- Email: [email protected]
On the Role of the General Control Problem in Control Engineering Education
Copyright © IFAC Advances in Control Education, Istanbul, Turkey, 1997
ON THE ROLE OF THE GENERAL CONTROL PROBLEM IN CONTROL ENGINEERING EDUCA nON
lan K Craig
Department of Electrical and Electronic Engineering. University of Pretoria, Pretoria, 0002, South Africa. [email protected]
Abstract: This paper describes t.wo cont.rol courses at the University of Pretoria., one at undergraduate and the other at graduate level. which are structured "ithin the General Control Problem framework. This framework includes all the steps that are required to establish a controller for a physical plant, from needs definition to controller evaluation. These courses are aimed at improving the quality and preparedness of graduates. A graduate who understands this framework is more likely to succeed in coDtrolling a physical system than a graduate who has a detailed knowledge of advanced coDtrol techniques, but does not know how or where this knowledge might be useful. Copyright © 1998 IFAC
Keywords: control education, industrial control. automatic control systems, control applications
Control education at the University of Pretoria and elsewhere, have traditionally focused on only a part of the GCP. i.c. on acti,ities which occur within the mathematical world. In tiils mathematical world. the student is presented ""ith a linear time invariant plant model which is used for simulation and controller design purposes. Students usually only haye a yague idea ",'here this model comes from and what its limitations arc. Controllers arc then designed for these idealized models, and "implemented" in CACE software or perhaps in a laboratory setting.
l. ThiTRODUCTION
Control engineering education has come a long way since control courses first appeared in electrical engineering curricula in the 1940s. Besides the many theoretical control courses on offer today, tile quality and preparedness of graduates are increasingly being enhanced by real-world laborator:v experiments and the use of sophisticated C.ACE (computer-aided control engineering) software. (Sec e.g. Kheir er al. (1996) for a comprehensive rc,iew of the status of control engineering education around the world.) This paper suggests ways in which the quality and preparedness of graduates can perhaps be improved even further.
Most students are therefore left \\ith tile impression that the activities performed in the mathematical world is what control engineering is all about. ~. This is of course only partially true. In maIlY real word applications, more tilan 90% of the effort (in terms of time and money) in establishing a control system lies outside this "\vorld. It is therefore not surprising that when graduates get to industry they are often disappointed to find that most of the ad ...anced controller design techniques they studied at university are almost neyer used ill practice. as control practitioners are too busy addressing the other 90% of the GCP.
Various processes are involved in establishing a control system for a physical plant, from defining tile needs that the control system has to satisfy. to tile tinal functional and economic evaluation of the controller after implementation. In this paper these processes are collectively referred to as the general control problem (GCP).
201
The control curriculum in the Department of Electrical and Electronic Engineering at the University of Pretoria was recentiy amended to include two courses which specifically sensitize students to the other 90% of tile GCP. These courses are discussed in more detail in this paper. For a complete description of tile control engineering curriculum at the Uni\'ersity of Pretoria, see Braae et al. (1996) .
• 2. THE GENERAL cO!',rrn.OL PROBLEM Various processes are invoh'ed in establishing a control system for a physical plant. These processes make many demands on tile control engineer and the engineering team, and often manifest themselves in a step by step design procedure. Skogestad and Postletwhaite (1996) divide tillS procedure into 14 steps, step 1 being "Study the system (plant) to be controlled and obtain initial information about control objectives", and step 14 "Test and validate the control system, and tune the controller on-line, if necessary". They go on to state that "Control courses and text books usually focus on steps 9 and 10 ... methods for controller design and control system analysis. Interestingly, many real control systems are designed 'without any consideration of these two steps."
•
A pictorial representation of the general control problem (GCP) is given in Figure l. This figure shows two worlds, a real world and a mathematical world. In the real world, plants are ill defined and often difficult to describe, hence the rugged boundaries for this world. In contrast tile matllematical world is generally well defined. hence tile circular boundary. Real World
•
Mathematical World
•
Oblala _ _ tieal model
Fig. 1. The general control problem.
- determine operational requirement that must be satisfied by the controller; - determine the role of tile operator, before and after the implementation of the controller; - determine which variables to measure and which to manipulate; - determine the control configuration and control law, - detennine whether requirements can be satisfied given budgetary constraints. Obtain an adequate mathematical model for the plant This step is required in the GCP to go from the real world to the mathematical world. This model has to be as simple as possible, but no simpler, such that the operational requirements can still be met. The process used to obtain a model would depend e.g. on whether the plant exists or is still being designed, and whether the plant dynamics are well understood. Generally, both modeling and system identification will be used. Controller design and control system analysis These tasks take place in the mathematical world. The plant model is used to design a controller according to the specifications determined in the plant analysis phase. It is critical to understand the limitations of the plant model (i.e. how different is it from the plant in the real world?) and the impact that these limitations have on the controller design and analysis. Controller implementation This step is required in the GCP to go from the mathematical world to the real world. Once the designer is satisfied that the controller works well in the mathematical world. and that it has a good chance of meeting the required specifications in the real world as well. tile controller can be implemented. Controller evaluation Once implemented in the real world. tile controller needs to be evaluated. bOtil functionally and economically. Does tile closed-loop system in the real world meet tile required specifications? Was tile controller operationalized \\;tltin time and \\ithin budget? What is the net present value of the project (if applicable)?
The GCP concept which can be applied to many problems and processes, provides a useful framework \\ithin which to structure contr01 engineering courses. People active in industr}· who are well versed in this framework and automatic control concepts (from management to operator). \\ill be able 10 pro\ide innoyative solutions to practical problems.
In tilC real world there is a real plant which has to bc controlled according to certain specifications within a limited budget (time and money). In order to do tilis. the controller designer has to do tile follo\\ing : • Analyze the plant for control purposes - obtain process knowledge: - classify process variables:
202
identification framework, as described in Chapter 1 of Ljung, and are given the theoretical framework used for calculating ARX models. These concepts are then enforced through laboratory and computer exercises. In the computer exercises, students are given real plant data, for which they have to obtain and evaluate ARX models using the MatIab System Identification Toolbox (Ljung, 1991). In the laboratory, they have to obtain transfer function models for a windtunnel-hairdryer system, similar to that described in Kheir et al. (1996). In this experiment step. sine and PRBS signals are used as plant inputs.
3. COl\TTROL COlJRSES AND TIlE GCP FRAMEWORK There are two control courses at the University of Pretoria dealing specifically \\ith the GCP, one at final year undergraduate level. and the other at graduate level. 3.1 Undergraduate course A second semester final year course, which is at present compulsory for all electronic engineering students in the Depanment of Electrical and Electronic Engineering, is structured around the GCP. The course consists of the following modules : • an introduction to the GCP; • system analysis for control purposes; • obtaining a mathematical model for control yia modeling and system identification: • the design and simulation of a controller and the choice of a suitable design technique; • the implementation of a control system using a computer: • the functional and economic evaluation of a control system: • guest lectures by control practitioners from industry.
The course is concluded with guest lectures by control practJtJoners from e.g. the mineral processing and aeronautical industries. By this time, because of the entrenchment of the GCP. students are in a position to appreciate what the guest lectures are talking about even though they cover vastly different applications. These lectures therefore serve to enforce what has been taught throughout the course. 3.2 Graduate course The aim of this course is to further entrench the GCP framework and at the same time, to act as a vehicle for developing case studies on industrial control applications in South Africa. In doing so. the approach folJO\ved in this course is somewhat unconventional. Students are given a specific industrial process to focus on.. and a major part of the course is devoted to studying and understanding the process. The mono which applies is that "if you do not understand the process, you do not have a control problem".
The prerequisites for this particular course is a standard introductory control course in which a tex1 such as Kou (1995) is used. A knowledge of the description of systems in discrete time and statespace is required. The basic concepts of the GCP can and should however be introduced even in an introductory control course. The GCP framework is enforced via repetition and various examples. Students are tested on the GCP in each of the two semester tests as well as in the exanl. This is done for the follo\\ing reason. A graduate who is already familiar ",ith the GCP framework. is more likely to succeed in controlling some physical system, than a graduate who has a detailed knowledge of advanced control techniques. but does not know how or where this knowledge might be useful.
Some of the criteria for selecting an industrial process are: • the availability of local ex-perts from industry: • the existence of literature on the control of tillS process: • the willingness of a faculty member, who is an ex-pert on the process. to panicipate in the course. The course consists of the follo\\ing modules: • the GCP; • analysis of the process to be controlled; • current industrial practice for the control of tile chosen process: • control technique chosen for sohing the control problem; • plant practice as applied to the process (how is it normally operated?): • the implementation of the proposed technique on a control and simulation platform; • writing a case study on tile ··application".
Most of the modules covered in this course could easily justify a graduate level course on the subject. For example. system identification is usually taught at graduate level using a tex1 such as Ljung (1987). The problem then arises as to what should be taught in a more general course. In this undergraduate course at the University of Pretoria. system identification. together ~ith modeling. is covered as a means of obtaining a mathematical model for a physical process to be controlled. This step is required in the GCP to go from the real world to the mathematical world. Students are taught the system
203
(1996). Control systems engineering education. Automatica. 32, 147-166. Kou, B. C. (1995). A utomatic Control Systems. 7th ed. Prentice-Hall, Englewood Cliffs, NJ. Ljung, L (1987). System Identification: Theory for the user, Prentice-Hall, Englewood Cliffs, NJ. Ljung, L (1991). System Identification Toolbox User's Guide, The MathWorks, Natick, MA. Skogestad, S. and L Postletwhaite (1996). lo.1ultivariable Feedback Control. AnalysiS and design , Wile)', New York.
Students are expected to become yery familiar \\ith the process under study. In addition to lectures on the operation of the chosen process and the students' own literature survey, students are required to spend time on a chosen plant in which the process is used. Here they study typical operating procedures, and gain practical knowledge by interviewing operators and technical personnel. Each student is expected to pose tlle control problem in the form of a case study. and to fonnally present the problem fonnulation to a panel of ex-perts. The case study and presentation fonn the basis of evaluation for the course. Based on a literature survey and the feedback from the panel, students suggest possible solutions to the control problem. Students are also expected to quantitatively test their proposed solutions on a Control and Simulation Platform, which simulates tlle implementation emironment found on many plants.
4. CONCLUSIONS A General Control Problem framework. which includes all the steps that are required to establish a controller for a physical plant has been described. An undergraduate and graduate control course at the University of Pretoria which are structured \\ithin this frame·work. were discussed. These courses are aimed at improving the quality and preparedness of graduates. Graduates who understand this framework will in practice be advantaged over graduates who have a detailed knowledge of ad,·anced control techniques. but do not know how or where this knowledge might be applied.
5. ACKNOWLEDGEMENTS The financial support of ASA Systems Automation (Pty) Ltd .. tlle Process Control Systems Division of AMS (Pty) Ltd .. the South African Department of Trnde and IndustI). and the South African Foundation for Research Development, is gratefully acknowledged.
REFERENCES Braac. M.. E. Boje. IX. Craig, P.L De Vaal. J. Gouws. S.G. Mclaren, C.LE. Swartz. c.P.Ungercr. J.L van Niekerk. and B. Wigdoro\\itz (1996). Special Issue on Control Education : South Africa. IEEE Control .~l'stems }.1agazine. ]6.41-47. Kheir. N.A.. KJ . Astrom, D. Auslander. K.c. Cheok. G.F. Franklin. M. Masten.. and M. Rabins
204
ON THE ROLE OF THE GENERAL CONTROL PROBLEM IN CONTROL ENGINEERING EDUCA nON
lan K Craig
Department of Electrical and Electronic Engineering. University of Pretoria, Pretoria, 0002, South Africa. [email protected]
Abstract: This paper describes t.wo cont.rol courses at the University of Pretoria., one at undergraduate and the other at graduate level. which are structured "ithin the General Control Problem framework. This framework includes all the steps that are required to establish a controller for a physical plant, from needs definition to controller evaluation. These courses are aimed at improving the quality and preparedness of graduates. A graduate who understands this framework is more likely to succeed in coDtrolling a physical system than a graduate who has a detailed knowledge of advanced coDtrol techniques, but does not know how or where this knowledge might be useful. Copyright © 1998 IFAC
Keywords: control education, industrial control. automatic control systems, control applications
Control education at the University of Pretoria and elsewhere, have traditionally focused on only a part of the GCP. i.c. on acti,ities which occur within the mathematical world. In tiils mathematical world. the student is presented ""ith a linear time invariant plant model which is used for simulation and controller design purposes. Students usually only haye a yague idea ",'here this model comes from and what its limitations arc. Controllers arc then designed for these idealized models, and "implemented" in CACE software or perhaps in a laboratory setting.
l. ThiTRODUCTION
Control engineering education has come a long way since control courses first appeared in electrical engineering curricula in the 1940s. Besides the many theoretical control courses on offer today, tile quality and preparedness of graduates are increasingly being enhanced by real-world laborator:v experiments and the use of sophisticated C.ACE (computer-aided control engineering) software. (Sec e.g. Kheir er al. (1996) for a comprehensive rc,iew of the status of control engineering education around the world.) This paper suggests ways in which the quality and preparedness of graduates can perhaps be improved even further.
Most students are therefore left \\ith tile impression that the activities performed in the mathematical world is what control engineering is all about. ~. This is of course only partially true. In maIlY real word applications, more tilan 90% of the effort (in terms of time and money) in establishing a control system lies outside this "\vorld. It is therefore not surprising that when graduates get to industry they are often disappointed to find that most of the ad ...anced controller design techniques they studied at university are almost neyer used ill practice. as control practitioners are too busy addressing the other 90% of the GCP.
Various processes are involved in establishing a control system for a physical plant, from defining tile needs that the control system has to satisfy. to tile tinal functional and economic evaluation of the controller after implementation. In this paper these processes are collectively referred to as the general control problem (GCP).
201
The control curriculum in the Department of Electrical and Electronic Engineering at the University of Pretoria was recentiy amended to include two courses which specifically sensitize students to the other 90% of tile GCP. These courses are discussed in more detail in this paper. For a complete description of tile control engineering curriculum at the Uni\'ersity of Pretoria, see Braae et al. (1996) .
• 2. THE GENERAL cO!',rrn.OL PROBLEM Various processes are invoh'ed in establishing a control system for a physical plant. These processes make many demands on tile control engineer and the engineering team, and often manifest themselves in a step by step design procedure. Skogestad and Postletwhaite (1996) divide tillS procedure into 14 steps, step 1 being "Study the system (plant) to be controlled and obtain initial information about control objectives", and step 14 "Test and validate the control system, and tune the controller on-line, if necessary". They go on to state that "Control courses and text books usually focus on steps 9 and 10 ... methods for controller design and control system analysis. Interestingly, many real control systems are designed 'without any consideration of these two steps."
•
A pictorial representation of the general control problem (GCP) is given in Figure l. This figure shows two worlds, a real world and a mathematical world. In the real world, plants are ill defined and often difficult to describe, hence the rugged boundaries for this world. In contrast tile matllematical world is generally well defined. hence tile circular boundary. Real World
•
Mathematical World
•
Oblala _ _ tieal model
Fig. 1. The general control problem.
- determine operational requirement that must be satisfied by the controller; - determine the role of tile operator, before and after the implementation of the controller; - determine which variables to measure and which to manipulate; - determine the control configuration and control law, - detennine whether requirements can be satisfied given budgetary constraints. Obtain an adequate mathematical model for the plant This step is required in the GCP to go from the real world to the mathematical world. This model has to be as simple as possible, but no simpler, such that the operational requirements can still be met. The process used to obtain a model would depend e.g. on whether the plant exists or is still being designed, and whether the plant dynamics are well understood. Generally, both modeling and system identification will be used. Controller design and control system analysis These tasks take place in the mathematical world. The plant model is used to design a controller according to the specifications determined in the plant analysis phase. It is critical to understand the limitations of the plant model (i.e. how different is it from the plant in the real world?) and the impact that these limitations have on the controller design and analysis. Controller implementation This step is required in the GCP to go from the mathematical world to the real world. Once the designer is satisfied that the controller works well in the mathematical world. and that it has a good chance of meeting the required specifications in the real world as well. tile controller can be implemented. Controller evaluation Once implemented in the real world. tile controller needs to be evaluated. bOtil functionally and economically. Does tile closed-loop system in the real world meet tile required specifications? Was tile controller operationalized \\;tltin time and \\ithin budget? What is the net present value of the project (if applicable)?
The GCP concept which can be applied to many problems and processes, provides a useful framework \\ithin which to structure contr01 engineering courses. People active in industr}· who are well versed in this framework and automatic control concepts (from management to operator). \\ill be able 10 pro\ide innoyative solutions to practical problems.
In tilC real world there is a real plant which has to bc controlled according to certain specifications within a limited budget (time and money). In order to do tilis. the controller designer has to do tile follo\\ing : • Analyze the plant for control purposes - obtain process knowledge: - classify process variables:
202
identification framework, as described in Chapter 1 of Ljung, and are given the theoretical framework used for calculating ARX models. These concepts are then enforced through laboratory and computer exercises. In the computer exercises, students are given real plant data, for which they have to obtain and evaluate ARX models using the MatIab System Identification Toolbox (Ljung, 1991). In the laboratory, they have to obtain transfer function models for a windtunnel-hairdryer system, similar to that described in Kheir et al. (1996). In this experiment step. sine and PRBS signals are used as plant inputs.
3. COl\TTROL COlJRSES AND TIlE GCP FRAMEWORK There are two control courses at the University of Pretoria dealing specifically \\ith the GCP, one at final year undergraduate level. and the other at graduate level. 3.1 Undergraduate course A second semester final year course, which is at present compulsory for all electronic engineering students in the Depanment of Electrical and Electronic Engineering, is structured around the GCP. The course consists of the following modules : • an introduction to the GCP; • system analysis for control purposes; • obtaining a mathematical model for control yia modeling and system identification: • the design and simulation of a controller and the choice of a suitable design technique; • the implementation of a control system using a computer: • the functional and economic evaluation of a control system: • guest lectures by control practitioners from industry.
The course is concluded with guest lectures by control practJtJoners from e.g. the mineral processing and aeronautical industries. By this time, because of the entrenchment of the GCP. students are in a position to appreciate what the guest lectures are talking about even though they cover vastly different applications. These lectures therefore serve to enforce what has been taught throughout the course. 3.2 Graduate course The aim of this course is to further entrench the GCP framework and at the same time, to act as a vehicle for developing case studies on industrial control applications in South Africa. In doing so. the approach folJO\ved in this course is somewhat unconventional. Students are given a specific industrial process to focus on.. and a major part of the course is devoted to studying and understanding the process. The mono which applies is that "if you do not understand the process, you do not have a control problem".
The prerequisites for this particular course is a standard introductory control course in which a tex1 such as Kou (1995) is used. A knowledge of the description of systems in discrete time and statespace is required. The basic concepts of the GCP can and should however be introduced even in an introductory control course. The GCP framework is enforced via repetition and various examples. Students are tested on the GCP in each of the two semester tests as well as in the exanl. This is done for the follo\\ing reason. A graduate who is already familiar ",ith the GCP framework. is more likely to succeed in controlling some physical system, than a graduate who has a detailed knowledge of advanced control techniques. but does not know how or where this knowledge might be useful.
Some of the criteria for selecting an industrial process are: • the availability of local ex-perts from industry: • the existence of literature on the control of tillS process: • the willingness of a faculty member, who is an ex-pert on the process. to panicipate in the course. The course consists of the follo\\ing modules: • the GCP; • analysis of the process to be controlled; • current industrial practice for the control of tile chosen process: • control technique chosen for sohing the control problem; • plant practice as applied to the process (how is it normally operated?): • the implementation of the proposed technique on a control and simulation platform; • writing a case study on tile ··application".
Most of the modules covered in this course could easily justify a graduate level course on the subject. For example. system identification is usually taught at graduate level using a tex1 such as Ljung (1987). The problem then arises as to what should be taught in a more general course. In this undergraduate course at the University of Pretoria. system identification. together ~ith modeling. is covered as a means of obtaining a mathematical model for a physical process to be controlled. This step is required in the GCP to go from the real world to the mathematical world. Students are taught the system
203
(1996). Control systems engineering education. Automatica. 32, 147-166. Kou, B. C. (1995). A utomatic Control Systems. 7th ed. Prentice-Hall, Englewood Cliffs, NJ. Ljung, L (1987). System Identification: Theory for the user, Prentice-Hall, Englewood Cliffs, NJ. Ljung, L (1991). System Identification Toolbox User's Guide, The MathWorks, Natick, MA. Skogestad, S. and L Postletwhaite (1996). lo.1ultivariable Feedback Control. AnalysiS and design , Wile)', New York.
Students are expected to become yery familiar \\ith the process under study. In addition to lectures on the operation of the chosen process and the students' own literature survey, students are required to spend time on a chosen plant in which the process is used. Here they study typical operating procedures, and gain practical knowledge by interviewing operators and technical personnel. Each student is expected to pose tlle control problem in the form of a case study. and to fonnally present the problem fonnulation to a panel of ex-perts. The case study and presentation fonn the basis of evaluation for the course. Based on a literature survey and the feedback from the panel, students suggest possible solutions to the control problem. Students are also expected to quantitatively test their proposed solutions on a Control and Simulation Platform, which simulates tlle implementation emironment found on many plants.
4. CONCLUSIONS A General Control Problem framework. which includes all the steps that are required to establish a controller for a physical plant has been described. An undergraduate and graduate control course at the University of Pretoria which are structured \\ithin this frame·work. were discussed. These courses are aimed at improving the quality and preparedness of graduates. Graduates who understand this framework will in practice be advantaged over graduates who have a detailed knowledge of ad,·anced control techniques. but do not know how or where this knowledge might be applied.
5. ACKNOWLEDGEMENTS The financial support of ASA Systems Automation (Pty) Ltd .. tlle Process Control Systems Division of AMS (Pty) Ltd .. the South African Department of Trnde and IndustI). and the South African Foundation for Research Development, is gratefully acknowledged.
REFERENCES Braac. M.. E. Boje. IX. Craig, P.L De Vaal. J. Gouws. S.G. Mclaren, C.LE. Swartz. c.P.Ungercr. J.L van Niekerk. and B. Wigdoro\\itz (1996). Special Issue on Control Education : South Africa. IEEE Control .~l'stems }.1agazine. ]6.41-47. Kheir. N.A.. KJ . Astrom, D. Auslander. K.c. Cheok. G.F. Franklin. M. Masten.. and M. Rabins
204