Andy Packard, University of California at Berkeley
Gary Balas, University of Minnesota
This workshop presents the theory of H-infinity based linear parameter varying control design techniques and shows how to apply these techniques to real-world control problems.
This two-day workshop presents the latest results in "parameter-dependent" control and their application to real world control problems. Parameter- dependent systems are systems whose state-space descriptions are known functions of time-varying parameters. However, the time variation of each of the parameters is not known in advance (but is assumed to be measurable in real-time). This type system is called linear, parameter varying (LPV). The controller is restricted to be a linear system, whose state-space entries depend causally on the parameter's history. The goals of feedback include stabilization and performance improvement, and these goals can be achieved by reformulating the parameter-dependent control design into (1) finite- dimensional, convex feasibility problems which can be solved exactly, and (2) infinite-dimensional convex feasibility problems which can be solved approximately. This control problem formulation constitutes a type of gain-scheduling problem.
This workshop provides a brief background on robust control, linear matrix inequalities (LMIs), and solutions of LMIs using convex optimization techniques. The underlining theory for LPV systems is presented as well as the solution to several gain-scheduled control problems that arise in the LPV framework. Applications of these methods are presented for the F-14 lateral- directional axis powered-approach flight control system, missile autopilots, and process control examples.
SCHEDULE: MONDAY, JUNE 2
SCHEDULE: TUESDAY, JUNE 3
Giorgio Rizzoni and Steve Yurkovich The Ohio State University Center for Automotive Research
This tutorial introduces practicing engineers and university researchers to the essential aspects of modeling and control of automotive powertrains, with particular focus on internal combustion engines and their integration within the total automotive powertrain.
This workshop recognizes both the significance of modeling and control of automotive powertrains to this growing area of engineering and its central role in the automotive industry. Emphasis is given to the integration of many different aspects of engineering, including: dynamics of mechanical, fluid, and thermodynamic systems; sensor and actuator technology; and feedback controls. Primary emphasis is given to dynamics and control of fuel-injected, spark-ignited, internal combustion engines, while integration with the complete powertrain (torque converter and transmission) is also addressed.
Part 1 (Powertrain Dynamics) presents the major dynamic phenomena that characterize powertrain behavior: intake and exhaust air flow dynamics; fuel system dynamics; combustion and emissions; crankshaft dynamics; transmission and driveline; and vehicle longitudinal dynamics. Emphasis is placed on explaining the interaction between subsystems, and the importance of considering the entire vehicle system when assessing the impact of performance of a subsystem on overall system behavior. A computer simulation of the overall powertrain system is also discussed.
Part II (Powertrain Control) begins with an overview of implementation issues relevant to automotive control systems in general, such as typical processor capabilities, typical noise sources, common disturbances, and production hardware. Specific models for control design, with attention to implementation in MATLAB and SimuLink for simulation are then developed. Control studies focus primarily on the idle speed control problem and the air-to-fuel ratio control problem; various other control objectives and issues are also addressed, such as exhaust gas recirculation and cold-start. Linear techniques which are discussed and compared range from simple PID to frequency-domain design techniques, to state variable methods and optimal control. Nonlinear techniques focus on sliding mode control and fuzzy logic, as well as adaptation mechanisms such as supervisory control.
SCHEDULE: MONDAY, JUNE 2
Part I POWERTRAIN DYNAMICS
Part II POWERTRAIN CONTROL
Organizers:
Datta Godbole, John Lygeros, Shankar Sastry
University of California, Berkeley
Presenters:
Datta Godbole, Alex Gollu, Tom Henzinger,
John Lygeros, Shankar Sastry, Pravin Varaiya
University of California at Berkeley
This workshop discusses new results in modeling, control design, verification and simulation of multi-agent, large scale, hybrid dynamical systems. This workshop is intended primarily for university researchers interested in hybrid control and computer aided verification, but it is also of interest to practitioners in the areas of automated highway systems and air traffic management.
The demand for increased levels of automation and system integration have forced control engineers to deal with increasingly larger and more complex systems. At the same time, recent technological advances, such as faster computers as well as cheaper and more reliable sensors have made it possible to extend the practical applications of control to systems that were previously impossible. Such large scale systems are typically controlled by a combination of continuous and discrete event controllers, thus making the overall closed loop system hybrid in behavior. This workshop presents new analytical tools and methods for design and analysis of hybrid systems. Hybrid system design and verification methodologies are presented based on game theory and optimal control; hybrid system verification tools are based on abstraction of continuous dynamical systems into automata and simulation tools designed for large scale multi-agent hybrid systems. These new methods are demonstrated for two multi-agent scarce resource system examples, namely: Automated Highway Systems (AHS) and Air Traffic Management Systems (ATMS).
SCHEDULE: MONDAY, JUNE 2
K.S. Narendra
Center for Systems Science, Yale University
This workshop shows how concepts and methods developed in systems theory and artificial neural networks can be suitably combined for the intelligent control of complex dynamical systems in the presence of uncertainty. From a systems theoretic point of view, neural networks can be considered as practically implementable convenient parametrizations of nonlinear maps. Such networks are ideally suited to cope with computational complexity, nonlinearity and uncertainty - three categories of difficulties encountered in intelligent control. Neural networks have been very successful in pattern recognition problems as well as mimicking rule-based expert systems. During the past five years, methods have also been developed to use them effectively for the identification and control of nonlinear dynamical systems.
Pattern recognition, learning, adaptive control and robust control are all applicable in disjoint contexts. In this workshop, we adopt the perspective that when these advanced capabilities are joined together in special ways, they can result in complex systems that respond appropriately to very challenging environments, and even in situations for which they have not been explicitly designed. An important question that has to be addressed is how to do this reliably in a general context. The lectures in the workshop will attempt to provide a uniform systems architecture and a unified design methodology for reliable interconnections.
This workshop begins with basic introductory material, gradually develops methods for identification and control of nonlinear systems, disturbance rejection, and multivariable control, and deals with problems in intelligent control. Applications of the methods to industrial problems are considered towards the close of the workshop. It is intended for industrial practitioners involved in modeling and control of nonlinear systems from input-output data as well as academic researchers interested in recent theoretical advances in neural network-based control.
SCHEDULE: TUESDAY, JUNE 3
R.E. Skelton, Purdue University
K.M. Grigoriadis, University of Houston
T. Iwasaki, Tokyo Institute of Technology
This workshop presents a new unified formulation of linear control theory, develops new control system analysis and design tools (based on elementary matrix algebra), and provides ready-to-use numerical algorithms to solve a large class of fixed-order robust control problems. It is intended for educators, researchers, industrial practitioners, and graduate students interested in a state-space based unified approach and associated computational methods for designing low-order robust controllers.
Modern state-space control theory has matured whereby an abundance of design techniques are available to treat a variety of performance objectives for continuous-time and discrete-time uncertain linear systems. However, the formalism and the mathematical tools used to approach these control problems are often fundamentally different, resulting in disorganized and inconsistent formulations of the design problems and distinct numerical techniques for their solution. This workshop presents a unified approach to formulate a large class of linear robust control design problems using elementary linear algebra. Stabilization, covariance upper bound, H-2, H-infinity, positive real and other control design problems are reduced to the same linear algebra problem, providing necessary and sufficient conditions for solvability in terms of Linear Matrix Inequalities and a parametrization of all admissible controllers. Model-order reduction problems and filtering problems will also be treated following the same unified formulation.
A fundamental limitation of modern control theory is its inability to characterize and design low-order controllers for linear uncertain systems. A great advantage of the unified formulation presented in this workshop is that fixed-order controllers can be easily characterized by including a nonconvex rank constraint in the LMI formulation of the design problem. The second major topic of this workshop is numerical algorithms to treat static output-feedback and fixed-order control design problems formulated with the unified framework. Numerical techniques based on alternating projections will be discussed, and MATLAB software for fixed-order control design, developed by the organizers, will be presented. Case studies and numerical examples are presented to illustrate the techniques and numerical algorithms. Convergent algorithms are also presented to address the integrated plant and controller design problem.
SCHEDULE: TUESDAY, JUNE 3
Notes: Excerpts from the book A UNIFIED ALGEBRAIC APPROACH TO LINEAR CONTROL DESIGN by R. Skelton, T. Iwasaki and K. Grigoriadis (Taylor & Francis, to appear in 1996).
Wallace E. Larimore; Adaptics, Inc.
Dale E. Seborg; University of California, Santa Barbara
Nancy Kirkendall, George Washington University and Office of
Management and Budget
Over the past several years, computational methods and software have been developed to reliably identify system dynamics from input/output data with optimal statistical accuracy. These automatic methods apply to a very general class of linear systems including multi-input/multi-output, state and measurement noise disturbances, unknown feedback, unknown state order, and possibly unstable or highly resonant dynamics. Existing methods for high accuracy identification such as Box/Jenkins and prediction error methods are problematic in that they are both computationally unreliable and involve a tedious toolbox approach requiring graduate level training.
The automatic methods presented in this workshop are fundamentally different and involve direct determination of the system states, i.e. system rank, using stable singular value decomposition (SVD) computations. Optimal rank selection based on canonical variate analysis (CVA) is related to partial least squares (PLS) and principal component analysis (PCA) methods. Statistical order selection methods are described which give optimal determination of state order. The state space dynamics are determined by simple multivariate regression. The concepts are presented in a direct first principles way that is appropriate for advanced undergraduate and graduate curriculum so that automated system identification can be made much more accessible to those in most need of using it. This advance in system identification has major implications for analysis, system monitoring, and design and implementation of control systems for many applications including aerospace systems and industrial process control. Several such examples are presented including an industrial recovery boiler, stirred tank reactor, autothermal reactor, distillation column, and on-line adaptive control of aircraft wing flutter. Automated system identification methods are compared with alternative approaches in terms of model types considered, required user knowledge, computational requirements and reliability, and results of model fitting using simulated data sets.
The intended audience includes those wishing to do model identification on applications data, those wishing an introduction to the concepts of automated system identification, those considering teaching an undergraduate or graduate course on the subject, or those with more advanced background. Material based on a draft textbook on automated system identification including MATLAB compatible software will be included in the course.
SCHEDULE: TUESDAY, JUNE 3
John Doyle, California Institute of Technology
Randy Freeman, University of California, Santa Barbara
Miroslav Krstic, University of California, Santa Barbara
This workshop is patterned after an invited session presented at the 1996 IEEE Conference on Decision and Control (CDC). A group of speakers, representing a variety of popular nonlinear design techniques, will be invited to present a brief tutorial of their nonlinear control method; this review will then be followed by the results achieved with the method when applied to some common benchmark examples.
The participants are not able to choose example problems to suit their selected control method, and the examples will be deliberately created so that each method will have severe difficulties with at least one of the benchmark examples. John Doyle serves as independent judge who coordinates creation of the benchmark examples. The goal is to provide both the participants and the audience a much clearer and objective view of the relative merits of various nonlinear methods.
The benchmark examples are generated in 2 ways: 1) analytically and 2) from physical experiments. The analytically generated examples will be nonlinear optimal control problems, with a novel twist. The examples will be generated with a "converse Hamilton-Jacobi-Bellman" method (CoHJB). Starting with the cost and optimal value function V, CoHJB solves HJB PDEs "backwards" algebraically to produce nonlinear dynamics and optimal controllers and disturbances. Although useless for design, it is a great procedure for generating benchmark examples. It is easy to use, computationally tractable, and generates essentially all possible nonlinear optimal control problems. The optimal control and disturbance are then known and can be used to study actual design methods, which must start with the cost and dynamics without knowledge of V. Experience so far with this method of generating examples has been excellent.
A second set of benchmark examples will use nonlinear control experiments available at Caltech, and particularly a thrust-vectored flight control experiment. While it may not be feasible for all participants to implement controllers on this example, we expect to have both simulation studies and actual experimental results for most of the design methods. For simulation studies, we will use a variety of nonlinear robustness analysis tools that have been developed recently at Caltech.
Anticipated nonlinear control design methods used for this session include,
Organized by: Rolf Isermann, Technical University Darmstadt
Presenters: M. Ayoubi, Darmstadt; M. Blanke, Aalborg; P. Frank, Duisburg; D. Fuessel, Darmstadt; J. Gertler, Fairfax; R. Isermann, Darmstadt
The goal of this workshop is to review the basic principles of supervision methods including model based fault detection and fault diagnosis and appropriate actions to improve reliability and safety of technical systems. The introductory lecture describes classical and advanced methods of monitoring, automatic protection, and supervision; the generation of analytical and heuristic symptoms; and an overview of fault detection and fault diagnosis methods. This is then followed by reviewing the most important model based fault detection methods (via parameter estimation, state observers, and parity approaches) and by describing change detection methods. Based on analytical (model based) and heuristic (operator observed) symptoms, a fault diagnosis can be performed by using classification methods or fault-symptom trees and methods of approximate reasoning including fuzzy logic and neuronal networks. Then redundancy and reconfiguration schemes are described to cope with faults and to avoid process malfunctions and failures.
Application examples are used throughout the workshop, showing practical results obtained e.g. for machine tools, robots, engines, automobiles, actuators, and sensors.
This workshop is intended for industrial practitioners, university researchers in engineering, and engineering students.
SCHEDULE: SATURDAY, JUNE 7
Updated 1/28/97