- Email: [email protected]
Self-tuning systems control and signal processing
Book Reviews
References Daniel, R. W. (1990). Chaos in high performance digital robot force controllers. Proc. of 1990 IEEE Conf. on Decision and Control, Honolulu, HI, 5-7 December 1990, pp. 1951-1953. About the reviewer Dr Daniel is currently a university lecturer in Robotics at Oxford University, U.K. He received his B.Sc. in Electrical
1287
Engineering from Brunel University, London in 1978 and his Ph.D. from Cambridge University, U.K. in 1982. His main interests lie in the design of high performance servo mechanisms with applications to robotics. Much of his work is aimed at remote operations by tele-manipulators and is currently working on high fidelity force reflection systems. Dr Daniel is Co-Editor-in-Chief of Mechatronics, published by Pergamon Press.
Self-tuning Systems Control and Signal Processing* R. E. Wellstead Reviewer: SEIICHI SHIN Department of Mathematical Engineering and Information Physics, University of Tokyo, Hongo, Bunkyo-ku, Tokyo 113, Japan. AN EXCITING BOOK has been published for users and researchers in the fields of control and signal processing. This is a jewelry box of know-how of the self-tuning systems in the practice. This book is written based on the results done over a 15 year period in Control System Center, University of Manchester, Institute of Science and Technology, U.K. As is well known, "tuning" is the main work of practical application of control and signal processing, though the modern theory of these fields offers "design", which requires the mathematical fundamentals. The controller is obtained with the theory as a solution of mathematical equations, which are derived from strict mathematical descriptions of the plant and the control objective. However, there are always modeling errors between the true plant and its mathematical model and it is really difficult to specify the control objective as mathematical equations. These two problems annoy users every day and symbolize a gap between the practice and the theory of control. Tuning is the users' solution of the problems. The tuning absorbs the modeling error and, after many trials of tuning, the users find a trade-off point among control quality, cost, and so on. An auto-tuning system is a dream of the user, because its saves engineers' labour and achieves high quality control/signal processing when mismatch of controller/signal processor is a main drawback of the control/signal processing system. This book is written about the self-tuning systems from a user's viewpoint by two famous self-tuning system researchers, so that users can get the overview of the self-tuning systems easily. Moreover, researchers may be stimulated as the book points out what is necessary for the practical use of the self-tuning systems. Though the combination of control and signal processing in this volume is heavy for readers (even for me as a reviewer) we can take a general view of the self-tuning technique. Specifically the self-tuning technique used in signal processing shows us a future direction of the self-tuning control research, since the control is more complicated than the signal processing so that the self-tuning technique is advanced in the signal processing. The book consists of four major parts. The aim of the book and the mutual relations of the four parts are described in the Preface. The first chapter, Introduction, presents the basic concept used in the four parts, that is, control, signal processing, self-tuning, and adaptation. A comparison is also * Self-tuning Systems Control and Signal Processing by P. E. Weflstead and M. B. Zarrop. John Wiley (1991). ISBN 0-471-92883-6, ISBN 0-471-93054-7 (paperback).
and M. B, Zarrop given between the self-tuning systems and their rivals. Several simple examples given here are the practical applications of the self-tuning systems, including the engine control, the control of a lime kiln, adaptive echo cancelling, and self-tuning vibration simulation. Readers can, without any mathematical background, get valuable information from this chapter. Part 1 is "System Identification for Self-tuning," where the basis of models of system and signal, and recursive estimation techniques are described, including initial parameter and covariance settings, test signals, forgetting factor, and computational alternatives for recursive estimation. Some simple PASCAL programme of recursive estimation algorithms are also attached here for the understanding and the ease of the practical application. This part also includes the Ordinary Differential Equation (ODE) approach, which is a useful tool for analysing the self-tuning systems. However, it is much too complicated for practical users. In the book, an essence of ODE is extracted and engineers can determine how to use ODE to analyse their own self-tuning system. Part 2 is devoted to "Self-tuning Controllers". First, the pole assignment controller is described based on the transfer function approach, where the Diophantine equation plays a key role. Adding to the pole assignment, minimum variance control and multistage predictive control are explained in detail. These control methods are very popular and effective in practical applications. Moreover, controls of a superheater attemperator, an active suspension system, and a robot, are illustrated here as applications of the self-tuning systems. Part 3 treats "Self-tuning Signal Processing", where prediction and filtering with the self-tuning ability are discussed. This part includes the self-tuning systems of prediction of hot metal quality in a blast furnace, interference cancelling filters, and sensor signal conditioning. Part 4 is a collection of special topics pertaining to self-tuning systems. This part consists of individual three hot topics: two-dimensional self-tuning algorithms, self-tuning extremum engine electronic control, and frequency domain self-tuning. The first topic is prediction and smoothing with two-dimensional ARMA models. The self-tuning recovers images corrupted by noises and artificial patterns. The second topic concerns seeking the optimal point of the spark ignition timing of a gasoline engine with a tunable map, which is tuned automatically. This also appeared briefly in the Introduction of this book, where the adaptation map achieves 9% improvement of fuel consumption comparing with the fixed map. The last topic is identification in the frequency domain. As is known, modern control theory is developed in the time domain, because theorists can get very simple solutions of the control problem with the exact model of the plant and the strict mathematical definition of the control objective. However, the modelling and grasping of plant properties are easier in the frequency domain. This is one of the superior
1288
Book Reviews
points of the classical control theory which is based on the frequency domain description of the plant, so that the H-infinity control theory has been actively studied in the recent years. It is the design method in the frequency domain based on the modern control theory. However, the H-infinity theory at the present time lacks the identification of the frequency domain. The topics treated here are, therefore, very interesting for the users intending to apply the H-infinity control theory, too. The topics also include an application to test civil engineering structures. All chapters contain the sections of 'Outline and Learning Objectives', 'Summary', 'Problems', and 'Notes and References', so that one can read this book very easily even without being acquainted with self-tuning systems. A variety of examples in the book may help the reader's understanding of the self-tuning systems. Now, open the jewelry box with your own hand.
About the reviewer Seiichi Shin received degrees of Bachelor, Master, and Doctor of Engineering all from University of Tokyo, in 1978, 1980, and 1987, respectively. From 1980 to 1988, he worked at the University of Tokyo, as an associate researcher in the former, and as an assistant professor in the latter. He became an associate professor of Institute of Information Sciences and Electronics, University of Tsukuba in 1988. In April 1992, he moved to the Department of Mathematical Engineering and Information Physics, University of Tokyo. His main interest is research on control theory and its application. It includes adaptive control, time delay systems, nonlinear systems, neural network, and distributed control systems. He is a member of IEEE, SICE, ISCIE, IEE of Japan, Japan SIAM, and JSME.
Temporal Logic for Real-time Systems* Jonathan S. Ostroff Reviewer: L. MOTUS Estonian Academy of Sciences, Institute of Cybernetics, Akadeemia tee 21, Tallinn 200108, Estonia. THE FIRST BOOK in this series elaborates the fair transition system idea (proposed by computing scientists), extends its application domain to the case of metric time, and suggests a new formalism--real-time temporal logic--for analysing the state sequences generated by the instantiations of the proposed extended fair transition system. The application domain of the presented methods is real-time control, more specifically, control of discrete event dynamic systems, e.g. flexible manufacturing, sequencing control and, of course, computer software. The author of this book follows the increasingly popular trend of blending different formalisms. This book mixes successfully logical models of discrete event dynamic systems (i.e. temporal logic and finite state machines) with algebraic models (i.e. communicating sequential processes)--the taxonomy of models is given by Ho in his plenary paper at the l l t h IFAC World Congress in 1990. An excellent feature of this book is a nice linking, explained in a few words, of many different techniques used in computing science so that a control engineer can see the connections between different techniques and methods. As an example, just note how elegantly the interconnections are explained between fair transition systems (proposed by Pnueli and Manna), the UNITY approach to concurrent programs (proposed by Chandy and Misra) state machines, communicating sequential processes and temporal logic. It is true, though, that a fair amount of background knowledge is needed to be able fully to enjoy the author's style. In the following the book is reviewed, chapter by chapter, subjective merits and demerits are listed for each chapter. Chapter 1, the Introduction, defines the basic notions used in the book. The author does not use the term discrete event dynamic systems, instead he uses real-time discrete event processes. Another definition that I am not very happy with is real-time--I cannot agree that real time is just a metric time. This certainly is one of its possible interpretations. However, for control engineers it is more important that the concept of real-time enables us to match different time counting methods used in co-operating dynamic systems that form a real-time system--we need more than metric properties of time to do that. Discrete event dynamic systems * Temporal Logic for Real-time Systems by Jonathan S. Ostroff. Research Studies Press Ltd, John Wiley (1989). ISBN 086380 0866, ISBN 0471 92402 4.
are in this book described by extended state machines (ESMs) which are a mixture of modified finite state machines, channel concepts of CSP (communicating sequential processes) language and the method of partial weakest preconditions/strongest postconditions. Chapter 2 is ESM Syntax which gives more details about extended state machines which are obtained from finite automata by: • Extending the type list of state variables (activity variable (next state variable), data variable and time variable are added). • Adding typical CSP (Communicating Sequential Processes) language properties for describing communication and synchronized actions of separate extended state machines (ESMs). The ESM syntax is illustrated in two examples--the two trains sharing a track, and transforming a piece of CONIC (a language for distributed computer system description) code to ESM notation. The ESM is based on the interleaving model of concurrency and assumes "...a precise behavioural description of all crucial features including plant dynamics, concurrency, nondeterminism, synchronization and realtime". This chapter leaves me with a feeling that the suggested methodology is much better suited for code verification than for system specification. Chapter 3, the ESM semantics introduces the notion of state sequences and trajectories of the extended state machine. This is important since the set of legal trajectories is the control systems description used for further study by temporal logic. When reading this chapter I could not avoid uncomfortable feeling that weak fairness and time bounds on the state-to-state transitions may cause a new symbiosis that cannot be called weak fairness any more--a transition cannot be continuously enabled under the time constraints. May be one should consider "a bounded-delay" fairness in such case. This problem is not studied in the book. Chapter 4, Real-time Temporal Logic gives a brief overview of Manna-Pnueli temporal logic. The Ostroff's real-time temporal logic (RTFL) is then obtained by extending a definition of formula in Manna-Pnueli temporal logic by adding specific ESM variables. Three typical problems explorable in ESM/R'I'TL framework are: • Verification problem of checking whether the generated trajectories satisfy the given requirements specification. • Synthesis problem of finding a controller's algorithm provided the plant has been described as an ESM and requirements specification is given.
References Daniel, R. W. (1990). Chaos in high performance digital robot force controllers. Proc. of 1990 IEEE Conf. on Decision and Control, Honolulu, HI, 5-7 December 1990, pp. 1951-1953. About the reviewer Dr Daniel is currently a university lecturer in Robotics at Oxford University, U.K. He received his B.Sc. in Electrical
1287
Engineering from Brunel University, London in 1978 and his Ph.D. from Cambridge University, U.K. in 1982. His main interests lie in the design of high performance servo mechanisms with applications to robotics. Much of his work is aimed at remote operations by tele-manipulators and is currently working on high fidelity force reflection systems. Dr Daniel is Co-Editor-in-Chief of Mechatronics, published by Pergamon Press.
Self-tuning Systems Control and Signal Processing* R. E. Wellstead Reviewer: SEIICHI SHIN Department of Mathematical Engineering and Information Physics, University of Tokyo, Hongo, Bunkyo-ku, Tokyo 113, Japan. AN EXCITING BOOK has been published for users and researchers in the fields of control and signal processing. This is a jewelry box of know-how of the self-tuning systems in the practice. This book is written based on the results done over a 15 year period in Control System Center, University of Manchester, Institute of Science and Technology, U.K. As is well known, "tuning" is the main work of practical application of control and signal processing, though the modern theory of these fields offers "design", which requires the mathematical fundamentals. The controller is obtained with the theory as a solution of mathematical equations, which are derived from strict mathematical descriptions of the plant and the control objective. However, there are always modeling errors between the true plant and its mathematical model and it is really difficult to specify the control objective as mathematical equations. These two problems annoy users every day and symbolize a gap between the practice and the theory of control. Tuning is the users' solution of the problems. The tuning absorbs the modeling error and, after many trials of tuning, the users find a trade-off point among control quality, cost, and so on. An auto-tuning system is a dream of the user, because its saves engineers' labour and achieves high quality control/signal processing when mismatch of controller/signal processor is a main drawback of the control/signal processing system. This book is written about the self-tuning systems from a user's viewpoint by two famous self-tuning system researchers, so that users can get the overview of the self-tuning systems easily. Moreover, researchers may be stimulated as the book points out what is necessary for the practical use of the self-tuning systems. Though the combination of control and signal processing in this volume is heavy for readers (even for me as a reviewer) we can take a general view of the self-tuning technique. Specifically the self-tuning technique used in signal processing shows us a future direction of the self-tuning control research, since the control is more complicated than the signal processing so that the self-tuning technique is advanced in the signal processing. The book consists of four major parts. The aim of the book and the mutual relations of the four parts are described in the Preface. The first chapter, Introduction, presents the basic concept used in the four parts, that is, control, signal processing, self-tuning, and adaptation. A comparison is also * Self-tuning Systems Control and Signal Processing by P. E. Weflstead and M. B. Zarrop. John Wiley (1991). ISBN 0-471-92883-6, ISBN 0-471-93054-7 (paperback).
and M. B, Zarrop given between the self-tuning systems and their rivals. Several simple examples given here are the practical applications of the self-tuning systems, including the engine control, the control of a lime kiln, adaptive echo cancelling, and self-tuning vibration simulation. Readers can, without any mathematical background, get valuable information from this chapter. Part 1 is "System Identification for Self-tuning," where the basis of models of system and signal, and recursive estimation techniques are described, including initial parameter and covariance settings, test signals, forgetting factor, and computational alternatives for recursive estimation. Some simple PASCAL programme of recursive estimation algorithms are also attached here for the understanding and the ease of the practical application. This part also includes the Ordinary Differential Equation (ODE) approach, which is a useful tool for analysing the self-tuning systems. However, it is much too complicated for practical users. In the book, an essence of ODE is extracted and engineers can determine how to use ODE to analyse their own self-tuning system. Part 2 is devoted to "Self-tuning Controllers". First, the pole assignment controller is described based on the transfer function approach, where the Diophantine equation plays a key role. Adding to the pole assignment, minimum variance control and multistage predictive control are explained in detail. These control methods are very popular and effective in practical applications. Moreover, controls of a superheater attemperator, an active suspension system, and a robot, are illustrated here as applications of the self-tuning systems. Part 3 treats "Self-tuning Signal Processing", where prediction and filtering with the self-tuning ability are discussed. This part includes the self-tuning systems of prediction of hot metal quality in a blast furnace, interference cancelling filters, and sensor signal conditioning. Part 4 is a collection of special topics pertaining to self-tuning systems. This part consists of individual three hot topics: two-dimensional self-tuning algorithms, self-tuning extremum engine electronic control, and frequency domain self-tuning. The first topic is prediction and smoothing with two-dimensional ARMA models. The self-tuning recovers images corrupted by noises and artificial patterns. The second topic concerns seeking the optimal point of the spark ignition timing of a gasoline engine with a tunable map, which is tuned automatically. This also appeared briefly in the Introduction of this book, where the adaptation map achieves 9% improvement of fuel consumption comparing with the fixed map. The last topic is identification in the frequency domain. As is known, modern control theory is developed in the time domain, because theorists can get very simple solutions of the control problem with the exact model of the plant and the strict mathematical definition of the control objective. However, the modelling and grasping of plant properties are easier in the frequency domain. This is one of the superior
1288
Book Reviews
points of the classical control theory which is based on the frequency domain description of the plant, so that the H-infinity control theory has been actively studied in the recent years. It is the design method in the frequency domain based on the modern control theory. However, the H-infinity theory at the present time lacks the identification of the frequency domain. The topics treated here are, therefore, very interesting for the users intending to apply the H-infinity control theory, too. The topics also include an application to test civil engineering structures. All chapters contain the sections of 'Outline and Learning Objectives', 'Summary', 'Problems', and 'Notes and References', so that one can read this book very easily even without being acquainted with self-tuning systems. A variety of examples in the book may help the reader's understanding of the self-tuning systems. Now, open the jewelry box with your own hand.
About the reviewer Seiichi Shin received degrees of Bachelor, Master, and Doctor of Engineering all from University of Tokyo, in 1978, 1980, and 1987, respectively. From 1980 to 1988, he worked at the University of Tokyo, as an associate researcher in the former, and as an assistant professor in the latter. He became an associate professor of Institute of Information Sciences and Electronics, University of Tsukuba in 1988. In April 1992, he moved to the Department of Mathematical Engineering and Information Physics, University of Tokyo. His main interest is research on control theory and its application. It includes adaptive control, time delay systems, nonlinear systems, neural network, and distributed control systems. He is a member of IEEE, SICE, ISCIE, IEE of Japan, Japan SIAM, and JSME.
Temporal Logic for Real-time Systems* Jonathan S. Ostroff Reviewer: L. MOTUS Estonian Academy of Sciences, Institute of Cybernetics, Akadeemia tee 21, Tallinn 200108, Estonia. THE FIRST BOOK in this series elaborates the fair transition system idea (proposed by computing scientists), extends its application domain to the case of metric time, and suggests a new formalism--real-time temporal logic--for analysing the state sequences generated by the instantiations of the proposed extended fair transition system. The application domain of the presented methods is real-time control, more specifically, control of discrete event dynamic systems, e.g. flexible manufacturing, sequencing control and, of course, computer software. The author of this book follows the increasingly popular trend of blending different formalisms. This book mixes successfully logical models of discrete event dynamic systems (i.e. temporal logic and finite state machines) with algebraic models (i.e. communicating sequential processes)--the taxonomy of models is given by Ho in his plenary paper at the l l t h IFAC World Congress in 1990. An excellent feature of this book is a nice linking, explained in a few words, of many different techniques used in computing science so that a control engineer can see the connections between different techniques and methods. As an example, just note how elegantly the interconnections are explained between fair transition systems (proposed by Pnueli and Manna), the UNITY approach to concurrent programs (proposed by Chandy and Misra) state machines, communicating sequential processes and temporal logic. It is true, though, that a fair amount of background knowledge is needed to be able fully to enjoy the author's style. In the following the book is reviewed, chapter by chapter, subjective merits and demerits are listed for each chapter. Chapter 1, the Introduction, defines the basic notions used in the book. The author does not use the term discrete event dynamic systems, instead he uses real-time discrete event processes. Another definition that I am not very happy with is real-time--I cannot agree that real time is just a metric time. This certainly is one of its possible interpretations. However, for control engineers it is more important that the concept of real-time enables us to match different time counting methods used in co-operating dynamic systems that form a real-time system--we need more than metric properties of time to do that. Discrete event dynamic systems * Temporal Logic for Real-time Systems by Jonathan S. Ostroff. Research Studies Press Ltd, John Wiley (1989). ISBN 086380 0866, ISBN 0471 92402 4.
are in this book described by extended state machines (ESMs) which are a mixture of modified finite state machines, channel concepts of CSP (communicating sequential processes) language and the method of partial weakest preconditions/strongest postconditions. Chapter 2 is ESM Syntax which gives more details about extended state machines which are obtained from finite automata by: • Extending the type list of state variables (activity variable (next state variable), data variable and time variable are added). • Adding typical CSP (Communicating Sequential Processes) language properties for describing communication and synchronized actions of separate extended state machines (ESMs). The ESM syntax is illustrated in two examples--the two trains sharing a track, and transforming a piece of CONIC (a language for distributed computer system description) code to ESM notation. The ESM is based on the interleaving model of concurrency and assumes "...a precise behavioural description of all crucial features including plant dynamics, concurrency, nondeterminism, synchronization and realtime". This chapter leaves me with a feeling that the suggested methodology is much better suited for code verification than for system specification. Chapter 3, the ESM semantics introduces the notion of state sequences and trajectories of the extended state machine. This is important since the set of legal trajectories is the control systems description used for further study by temporal logic. When reading this chapter I could not avoid uncomfortable feeling that weak fairness and time bounds on the state-to-state transitions may cause a new symbiosis that cannot be called weak fairness any more--a transition cannot be continuously enabled under the time constraints. May be one should consider "a bounded-delay" fairness in such case. This problem is not studied in the book. Chapter 4, Real-time Temporal Logic gives a brief overview of Manna-Pnueli temporal logic. The Ostroff's real-time temporal logic (RTFL) is then obtained by extending a definition of formula in Manna-Pnueli temporal logic by adding specific ESM variables. Three typical problems explorable in ESM/R'I'TL framework are: • Verification problem of checking whether the generated trajectories satisfy the given requirements specification. • Synthesis problem of finding a controller's algorithm provided the plant has been described as an ESM and requirements specification is given.