for a two-input/two-output system). However one is free to restrict the procedure of feedforward compensation to the most important variables only, leading to a much simpler control system. The term "feedforward" ;s somewhat misleading because it is feedforward action from an internal disturbance (internal in the MIMO system) rather than from an external disturbance for which feedforward is a well-known technique. This observation has some importance concerning the stability
of the control system. The duration of the disturbance can be reduced to a large amount by the control stra~ : time-optimal control. When working with single-loop controllers on a coupled MIMO process it is worthwhile using timeoptimal controllers instead of PI controllers as is demonstrated in figure 2 (D'Hulster, De Keyser, Van Cauwenberghe; 1980)
Strategies for the control of coupled MIMO processes The ideas introduced above are summarized in table I together with some further characteristics. Dependent on the importance of each physical process output and its effect on paper quality and economical profits one has to choose for non-interaction or for interaction with limited amplitude and/or limited duration disturrances. Complexity of the control system is the price one has to pay when evolving from single-loop control over feedforward compensation to multivariable control strategies.
Conclusion When all process inputs and outputs are strongly coupled and all process variables are of equal importance in the control philosophy, multivariable control seems to be a very interesting and quite general design technique. Purely multivariable control strategies are however not the only candidates for controlling coupled MIMO processes in practice. Sometimes a priori knowledge about the process to be controlled can be used to separate less important output signals or even subsystems to obtain a less complex control system. Feedforward compensation techniques can be used to obtain such a more flexible control structure. Even when restricting the control system to a set of single-loop controllers, disturbances due to interaction in the MIMO process can be effectively rejected by using more powerful, e.g. time-optimal, strategies. It should be emphasized that time-optimal as well as decoupling or compensating techniques require the availability of a oood process model. It is interesting to incorporate them in a self-tuning or addptive control environment.
References DE KEYSER, R. (1980). Design and evaluation of self-tuning controllers. PhD. thesis, Automatic Control Laboratory, State
Benefits from Multivariable Control in the PRP-Industry
473
University of Ghent, Belgium (in Dutch) D1HULSTER F.M., R.M.C. DE KEYSER, A.R. VAN CAUltJErJBERGHE (1980). App 1i ca ti on of several parameter-adaptive controllers to a paper machine headbox. 4th IFAC PRP Conference. Ghent. LEBEAU B., R. ARRESE, S. BAUDUIN, R. GROBET, C. FOULARD (1980). Non interacting multivariable paper machine headbox control some comparisons with classical loops. 4th IFAC PRP Conference, Ghent. FB regulator
Setpoint
Ul
Ul
+
Process Process Process
Process Process +
Setpoi nt
U2
U2
+
Process
F8 reg ula tor
Y2
FF compensation
Fig. 1.
Feedforward from "intelnal disturbances".
t (s)
t (5) 1000
1500
Fig. 2. : Time-optimal comp.:lred to PI-control.
1000
1500
c.
474
Foulard
Time Optimal (single loop)
PI
(single loop)
x simple and robust
~l1ltivariable
x compensation of interac- x compensation of interaction tion x a priori knowledge about x still interaction, but of x easy to incorporate a very short duration so priori knowledge about the process does in genethat production loss is the process leading to a ral not lead to a more negligible more simple control simple control system structure
x interaction may result in x coupling between process production loss variables requires special attention to the stabil i ty problem
Table I
... (pseudo-multivariable)
l~ceJror ~r~ cu.~ _<~nsation
x feedforward from "internal" disturbances requires special attention to the stability problem
Strategies for the control of coupled
CONTRIBUTION by Dr. S. HEM (Reed International Consultants Ltd, Aylesford, Maidstone, Kent, UK) I shall forego the practice of pointing out the great importance of the subject of this round-table discussion in general terms. Instead I will describe briefly some of the multivariable control systems developed at RIC for use in Reed Mills as well as outside companies. Numerous multiple input and output control systems are designed on a single-loop basis and often the result is quite satisfactory. However, we have experienced considerable difficulty, on occasions, in obtaining a reasonably responding system using the single loop approach. It was only after proper consideration of the interactive nature of the system that a good result was achieved. The most obvious example of a multivariable control system in the paper industry is of course the interactive control of basis weight and moisture. This type of control is now so common on paper machines that very little that is new can be said about it. Nevertheless, we have been able to improve this control system's performance by the development of special multirate sampleddata control algorithms that allow greater stability at higher overall loop gains. The method has been extended to the general multivariable situation employing the stochastic Linear-Quadratic-Gaussian (LQG) procedure and also allows for the use of the Kalman filter for estimation of inaccessible state variahles. It should be noted, however, that multivariable control systems which have been designed using the LQG procedure can become unstable for a failure in one of the instruments. Special precautions must therefore by employed to guard against this situation.
MI~ln
complex control strategies (e.g. large dimensionality, reconstruction of ) state variables, x elegant solution of the stability problem ~
...
processes.
Another multivariable control system employed on some of our paper machines operates on cross machine direction profile data of certain paper properties. The control system produces a number of control actions within given constraints in a highly interactive system and performs an on-line iterative process of constrained optimization. In particular, systems for moisture profile control using plain presses, swimming rolls and Nipco rolls have been successfully implemented. The application of moisture profile control and profile management in general represent an economic and operational benefit at least as large as the one achieved with the machine direction control system. ~t is int:resting to note the resurgence of lnterest ln the design of headbox control systems. This problem is particularly interesting for me since I made a theoretical study of the dynamic behaviour of a headbox nearly 20 years ago. ~1ultivariable control theory has now developed to the point that the interaction problems can be understood and in turn makes it possible to design highly sophisticated and worthwhile control systems for this unit process of the paper machine.
Multivariable control systems have been designed using a variety of approaches, the LQG problems has been mentioned, inverse Nyquist array, characteristic loci techniques or pole assignment are other very successful methods, all with their own particular merits, too numerous to go into here. Finally, I would like to say that although we are moving towards systematic procedures and techniques for modelling and computer-aided controller design there will always be a need for the subjective element of creative engineering and know-how required for understanding the real-world aspect of the systems that must be controlled.
Benefits from Multivariable Control in the PRP-Industry
475
CONTRIBUTION by S. SUH (St. Regis Paper Co., West Nyack , USA)
As you may know, the paper industry is no stranger to the subject of multivariable control and has been exposed to this since early in 1960. The papermaking process is a prime example of a process with multiple interacting variables as shown in Figure 1. In addition to papermachine control, we at St. Regis have had considerable experience with multivariable control in other process areas such as Kamyr digester and utility plant, see Figures 2 and 3. We also recognize that there are other areas within a Kraft mill that could be easily considered for multiloop control. Table 1 shows such a 1i st.
MULTIVARIABLE CONTROL OF PAPERMACHINES
STOCK FLOW STEAM PRESSURE OR MACH INE SPEED
WEIGHT PAPERMAKI NG PROCESS
MOISTURE
MULTI VARIABLE CONTROL OF KAMYR DIGESTERS I have so far limited my opening remarks to the potential areas of application. Let me now move on to the benefit side of multivariable control which is the theme of this TEMPERATURE COOKING round-table session. The benefits we have KAPPA CHEMICALS PROCESS enjoyed at St. Regis through multi variable TIME control are substantial and readily acknowledged by the management. They include improved quality, fuel efficiency and production rate, and reduced raw material usage. However, this was attained at the added expense of manpower and capital equipment. For this MULTIVARIABLE OPTIMIZATION CONTROL OF reason, one should conduct an economic analysis of "cost vs. benefit before committing STEAM AND ELECTRICITY GENERATION oneself to multivariable control. It should also be mentioned here that it is sometimes possible to reduce a multi loop problem to a single-input/output problem and still reap ~ STEAM t--H_I_GH_.-.-, ELECTRICITY PROCESS STEA,& a good portion of the total benefit. PRESSURE GENERATION ERATION E GN STEAM AIR ~ ELECTRICITY I recognize that the advent of cheaper and faster computers, state-space method and advanced computer control strategies has PURCHASED enabled us to implement multi loop control more simply and inexpensively as compared to the 60 IS. However, it is getti ng di ffi cul t to recruit trained people who can work and provide adequate maintenance on the multiTable variable control systems installed in the plant level. The maintenance aspect of multivariOther areas of interest able control, which is skilled-labor intensive, should also be taken into consideration f recovery boilers when dealing with multivariable control. l power boilers The point I want to make here is that in our ~ lime kiln/causticizing plant industry, multivariable control is looked l evaporators at from the return-an-investment point of view. l refi ners t washers ll
c.
476
Foulard
CONTRIBUTION by B. LEBEAU (Centre Techniwue de 1 'Industrie des papier, cartons et celluloses, Grenobles, France) If we lookat the hea~box control system, it is well-known that we have a multivariable system with the following classical inputs:
4.
Does multivariable control give very significant economic returns compared with classical one?
5.
Is educational level sufficient for industrial people to use new control technique?
I think that the round table can give some answers to these different questions.
- air valve - fan pump speed and the following outputs
CONTRIBUTION by P. URONEN (Oulu University, Finland)
- 1eve 1 - total pressure. As you have seen earlier we have a strong interaction between these two loops. For example, if we change the fan pump speed, we do not only change the total head but we modify the level. So it is well-known that for the control of these two loops, we can have two following single variable control structures. Several people have proposed to decouple these two loops by means of decoupled compensators designed generally by using transfer function representation. Moreover, if we look on a paper machine, at modifications of the total head, we have disturbances on the basis weight. And in this case, it is necessary to change the stock valve in a coordinating way to keep constant substance flowrate at the slice opening. So, we increase the dimension of the control and we obtain a 3-inputs/3-outputs system. And finally, it is also well-known that we have a fourth input for the headbox which is the slice opening. This input also has a very strong interaction with the outputs. So, a good control system of the headbox must take into account the multivariable nature of this process and for example, we can imagine new multivariable control of the headbox. So question can be the following: 1. What kind of automatic control techniques
must we use to solve this problem? We have seen that for a 2-inputs/2-outputs system it is possible to use classical single variable techniques but with a 4-inputs/4-outputs system what kind of control must we use? 2. What kind of criterion must we use to the design of the control structure which takes into account industrial objectives? 3. What is the better control design which gives simple implementation on the process for industrial mill people?
It is a well-known fact that systematic methodology for the analysis and synthesis of multivariable control systems is available today. There also exist designing program packages, i.e., CAD for multivariable control systems based either on frequency-domain or time-domain methods. So the theory of multivariable control has reached a certain level of maturity. Another fact is that a complicated PRP process, for example, a recovery boiler or paper machine, is a system with many interacting cause-effect relations; thus the dynamic modelling and control of this kind of system is very difficult and basically it should be handled by stochastic and multivariable methods. It is also clear that in order to control this kind of processes, several sensors and actuators are necessary, and that singleloop approach can not give the best possible solution. So there exist the theoretical methods and practical needs for multivariable control in PRP industries, but why are these methods not applied to a larger extent? Obviously, there are definitEly benefits and advantages to be gained by application of multivariable and other so-called "mo dern" control methods in PRP industries; decoupling of interactions for example, in paper machine control systems, mass balance control at recovery boilers, level control of a Kamyr-digester etc., are just a few simple real applicatons of multivariable control in the pulp and paper industry. With multivariable control, the following advantages can be achieved : - better overall performance, i .a. closer control - compensation for interactions of single contro 1 loops, strong non-l i neari ti es, long time-delays etc. - better control of batch type and sequencing - control of not directly measured variables - control of processes which are very difficult to handle with SISO-control strategy - optimization and coordination in larger meaning - fault detection, cross checking and monitoring.
Benefits from Multivariable Control in the
The barriers and restrictions for the use of these methods are : -these methods are more sophisticated, more complex and more difficult to learn and use - the direct economic benefits from using them are difficult to show a priori. - shortage of skilled people in industry - updating and maintaining of the system is more difficult - more complex back-up system and difficulties in manual operations - reluctance against new ideas. I feel that together with distributed digital control systems, these methods will be increasingly more applied and used; we must start from simple and reliable applications with an evolutionary process towards more sophisticated systems. It is a typical diffusional process of new technology and innovation. by T. TAKAr1ATSU and I. HASHIMOTO (Kyoto University, Dept. of Chemical Eng., Japan)
CONTRIBUTIOI~
In recent years, the rapid increase in raw material and energy costs, especially the enormous increase of energy cost since the oil embargo of 1973, has caused a great change in the philosophy of process system design and forced process designers to utilize energy and other natural resources more efficiently. Intermediate storage tanks are made smaller or eliminated completely to reduce inventory and capital cost. This, however, results in more sophisticated and highly coupled process systems. Design margins are also reduced for improving the static design performance, which causes loss of flexibility and operability in the process. In order to minimize the energy consumption, heat integration is throughly performed by connecting heat producing units the heat of which was previously wasted, and heat consuming units which were previously uncoupled. Energy integration can save a lot of energy from the static viewpoint. But from the dynamic viewpoint, it makes the process a highly interconnected multivariable system which loses flexibility, operability and controlability. These lost favorable dynamic atributes have to be recovered by implementing the good quality control. This offers strong impetus for creating new advance in multivariable control techniques. Here, we choose a heat integrated distiTIation column system as an example for further discussion. Distillation is most widely used for the separation of products in refineries or many chemical processes. It is a typical energy consuming process, and it presents as a prime candidate for the improvement in energy efficiency.
~KP-Industry
477
Significant efforts have been devoted to the application of modern multivariable control techniques such as decoupling control, disturbance rejection control, the inverse Nyquist array method, and so forth, to a binary distillation column control, that is to control the compositions of both the top and bottom streams simultaneously. The separation of multi-component mixtures requires a network of distillation columns which includes material inteqration. This system provides a very challenging dynamic problem to which multivariable control theories have to be intensively applied. Let's note the prominent problems in designing the control system of a highly interacted multivariable system. 1. How to build dynamic model with the necessary and sufficient accurQCY based on which the control system design can be dealt with: The selection of the model is most crucial in applying any multivariable control technique. The precise physical model is generally too complex to be used for designing control systems. Simple "black box" models between input and outputs are, therefore, favored, but they often lead to poor control po1i ci es . 2.
How to design control systems based on a limited number of measuring points: It is very common in actual chemical processes that only some of the state variables are available for utilization. Even when it is possible to feedback all of the state variables, it is not unusual to utilize only some of them due to various restrictions or for economic reasons.
3.
How to select the best palrlng between manipulating and output variables; The relative gain array method selects favorable pairing. But this pairing can often lead to poor control, as pointed out by several researchers. More intensive research efforts will be necessary to resolve this problem.
4.
Rescrutinization of multivariable control techniques and their extension in order to obtain more robust control : Every technique has some advantages and disadvantages. A definitive method which can be applied to any multivariable process control and lead to robust control with high quality, has not been developed yet. By applying multivariable control techniques to more realistic problems, the applicability of each technique has to be rescrutinized and at the same time the extension of the theory has to be done, if necessary.
c.
478
5.
Foulard
Trade-off between high-quality control system design and optimal steady-state design: The high energy cost provides a very strong incentive to practice the heat integration which makes the process system more complex and more dynamically interacting. The more complex and interacting the system becomes, the more difficult it is to design the optimal system from both the steady state and dynamic points of view. The problem is how to reconcile the steady-state design with the control system design. In other words, a method has to be developed to incorporate the control structure design as part of the steady state design optimization. The research relating this topic has just started and much effort has to be devoted to resolve this problem.
As discussed in the above, modern multivariable control techniques seem to offer considerable promise in improving the quality of process control, but it has not yet been widely implemented on industrial scale processes. It is not easy to assess to costs of engineering manpower, control hardware and instrumentations against the potential benefits of modern multivariable control techniques in the industrial field. The control system which can secure the optimal operation of a large complex process system, must have a decentralized structure. The control system of each subsystem would be designed by a multivariable control technique by taking into account their existing strong interactions and using ~icro- and minicomputers. The coordination between subsystems could be performed by a large computer so as to optimize the overall objective. There have been few applications of hierarchy or decentralized control in real industrial processes. We must endeavour to exploit advances in this direction in order to provide a far better control system for the real complex and large chemical system. DISCUSSION
are very large, it is possible to break them down into smaller multivariable subsystems. This is a very i~portant part when you have very large systems. On the second point about safety and on computer frontiers that is something we met quite a lot in the paper industry. What do you do when a pump fails? Perhaps you should consider the control systems as part of the system on the same basis that you could consider the other hardware. If it is down the plant is down. But I would say that probably computers are more reliable than pumps and all the other hardware together. ~Q~~~~~
(by Prof. Brandenburg, Technical University Munich, FRGermany) I would like to give short information concerning the electrical drives in continuous plants of PRP-industries. They are coupled by the moving web in such a way that a multivariable system results. Therefore control synthesis must be carried out with the methods of the multivariable theory. However in many practical circumstances simplifications are possible due to fast speed control loops, by which the drives are approximately decoupled. The dynamical behaviour becomes very poor, if certain conditions are not fulfilled by the mechanical system. Then state controllers with observers are favorable. The problem we are working on is to get state controllers which are simple enough to be applied in industrial plants. ~Q~~~Q~
(by Prof. Uronen) I fully agree with those points made by the representative of the chemical industry. These special safety questions may be there even much more critical than in the pulp and paper industry because there are hazardous areas and explosive materials handled in destillation columns, etc. Potentially there are big production losses in pulp and paper industry but in addition of that in systems of petrochemical industries there are other risks. I fully agree that these multivariable systems need computer systems with very good redundancy and back-up systems for difficult and hazardous situations. This is really one of the biggest questions here. g~~~!iQ~
g~~~!iQ~
(by C. Van Rijn, Koninklijke Shell Laboratorium, the Netherlands) In the petrochemical industry the scope for multivariable control is clearly visible. In this respect, I like to make two points - the identification of MIMO systems of appreciable dimensions requires considerable effort (MIMO identification) - the aspect of loosing the computer faces the operator with different backup situations. May I have the opinion of the panel on these?
(by Prof. J. Van De Vegte, University of Toronto, Canada) Multivariable control techniques are only now gradually being simplified sufficiently that those who have to implement them are willing to consider them. State estimation or Kalman filtering are in general too complex for practical systems. Output feedback and approximate noninteraction provide promise to sufficiently simple control schemes, with dynamic compensation restricted to fixed low order.
~Q~~~~
~Q~~~~!
(by dr. Hem) I think, on the first aspect where systems
(by dr. De Keyser) I fully agree with the comment of Prof.
Benefits from Multivariable Control in the PRP-Industry
Van de Vegte. Indeed we should not use multivariable control if we can do it with singleloop controllers. This is what I said before. Also decoupling control, I should not use it in practice, perhaps approximate decoupling. However we should not stay with the singleloop controls with PI-regulators as we do now. We should go something in between let us say we should use single-loop controllers with more complicated control algorithms, with more powerful algorithms instead of PI. That will be already much better than the methods we are using now. But I agree we should not go to multivariable control except for some special processes as perhaps electrical drives and some other processes but we should not generalize. ~Q~~~0~
(by dr. Hem) It is obviously complicated but using advanced technology ought to be unobtrusive and really no one is questioning what goes no inside a television set in order to use it. The same thing could be said for control systems, provided they are made easy to operate for the operators and that is obviously a question of design. The fact that one may use advanced technology to operate the system should not ~o away with that. As you perhaps suggest we lmplemented several multivariable control systems including Kalman filters. There is no need to tell the operator that there is a Kalman filter in the loop. He would not know what you talk about, but it is perfectly possible to incorporate advanced technology without having to tell everybody details. ~Q~~~Q~
(by Prof. C. Foulard) I would like to give a little comment about this intervention. First I think that what w~s complicated in the beginning can be very slmple after some years, e.g. computers were complicated 20 years ago and a laser e.g. also was complicated and now we have lasers for printing. So this can change. I want also to comment on the quasi-decoup1ing system, the output feedback, mu1tivariable control, etc. There are answers to all these questions now and e.g. the structures which were presented by dr. Lebeau in his paper on headbox control solve these proble~s. I think not all the problems, but this one.
479
~Q~~~Q~
(by dr. Olsen) I just like to make some comments on this question. You design a control system either monovariable or multivariable to suit your specifications. If the specifications are operator acceptance you design the control system to achieve operator acceptance e.g. instead of going directly for DDC-control ~ou use a man in the loop with graphical dlsplays to guide his operation of the plant e.g. If the algorithm within the computer, guiding the operator, is based on multivariable control theory, or if it is based on monovariable control theory, does not affect the operator as long as he designs a system w~ich is guiding him in a way acceptable to hlm. The same way in cooperation with the management, to get user acceptance of the system you agree on the specifications of the system. Decoupling is one way to go if decoupling is important to you. But you pay for decoupling e.g. by large variations in your control inputs. If you choose decoupling, you might see motors going very often full speed or lower speed. If you like to have few and small variations in your process control variables, then you might ca~l f?r linear quadratic feedback with heavy welghtlng of your control variations. These are only different ways of designing control systems. To sum up I would just say that I he~v~ly disagree with some of the very heavy 0plnlons for one or the other method. I think that you should choose the method for the application. ~Q~~~~~
(by dr. Lebeau) I just want to make a remark on some of the comments which have been made and particularly this one of somebody, saying when a contr?l scheme stops then the plant stops. That lS not acceptable for an industrial operation. The same person said we have not to know what is in a television set. You have not to know it when you have not to support a stop of this television set, but if there is something wrong, then we can accept to have a quarter of the video-screen but we cannot accept to have nothing. I think this remark explains why it is very difficult to implement multivariable control actually. It is because it is not at the state of maturity as you have heard before. It is not clear enough to be understood by people as we have in the mills and it is those people who take back a process when something fails. That is why we refuse it.
© IFAC
Session 15
PRP 4 Automation, Ghent, Belgium 1980
BENEFITS FROM MULTIVARIABLE CONTROL IN THE PRP-INDUSTRY Chairman:
C. Foulard (Grenoble, France)
Multivariable control theory was developed since about 30 years. But it takes a long time to be able to define robust control structures and also to implement such controllers. Both constraints were solved since 10 years until today. We can define robust multivariable decoupled control structures and we can use computers or microcomputers for realization.
Multivariable control gives important benefits and is a necessity to solve specific problems (for instance decouplin9 or quasidecoupling, stability, high level performances ... ). Some unsolved problems have to be taken into account and we can estimate to have solutions (for instance for failures). The chairman can observe that such a division is classical in the history of control theory and its applications: use of digital computers 10 - 15 yea~ ago, use of state variables, use of optimization techniques.
Having now some experience, we can ask if it is interesting or not to use multivariable control instead of single variable loops. The question is not trivial because we can observe that very often single variable control is used and proposed by computer and controller manufacturers.
CONTRIBUTION by R.M.C. DE KEYSER (Automatic Control Laboratory, University of Ghent, Belgium)
So the chairman asked to 7 authors of papers in the PRP Congress concerned with control applications to give their opinion on that problem.
Abstract The benefits and the realization possibilities of multivariable control in industrial practice are considered. It is stressed that a pure multivariable control approach of coupled MIMO processes is theoretically attractive but leads to control strategies that are far more complex than a combination of single loop controllers.
These authors were successively: R. DE KEYSER, Ghent University, Belgium S. HEM, Reed International Consultants Ltd., UK S. SUH, St. Regis Paper Co., USA B. LEBEAU, Centre Technique du Papier, France Th. OLSEN, Sintef, Norway T. TAKAt,1ATSU, Kyoto Uni vers i ty, Japan P. URONEN, Oulu University, Finland
Some aspects of the control of coupled MIMO Processes
After these presentations a discussion during 20 minutes took place, based on 4 questions and for each several answers. We can summarize the discussion dividing participants into two main groups: - multivariable control techniques are 07 too high level abstraction to be acceptable in industry; there are also problems linked to their applications: reliability, identification, failures of a part of instrumentation consequently this g\oup prefer to use single variable technlques.
471
We will bring out these aspects using a specific example of a coupled r~I~~O system in the PRP environment: the control of a pressurized paper head box. In its most general configuration the paper head box can be considered as a system with three inputs (air valve, pump speed, thick stock valve) and three outputs (total head at the slice, liquid level, consistency). The thick stock valve-consistency subsystem is not always present but it might be of great use in a slave control loop of the basis weight controller on machines with variable speed (Lebeau et al., 1980). Most of the inputs and outputs of the above system are cross-coupled. This cross-coupling has
472
c.
Foulard
two drawbacks as far as control performance is concerned: - the stability problem of the global system is by far more complex to analyse than that of three non-interacting single loops even if the coupled MIMO process is controlled by three single loop controllers. However this difficulty should not be overestimated in control applications: most coupled MIMO processes are until now controlled by single loop (PI) controllers with reasonable although suboptimal control performance. It might however be suspected (and it was experienced during simulation studies) that the use of more powerful control strategies (e.g. time optimal strategies, model following strategies, minimum variance strategies) will also emphasize the stability problem. - the interaction may lead to a degradation of control performance especially on machines with speed or grade changes. Speed variation e.g. requires an adaptation of the total head pressure at the slice. This can be done by controlling the pump speed but as a by-effect the consistency varies leading to a disturbance in the basis weight. When the basis weight falls outside the specification range this means a production loss. However to evaluate the economical loss we should further investigate the characteristics of the disturbance. The amount of production loss is dependent on ~ the magnitude of the disturbance ~ the duration of the disturbance. The magnitude of the disturbance can be influenced by the control structure: multivariable or pseudo-mult1var;able control. The amplitude of the disturbance can even be brought back to zero when using decoupl;ng control. However due to some drawbacks pure dynamic decoupling does not seem to be a robust method for practical control. Strategies based on multivariable control theory are powerful but rather complex. Every output and every input of the MI~10 process ; s used in a coup 1ed t'lIt10 regul ator even if some subsystems of the physical process are noninteracting (e.g. air valve and consistency) or even if some process outputs are less important (e.g. liquid level). The e~eldo multivariable strategy is more flex1b e : it 1S a logical extension of the single loop structure for which the coupling in the process might be compensated by what is called "feedforward from the other input/output signals" (De Keyser; 1980) (see figure 1 for a two-input/two-output system). However one is free to restrict the procedure of feedforward compensation to the most important variables only, leading to a much simpler control system. The term "feedforward" ;s somewhat misleading because it is feedforward action from an internal disturbance (internal in the MIMO system) rather than from an external disturbance for which feedforward is a well-known technique. This observation has some importance concerning the stability
of the control system. The duration of the disturbance can be reduced to a large amount by the control stra~ : time-optimal control. When working with single-loop controllers on a coupled MIMO process it is worthwhile using timeoptimal controllers instead of PI controllers as is demonstrated in figure 2 (D'Hulster, De Keyser, Van Cauwenberghe; 1980)
Strategies for the control of coupled MIMO processes The ideas introduced above are summarized in table I together with some further characteristics. Dependent on the importance of each physical process output and its effect on paper quality and economical profits one has to choose for non-interaction or for interaction with limited amplitude and/or limited duration disturrances. Complexity of the control system is the price one has to pay when evolving from single-loop control over feedforward compensation to multivariable control strategies.
Conclusion When all process inputs and outputs are strongly coupled and all process variables are of equal importance in the control philosophy, multivariable control seems to be a very interesting and quite general design technique. Purely multivariable control strategies are however not the only candidates for controlling coupled MIMO processes in practice. Sometimes a priori knowledge about the process to be controlled can be used to separate less important output signals or even subsystems to obtain a less complex control system. Feedforward compensation techniques can be used to obtain such a more flexible control structure. Even when restricting the control system to a set of single-loop controllers, disturbances due to interaction in the MIMO process can be effectively rejected by using more powerful, e.g. time-optimal, strategies. It should be emphasized that time-optimal as well as decoupling or compensating techniques require the availability of a oood process model. It is interesting to incorporate them in a self-tuning or addptive control environment.
References DE KEYSER, R. (1980). Design and evaluation of self-tuning controllers. PhD. thesis, Automatic Control Laboratory, State
Benefits from Multivariable Control in the PRP-Industry
473
University of Ghent, Belgium (in Dutch) D1HULSTER F.M., R.M.C. DE KEYSER, A.R. VAN CAUltJErJBERGHE (1980). App 1i ca ti on of several parameter-adaptive controllers to a paper machine headbox. 4th IFAC PRP Conference. Ghent. LEBEAU B., R. ARRESE, S. BAUDUIN, R. GROBET, C. FOULARD (1980). Non interacting multivariable paper machine headbox control some comparisons with classical loops. 4th IFAC PRP Conference, Ghent. FB regulator
Setpoint
Ul
Ul
+
Process Process Process
Process Process +
Setpoi nt
U2
U2
+
Process
F8 reg ula tor
Y2
FF compensation
Fig. 1.
Feedforward from "intelnal disturbances".
t (s)
t (5) 1000
1500
Fig. 2. : Time-optimal comp.:lred to PI-control.
1000
1500
c.
474
Foulard
Time Optimal (single loop)
PI
(single loop)
x simple and robust
~l1ltivariable
x compensation of interac- x compensation of interaction tion x a priori knowledge about x still interaction, but of x easy to incorporate a very short duration so priori knowledge about the process does in genethat production loss is the process leading to a ral not lead to a more negligible more simple control simple control system structure
x interaction may result in x coupling between process production loss variables requires special attention to the stabil i ty problem
Table I
... (pseudo-multivariable)
l~ceJror ~r~ cu.~ _<~nsation
x feedforward from "internal" disturbances requires special attention to the stability problem
Strategies for the control of coupled
CONTRIBUTION by Dr. S. HEM (Reed International Consultants Ltd, Aylesford, Maidstone, Kent, UK) I shall forego the practice of pointing out the great importance of the subject of this round-table discussion in general terms. Instead I will describe briefly some of the multivariable control systems developed at RIC for use in Reed Mills as well as outside companies. Numerous multiple input and output control systems are designed on a single-loop basis and often the result is quite satisfactory. However, we have experienced considerable difficulty, on occasions, in obtaining a reasonably responding system using the single loop approach. It was only after proper consideration of the interactive nature of the system that a good result was achieved. The most obvious example of a multivariable control system in the paper industry is of course the interactive control of basis weight and moisture. This type of control is now so common on paper machines that very little that is new can be said about it. Nevertheless, we have been able to improve this control system's performance by the development of special multirate sampleddata control algorithms that allow greater stability at higher overall loop gains. The method has been extended to the general multivariable situation employing the stochastic Linear-Quadratic-Gaussian (LQG) procedure and also allows for the use of the Kalman filter for estimation of inaccessible state variahles. It should be noted, however, that multivariable control systems which have been designed using the LQG procedure can become unstable for a failure in one of the instruments. Special precautions must therefore by employed to guard against this situation.
MI~ln
complex control strategies (e.g. large dimensionality, reconstruction of ) state variables, x elegant solution of the stability problem ~
...
processes.
Another multivariable control system employed on some of our paper machines operates on cross machine direction profile data of certain paper properties. The control system produces a number of control actions within given constraints in a highly interactive system and performs an on-line iterative process of constrained optimization. In particular, systems for moisture profile control using plain presses, swimming rolls and Nipco rolls have been successfully implemented. The application of moisture profile control and profile management in general represent an economic and operational benefit at least as large as the one achieved with the machine direction control system. ~t is int:resting to note the resurgence of lnterest ln the design of headbox control systems. This problem is particularly interesting for me since I made a theoretical study of the dynamic behaviour of a headbox nearly 20 years ago. ~1ultivariable control theory has now developed to the point that the interaction problems can be understood and in turn makes it possible to design highly sophisticated and worthwhile control systems for this unit process of the paper machine.
Multivariable control systems have been designed using a variety of approaches, the LQG problems has been mentioned, inverse Nyquist array, characteristic loci techniques or pole assignment are other very successful methods, all with their own particular merits, too numerous to go into here. Finally, I would like to say that although we are moving towards systematic procedures and techniques for modelling and computer-aided controller design there will always be a need for the subjective element of creative engineering and know-how required for understanding the real-world aspect of the systems that must be controlled.
Benefits from Multivariable Control in the PRP-Industry
475
CONTRIBUTION by S. SUH (St. Regis Paper Co., West Nyack , USA)
As you may know, the paper industry is no stranger to the subject of multivariable control and has been exposed to this since early in 1960. The papermaking process is a prime example of a process with multiple interacting variables as shown in Figure 1. In addition to papermachine control, we at St. Regis have had considerable experience with multivariable control in other process areas such as Kamyr digester and utility plant, see Figures 2 and 3. We also recognize that there are other areas within a Kraft mill that could be easily considered for multiloop control. Table 1 shows such a 1i st.
MULTIVARIABLE CONTROL OF PAPERMACHINES
STOCK FLOW STEAM PRESSURE OR MACH INE SPEED
WEIGHT PAPERMAKI NG PROCESS
MOISTURE
MULTI VARIABLE CONTROL OF KAMYR DIGESTERS I have so far limited my opening remarks to the potential areas of application. Let me now move on to the benefit side of multivariable control which is the theme of this TEMPERATURE COOKING round-table session. The benefits we have KAPPA CHEMICALS PROCESS enjoyed at St. Regis through multi variable TIME control are substantial and readily acknowledged by the management. They include improved quality, fuel efficiency and production rate, and reduced raw material usage. However, this was attained at the added expense of manpower and capital equipment. For this MULTIVARIABLE OPTIMIZATION CONTROL OF reason, one should conduct an economic analysis of "cost vs. benefit before committing STEAM AND ELECTRICITY GENERATION oneself to multivariable control. It should also be mentioned here that it is sometimes possible to reduce a multi loop problem to a single-input/output problem and still reap ~ STEAM t--H_I_GH_.-.-, ELECTRICITY PROCESS STEA,& a good portion of the total benefit. PRESSURE GENERATION ERATION E GN STEAM AIR ~ ELECTRICITY I recognize that the advent of cheaper and faster computers, state-space method and advanced computer control strategies has PURCHASED enabled us to implement multi loop control more simply and inexpensively as compared to the 60 IS. However, it is getti ng di ffi cul t to recruit trained people who can work and provide adequate maintenance on the multiTable variable control systems installed in the plant level. The maintenance aspect of multivariOther areas of interest able control, which is skilled-labor intensive, should also be taken into consideration f recovery boilers when dealing with multivariable control. l power boilers The point I want to make here is that in our ~ lime kiln/causticizing plant industry, multivariable control is looked l evaporators at from the return-an-investment point of view. l refi ners t washers ll
c.
476
Foulard
CONTRIBUTION by B. LEBEAU (Centre Techniwue de 1 'Industrie des papier, cartons et celluloses, Grenobles, France) If we lookat the hea~box control system, it is well-known that we have a multivariable system with the following classical inputs:
4.
Does multivariable control give very significant economic returns compared with classical one?
5.
Is educational level sufficient for industrial people to use new control technique?
I think that the round table can give some answers to these different questions.
- air valve - fan pump speed and the following outputs
CONTRIBUTION by P. URONEN (Oulu University, Finland)
- 1eve 1 - total pressure. As you have seen earlier we have a strong interaction between these two loops. For example, if we change the fan pump speed, we do not only change the total head but we modify the level. So it is well-known that for the control of these two loops, we can have two following single variable control structures. Several people have proposed to decouple these two loops by means of decoupled compensators designed generally by using transfer function representation. Moreover, if we look on a paper machine, at modifications of the total head, we have disturbances on the basis weight. And in this case, it is necessary to change the stock valve in a coordinating way to keep constant substance flowrate at the slice opening. So, we increase the dimension of the control and we obtain a 3-inputs/3-outputs system. And finally, it is also well-known that we have a fourth input for the headbox which is the slice opening. This input also has a very strong interaction with the outputs. So, a good control system of the headbox must take into account the multivariable nature of this process and for example, we can imagine new multivariable control of the headbox. So question can be the following: 1. What kind of automatic control techniques
must we use to solve this problem? We have seen that for a 2-inputs/2-outputs system it is possible to use classical single variable techniques but with a 4-inputs/4-outputs system what kind of control must we use? 2. What kind of criterion must we use to the design of the control structure which takes into account industrial objectives? 3. What is the better control design which gives simple implementation on the process for industrial mill people?
It is a well-known fact that systematic methodology for the analysis and synthesis of multivariable control systems is available today. There also exist designing program packages, i.e., CAD for multivariable control systems based either on frequency-domain or time-domain methods. So the theory of multivariable control has reached a certain level of maturity. Another fact is that a complicated PRP process, for example, a recovery boiler or paper machine, is a system with many interacting cause-effect relations; thus the dynamic modelling and control of this kind of system is very difficult and basically it should be handled by stochastic and multivariable methods. It is also clear that in order to control this kind of processes, several sensors and actuators are necessary, and that singleloop approach can not give the best possible solution. So there exist the theoretical methods and practical needs for multivariable control in PRP industries, but why are these methods not applied to a larger extent? Obviously, there are definitEly benefits and advantages to be gained by application of multivariable and other so-called "mo dern" control methods in PRP industries; decoupling of interactions for example, in paper machine control systems, mass balance control at recovery boilers, level control of a Kamyr-digester etc., are just a few simple real applicatons of multivariable control in the pulp and paper industry. With multivariable control, the following advantages can be achieved : - better overall performance, i .a. closer control - compensation for interactions of single contro 1 loops, strong non-l i neari ti es, long time-delays etc. - better control of batch type and sequencing - control of not directly measured variables - control of processes which are very difficult to handle with SISO-control strategy - optimization and coordination in larger meaning - fault detection, cross checking and monitoring.
Benefits from Multivariable Control in the
The barriers and restrictions for the use of these methods are : -these methods are more sophisticated, more complex and more difficult to learn and use - the direct economic benefits from using them are difficult to show a priori. - shortage of skilled people in industry - updating and maintaining of the system is more difficult - more complex back-up system and difficulties in manual operations - reluctance against new ideas. I feel that together with distributed digital control systems, these methods will be increasingly more applied and used; we must start from simple and reliable applications with an evolutionary process towards more sophisticated systems. It is a typical diffusional process of new technology and innovation. by T. TAKAr1ATSU and I. HASHIMOTO (Kyoto University, Dept. of Chemical Eng., Japan)
CONTRIBUTIOI~
In recent years, the rapid increase in raw material and energy costs, especially the enormous increase of energy cost since the oil embargo of 1973, has caused a great change in the philosophy of process system design and forced process designers to utilize energy and other natural resources more efficiently. Intermediate storage tanks are made smaller or eliminated completely to reduce inventory and capital cost. This, however, results in more sophisticated and highly coupled process systems. Design margins are also reduced for improving the static design performance, which causes loss of flexibility and operability in the process. In order to minimize the energy consumption, heat integration is throughly performed by connecting heat producing units the heat of which was previously wasted, and heat consuming units which were previously uncoupled. Energy integration can save a lot of energy from the static viewpoint. But from the dynamic viewpoint, it makes the process a highly interconnected multivariable system which loses flexibility, operability and controlability. These lost favorable dynamic atributes have to be recovered by implementing the good quality control. This offers strong impetus for creating new advance in multivariable control techniques. Here, we choose a heat integrated distiTIation column system as an example for further discussion. Distillation is most widely used for the separation of products in refineries or many chemical processes. It is a typical energy consuming process, and it presents as a prime candidate for the improvement in energy efficiency.
~KP-Industry
477
Significant efforts have been devoted to the application of modern multivariable control techniques such as decoupling control, disturbance rejection control, the inverse Nyquist array method, and so forth, to a binary distillation column control, that is to control the compositions of both the top and bottom streams simultaneously. The separation of multi-component mixtures requires a network of distillation columns which includes material inteqration. This system provides a very challenging dynamic problem to which multivariable control theories have to be intensively applied. Let's note the prominent problems in designing the control system of a highly interacted multivariable system. 1. How to build dynamic model with the necessary and sufficient accurQCY based on which the control system design can be dealt with: The selection of the model is most crucial in applying any multivariable control technique. The precise physical model is generally too complex to be used for designing control systems. Simple "black box" models between input and outputs are, therefore, favored, but they often lead to poor control po1i ci es . 2.
How to design control systems based on a limited number of measuring points: It is very common in actual chemical processes that only some of the state variables are available for utilization. Even when it is possible to feedback all of the state variables, it is not unusual to utilize only some of them due to various restrictions or for economic reasons.
3.
How to select the best palrlng between manipulating and output variables; The relative gain array method selects favorable pairing. But this pairing can often lead to poor control, as pointed out by several researchers. More intensive research efforts will be necessary to resolve this problem.
4.
Rescrutinization of multivariable control techniques and their extension in order to obtain more robust control : Every technique has some advantages and disadvantages. A definitive method which can be applied to any multivariable process control and lead to robust control with high quality, has not been developed yet. By applying multivariable control techniques to more realistic problems, the applicability of each technique has to be rescrutinized and at the same time the extension of the theory has to be done, if necessary.
c.
478
5.
Foulard
Trade-off between high-quality control system design and optimal steady-state design: The high energy cost provides a very strong incentive to practice the heat integration which makes the process system more complex and more dynamically interacting. The more complex and interacting the system becomes, the more difficult it is to design the optimal system from both the steady state and dynamic points of view. The problem is how to reconcile the steady-state design with the control system design. In other words, a method has to be developed to incorporate the control structure design as part of the steady state design optimization. The research relating this topic has just started and much effort has to be devoted to resolve this problem.
As discussed in the above, modern multivariable control techniques seem to offer considerable promise in improving the quality of process control, but it has not yet been widely implemented on industrial scale processes. It is not easy to assess to costs of engineering manpower, control hardware and instrumentations against the potential benefits of modern multivariable control techniques in the industrial field. The control system which can secure the optimal operation of a large complex process system, must have a decentralized structure. The control system of each subsystem would be designed by a multivariable control technique by taking into account their existing strong interactions and using ~icro- and minicomputers. The coordination between subsystems could be performed by a large computer so as to optimize the overall objective. There have been few applications of hierarchy or decentralized control in real industrial processes. We must endeavour to exploit advances in this direction in order to provide a far better control system for the real complex and large chemical system. DISCUSSION
are very large, it is possible to break them down into smaller multivariable subsystems. This is a very i~portant part when you have very large systems. On the second point about safety and on computer frontiers that is something we met quite a lot in the paper industry. What do you do when a pump fails? Perhaps you should consider the control systems as part of the system on the same basis that you could consider the other hardware. If it is down the plant is down. But I would say that probably computers are more reliable than pumps and all the other hardware together. ~Q~~~~~
(by Prof. Brandenburg, Technical University Munich, FRGermany) I would like to give short information concerning the electrical drives in continuous plants of PRP-industries. They are coupled by the moving web in such a way that a multivariable system results. Therefore control synthesis must be carried out with the methods of the multivariable theory. However in many practical circumstances simplifications are possible due to fast speed control loops, by which the drives are approximately decoupled. The dynamical behaviour becomes very poor, if certain conditions are not fulfilled by the mechanical system. Then state controllers with observers are favorable. The problem we are working on is to get state controllers which are simple enough to be applied in industrial plants. ~Q~~~Q~
(by Prof. Uronen) I fully agree with those points made by the representative of the chemical industry. These special safety questions may be there even much more critical than in the pulp and paper industry because there are hazardous areas and explosive materials handled in destillation columns, etc. Potentially there are big production losses in pulp and paper industry but in addition of that in systems of petrochemical industries there are other risks. I fully agree that these multivariable systems need computer systems with very good redundancy and back-up systems for difficult and hazardous situations. This is really one of the biggest questions here. g~~~!iQ~
g~~~!iQ~
(by C. Van Rijn, Koninklijke Shell Laboratorium, the Netherlands) In the petrochemical industry the scope for multivariable control is clearly visible. In this respect, I like to make two points - the identification of MIMO systems of appreciable dimensions requires considerable effort (MIMO identification) - the aspect of loosing the computer faces the operator with different backup situations. May I have the opinion of the panel on these?
(by Prof. J. Van De Vegte, University of Toronto, Canada) Multivariable control techniques are only now gradually being simplified sufficiently that those who have to implement them are willing to consider them. State estimation or Kalman filtering are in general too complex for practical systems. Output feedback and approximate noninteraction provide promise to sufficiently simple control schemes, with dynamic compensation restricted to fixed low order.
~Q~~~~
~Q~~~~!
(by dr. Hem) I think, on the first aspect where systems
(by dr. De Keyser) I fully agree with the comment of Prof.
Benefits from Multivariable Control in the PRP-Industry
Van de Vegte. Indeed we should not use multivariable control if we can do it with singleloop controllers. This is what I said before. Also decoupling control, I should not use it in practice, perhaps approximate decoupling. However we should not stay with the singleloop controls with PI-regulators as we do now. We should go something in between let us say we should use single-loop controllers with more complicated control algorithms, with more powerful algorithms instead of PI. That will be already much better than the methods we are using now. But I agree we should not go to multivariable control except for some special processes as perhaps electrical drives and some other processes but we should not generalize. ~Q~~~0~
(by dr. Hem) It is obviously complicated but using advanced technology ought to be unobtrusive and really no one is questioning what goes no inside a television set in order to use it. The same thing could be said for control systems, provided they are made easy to operate for the operators and that is obviously a question of design. The fact that one may use advanced technology to operate the system should not ~o away with that. As you perhaps suggest we lmplemented several multivariable control systems including Kalman filters. There is no need to tell the operator that there is a Kalman filter in the loop. He would not know what you talk about, but it is perfectly possible to incorporate advanced technology without having to tell everybody details. ~Q~~~Q~
(by Prof. C. Foulard) I would like to give a little comment about this intervention. First I think that what w~s complicated in the beginning can be very slmple after some years, e.g. computers were complicated 20 years ago and a laser e.g. also was complicated and now we have lasers for printing. So this can change. I want also to comment on the quasi-decoup1ing system, the output feedback, mu1tivariable control, etc. There are answers to all these questions now and e.g. the structures which were presented by dr. Lebeau in his paper on headbox control solve these proble~s. I think not all the problems, but this one.
479
~Q~~~Q~
(by dr. Olsen) I just like to make some comments on this question. You design a control system either monovariable or multivariable to suit your specifications. If the specifications are operator acceptance you design the control system to achieve operator acceptance e.g. instead of going directly for DDC-control ~ou use a man in the loop with graphical dlsplays to guide his operation of the plant e.g. If the algorithm within the computer, guiding the operator, is based on multivariable control theory, or if it is based on monovariable control theory, does not affect the operator as long as he designs a system w~ich is guiding him in a way acceptable to hlm. The same way in cooperation with the management, to get user acceptance of the system you agree on the specifications of the system. Decoupling is one way to go if decoupling is important to you. But you pay for decoupling e.g. by large variations in your control inputs. If you choose decoupling, you might see motors going very often full speed or lower speed. If you like to have few and small variations in your process control variables, then you might ca~l f?r linear quadratic feedback with heavy welghtlng of your control variations. These are only different ways of designing control systems. To sum up I would just say that I he~v~ly disagree with some of the very heavy 0plnlons for one or the other method. I think that you should choose the method for the application. ~Q~~~~~
(by dr. Lebeau) I just want to make a remark on some of the comments which have been made and particularly this one of somebody, saying when a contr?l scheme stops then the plant stops. That lS not acceptable for an industrial operation. The same person said we have not to know what is in a television set. You have not to know it when you have not to support a stop of this television set, but if there is something wrong, then we can accept to have a quarter of the video-screen but we cannot accept to have nothing. I think this remark explains why it is very difficult to implement multivariable control actually. It is because it is not at the state of maturity as you have heard before. It is not clear enough to be understood by people as we have in the mills and it is those people who take back a process when something fails. That is why we refuse it.