Fundamentals and Applications

Constantine H. Houpis Steven J. Rasmussen
Graduate School of Engineering Air Force Institute of Technology Wright-Patterson Air Force Base, Ohio




Library of Congress Cataloging-in-Publication Data
Houpis, Constantine S. Quantitative feedback theory: fundamentals and applications / Constantine H. Houpis, Steven J. Rasmussen. p. cm. — (Control engineering; 3) Includes bibliographical references (p. ). ISBN 0-8247-7872-3 (alk. paper) 1. Feedback control systems. I. Rasmussen, Steven J. II. Title. III. Series: Control engineering (Marcel Dekker); 3. TJ216.H69 1999 629.8'3—dc21 99-26669 CIP
Marcel Dekker, Inc., and the authors make no warranty with regard to the accompanying software, its accuracy, or its suitability for any purpose other than as described in the book. This software is licensed solely on an "as is" basis. The only warranty made with respect to the accompanying software is that the diskette medium on which the software is recorded is free of defects. Marcel Dekker, Inc., will replace a diskette found to be defective if such defect is not attributable to misuse by the purchaser or his agent. The defective diskette must be returned within 10 days to: Customer Service, Marcel Dekker, Inc., P.O. Box 5005, Cimarron Road, Monticello, NY 12701, (914) 796-1919.

This book is printed on acid-free paper.
Headquarters Marcel Dekker, Inc. 270 Madison Avenue, New York, NY 10016 tel: 212-696-9000; fax: 212-685-4540 Eastern Hemisphere Distribution Marcel Dekker AG Hutgasse 4, Postfach 812, CH-4001 Basel, Switzerland tel: 41-61-261-8482; fax: 41-61-261-8896

World Wide Web http://www.dekker.com
The publisher offers discounts on this book when ordered in bulk quantities. For more information, write to Special Sales/Professional Marketing at the headquarters address above.

Copyright © 1999 by Marcel Dekker, Inc. AH Rights Reserved.
Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming, and recording, or by any information storage and retrieval system, without permission in writing from the

Current printing (last digit) 10 9 8 7 6 5 4 3 2 1

A Series of Reference Books and Textbooks Editor NEIL MUNRO, PH.D., D.SC. Professor Applied Control Engineering University of Manchester Institute of Science and Technology Manchester, United Kingdom

1. Nonlinear Control of Electric Machinery, Darren M. Dawson, Jun Hu, and Timothy C. Burg 2. Computational Intelligence in Control Engineering, Robert E. King 3. Quantitative Feedback Theory: Fundamentals and Applications, Constantine H. Houpis and Steven J. Rasmussen Additional Volumes in Preparation Self-Learning Control of Finite Markov Chains, A. S. Poznyak, K. Najim, and E. Gomez-Ramirez Robust Control and Filtering for Time-Delay Systems, Magdi S, Mahmoud

Series Introduction
Many textbooks have been written on control engineering, describing new techniques for controlling systems, or new and better ways of mathematically formulating existing methods to solve the ever-increasing complex problems faced by practicing engineers. However, few of these books fully address the applications aspects of control engineering. It is the intention of this new series to redress this situation. The series will stress applications issues, and not just the mathematics of control engineering. It will provide texts that present not only both new and wellestablished techniques, but also detailed examples of the application of these methods to the solution of real-world problems. The authors will be drawn from both the academic world and the relevant applications sectors. There are already many exciting examples of the application of control techniques in the established fields of electrical, mechanical (including aerospace), and chemical engineering. We have only to look around in today's highly automated society to see the use of advanced robotics techniques in the manufacturing industries; the use of automated control and navigation systems in air and surface transport systems; the increasing use of intelligent control systems in the many artifacts available to the domestic consumer market; and the reliable supply of water, gas, and electrical power to the domestic consumer and to industry. However, there are currently many challenging problems that could benefit from wider exposure to the applicability of control methodologies, and the systematic systems-oriented basis inherent in the application of control techniques. This new series will present books that draw on expertise from both the academic world and the applications domains, and will be useful not only as academically recommended course texts but also as handbooks for practitioners in many applications domains. This latest volume, Quantitative Feedback Theory, is another excellent contribution to this series. Dr. Houpis and Mr. Rasmussen are to be commended.
Neil Munro

The objective of this text is to bridge the gap between the scientific (theoretical) and engineering methods of Quantitative Feedback Theory (QFT) by applying this multivariable robust control system design technique to real-world problems. Thus, the engineer is at the interface of the real-world with the body of knowledge and theoretical results available in the technical literature. Professor Isaac M. Horowitz, the developer of the QFT technique, has continually stressed the transparency of QFT; that is, the ability to visually relate the implementation of the design parameters to the real-world problem, from the onset of the design and throughout the individual design steps. Therefore, it is the purpose of this text to enable and enhance the ability of the engineering student and the practicing engineer to bridge the gap between the scientific and engineering methods. In order to accomplish this goal, the text is void of theorems, corollaries, and/or theoretical lemmas. In other words, this textbook stresses the engineering approach and not the scientific approach - this is a textbook for engineers. Professor Horowitz began developing the Quantitative Feedback Theory (QFT) in early 1960. Since then great strides have been made in exploiting the full potential of the QFT technique. It is a frequency domain technique for designing a class of control systems for nonlinear plants. The abbreviation QFT should not be confused with the physics topic Quantum Field Theory (QFT). The catalyst that has propelled Horowitz's QFT to the level of being a major robust multiple-input multiple-output (MIMO) control system design method has been the development and availability of viable QFT computer-aided-design (CAD) packages. Through the close collaboration of Professor Horowitz with Professor C. H. Houpis and his graduate students, during the 1980s and the early part of the 90s, successful QFT designs involving structured parametric uncertainty had been completed and published by the Air Force Institute of Technology (AFIT) MS thesis students and faculty. During this period, the first multiple-input singleoutput (MISO) and MIMO QFT CAD packages were developed at AFIT. Another major accomplishment was the successful implementation and flight test of two QFT designed flight control systems, by Captains S.J. Rasmussen and S.N. Sheldon, of the Air Force Wright Laboratory, for the LAMBDA unmanned research vehicle in 1992 and 1993. Also, on April 28, 1995, Dr. Charles Hall of North Carolina State University, announced that four successful flight tests of QFT flight controllers were accomplished.^3 Based upon these solid accomplishments, an aerospace engineering firm began applying the QFT design method in
1995. Other individuals throughout the world are also applying QFT to design

real-world robust control systems. Professor D. S. Bernstein ably points out the power of the frequency domain analysis, not only for linear systems, but also for nonlinear systems.76

Refinements based upon the class testing of the first and second editions are incorporated in this text. Extensive use is made of the MISO and MIMO QFT CAD packages to assist the reader in understanding and applying the QFT design technique. for a MIMO system. Chapters 2 through 7 present the fundamentals of the QFT technique and the associated design procedure for the tracking control problem. and illuminating account of those elements that have relevance in the analysis and design of robust control systems and in bridging the gap between the scientific and the engineering methods. This is followed by presenting an overview of QFT: the design objectives. calculations. aimed to provide students and practicing engineers with a document that presented QFT in a unified and logical manner. and QFT basics. Much of the material in this text is based upon the numerous articles written by Professor Horowitz. the QFT technique was initially developed for MISO control systems and was then extended to MIMO control systems. The authors have exerted meticulous care with explanations. Extensive use of the MISO and MIMO QFT CAD packages (see App. design overview. The remaining chapters focus on bridging the gap between theory and the real world by presenting engineering rules and the factors that are involved. performance specifications. Chapter 3 presents the fundamentals of the QFT design technique for the analog MISO control system. This is followed by an extension of the technique to handle external disturbances. the regulator control problem. like the first edition. and.vi Preface In 1986 Professor Houpis published a technical report41 which was the first attempt to bring under one cover the fundamentals of QFT. etc. such as simulations. that are important in implementing a successful robust control system design. tables. diagrams. This is followed by the extension of this technique to MISO discrete-time control systems in Chapter 4. and the numerous lectures that he presented at the Air Force Institute of Technology. Thus. A-C) is made for the MISO and MIMO examples throughout the text Chapter 2 discusses the reasons why feedback is required to achieve the desired system performance. comprehensive. The chapter concludes with an insight into the QFT technique and the benefits of applying this technique. and symbols. The reader is shown how to make intelligent realworld assumptions based upon mathematics and/or on a sound knowledge of the characteristics and the operating scenario of the plant to be controlled The text provides a strong. an mxm MIMO control system can be represented by m2 MISO equivalents. along with his colleagues. This is accomplished by including appropriate examples in each chapter and problems at the end of each chapter in order to reinforce the fundamentals. implementation. The second edition2of this TR brought the material up to the state-of-the-art.. . what is structured parametric uncertainty and its Bode plot and Nichols Chart representations. As Horowitz and his colleagues have shown. Both analog and sampled-data (discrete-time) MISO and MIMO feedback control systems are covered in detail. As a result.

Chapter 10 discusses the factors that are involved in a control system design cycle. The Binet-Cauchy formula is applied to determine if a minimumphase (m.) effective plant (detP) is achievable. sensitivity analysis.Preface vii Chapter 5 begins with an introduction to MIMO plants having structured parametric uncertainty. Aspects such as performance tolerances. These rules attempt to bridge the gap between QFT and the real-world problems. Thus. the regulator control problem. The QFT Method 1 design technique is discussed in detail in Chapter 6..e. which are based on both the state- . and upon many years of applying the QFT robust control system design technique to many real-world nonlinear problems. Based upon this formulation. The aspect of diagonal dominance which is a requirement for Method 1 is presented. From the state-space equations the corresponding plant and disturbance matrices are derived and the corresponding block diagram representation is shown. factors that must be kept in mind from the onset of the design process and which bridge the gap between theory and the real world. There are a number of factors that contribute to making the decision that a satisfactory control system design has been achieved. equilibrium and tradeoffs. This is followed by the introduction to the QFT MIMO compensated system formulation and the development of the effective MISO equivalents of a MIMO system. This method has the advantage of reducing the amount of over-design inherent in Method 1 and is applied when the diagonal dominance condition is not satisfied. some universal design features. The remaining portion of the chapter adapts the MISO analog QFT design technique of Chapter 3 to the QFT robust design (Methods 1 and 2) of MIMO control systems containing structured parametric uncertainty. this text includes a section entitled QFT Standard Symbols & Terminology. Engineering Rules are presented in Chapter 9. and the determination of bounds are discussed Chapter 7 thoroughly presents the details of Method 2. This QFT regulator design technique is applied to a real-world design example. The most effective way of expediting the transfer of QFT knowledge and its corresponding state-of-the-art material is for all authors on this subject to adhere to a standard list of QFT symbols. In Chapter 8 the QFT technique is extended to the design of MIMO control systems with external disturbance inputs. The remaining chapters enhance the emphasis of this text: to bridge the gap. high frequency and stability analyses. Throughout the preceeding chapters the elements that contribute to the transparency of QFT have been stressed. Design equations for the 2x2 and the 3x3 MIMO systems are presented with corresponding design guidelines. where applicable. i. There are many worthwhile robust multivariable control system design techniques available in the technical literature. simplification of the single-loop structure. the QFT m2 MISO effective loop equations are derived for the regulator case. Based upon these elements.p.

aeronautical. but also with a degree of control effector failures. J. J. Professor Emeritus of Electrical Engineering. for the encouragement they have provided in the preparation of this text. The authors wish to express their appreciation for the support and encouragement of Professor M. It incorporates the concept of designing a robust control system that maintains the desired system performance. The text has been thoroughly class-tested. etc? For some techniques the designer is assisted by available computer-aided-design (CAD) packages. and Professor Pachter. The authors believe that this method has proven its applicability to the design of practical MISO and MIMO control systems with low order compensators (controllers) with minimal gain.viii Preface variable approach and on conventional control theory. not only over a prescribed region of plant parameter uncertainty. unambiguous. The computer-aided-design (CAD) packages of App. Glass of Purdue University for review of the TOTAL-PC CAD software that accompanies this volume. A through C are available to assist a control engineer in applying the QFT design method The use of these QFT CAD packages are stressed throughout the text. Air Force Institute of Technology. D'Azzo. The control engineer must have a sufficiently broad perspective to be able to apply the appropriate technique to the right design problem. Some of the questions that the control engineer must keep in mind (see Chapter 10) in selecting a design method are: (a) can it solve a real-world problem? (b) is the method computationally intensive? (c) can it handle structured parametric uncertainty? (d) which method yields the lowest order compensator or controller? (e) will the design method result in a control system that can be implemented on the target hardware. This textbook provides students of control engineering and the practicing control engineer with a clear. and relevant account of the QFT technique. Included are examples of feedback control systems in various areas of practice (electrical. Appreciation is expressed to Dr.) while maintaining a strong basic QFT text that can be used for study in any of the various branches of engineering. The text is arranged so that it can also be used for self-study by the engineer in practice. etc. This text presents a control system design based on quantitative feedback theory (QFT) which is a very powerful design method when plant parameters vary over a broad range of operating conditions. Pachter. mechanical. during the preparation of the 1995 technical report. Special appreciation is expressed to Professor Horowitz for the association and his collaboration with Professor Houpis during the period of 1981 through . The applicability of each design technique may be limited to certain classes of design problems. The authors wish to thank John W. thus enhancing its value for classroom and self-study use.41 His wealth of knowledge of the flight control area enhanced its value and his comments with respect to this text are appreciated The authors express their thanks to those students who have used this book and to the faculty who have reviewed it for their helpful comments and recommendations. at the Air Force Institute of Technology.

McLean of the University of Southampton was of immense value and was greatly appreciated by the authors. Air Force Institute of Technology Research associate. Further acknowledgment must be made of the support of Dr. His perception and insight have contributed to the clarity and rigor of the presentation. U.S. during the 1980's. E. Sheldon and the support given by the European Office of Aerospace Research and Development of the U. D. R. Rasmussen .S. Morris and Mr. Rubertus of the Air Force Research Laboratory. Flinn. Davis (retired). Houpis Steven J. for his work on improving and maintaining the TOTAL-PC CAD package. and Mr.N. L. J. Air Force Wright Aeronautical Laboratories. Acknowledgment of Mr. Air Force Office of Scientific Research during these past years. and his colleagues Mr. The thorough review of the manuscript by Professor D.Preface ix 1992. The personal relationship with him has been a source of inspiration and deep respect. Appreciation is also expressed to Dr. Mr. Ramage. Rubertus is made for the support and encouragement in developing and extending the QFT technique. J. Constantine H. S. M. This support and encouragement was maintained by Mr. Ewing.


2 Structured Parametric Uncertainty: A Basic Explanation 2-3.1 QFT Design Objective 2-3.5 VISTA F-16 Flight Control System (Including Configuration Variation) 1-6.3 Welding Control Systems 1-6.2 Idle Speed Control for Automotive Fuel Injected Engine 1-6.3 Control System Performance Specifications 2-3.2 A Simple Mathematical Description 2-3.Contents Series Introduction iii Preface QFT Standard Symbols & Terminology v xix Chapter 1 INTRODUCTION 1-1 Introduction 1-2 The Engineering Control Problem 1 -3 Quantitative Feedback Theory (QFT) 1-4 Control Theory Background 1-5 Definitions and Symbols 1-6 QFT Applications 1-6.5 QFT Basics 1 1 1 5 6 7 7 7 7 9 9 10 12 12 13 14 14 16 16 16 19 19 19 19 21 22 23 24 xi .6 Design of Flight Control Laws for Aircraft with Flexible Wings Using Quantitative Feedback Theory 1- Operational Amplifiers (Op-Amp) 1-6.1 QFT and Robust Process Control 1-6.4 Control System for an Actuator Plant 1-6.9 Wastewater Treatment Control System 1-7 Outline of Text Chapter 2 INTRODUCTION TO QFT 2-1 Quantitative Feedback Theory 2-2 Why Feedback? 2-3 QFT Overview 2-3.7 Robot Controllers 1-6.4 QFT Design Overview 2-3.1 A Simple Example 2-3.

1 Tracking Models 3-4.): Case 1 3.1 Open-Loop Plant 2-4. Plant 4-4 Discrete MISO Module with Plant Uncertainty 92 93 93 94 97 101 103 106 .to z-Plane Transformation: Tustin Transformation 4-3 Nonminimum Phase Analog Plant 4-3.1 Analog QFT Design Procedure for a n.3 Results of Applying the QFT Design Technique 2-4.and w -Domain Transformation 4.3 5.) 3-7 Nominal Plant 3 -8 U-Contour (Stability Bound) 3-9 Tracking Bounds Bpfjaj.6 QFT Design 2-4 Insight to the QFT Technique 2-4.2 Closed-Loop Formulation 2-4.1 w.(s). JPQ'o).): Case 2 3-12 The Composite Boundary B0(ja>?) 3-13 Shaping of L0(jco) 3-14 Guidelines for Shaping L0(fco) 3-15 Design of the Prefilter 3-16 Basic Design Procedure for a MISO System 3 -17 Design Example 1 3-18 Design Example 2 3-19 Template Generation for Unstable Plants 3-20 Summary Chapter 4 DISCRETE QUANTITATIVE FEEDBACK TECHNIQUE 36 37 41 42 42 45 45 47 52 57 61 62 68 70 72 75 85 88 91 92 4-1 Introduction 4-2 Bilinear Transformation 4-2.4 Insight to the Use of the Nichols Chart (NC) in the QFT Technique 2-5 Benefits of QFT 2-6 Summary Chapter 3 THE MISO ANALOG CONTROL SYSTEM 3-1 Introduction 25 27 27 28 28 28 31 32 33 33 3-2 The QFT Method (Single-Loop MISO System) 3-3 Design Procedure Outline 34 35 3-4 Minimum-phase System Performance Specifications 3-4.11 Disturbance Bounds BD(jo).) on the NC 3-10 Disturbance Bounds BD(ja).2 Disturbance Rejection Models 3-5 JLTI Plant Models 3 -6 Plant Templates of P.2.m.xii Contents 2-3.p.2 s-Plane and w-Plane Relationship 4-2.

2 Simple PCT Example 4-8.2 Performance Bounds 5-6.3 Conclusions 107 110 113 113 115 118 119 123 126 127 133 136 137 139 141 145 145 145 146 146 146 147 147 149 149 Chapter 5 MULTIPLE-INPUT MULTIPLE-OUTPUT (MIMO) PLANTS: STRUCTURED PLANT PARAMETER UNCERTAINTY 5-1 Introduction 5-2 The MIMO Plant 5-3 Introduction to MIMO Compensation 5-4 MIMO Compensation 5-5 Introduction to MISO Equivalents 5-5.1 Minimum-Phase. and Unstable P(s) 4-11.3 Bounds B(jv) on L0(jv) 4-5.2 Digital Controller Implementation 4-11.4 The PCT System of Fig.3 QFT Design Method 1 5-6.4 Nonminimum Phase L0(w) 4-5.1 Effective MISO Equivalents 5-6 Effective MISO Loops of the MIMO System 5-6.8 Design of the Prefilter F(w) 4-6 Simulation 4-7 Basic Design Procedure for a MISO Sampled-Data Control System 4-8 QFT Technique Applied to the Pseudo-Continuous-Time (PCT) System 4-8.Contents xiii 4-5 QFT w-Domain DIG Design 4-5. Nonminimum Phase.4 QFT Design Method 2 5-6.3 The Sampled-Data Control System Example 4-8.7 Error in the Design 4-5.8 4-8. 4.1 Example: The 2x2 plant 5-6.5 Summary 5-7 Constraints on the Plant Matrix 5 -8 Basically Non-Interacting (BNIC) Loops 5-9 Summary 150 155 157 158 160 165 165 169 173 173 175 176 181 182 .6 o)s = 120 Is too Small 4-5.2 Plant Templates 4-5.1 Closed-Loop System Specifications 4-5.1 Introduction to Pseudo-Continuous-Time System DIG Technique 4-8.5 PCT Design Summary 4-9 Applicability of Design Technique to Other Plants 4-10 Designing L(w) Directly 4-11 Summary 4-11.5 Synthesizing Lmo(w) 4-5.

THE SINGLE-LOOP (MISO) EQUIVALENTS 6-1 Introduction 6-2 Design Example 6-2.2 Applications of Sec.3 Inherent Constraints 7-9 Nondiagonal G 7-10 Achievability of a m.5 Scope 8-3.1 Trade-Off In High Frequency Range 6-5. Effective Plant det Pe 7-11 Summary 213 214 216 217 218 219 221 222 222 225 226 227 228 234 Chapter 8 MEV1O SYSTEM WITH EXTERNAL DISTURBANCE INPUTS 235 8-1 Introduction 235 8-2 MDVIO QFT with External (Input) Disturbance 8-3 An External Disturbance Problem 8-3. Sensitivity" in Single-Loop Design 7-8.1 Conditions For "a.1 Performance Tolerances 6-2.1 to Design Method 2 7-8.xiv Contents Chapter 6 DESIGN METHOD 1 .3 Examples-Bounds Determination 6-6 Templates: Special Case 6-7 Summary 183 183 184 185 186 189 192 194 195 199 202 203 212 212 Chapter 7 MIMO SYSTEM DESIGN METHOD 2 -MODIFIED SINGLELOOP EQUIVALENTS 213 7-1 Introduction 7-2 Design Equations For The 2x2 System 7-3 Design Guidelines 7-4 Reduced Overdesign 7-5 3x3 Design Equations 7-6 Example: 3x3 System Design Equations 7-7 mxm System: m > 3 7-8 Conditions for Existence of a Solution 7-8.6 Methodology 235 242 243 243 243 244 244 244 .s. 7-8.1 Aerial Refueling Background 8-3.4 Design Objectives 8-3.2 Problem Statement 8-3.2 Sensitivity Analysis 6-2.3 Simplification of the Single-Loop Structures 6-3 High Frequency Range Analysis 6-4 Stability Analysis 6-5 Equilibrium and Trade-Offs 6-5.2 Some Universal Design Features 6-5.p.3 Assumptions 8-3.

12 E.6 Minimum Order Compensator (Controller)(MOC) 9-3.4 Design Techniques 9-3.'s 9-3.1 E.1 Linear Simulations 8-8.1C-135B Modeling 8-4.3 Channel 2 Loop Design (Method 1 "Loop 1" Design Procedure) 8-7. 11 Asymptotic Results 9-3.7 Minimum Compensator Gain 9-3.2 Loop Shaping 8-7.3 E.R.R.7 E.13 E. q.11 E.R. 10 Optimization and Simulation Run Time 9-3.p. 13 Nonideal Step Function for Simulation 9-4 Nonlinearities .9 Nominal Plant Determination 9-3.R..10 E.2 am.6 Closed Loop Lm Plots 8-8 Air-to-Air Refueling Simulations 8-8.9 E. 8 Basic mxm Plant P Preconditioning 9-3.7 Overview of the Aerial Refueling Design Problem 8-4 Air-to-Air Refueling PCS Design Concept 8-4.5 Channel 3 Loop Design (A SISO Design) 8-7.5 E.R.R.R.Contents xv 8-3.The Engineering Approach 9-5 Plant Inversion 9-6 Invertibility 9-7 Pseudo-Continuous-Time (PCT) System 9-8 Bode's Theorem 9-9 The Control Design Process 9-10 Summary 265 267 271 271 272 272 273 274 274 274 275 275 276 276 276 276 277 278 280 280 282 283 284 .R. 12 Controller Implementation 9-3.2 E.2 Nonlinear Simulations 8-9 Tracking/Regulator MfMO System 8-10 Summary Chapter 9 NOW THE "PRACTICING ENGINEER TAKES OVER" 245 245 245 246 247 248 250 250 252 252 253 255 258 259 259 260 262 264 265 9-1 Introduction 9-2 Transparency of QFT 9-3 Body of Engineering QFT Knowledge 9-3.4 E.R.3 Templates 9-3.R.4 Channel 1 Loop Design (Method 2 "Loop 2" Design Procedure) 8-7.R.8 E. 1 Weighting Matrix 9-3.R.R.1 Disturbance Rejection Specification 8-7.5 QFT Method 2 9-3.2 Disturbance Modeling 8-5 Plant and Disturbance Matrices 8-6 Control Problem Approach 8-7 The QFT Design 8-7.6 E.

1 Lambda Bending Example 10-7.9 Engineering Interactive Simulation 10-2.10 Hardware-in-the-Loop Simulation/Implementation 10-2.2 Unmodeled Behavior 10-8 Summary APPENDIXES Appendix A: MTMO QFT CAD PACKAGE 288 288 289 289 289 290 290 290 290 291 291 292 294 296 297 298 300 300 301 301 302 305 307 307 A-l Introduction A-2 Introduction: Overview of Multivariable Control A-3 Continuous-Time vs.5 QFT Control System Design 10-2.2 Performance Specifications 10-2.2 Second Design Cycle 10-3.Contents Chapter 10 THE DESIGN AND IMPLEMENTATION PROCESS FOR A ROBUST CONTROL SYSTEM 10-1 Introduction 10-2 Control System Design Process 10-2.4 Fourth Design Cycle 10-4 Selection of Design Envelope 10-5 Control System Implementation Issues 10-6 Hardware/Software Consideration 10-7 Bending Modes 10-7.3 Dynamics Model 285 285 286 286 286 288 10-2. Discrete-time Design A-4 Overview of the Multivariable External Disturbance Rejection Problem A-5 Open-loop Structure A-6 Formation of Plant Models for Tracking Control Problems A-7 Inverse of Pe A-8 MISO Loops of the Tracking Control Problem A-9 MISO Loops of External Disturbance Rejection Problem A-10 Q Matrix Validation Checks 307 308 308 311 311 314 315 315 317 318 .1 First Design Cycle 10-3.6 Linear Simulation 10-2.12 System Test 10-2.4 Control Authority Allocation 10-2.11 Operator-in-the-Loop Simulation 10-2.3 Third Design Cycle 10-3.1 Functional Requirements 10-2.8 Engineering Visualization 10-2.13 Redesign 10-3 Design Process Example 10-3.7 Nonlinear Simulation 10-2.

2 Design (Improved Method) for L2(s) E-3.6 Composite Bounds A-14 Compensator Design A-15 Prefilter Design A-16 Design Validation A-17 Summary Appendix B: TOTAL-PC CAD PACKAGE B-l Introduction B-2 Overview of TOTAL-PC B-3 QFT CAD Package Appendix C: TOTAL-PC: DISCRETE QFT DESIGN PROCESS C-1 Discrete QFT Design (vc'.3 External Disturbance Rejection Performance Specifications A-12.2 Determination of Boundaries B^jcoi) D-3.1L.4 Design of F(s) D-3.3 Determination of L0(s) D-3.1 Stability Bounds A-13.3 Prefilter£2(s) Design ' 341 341 341 342 342 343 343 346 346 347 347 348 348 348 3 51 358 .4 Allocated Tracking Bounds A-13.5 Simulation of Design Appendix E: MTMO DESIGN EXAMPLE E-l Design Problem E-2 Problem E-3 Solution E-3.Contents xvii A-11 Improved Method A-12 Specifications A-12.2 Tracking Performance Specifications A-12.1 Stability Specifications A-12.5 External Disturbance Rejection Bounds A-13.3 Gamma Bounds A-13.4 Gamma Bound Specifications A-13 Bounds on the NC A-13.and z-domain) 318 319 319 319 320 320 320 321 321 322 322 323 324 324 325 326 327 328 328 328 329 335 335 Appendix D: MISO DESIGN EXAMPLE D-1 MISO Design Problem D-2 Problem D-3 Solution D-3.o(s) Design E-3.1 Template Generation D-3.2 Cross-Coupling Bounds A-13.

xviii Contents E-4 Simulation E-5 Summary References Problems Answers to Selected Problems Index 359 364 365 370 386 394 .

and optimal bounds on LQco. a.I.TC The disturbance. tracking. av = Lm Ttj by = Lm Tij The desired modified upper tracking bounds for the MIMO system b'u = bit .s.) for the MISO system Bh B Ultra high frequency boundary (UHFB) for analog design Ultra high frequency boundary (UHFB) for discrete design LmTR The Lm of the desired tracking control ratio for the upper bound of the MISO system six .QFT Standard Symbols & Terminology SYMBOL Op The specified peak magnitude of the disturbance response for the MISO system Arbitrarily large Arbitrarily small The desired lower tracking bounds for the MIMO system The desired upper tracking bounds for the MIMO system The desired modified lower tracking bounds for the MIMO system: a'H = au + TC Z.

g. Allotted portion of the ij output due to a cross-coupling effect for a MIMO system The (upper) value of Lm 7X/».e. h... i.) for MISO system The dB difference between the augmented bounds of Bu and BL in the high frequency range for a MISO system &(/'a>i) AT ctj D 3) = {Z>} F. for a MISO system The difference between bu and a. Ar = ba .xx Symbols BL = LmTR The Lm of the desired tracking control ratio for the lower bound of the MISO system Bs BW Stability bounds for the discrete design Bandwidth ra.as The interaction or cross-coupling effect of a MIMO system MISO and MDVIO system external disturbance input Script cap dee to denote the set of external disturbance inputs for a MIMO system 9 = {D} The prefilter for a MISO system and the mxm prefilter matrix for a MDVIO system respectively Figures of merit (FOM) (see Ref.j} The dB difference between Bv and BL for a given co.f. respectively.G = {g.} High frequency High gain h. ..F= {ftj} FOM G. For a diagonal matrix G = {g. 1) The compensator or controller for a MISO system and the mxm compensator or controller matrix for a MIMO system.

k kT /I Lo.2. and a single output MISO Mi>MLi The specified closed-loop frequency domain overshoot constraint for the MISO system and for the /* loop of a MBVIO system.. a system having one tracking input. respectively. respectively Yij J A function only of the elements of a square plant matrix The number of plant transfer functions for a MISO system or plant matrix for a MIMO system that describes the region of plant parameter uncertainty where i = 1.7.Symbols xxi Y.. This overshoot constraint may be dictated by the phase margin angle for the specified loop transmission function m..p.J denotes the particular plant case in the region of plant parameter uncertainty A running index for sampled-data systems where k = 0... respectively Left-half-plane Linear-time-invariant LHP LTI MIMO Multiple-input multiple-output. L0i The sampled time The excess of poles over zeros of a transfer function The optimal loop transmission function for the MISO system and the /'* loop of the MIMO system. more than one tracking and/or external disturbance inputs and more than one output Multiple-input single-output. 2... Minimum-phase . YI The phase margin angle for the MISO system and for the /* loop of the MEMO system. one or more external disturbance inputs.

is the transfer function relating the /'* output to the j* input for plant 0)0. respectively Sampling frequency MISO plant with uncertainty rnxe MIMO plant matrix where (py).xxii Symbols n.p.e. P Pi = {(Pij) } case i 9 Script cap pee to denote a set that represents the plant uncertainty for J cases in the region of plant uncertainty. is not a square plant matrix and Wv& an Sxm weighting or a squaringdown matrix Quantitative feedback design based on quantitative feedback theory An mxm matrix whose elements are given by Pd Pf P~l = {(/?*).0)0 a>..m.} for a MIMO system mjxm MIMO external disturbance matrix Plant model for a tracking and external disturbance input system which is partitioned to yield Pe and Pd The inverted plant matrix for plant case i where e = m The mxm effective plant matrix when P. £P= {P. Nonminimum-phase NC a/i. i.} p = pjy e ' QFD S = {(<?!/)} . Nichols Chart The symbol for bandwidth frequency of the models for 7^7^ and r={^} Phase margin frequency for a MISO system and for the /'"' loop of a MIMO system.

} Script cap tee in conjunction with P or (c^) denotes a template. 3..) frequency. 3R = {TR{} for the MISO system and 3^..e. R.} for the MISO system and 3C = {(TCiJ ). i.R= {r. respectively Right-half-plane T Sampling time 3Pd'a>. i.e.} RHP Script cap que to denote a set that represents the plant uncertainty for a MIMO system.} for the MDVIO 3D system .Symbols xxiii S. } for the MIMO system The script cap tee denotes the set that represents the disturbance control ratios for J cases.. for a MISO and MIMO plants respectively The desired MISO tracking control ratio that satisfies the specified upper bound FOM The desired MISO tracking control ratio that satisfies the specified lower bound FOM TRu TRL TD The desired MISO disturbance control ratio which satisfies the specified FOM TRt> TDl TI = {(tij)t) The MISO tracking and disturbance control ratios for case i The mxm MIMO tracking control ratio matrix for plant case i Td = {(td. i. -3PJ/®.e. = {(7^ ). 3D = {TD.)i} -35? The mxm MEVIO external disturbance control ratio matrix The script cap tee denotes the set that represents the tracking control ratios for J cases.e.) and 3Q(ja). = The tracking input for a MISO system and the tracking input vector for a MIMO system. i.

{Lm Pmax . where yt.* loop template dB limiting value for a MEMO plant The weighting or squaring-down or mixer matrix w'-domain variable'. the use of u and v must be interpreted in context Y.xxiv Symbols Ttj A set of assigned tolerances on tu where du and b'u and TCiJ are the assigned tolerances for tracking and cross-coupling rejection. respectively. respectively The mxl controller input vector U UHFB v. v V The ultra high frequency boundary The MISO prefilter output and the mxl MEMO prefilter output vector.Lm (q^min} is the . W= {Wjj} w'=u +jv linv-^ {Lm (?.)mar . Y = {y.Lm Pmin} is the dB hmiting value for a MISO plant V.^.= yr _ + yClJ The mxl plant output vector Is that portion of the / * output due to the j* input Is that portion of the /'* output due to cv (cross-coupling effect or interaction of the other loops) y yri ycij ..j} The output of a MISO system and the output matrix of a MIMO system. respectively linia.

efficient robotic assembly lines. A control-theory design approach. 3.1 INTRODUCTION 1-1 INTRODUCTION Automatic control systems permeate life in all advanced societies today. safe. 2. The performance specifications establish the control problem. 1-2 THE ENGINEERING CONTROL PROBLEM1 In general. Such systems act as catalysts in promoting progress and development propelling society into the 21st century. A set of performance specifications is established. 4. Technological developments have made it possible for high-speed bullet trains. and efficient environmentally friendly pollution controls for factories. A set of differential equations that describe the physical system is formulated or a system identification technique is applied in order to obtain the plant model transfer functions. sophisticated civilian and military aircraft. The successful operation of all of these systems depends on the proper functioning of the large number of control systems used in such ventures. comfortable and efficient automobiles. aided by available computer-aideddesign (CAD) packages or specially written computer programs. involves the following: . exotic vehicles capable of exploration of other planets and outer space. a control problem can be divided into the following steps: 1. This development has evolved more complex systems in which nonlinearities need to be more effectively addressed in the design process.

based upon a knowledge of what can be reasonably expected of the plant that is being controlled. The intent of this figure is to give the reader an overview of what is involved in achieving a successful and practical control system design. For example. if the performance of interest is the steady-state response. in many cases the nonlinearity is small enough to be neglected. (c) For plants having structured parameter uncertainty. Performing a simulation of the designed nonlinear system. A basic system has the minimum amount of equipment necessary to accomplish the control function. then the simulation run-time must be considerably longer. during the design process one must keep the real world in mind For instance. (b) If the performance of the original system does not meet the required specifications a control design method is selected that will improve the system's response. one must be able to interpret the results obtained. 6. Design of the system to obtain the desired performance is the control problem.2 Chapter 1 (a) The performance of the basic (original or uncompensated) system is determined by application of one of the available methods of analysis (or a combination of them). Parametric uncertainty is present when parameters of the plant to be controlled vary during its operation. A computer is generally used in the design. This textbook only considers nonlinear systems having structured parametric uncertainty. or the limits of operation are small enough to allow a linear analysis to be used. Another real- . 5. Implementing and testing the actual system. While accomplishing a practical control system design. the simulation run time may need to be only 5 s since by that time a pilot would have instituted a new command signal. in performing a time simulation of an air- craft's transient response to a pilot's maneuvering command to the flight control system. depending upon the complexity of the system. If within this 5 5 window the per- formance specifications are satisfied. However. in performing the simulations. based upon the nonlinear design technique that is chosen. 1. or on the actual system. then it will be deemed that a successful design has been achieved. as explained in Chapter 2. final adjustments can be made in a simulation. is to implement a control system which meets the functional requirements. to take into account the nonlinearities that were neglected.1. the Quantitative Feedback Theory (QFT) design technique may be used. The necessary basic equipment is then assembled into a system to perform the desired control function. After a control system is synthesized to achieve the desired performance. The aspects of this figure that present the factors that help in bridging the gap between theory and the real-world are addressed in the next paragraph. Although most systems are nonlinear. The essential aspects of the control system design process are illustrated in Fig. besides achieving a satisfactory theoretical robust design. the designer must keep in mind that the goal of the design process. In other words.

open and closed-loop Engineering Visualization Used to better understand results of simulations and System tests Design can be broken down by open or dosedloop operating condition Engineering Interactive Simulation •User supplies commands and then can react to resulting dynamic behavior •Gives a better understanding of control system operation Hardware-in-the-Loop Simulation/Implementation •Real-time operation of control algorithm •Noise corrupted measurements available for feedback •Computation cycle time/Sampling Rate •Quantization Error. Included are the operational goals of the controlled system and its operating environment Dynamics Model Mathematical model of the system to he controlled. performance.Introduction Performance Specifications These are the mathematical expressions which represent the functional requirements. Warping Fig. . Control Authority Allocation Algorithm to optimize the use of control effectors for efficiency and decoupling OFT Control System Design Produces control algorithms for test and implementation Linear Simulation Implement and test candidate control algorithm using Simplified Dynamic Robust Stability. 1. and disturbance boundaries a function of frequency System T Tryout of controlled system under actual operating conditions Nonlin ear Simulation •Nonlii ear Functions •Nonlir ear Dynamics Imbedded Performance Specifications Ability to manipulate closed-loop response as a function of frequency 1 C Human in the Loop Simulation •Visual cues •Moving base Concurrent Bode Plots. Functional Requirements This is the problem statement.1 The QFT control system design process: bridging the gap.

4 Chapter 1 world factor is control authority allocation. Once the initial design has been accomplished. This allocation must be based upon a thorough knowledge of the plant that is being controlled and the conditions under which the plant will operate. The designer can use this insight for such things as picking out key frequencies to use during the design process. all of the / loop transmission functions can be plotted on a Nichols chart to analyze the results of applying the designed compensator (controller) to the nonlinear system. The last two elements of the QFT design process that help bridging the gap is the relation of the controlled system's behavior to the frequency domain design and the operating condition. based on knowledge of the real world system. The next element of QFT consist of the design boundaries. the designer uses boundaries plotted on the Nichols chart. In order to be a successful control system designer.1. an individual must be fully cognizant of the role corresponding to each aspect illustrated in Fig. A key element of QFT is embedding the performance specifications. To obtain a successful control design. During the actual loop shaping process. Bridging the gap. the designer may determine that meeting performance boundaries below a certain frequency is not important. and as a tool to compare the nonlinear system with the desired performance boundaries. 1. hardware-in-the-loop and system tests must be performed to check for unmodeled effects not taken into account during the design and implementation phases. the controlled system must meet all of the requirements during simulation and . Linear and nonlinear simulations are very helpful in early evaluation of the controlled system. Another element of the QFT design process. but if the system is to operate in the real word. The size of the templates indicate whether or not a robust design is achievable. These boundaries are only guidelines and the designer can exercise engineering judgement to determine if all the boundaries are critical or if some of the boundaries are not important. rate saturation of the output effectors will significantly affect the achievement of the functional requirements. is enhanced by the transparency of the metrics depicted by the oval items in the interior of the QFT design process. the manner in which the avail- able control power is assigned to the control effectors. as an indicator of potential problems such as nonminimum phase behavior. as illustrated in Fig. then the templates can be used as a metric in the reformation of the control design problem.1. If a robust design is not achievable. These relationships enable the designer to better analyze simulation or system test results for problems in the control design. For example. at the onset of the design process. is the ability to concurrently analyze frequency responses of the J linear-time invariant (LTT) plants that represent the nonlinear dynamical system throughout its operating environment. that is. but it is important to meet the disturbance rejection boundaries below that frequency. Another important element is the creation of templates at various frequencies. This establishes design goals that enhance and expedite the achievement of a successful design. This gives the designer a first look at any areas of the design that may present problems during simulation and implementation. 1. Position saturation and even more dramatically. This gives the designer insight into the behavior of the system.

The solutions to these single-loop problems represent a solution to the MIMO plant. QFT has been developed for control systems which are both linear and nonlinear. and (b) a set 3>= {P} of possible plants which include the uncertainties. QFT provides many metrics that provide the link between the control design process and real world implementation. 1. uncertain multiple-input single-output (MISO) and multiple-input multiple-output (MEMO) plants. QFT quantitatively formulates these two factors in the form of (a) the set •3/i= {TR} of acceptable command or tracking input-output relationships and the set -3c= {TD} of acceptable disturbance input-output relationships. The MIMO synthesis problem is converted into a number of single-loop feedback problems in which parameter uncertainty. It is not necessary to consider the complete system characteristic equation. this is the transparency of QFT. 1. The objective is to guarantee that the control ratio TK = Y/R is a member of 3S and TD = Y/D is a member of 3o. It has also been extended to some classes of uncertain distributed systems (in which the plant is described by partial differential equations) and the feedback and specifications are also distributed The representation of a MEV1O plant with e inputs and m outputs is shown in Fig. . for all plants P which are contained in f >. If the controlled system fails any of the simulation or system tests. using the design elements of QFT. and performance tolerances are derived from the original MIMO problem.2 A MMO plant. 1-3 QUANTITATIVE FEEDBACK THEORY (QFT)2-7-14-18-24-2"2'38 QFT is a unified theory that emphasizes the use of feedback for achieving the desired system performance tolerances despite plant uncertainty and plant disturbances. the designer can trace that failure back through the design process and make necessary adjustments to the design. The QFT synthesis technique for highly uncertain linear time-invariant MEMO plants has the following features: 1. external disturbances.Introduction system test. and for both output and internal variable feedback. Fig. then. time-invariant and time-varying. continuous and sampled-data.2.

3. 1.6 Chapter 1 2. QFT is a frequency-domain technique that is applied in this text to MISO systems. 14 CONTROL THEORY BACKGROUND For an understanding of the analog QFT design technique. Single-input single-output (SISO) linear-time-invariant (LTI) systems 2. to find compensation functions which guarantee that the performance tolerances for each MISO problem are satisfied for all P in £P. MIMO LTI systems. The design is tuned to the extent of the uncertainty and the performance tolerances. The principal properties of QFT and existence theorems are presented along with other important aspects of this technique. 5. . Professor Horowitz's paper Survey of Quantitative Feedback Theory (QFT)52 is an excellent presentation of his robust multivariable control system design technique. MTMO nonlinear systems. Any other comparable textbooks covering similar material (conventional control theory and introductory state-space theory) are satisfactory.e. A good understanding of the material in Chapters 1 through 6 and Chapters 11 through 13 of Ref. Distributed systems. i. The performance specifications for each individual closed-loop system transfer function and for all the closed-loop disturbance response functions must be specified. The amount of feedback designed into the system is then tuned to the desired performance sets 3K and 3D and the given plant uncertainty set &. The objective is to solve the MISO problems. 6. as covered by the first 14 chapters of Ref. SISO nonlinear systems.. Also. This design technique is applicable to the following problem classes: 1. They are rigorously converted to equivalent class 3 systems whose solutions are guaranteed to work for a large class. 13 is required for the study of the discrete-time QFT design technique. it is necessary that the reader have a good foundation of fundamental control theory. Sampled-data systems as well as continuous systems for all of the preceding. Problem classes 3 and 4 are converted into equivalent sets of MISO systems to which the QFT design technique is applied. time-varying and nonlinear uncertain plant sets can be converted into equivalent MISO LTI plant problems to which the MISO frequency-domain technique can be readily applied and where the fundamental tradeoffs are highly visible. 4. These are rigorously converted to equivalent class 1 systems whose solutions are guaranteed to work for a large problem class.

The following sections highlight some of the typical areas of application such as: process control systems.1 QFT AND ROBUST PROCESS CONTROL9 Professor P. Thus. The section at the beginning of this text entitled QFT Standard Symbols & Terminology is consistent with Professor Horowitz's writings. they are manipulated to achieve an appropriate mathematical format.1. i.Introduction 7 LTI mathematical models that represent the region of structured parametric uncertainty are used in the linear analysis presented in this text. wastewater treatment control system. mechanical. This technique helps the designer to spot similarities based upon previous experience. Once a physical system has been described by a set of mathematical equations. the subsequent method of analysis is independent of the nature of the physical system. welding control systems. design of robot controllers.2 IDLE SPEED CONTROL FOR AUTOMOTIVE FUEL INJECTED ENGINE42'44 The following is the abstract from the paper entitled "Robust Controller Design and Experimental Verification of 1C. etc. aircraft propulsion control systems. actuator control system.e. India have tackled nine problems falling into various process control classes which have been successfully solved The specific problems with corresponding problem classes are listed in Table 1. Nataraj and his graduate students of The Indian Institute of Technology. idle speed control for an automotive fuel injected engine. Some of these examples are described in the following sections. V. 1-6 QFT APPLICATIONS The advent of the QFT computer-aided-design (CAD) packages1'13'15-36 at the Air Force Institute of Technology in the late 80's and early 90's accelerated the application of QFT to many real-world problems. 1-5 DEFINITIONS AND SYMBOLS As is the normal procedure in undertaking the study of a new subject matter. it does not matter whether the system is electrical. one quite often desires to refer to the technical literature for a broadening perspective in enhancing his or her understanding of the material. Engine Speed Control" by Dr. 1-6. a most effective way of enhancing this understanding and expediting the state-of-the-art is for all authors on QFT to adhere to a standard list of definitions and symbols. M. aircraft flight control systems. and operational amplifier control systems. Bombay. 1-6.. A Fran- . When this has been done. S.

LTI. LTI.. 1. Dist . 2x2 MIMO (2-DOF) Dist.minimum-phase.Degree-of-Freedom col . nonlinear. 3x3 MMO (2-matrix-DOF) Lumped. 2x2 MIMO. Purdue University. Hamilton. K. SISO. mp (2-matrix-DOF) Lumped. DOF .3 Fuel injected engine.Distributed Fig. 1 System Cascade of 5 CSTRs Problem Class Lumped. LTI .column. nonlinear.Continuous stirred tank reactor(s). mp. 2x2 MMO. mp. nonlinear. SISO (2-DOF) Notation: CSTR . The goal of this investigation is to design a robust feedback .. chek and G. mp (2-matrix-DOF) Lumped.Linear-time-invariant. linear. SISO (2-DOF) Lumped. SISO (2-DOF) bed catalytic reactor Exothermic CSTR Exothermic CSTR Isothermic-fixed Heat equation Dist. SISO (2-DOF) Lumped. linear. Multiple-loop (6-DOF) 2 3 4 5 6 7 8 9 Distillation col. LTI.Chapter 1 Table 1.44 Presented in this paper is the robust idle speed control of a Ford 4. with sidestreams Fluidized catalytic cracker High purity distillation Isothermal CSTR Lumped. LTI.6L V-8 fuel injected engine.1 Process control classes No. School of Mechanical Engineering.

This is the first time feedback control had been successfully applied to pinch welding. Issues complicating the controller design include the nonlinear nature of the engine dynamics. and with 200percent changes in part thickness. Bently. For both applications Mr. An experimental verification of the proposed controller. This facilitates welding assemblies with limited access to the underside of the weld.3. A. Sandia National Laboratories has published the following two papers9: "Arc Welding Penetration Control Using Quantitative Feedback Theory. which is the power element. This control design problem . 1-6. The system also compensates for sharp discontinuities in heat sinking and arc length.3 WELDING CONTROL SYSTEMS Mr. 1. The latter governs the piston.5 to 6 in/min (0. "Pinch Weld Quality Control Using Quantitative Feedback Theory.64-2. In this application the mechanical feedback link commonly used in actuators is opened and replaced by a sensor and an electronic controller which drives the valve. The authors show in their paper that they met all the design objectives and have achieved excellent results. A correlation between weld quality and electrode displacement is established for constant force. utilizing the nonlinear plant.Introduction 9 controller that maintains the idle speed within a 750 rpm tolerance of about 600 rpm despite a 20 Nm step torque disturbance delivered by the power steering pump. Process Development and Fabrication Division." involves a control system that is designed for arc weld penetration control. The controlled input is the by-pass air valve which is subjected to an output saturation constraint. E. 1-6. and the saturation constraint of the by-pass air valve. the induction-to-power delay of the manifold filling dynamics. Welds of constant penetration have been demonstrated in tests with travel speeds varying from 1. The system is sensitive enough to use a fiber-optic cable for transmitting the light from the weld to the sensor. Bently met all the design objectives and achieved excellent results. The feedback signal is obtained by measuring the amount of visible and near-infrared light emitted from the back side of the weld. is included. The authors' control system is based on controlling an engine such as the one shown in Fig. The paper entitled. The QFT designed control system is capable of producing repeatable welds of consistent thickness (and thus consistent quality).4 CONTROL SYSTEM FOR AN ACTUATOR PLANT45 Actuators can be regarded as feedback control systems in their own right.54 mm/sec). with wide variations in weld parameters." involves a system that is based on electrode displacement feedback that greatly improves the quality control of the pinch welding process.

flight control engineers have taken a conservative.4 Hydraulic actuator. thus reducing the number of transitions required. 1. and (2) the manufacturing tolerances of actuators. within and along the border of the flight envelope plot were selected. see Fig. 1. such that when an actuator needs to be replaced. Second. These two factors result in the structured parametric uncertainty of the hydraulic actuator. individual compensator designs were accomplished for each of these points. that: (1) takes into account the aging of some of the actuator components over its expected lifetime. which for this design were points representing airspeed vs altitude. . 1-6. the design process can be made more efficient. 1.4. Third. Fig.5 was accomplished by an Air Force Institute of Technology student who is an F-16 pilot. brute force approach to designing a full envelope PCS for an aircraft. Traditionally. The resultant QFT design of the actuator's control system achieved the desired degree of robustness. the overall control system's robustness is maintained by the replacement. smooth transitions between these compensators must be engineered Making the transitions imperceptible to the pilot is very difficult and time-consuming because each airspeed-altitude design point can be approached from an infinite number of initial conditions. Obviously. if the number of the design points can be reduced. and the resulting PCS is less complex.5 VISTA F-16 FLIGHT CONTROL SYSTEM (INCLUDING CONFIGURATION VARIATION)46 The design of the robust flight control system (PCS) for the VISTA F-16 of Fig. many design points. First. He was able to utilize his real-world knowledge of the aircraft and its handling qualities to achieve the desired robust PCS.10 Chapter 1 entails the design of a robust controller for an actuator.

In summary. hence. actuator saturation or . "real-world" problems have been dismal failures. it does give the designer complete control over the gain and the order of the compensator. full envelope design.e. Either the compensator is of too high order. any sensor noise present is accentuated by this gain. the resulting compensator is impractical to implement. A robust control design technique that overcomes the aforementioned pitfalls is the Quantitative Feedback Theory (QFT) design technique.5 VISTA F-16 A way to reduce the number of necessary design points is to apply a robust control design technique to the problem. many previous attempts at applying robust control design algorithms to practical. time-consuming endeavor. QFT offers the ability of incorporating enough robustness to simplify the design process and the resulting FCS. or its gain is too large to accommodate "real-world" nonlinearities. limiting robustness in FCS design results in a convoluted. if a decision is made to decrease or limit the order or gain of a compensator. if not the whole. Also. 1. the synthesis of those FCS's has been a costly. Unfortunately.Introduction 11 Fig. but not so much robustness that the resulting FCS is impractical to implement due to violation of physical limitations imposed by the "real-world" (i. flight envelope.46 This is because. the performance tradeoffs due to this action can be clearly seen by the designer.. complex. Thus. In addition. Although a QFT design effort could very easily result in a compensator of high order and of high gain. The typical reason for these poor results is that the robust design is synthesized in the essentially noiseless world of the digital computer. although excellent FCS's have been designed for aircraft using traditional design methods. A compensator synthesized using robust control principles should be able to handle large parts of. and then validated on the digital computer through the use of small signal linear simulation. QFT is not constrained to produce an impractical compensator. although the problem is well posed.

Also. The proposed design incorporated load alleviation concepts to reduce the load on the wings.6. A simulation47. the model-based QFT (MBQFT) design technique is applied which resulted in controllers that are implemented by a series of simple backwards difference equations. This robust design technique increased tracking accuracy by up to a factor of four over the model-based controller performance baseline. 1-6. or design the lateral/directional flight control system cognizant of the wing's flexibility. Two avenues of approach are possible: stiffen the wings. while at the same time meeting military specifications for roll maneuvers.variable control system design technique is required which solves the payload and dynamics uncertainty. QFT has the feature of utilizing the control system designer's knowledge of the "real-world" characteristics of the plant.an F-16 pilot. It illustrated the benefits of designing flight control systems with the QFT robust control system design technique in contrast to the brute force approach of optimizing a flight control system for performance in expected configurations and then scheduling the gains.6 DESIGN OF FLIGHT CONTROL LAWS FOR AIRCRAFT WITH FLEXIBLE WINGS USING QUANTITATIVE FEEDBACK THEORY48 Aircraft composed of lightweight composite materials are extremely enticing since their structural weight is greatly reduced. thus avoiding adverse wing twisting. The additional degree of freedom afforded by the availability of redundant control surfaces and the optimal and coordinated use of all control surfaces allowed for the reduction of the load on the wings. was performed on the Lamars Simulator by the PCS designer . Payload invariance is a necessary component of human arm emulation. involving the nonlinear plant. the control of these aircraft is complicated by the resultant flexibility of the wings. thus losing some of the weight reduction benefits. 1. during the on-going design process in maximizing the ability to achieve the desired robust system performance. Model-based controllers require accurate knowledge of payload and drive system dynamics to provide good high speed tracking accuracy. and met the military specifications. A robust multi. etc. as verified by a nonlinear time simulation. However. The MBQFT tracking perform- . The design incorporated a weighting matrix to distribute generalized aileron and rudder commands to the five control surfaces available on the F-18. MBQFT high speed tracking accuracy was experimentally evaluated on the first three links of the PUMA-500 of Fig.7 ROBOT CONTROLLERS35 The ultimate objective in robotic arm control research is to provide human arm emulation. Thus. The excellent performance in these simulations demonstrated the viability of a QFT design approach in producing flight worthy aircraft control systems.12 Chapter 1 sensor noise amplification). The second approach was used in a QFT design of a lateral/directional robust flight control system for the sub-sonic flight envelope of an F-18 aircraft. 1-6.

To examine this paradigm. ance was robust to both unmodelled drive system dynamics and payload uncertainty.8 OPERATIONAL AMPLIFIERS (OP-AMP)49 The QFT robust design methodology was examined. 1. in order to develop an automated design approach for the analog domain. 1-6. the structured parametric uncertainty involving 741 (BJT) Op-Amp was successfully . The nonheuristic nature of the MBQFT design and tuning should allow application to a wide range of manipulators. applied.6 Robot arm. and incorporated into the operational-amplifier (Op-Amp) paradigm.Introduction 13 Fig.

A very important characteristic of mxm MIMO control systems is Influent Internal recycle Ml ^— D rP 0 : •£— br^SNtf ° 0 Effluent SNO i 0 °° 0° Elimination v^-- o SNH Elimination „ i -SNO i -DO Facultative volumes ^ ^T j Air flow -J-*a^||pS3s»" j Sludge blancket Wastage rate Sludge recycle Fig. etc. The second part consists of Chapters 9 through 10 which focuses on bridging the gap between theory and the real -world. The first part. 1-6. such as simulation. 1.. Thus. The controller developed is also suitable for low-cost microcomputer implementation. The QFT robust design methodology provided the insight for the design of a compensated Op. design overview. performance specifications. The first part of the text deals with presenting an overview of QFT: the design objectives. controlling the maximum concentration of pernicious substances. The control system design resulted in an unproved performance of the plant because the concentration levels obtained are nearer to those required by environmental law and a notable reduction in the running costs is produced.7. 1-7 OUTLINE OF TEXT The text is essentially divided into two parts. .) and the factors that are involved. This is accomplished by presenting Engineering Rules (E. 1. and QFT basics. what is structured parametric uncertainty and its Bode plot and Nichols chart representations. implementation. present the fundamentals of the QFT robust control system designs technique for the tracking and regulator control problems.7 Wastewater treatment plant. is to protect the water environment from negative effects produced by residual water.9 WASTEWATER TREATMENT CONTROL SYSTEM50 One of the most important objectives of a wastewater treatment plant (WWTP). A computer simulation of the QFT designed WWTP compensated control system met the desired performance specifications. shown in Fig.R. that are important in achieving successful and implementable control system designs.Amp which satisfied the closed-loop performance specifications. consisting of Chapters 2 through 8.14 Chapter 1 compensated by the QFT compensator design technique. the operation of the plant is notably more efficient.

they rely on. Select the system design which has the minimum complexity. The book uses the QFT CAD packages (see Apps.. To an extent depending on their own background and experience. This property is developed and is followed by a presentation of the fundamentals of the QFT technique and the associated design procedure.) that contribute to the transparency of the design process. plant matrices. 3. The second portion of the text focuses on bridging the gap between theory and the real world by presenting Engineering Rules (E. Compare the cost with the performance and select the fully justified system implementation. The following design policy includes factors that are worthy of consideration in the control system design problem: 1. Use proven design methods. 4. In closing this introductory chapter it is important to stress that feedback control engineers are essentially "system engineers. Use minimum specifications or requirements that yield a satisfactory system response. that are important in achieving a successful and implementable control system design. Perform a complete and adequate simulation and testing of the system. engineers in the various recognized branches of engineering to furnish them with the operating scenario of the plant and the transfer functions.. people whose primary concern is with the design and synthesis of an overall practical robust control system.e. . etc. and work closely with. implementation. such as simulations." i. 2. and/or system equations of various portions of a control system. A-C) for the examples contained in the text. The remaining portion of this part of the text is devoted to the extension of the QFT technique to MIMO tracking and regulator control systems. These CAD packages have been developed to expedite the analysis and design of robust control systems containing structured parametric uncertainty. This is followed by the factors which are involved. The applications include tracking and regulator control problems for MISO analog and discrete-time systems.R.Introduction 15 that they can be represented by m2 MISO equivalent systems.

and to the benefits of QFT. 2-2 WHY FEEDBACK?5132 To answer the question "Why do you need QFT?" consider first the following system.1 An open-loop system (basic plant). This compensated system produces the desired output as long as the plant D. 2. First presented is a discussion of why feedback is needed to achieve performance goals. a design overview. This is followed by a QFT overview of design objectives. as shown in Fig. and QFT basics and design. 16 . The plant P(s) responds to the input r(t) [R(s)] with the output y(f) [Y(s)] in the presence of disturbances d. If it is desired to achieve a specified system transfer function T(s) [= Y(s)/R(s)] then it is necessary to insert a prefilter. The last few sections are devoted to the insight provided by the QFT technique. 2.(t) [D.2.(s)] and dAf) \D2(s)\ (see Fig. an explanation of structured parametric uncertainty. whose transfer function is T(s)/P(s).(s) D2(s) R(s) A >o—*- P(s) J VC O—^ * VL Fig.INTRODUCTION TO QFT 2-1 QUANTITATIVE FEEDBACK THEORY This chapter is devoted to presenting an overview of QFT in order to enhance the understanding and appreciation of the power of the QFT technique which is presented in the following chapters. 2.1).

b. respectively: . This last item is important in order to avoid problems associated with high loop gains such as sensor noise amplification. does not change. see Fig. Maintains reasonably low loop gain (i.. and the disturbances are reflected directly into the output. Thus. 2. This type of system is sensitive to changes in the plant (or uncertainty in the plant). are. 2. k p Fig. 2.4a. 2.3 A closed-loop system.2 A compensated open-loop system. and there are no disturbances.e. that represents a plant with variable parameters. Incorporates information on the desired output tolerances. 2. ff. based upon the nominal plant P0 e 9.Introduction to QFT 17 R Fig. there is no plant uncertainty. Consider the control system of Fig.4a. Addresses at the onset all known plant variations. In designing a feedback control system. and high frequency uncertainties. This system has two inputs: r(t) the desired input signal to be tracked and d(t) an external disturbance input signal which is to be attenuated to have minimal effect ony(f). containing a plant uncertainty set. it is desired to utilize a technique that: a.3. The tracking and disturbance control ratios of Fig. it is necessary to feed back the output in order to reduce the output sensitivity to parameter variation and attenuate the effect of disturbances on the plant output. saturation. reduces the "cost of feedback"). c.

respectively: T = .IS Chapter! Y(s) ———— D(S) (2.4) (2.1) 7(5) —— (2. i. 2.5) 1 + GP0 1 + L0 D R 4 (a) Y D R D—— » ^ G ^0^ 9 (b) ^ 1 Fig.4b the tracking and disturbance control ratios are.: (2. 2.4a for the two cases: YK(s) \jft) =0 and YD(s) \r(t). GP0 L0 (2.4 Control systems: (a) open-loop uncompensated sytem.3) For the compensated system of Fig. (b) closed-loop compensated system.2) The sensitivity functions1 of the open-loop uncompensated system of Fig. .0 are identical.e. 2.

5. 3.Introduction to QFT 19 where L0 = GP0 is defined as the nominal loop transmission function. i.e. For the compensated system of Fig.4b.1 A Simple Example To illustrate 'What is Parametric Uncertainly?" consider the undergraduate laboratory experiment that involves hooking up the d-c shunt motor of Fig. The weekend room temperature was set at . 2. Further. (2. 2-3.2 STRUCTURED PARAMETRIC UNCERTAINTY: A BASIC EXPLANATION 2-3. The order of the compensator (controller) G can be minimized by incorporating1'41 the nominal plant P0 into L0 The reader is referred to the QFT literature for a more detailed analysis of the sensitivity function with respect to sensor noise and plant parameter uncertainty on system performance.: j Pod) P. A practical optimum design requires working to the limits of the system's performance specifications.3) readily reveals that the effect of changes of the uncertainty stiff (s) upon the output of the closed-loop control system is reduced by the factor 1/[1 + GP0 ] compared to the open-loop control system.W 1 + GPo j i +Lo Comparing Eq.3'6'40'54 2-3 QFT OVERVIEW2-3 2-3. consider that the students entered the laboratory on a cold January Monday morning to perform this experiment.2. The reasons52 for the choice of L0 are: 1. This reduction is an important reason why feedback systems are used. 2. The sensitivity function is very sensitive to the cost of feedback 2. (2.1 QFT DESIGN OBJECTIVE Design and implement robust control for a system with structured parametric uncertainty that satisfies the desired performance specifications. Horowitz has shown52 that a robust control system design is best achieved by working with L0 and not with the sensitivity function S.6) with Eq. the sensitivity functions for these two cases are also identical.

m.20 Chapter 2 50° F but was reset to 70° F when the students entered the room.p. Why the change in speecH • Due to the heating of the d-c shunt field by the field current . . with no adjustments to the applied voltage or of the field rheostat Rf. The students hurriedly hooked-up the motor and set the field rheostat to yield a speed ca = 1200 r. • Which in-rum decreased the value of // and in-turn resulted in the increase in speed since speed is inversely proportional to //./?/ will be at the instant a command is given to the system. Nonlinear Plant Vf Fig.5 d-c shunt motor.R/can vary • anywhere within the range Rfnin <Rf< R^ due to the variable environmental temperature and field current. the value of R/ increased.m. assuming Vf is constant. the parameter . • As a consequence./and the environment. there is uncertainty as to what the actual value of the parameter. during the operation of the motor. 2. • Therefore.p. Upon their return they found that the speed of the motor was now 1250 r. Upon accomplishing this phase of the experiment they took a one hour break hi order to allow the room to reach the desired temperature of 70°.

2 A Simple Mathematical Description The transfer function of a d-c servo motor. over the following range: K e (Kmin Kmax) and a <E (amia a^.. (i = 1.6. the plant parameter variations are described by Fig. 2..2. 2. Over the region of operation. is: Ka (2. utilized as a position control device. The Bode plots for these 6 LTI plants are shown in Fig.2. 2. Region of plant parameter uncertainty Fig.7) Vf(s) where the parameters K and a vary. in a position control system..Introduction to QFT 21 Thus.6 Region of plant parameter uncertainty. due to the operating scenario. The motor can be represented by six LTI transfer functions P. The shaded region in this figure represents the region of structured parametric uncertainty (region of plant uncertainty).. the parametric uncertainty is structured because the range of the variation of R/ is known and its effects on the relationship between V/snd CD can be modeled 2-3.J) at the points indicated on the figure.7. .

2. peak time. Bu and BL. The conventional time-domain figures of merit.8b. Assuming that the control system has negligible sensor noise and sufficient control effort authority. and /.. peak overshoot Lm Mm. 2. tp. as shown in Fig.8a. 2.8a.p. then for a stable linear-time-invariant (LTI) minimum-phase (m. are shown in Fig.3 CONTROL SYSTEM PERFORMANCE SPECIFICATIONS In many control systems the output y(t) must he between specified upper and lower bounds. y(t)u andy(f) L . 2-3.p.) plant a LTI compensator may be designed to achieve the desired control system performance specifications.Plant 4. /„ rise time.7 Bode plots of 6 LTI plants: the range of parameter uncertainty. settling time. and the frequency bandwidth Wh which are shown in Fig.5 -150 -160 -170 -180 I I I I I J___II 10 Frequency («) Fig. peak overshoot. 2. Corresponding system performance specifications in the frequency domain are. The case of nonminimum-phase (n. the upper and lower bounds respectively.) plant is discussed in Chapter 9.22 Chapter 2 -40 -60 10 Frequency (w) -90 -100 -110 I -120 I -13° S -140 01 ————— Plant 1. based upon a step input signal r(t) = R0 u. .6 — — — .m.2 ————— Plant 3.(f). They are: Mp. respectively.

(b) frequency domain response specifications.8 Desired system performance specifications: (a) time domain response specifications. 2. . 2-3. Bandwidth (b) Fig. • Using these representations to design a compensator (controller). 2. • Representing the nonlinear plant characteristics by a set of LTI transfer functions that cover the range of structured parametric uncertainty.8) by LTI transfer functions that form the upper Bv and lower BL boundaries for the design.4 QFT DESIGN OVERVIEW The QFT design objective is achieved by: • Representing the characteristics of the plant and the desired system performance specifications in the frequency domain.Introduction to QFT 23 (a) LmT. • Representing the system performance specifications (see Fig.

SP(jo).7. It represents the region of structured plant parametric uncertainty on the NC and are obtained for specified values of frequency. 9> = (Pi(s)} (t = 1. as shown in Fig. Six data points (log magnitude and phase angle) for each value of ».1 la.. A contour is drawn through the data points. and cross-coupling (for MTMO systems) boundaries on the Nichols chart in order to satisfy the performance specifications. Obtaining the stability.6. • J data points (log magnitude and phase angle).. 2. 2..24 Chapter! Reducing the effect of parameter uncertainty by shaping the open-loop frequency responses so that the Bode plots of the J closed-loop systems fall between the boundaries Bv and BL. To accomplish a QFT design: • The nonlinear plant is described by a set of J m. 2. LTI plants. are plotted on the Nichols chart (NC). 2-3..1 Ib. 2. where G(s) is a compensator.6 and 2.9. • The magnitude variation due to the plant parameter uncertainty..p. while simultaneously satisfying all performance specifications. for each value of coj. 2. This contour is referred to as a template. 2. . i. for the example shown in Fig. F(s) is a prefilter. which define the structured plant parameter uncertainty. for each value of frequency. and ^is the nonlinear plant. to plot the templates. within the bandwidth (BW) of concern.).e. are obtained. J). disturbance.p.2.p.. for cot. co = cat. 2. is depicted by the Bode plots of the LTI plants as shown in Fig. tracking. plants.m.5 QFT BASICS Consider the control system of Fig. that described the boundary of the region that contains all J points. co = COL. as shown in Fig.9 Compensated nonlinear system. Note: for MIMO systems the elements ptj of the mxm plant matrix Pcan be n.10 for the example of Figs.52 for MISO systems the discussion is restricted to m. )— G(s) I L Y 9 ^ v Fig.

2.12.8. 2. 2.) of Fig. and their corresponding Bode plots are shown in Fig. respectively. 2.J and BL. 2. • The system performance specifications are represented by LTI transfer functions. 2.12 as shown in Fig. the QFT robust design technique assures that the desired performance specifications are satisfied over the prescribed regi on of structured plant parametric uncertainty. 2-3. 2.). 2. and • results in the & (jot) of Fig. 2. of the compensated system.13. • results in the closed-loop frequency responses TLi shown in Fig.6 QFT DESIGN The tracking design objective is to (a) Synthesize a compensator G(s) of Fig.Introduction to QFT 25 Lm 5P(ja>2) 5p(jco3) 5P(jco4) dB Fig.9 that • Results in shifting and reshaping the TL[ responses in order that they lie within the Bv and BL boundaries in Fig. 10 by the upper and lower bounds B. that is: SL (jWi ) < SR (b) Synthesize a prefilter F(s) of Fig. being equal to or smaller than SJjco. Therefore.10 LTI plants.12. . 2. for each value of «% of interest.10 for the uncompensated system and that it is equal or less than SR{) 'co.9 that • results in satisfying the desired performance specifications of Fig.

(c) construction of the Nichols chart plant templates. . 2.26 Chapter 2 6 5 Phase (decrees) 6 5 Phase (deyees) (b) -O <u a 1 oj CO. 3 rad/sec. C06> C0 5 > . (b) template construction for ». Phase (c) Kg. CO.11 (a) Bode plots of 6 LTI plants.

13 Closed-loop responses: LTI plants with G(s) and F(s).1 OPEN-LOOP PLANT Consider the position control system of Fig. 2-4 INSIGHT TO THE QFT TECHNIQUE 2-4. 2.8) .Introduction to QFT 27 dB Fig. whose plant transfer function is given by Ka „ K' s(s+a) s(s+a) (2. 2. 2.9.12 Closed-loop responses: LTI plants with G(s). Lm dB Fig.

Once the proper loop shaping of L0(s) = G(s)P0(s) is accomplished.' = Ka and i = 1. the nominal loop transmission function is Lm L0 = Lm GP0 = LmG + LmP0 (2.13) ..For the nominal plant P0(jaj).9 is (2. and the selection of a nominal plant P0 from the J LTI plants.11) 2-4. 2-4..28 Chapter 2 where K. P(/a>). the loop transmission function is = LmGP=LmG + LmP (2.. The loop transmission L(s) is defined as (2.7. The last step of this design process is the synthesis of the prefilter that ensures that the Bode plots of TRi all he between the upper and lower bounds Bu and BL. J.4 INSIGHT TO THE USE OF THE NICHOLS CHART (NQ IN THE QFT TECHNIQUE3 This section is intended to provide the reader a review of the use of the NC and an insight as to how it applies to the QFT technique. The log magnitude variation due to the plant parameter uncertainty. (1) Open-Loop Characteristics ..(s) ' (2. for J = 6.12) whereas for all other plants.2. a synthesized G(s) is achieved that satisfies the desired performance specifications. 2.10) "' RL 1+ A The overall system control ratio Tx is given by: T (S) = F(s)L. 2.3 RESULTS OF APPLYING THE QFT DESIGN TECHNIQUE The proper application of the robust QFT design technique requires the utilization of the prescribed performance specifications from the onset of the design process. is depicted by the Bode plots in Fig.2 CLOSED-LOOP FORMULATION The control ratio TL of the unity-feedback system of Fig.9) 2-4.

from the plot of Lm L0(j(o) vs.15) and (2.ZP0 (2.) + LmP0(JcOi) + *(/'». 3P(ja>j). . given by Eq. the variation S^jco.17) (3) Parametric Variation NC Characteristics .9) G = IZO°.) = LmG(jco.14 • Consider point A on the template represents the nominal plant P =9>0(j(al). over the range of plant parameter variation is described by the template 3P(jco.) shown in Fig. ZL0 shown on the NC in Fig.Lm L0(ja>. where R(J6>) (2. 2. (2.and a-contours.13) to yield Lm L(jw. the closed-loop frequency response data may be obtained for plotting M0 and avs. given by Eq.As an example.(ZG + ZP0) = ZP .) in Lm L(jco. is substituted into Eq.15) • and in a vertical translation in the log magnitude value of P. its corresponding closed-loop frequency response is .Introduction to QFT 29 Thus. This template represents a region of plant parameter uncertainty for o)i as expressed mathematically by Eqs. co is shown. (2. InFig. consider that for Eq. C. for a given G(jco) and Pa(jco).) .151 aplot ofM0 vs.14 represents Lm P0 vs.14) • Translations are shown hi Fig.14) The expression Lm P\jco^ = Lm P0(jco.) (2.15) (2.The closed-loop system characteristics can be obtained.Lm P0(jca^ and its phase angle variation is given by ZAfl(/a%) = ZZ. From the loop transmission plot and its intersections with the M.) + Sp(ja)i). (2. obtained from Eq. (2.14). • Thus. (2.16) (2) Closed-Loop Characteristics . 2. then a variation in P results in: • a horizontal translation in the angle of P.ZZ0 = (ZG + ZP) . • Variation <3>(/oO of the plant. 2.7. from the nominal value for co = cot. and D. 2.a). 2.14 at points B. for co = cot.) is given by dp(jati = Lm L(fc^) .141 Also shown is a plot of a template. (2. 2.16). ZP0.) = Lm Pdco. whose contour is based upon the data obtained for co = cOj from Fig. If point A on the template in Fig. and in turn L(jcoi).

(j(o)'.)]^ = Lm a .30 Chapter 2 1*0* -16fl" -149* -125* -100" -SO* -SO* -46* -20* 0" Fig.) at co = coh i. LmMA=Lmj3 = For P(j(o)= P. 2.. point C in Fig. Note that point A represents the minimum value of Lm M(jo). Lm Tmin = [Lm M(jcu. 2..e. Lm Tnua = [Lm MQa).14 Nominal loop transmission plot with plant parameter area of uncertainty . the point C represents the largest value of Lm M(jco. its corresponding closed-loop frequency response is Lm Mc = Lm a = -2 dB These values are plotted in Fig.)]min = Also.) at co = «y.15.14. 2.

operating envelope. This graphical description of the effect of plant parameter uncertainty on the system's performance is the basis of the QFT technique.15 Closed-loop responses obtained from Fig.) for each value of frequency co = ». • This procedure is repeated to obtain the maximum variation S^w. • There can be one robust design for the full.) dB LmT min Fig.18) • When L0(s) is properly synthesized. with a range of plant parameter variation described by the template. • Design limitations are apparent up front and during the design process. 2-5 BENEFITS OF QFT The benefits of the QFT technique may be summarized as follows: • It results in a robust design which is insensitive to structured plant parameter variation. . 2.(-6) = 4 dB (2.14.Shown in Fig. 2. midway between points A and C.15). • From this figure it is possible to determine the variation in the control system's figures of merit due to the plant's parameter uncertainty.(/#0 ^ <5?(/'fi*). that corresponds to point E in Fig.14 which lies within the variation template 3P(ja%).15 is point E. 2. then <5. the maximum variation in Lm M. denoted by SL(j(a. according to the QFT design technique.Introduction to QFT LmT(ja>) M 31 0 -2 -4 -6 Lm M0 (j<o) = Lm Tmin (jo. Thus.). 2.Lmfi = -2. 2. within the desired B W (see Fig. for this example is SL(JO)i) = Lma .

2-6 SUMMARY The purpose of this text is to present the concepts of the QFT technique in such a manner that students and practicing engineers can readily grasp the fundamentals and appreciate its transparency in bridging the gap between theory and the real world. This is accomplished by limiting the scope of the presentation. to minimum-phase (m. • If necessary.p. Horowitz entitled Survey of Quantitative Feedback Theory (QFT)52 . read the excellent paper by Professor I. there is less development time for a full envelope design. such as existence theorems and nonminimum-phase (n.) plants. one can redesign for changes in the specifications quickly with the aid of the QFT CAD package (see App.p. For a more theoretical discussion of QFT. the reader is referred to the literature listed in the reference section of this text.m. • The achievable performance specifications can be determined early in the design process. for example. It is highly recommended that the reader. • As a consequence of the above benefits.) systems. • The structure of the compensator (controller) is determined up front. A). M.32 Chapter! • Design limitations are apparent up front and during the design process. at the conclusion of reading this text.

This chapter builds upon the introduction to QFT that is presented in Chap. In Chapter 5 the development of a suitable mapping is presented that permits the analysis and synthesis of a MIMO control system having m inputs and m outputs by a set of nf equivalent single-loop multiple-input singleoutput (MISO) control systems. 2. respectively. and a single output (a MISO system). a tracking and an external disturbance input. the external disturbance rejection problem. and system performance tolerances are derived from the original MIMO problem.) performance specifications is presented The remaining portion of the chapter is devoted to an in-depth development of the QFT technique as applied to the design of robust single-loop control systems having two inputs. The design is tuned to the extent of the uncertainty and the performance tolerances. This is accomplished by first presenting an introduction of the QFT technique using the frequency-response method as applied to a single-loop MISO system and an overview of the design procedure. 33 . It is devoted to presenting an in-depth understanding and appreciation of the power of the QFT technique. Chapter 8 presents the analysis and design of a MIMO control system with external disturbances. It is not necessary to consider the complete system characteristic equation.THE MISO ANALOG CONTROL SYSTEM 3-1 INTRODUCTION52 The multiple-input multiple-output (MIMO) synthesis problem is converted into a number of single-loop feedback problems in which parameter uncertainty. 5 through 7. The solutions to these single-loop problems represent a solution to the MIMO plant.e. i. crosscoupling effects..1 The in-depth development of the QFT technique as applied to the design of robust MIMO control systems for tracking inputs is presented in Chaps. Next a discussion of minimum-phase (m.p.

1 has two measurable quantities.2. G is the cascade compensator. The compensator G in Fig. 3. 10] and a is in the range [-2. then it represents a four DOF structure. 2].3) Fig. 3. r(f) and y(f). it is referred to as a two degree-of-freedom (DOF) feedback structure.34 Chapters 3-2 THE QFT METHOD (SINGLE-LOOP MISO SYSTEM) The feedback control system in which 3* represents the set of transfer functions which describe the region of plant parameter uncertainty. J (3.1 A feedback structure. The output y(t) is required to track the command input r(t) and to reject the external disturbances dt(t) and d£f)..1 The plant transfer function is P(s) = (3. If the two disturbance inputs are measurable. The region of parameter uncertainty of the plant is illustrated by Fig...1 is to be designed so that the variation of y(t) to the uncertainty in the plant P is within allowable tolerances and the effects of the disturbances rf/(r) and d2(t) on y(f) are acceptably small. the prefilter properties of F(s) must be designed to the desired tracking by the output XO of the input r(f). The uncertainty of the disturbance is denoted by a set of plant disturbances (3.2. The uncertainty of the plant transfer function is denoted by the set where i = 1. Also. .1.1. Example 3.1) and is illustrated by Example 3. Since the control system in Fig. 3.. 3. The actual design is closely related to the extent of the uncertainty and to the narrowness of the performance tolerances. 3. and Fis an input prefilter transfer function is represented in Fig.2) s(s + a) where the value of A' is in the range [1.

2 Plant parameter uncertainty. 3.) can be determined 3-3 DESIGN PROCEDURE OUTLINE The basic design procedure which is to be followed in applying the QFT robust design technique is outlined in this section. and thus. i.e. the principal challenge in the control system design is to relate the system performance specifications to the requirements on the loop transmission function L(s) = G(s)P(s) in order to achieve the desired benefits of feedback. This outline enables the reader to obtain an overall perspective of the QFT technique at the onset. The advantage of the frequency domain is that L(s) = G(s)P(s) is simply the multiplication of complex numbers.. for all P e £?and all D e 3). respectively.M1SO Analog Control System 35 Paramter space 9 Fig. and the acceptable closed-loop transmittances (control ratios) are denoted by (3. The following sec- .4) The design objective is to guarantee that TR(s) = Y(syR(s) and TD(s)=Y(s)/D(s) are members of the sets of acceptable 3R and 3D. at each «$. the optimal bounds on L(J(o. the desired reduction in sensitivity to plant uncertainty and desired disturbance attenuation. In the frequency do- main it is possible to evaluate L<JOJ) at every coi separately. In a feedback system.

Step 10: Synthesize the nominal loop transmission function L0(s) = G(s)P0(s) that satisfies all the bounds and the stability contour. Step 11: Based upon Steps 1 through 10. based upon the system's desired performance specifications in the time domain. Step 5: Select the nominal plant transfer function P0(s). The design procedure to accomplish this objective is as follows: Step 1: Synthesize the desired tracking model. F(s)G(s)P(s) F(s)L(s) T (s) = _AjL_LZ_AjL = _LZ_1Z = F(s)T(s) l + G(s)P(s) l + L(s) with d. 3.(0 = 0 (3. The following sections illustrate this design procedure. 3. respectively.6) TD 2 = — — — — — — = — — wttr(0 = rf. For the minimum-phase (m. and optimal bounds on the NC. (t) = 0 (3.p.1 so that the specified robust design is achieved for the given region of plant parameter uncertainty.) LTI MISO system of Fig. Step 2: Synthesize the desired disturbance model.36 Chapters lions present a detailed discussion which is intended to establish for the reader a firm understanding of the fundamentals of this technique. (t) = d. synthesize the prefilter F(s). Step 12: Simulate the system in order to obtain the time response data for each of the J plants. Step 3: Specify the J linear-time-invariant (LTI) plant models that define the boundary of the region of plant parameter uncertainty. Step 6: Determine the stability contour ([/-contour) on the NC.1 the control ratios for tracking and for disturbance rejection are. Steps 7-9: Determine the disturbance. tracking. The objective is to design the prefilter F(s) and the compensator G(s) of Fig. (0 = 0 (3. Step 4: Obtain plant templates. 3-4 MINIMUM-PHASE SYSTEM PERFORMANCE SPECIFICATIONS In order to apply the QFT technique it is necessary to synthesize the desired or model control ratio.7) 1 + G(s)P(s) 1+ L .5) P(s) P Tn1 = —————— = ——— 1 + G(s)P(s) 1+ L with r(t) = d. that pictorially describe the region of plant parameter uncertainty on the Nichols chart (NC). at specified frequencies.

any desired frequency bandwidth (BW) is achievable whereas for a n.)\ needs to be satisfied for a satisfactory design.) must also be specified and satisfied in the design process. The case of n. The time responses XOu andXOi. 3. plants is discussed in Chap.3. i. Thus. Note that for m.p. plants.4b) of TRu.p.m. 1. respectively.p. an acceptable response y(f) must lie between these bounds. 3. only the tolerance on \TR(jo). co are shown in Fig. Km) and over-damped (ta tn Km) responses. plants the BW that is achievable is limited The modeling of a desired transmittance T(s) is discussed in Ref. maximumlOT TD! =Lm Op. plants. plants.3 System time domain tracking performance specifications.p. so that Sdjcoft increases as Wj increases above the 0-dB crossing frequency <ocf (see Fig. 3-4.m. tolerances on Z T^jco.3. for m.i(t). This characteristic of Fig.p. This text deals only with m. . tp 4 tr.m. 8. of the tracking performance specifications.MISO Analog Control System 37 Note that for TDp the specified maximum value \y(tp)\ = Op.1 TRACKING MODELS The QFT technique requires that the desired tracking control ratios be modeled in the frequency domain to satisfy the required gain Km and the desired time domain performance specifications for a step input. that is. in this figure represent the upper and lower bounds. It is desirable to synthesize the control ratios corresponding to the upper and lower bounds TRu and TRL. the system's tracking performance specifications for a simple second-order system are based upon satisfying some or all of the step forcing function figures of merit (FOM) for under-damped (Mp. It should be noted that.e. 3. For n. respectively. due to dj(f) = u.p.. respectively. These are graphically depicted in Fig. plants. The Bode plots of the upper bound BU and lower bound BL for Lm TR(jaj) vs. is often used as the disturbance model specification.4.

(3.8) can be represented by an equivalent unity-feedback system so that . This characteristic of SR(J(O. (3.) decrease in magnitude as at increases (see Sec. = Lm TB = Lm B Augmented Model- -60 dB/dec (b) Fig. (b) The augmented models.8) where wn2 =p&2 and ts&Ts = 4/£a>n = 4l\uD\ (the desired settling time). To synthesize L0(s) it is necessary to determine the tracking bounds BpQ'a). 3.) ensures that the tracking bounds BR(j(o.4 Bode plots of TR (a) Ideal simple second-order models. 3-13. 3-9) which are obtained based upon SRdajj).38 Chapters LmTD B u= Lm TRU= Lm dB Ideal Model - -40 dB/dec (a) Bandwidth -20 dB/deo -12 ~-40 dB/dec 8R(j<»i) dB B. The control ratio TRu(s) of Eq. An approach to the modelling process is to start with a simple second-order model of the desired control ratio TRu having the form1 (Or.) (see Sec. 3-9). simplifies the process of synthesizing the loop transmission L0(s) = G(s)P0(s) as discussed in Sec.

Because the models for TRy and TRl are both second-order.R.3 and 3. The high-frequency range (hf) in Fig. where cob is the model BW frequency of BU. . For this system XO^ = Rofor r(t) = R<jU-i(t). 0<a> < coi. The achievement of the desired performance specification is based upon the BW. the time and frequency response plots of Figs. respectively. 7071 is defined as the system bandwidth frequency Wbv- The simplest over-damped model for TBL(S) is of the form (3. the high-frequency asymptotes in Fig. This desired increase in SR is achieved by changing BU and BL.4a. without violating the desired time-response characteristics. an increasing spread between BU and BL is required in the hf range (see Fig.MISO Analog Control System 39 T I D f \ * \s) eq\ > (S) = ———— = ———-———— ft Q\ VJ-"/ where 2 (3. which is determined by the intersection of the -12 dB line and the BU curve in Fig.8) satisfies the requirement thatXO^ = RoU-i(i) for r(t) = R<ju-i(t). 3. 3. 3. can then be drawn. 3. The frequency o% for which \Tpifi 0$ = 0.11) where 5[(5-(CTi + cr2)] and IM = -a1CT2/(cJi + CT2). that is.12) must increase with increasing frequency. 3.4b.4a is defined as a> > coi. Selection of the parameters o> and a2 is used to meet the specifications for ts and Kj. 4b). Sh=Bv-BL (3.4a have the same slope. Once the ideal models TRu(jco) and TRL(jco) are determined. Equation (3.10) The gain constant of this equivalent Type 1 transfer function Ge<j(s) is Kt = lim^sG^s)] = cq/2f. In addition to achieving the desired characteristic of an increasing magnitude of 4? of Bv for o)t > cocf.

z\) TRu (s) = —2———————5. The spread can be further increased by augmenting TRL with a negative real pole [see Eq. (3.\ very much less than |p7| and \p2\ in order to effectively have an under-damped response.14) to have the value of \a3\= a3 = a)n > a2 = a. and (b) then. 3.0i)(s . (3. one or more of these poles are moved right and/or left until the desired specifications are satisfied. (3. i. > \aD\.4b. (d) Depending on the performance specifications [see the paragraph below Eq.14) Thus. This additional pole lowers BL for this frequency range.= ————————— (3. as a first trial.4b) must be larger than the actual variation in the plant. the following procedure may expedite the design process: (a) first synthesize the second-order model of Eq.13) containing the zero at |zj| = a > wn that meets the desired FOM.er\)(s - K K TRL (s) = ——————————— = ———————————— (s + cti)(s +a2)(s +03) (s .. 3. (3. the magnitude of &(/($) increases as <». if necessary. <§>. (3. For this situation. increases above <%• The manner of achieving a SR(JCO) that increases with frequency is described below and is illustrated hi the example in Sec. This additional zero raises the curve BU for the frequency range above a>Cf. At high frequencies <% (see Fig. In order to minimize the iteration process in achieving acceptable models for TRJ(s) and TR[(s) which have an increasing &(/<»).14)] which is as close to the origin as possible but far enough away not to significantly affect the time response.13)] as close to the origin as possible without significantly affecting the time response. TRJis) may require two real poles and a zero "close" to the origin. (can/a)(s+a) (tal Ia)(s .40 Chapter3 by augmenting TR with a zero [see Eq. Note that the straight-line Bode plot is shown only for !RL.e.13) s + 2ga)ns+o)n (s .15)]. selecting the value of all three poles in the range specified above insures an increasing SR. This characteristic is the result of the Bode theorem which states that j^° LmSp d(o — 0 . select \z. For succeeding trials. 3-17.0-3) (3. the real pole must be more dominant than the complex poles. for these augmented models.CT2)(s . As illustrated by the slopes of the straight-line Bode plots in Fig. Other possibilities are as follows: (c) the specified values of tp and 4 for TRL may be such that a pair of complex poles and a real pole need to be chosen for the model response. select all three real poles of Eq.

. 3.4b). a dominant complexpole pair is not suitable for TRu.5 Bode plots of disturbance models for TD(Jca).15) For the case where y(t). and for: d2(t) \y(f)\ <Op for / > tx. At some high frequency co^cah (see Fig. the frequency domain disturbance specification is Lm TD(jco) < Lm Op over the desired specified BW (see Fig. a dominant real zero (z/ > pi) "close"' to /?/. the disturbance specification is represented by only an upper bound on the NC LmT. is to have an allowable "large"' overshoot followed by a small tolerable undershoot.2 DISTURBANCE REJECTION MODELS1 The simplest disturbance control ratio model specification is \TD(ja))\ = \Y(j a/)/DQ a))\ < Op. then lim SR (jca i) » BU . [the desired maximum magnitude of the output based upon a unit-step disturbance input [D of Fig. and a lar off pole p3 « p2. Thus.3.M1SO Analog Control System 41 Thus. Thus. 3. for di(f): \y(tT)\ <ap. 3.5). i. since \TRL\« 0. corresponding to TRu.(-00) = oo dB (3. a designer selects a pole-zero combination to yield the form of the desired time-domain response. 3-4. ^x__ Specified maximum magnitude of TD(j<o) dB The "upper" boundary Fig. The closeness of the zero dictates the value of Mp.1)]. a constant.e.. Thus. An acceptable overshoot with no undershoot for TRu can be achieved by TRjJ having two real dominant poles pi > p2. 8). the reduction in sensitivity SPT at the lower frequencies must be balanced by an increase in sensitivity at the higher frequencies (see Chap.

3. This region may be described by J LTI plants.2.7). the boundary points 1. That is. This is done in order to simplify the labeling in the figures associated with the discussion in this chapter.6. .10} and a e {1. WithZ..7). based upon desired performance specifications.6) and (3. to obtain 6 LTI plant models that adequately define the region of plant parameter uncertainty.6.42 Chapter 3 over the specified BW. which lie on its boundary. 5. and 6 are utilized. (3. 2.4.16) 1+L Region of plant parameter uncertainty Fig.. 3. 3-6 PLANT TEMPLATES OFP. Note: the numbered points around the contour in Fig. 3-5 /LTI PLANT MODELS The simple plant of Eq.. The region of plant parameter uncertainty is illustrated by Fig. (2. A detailed discussion on synthesizing disturbance-rejection models for Eqs. 1. = Lm F + Lm = Lm F + Lm T (3.6 are relabeled by letters as shown in Fig. J.10}. is used to illustrate the MISO QFT design procedure. 3.5) yields Lm'. Eq. as discussed in Chapter 2. to yield the upper bounds shown in Fig.2.(s). (3. where K e {1.5 for DI and D2 is given in Ref. where i = 1.6 Region of plant parameter uncertainty. = GP. 3. 3.

up to approximately the -12 dB value of the Bv plot in Fig. 3. for a Type 0 plant select o^ = 0 and for Type 1 or higherorder plants select o^^O.4b. To provide more details in obtaining templates. where a^ < m^.7 NC characterizing Eq. is &(LmTR) = LmTR -LmF = Lm (3. 3.< o)hR. In addition. A guideline for selecting the frequency range for the templates are to select three frequency values below and above the 0-dB crossing frequency <%.6 are selected. The first step in synthesizing an L0 is to make NC templates which characterize the variation of the plant uncertainty for various values of 00. 3.) and ZP(ja>. The boundary of the plant template on the NC can be obtained by mapping the boundary of the plant parameter uncertainty region. must restrict this change in TR so that the actual value of Lm TR always lies between Bv and BL of Fig.7) over the region of uncertainty. since F is LTI. the simple plant of Eq. over a frequency range en < co.7) is used whose region of plant uncertainty is depicted in Fig. 3. no less than an octave apart. . A number of points on the perimeter of Fig. (2.4b. and values of Lm P(j(o. (2. The change in TR due to the uncertainty in P. 3.) are obtained at each point.6.MISO Analog Control System 43 K=10 a=10 Fig.17) 1 +L The proper design of L = L0 and F.

For the plant of Eq.D^i and the shaded area is labeled 3PQ1).B. any plant in P can be chosen as the nominal plant4. A sufficient number of points must be selected so that the contour of 3P(ja>i) accurately reflects the region of plant uncertainty. A curve is drawn through the points A.7 represents a given flight condition (FC) at co = 0%. 3. (3. the template of Fig.11) and eventually approach a straight line of height FdB [see Eq.) = Lm [LQ'enyfL + L(jo)i)]. However. (2. T. is. the templates widen (angular width becomes larger) for increasing values of «%.7 represents LmP^a variation in P results in a horizontal translation in the angle of P and a vertical translation in the log magnitude value of P. as ». are plotted on a NC as is illustrated in Chapter 2.B. i.C.7).LmL0 = (LmG + LmP)-(LmG + LmP0) = (Lm P ..tbe data obtained from Eq. In addition to the points A. [P/i/eO'0*)]/^. the values of the M-contours at the intersections with the template are the values of the control ratio Lm TQco. A characteristic of these templates is that. (2.7. 2.LmP0) < SR(jo}.B. each point A. 3. PcQl) = 1001JM Z-95.D.e.ZP0 Thus. if point A in Fig. For the points A. starting from a "low value" of o)i.7) the values K = a = 1 represent the lowest point of each of the templates 3PQcoj) and may be selected as the nominal plant P0 for all frequencies. These data are plotted on the NC shown in Fig.B. 3. etc. for each value of frequency co = 0%. 3. With L= GPand L0 = GP0.6.C. The templates for other values of ct)i are obtained in a similar manner. and D.7 may be drawn on a plastic sheet (preferably colored) so that a plastic template for 3P(jl) can be cut and labeled.20)].PA(/l) = J2/2Z-13F.) dB and ZAP = ZP .B.T and PDQ1) = 10/JlOl Z-95. in Fig. For an aircraft.One of the flight conditions may be identified as the nominal plant P0. 3. The contour A. takes on larger values and approaches infinity they become narrower (see Fig.C.C.15) and (2. Then.C. at the frequency co = l.D in Fig. When G(jo>) represents a specific transfer function. 3.16): ~&p (j(Oi) = Lm L. For the template of L(J(OJ).7 can be converted into a template ofLQcoj) by translating it vertically by Lm GQ&i) and horizontally by ZG(jo)j).D marked in Fig.44 Chapter 3 These data. The range of values of TQfOj) for the entire range of parameter variation (K and a) can therefore be determined . A curve is drawn through these points and becomes the template 3PQa>i) at the frequency CDJ. it may also be necessary to include additional points on the perimeter. Prf 1) =5 J2-Z-13F. as given in Eqs. (2.

The formation of the [/-contour is discussed in this section. 3. the top portion. 3. On the NC this bound on Mr = ML (see Fig. it is a common practice to select.7): M-contour (Lm ML) (/-contour Fig. 3. 8). as given hiEqs. (3. .4) identify a minimum damping ratio ffor the dominant roots of the closed-loop system which corresponds to a bound on the value of Mp ~Mm. For the two cases of disturbance rejection depicted in Fig. The boundary of this region is referred to as the universal high-frequency boundary (UHFB) or stability bound.3) and in the frequency domain (see Fig.8 [/-contour construction (stability contour). Therefore. 3. efa. respectively. 3-8 {/-CONTOUR (STABILITY BOUND) The specifications on system performance in the time domain (see Fig. the U-contour. whenever possible.8) establishes a region which must not be penetrated by the templates and the loop transmission functions LjQa)) for all CD. because this becomes the dominating constraint on L(jo)).1 the control ratios are.MISO Analog Control System 45 3-7 NOMINAL PLANT While any plant case can be chosen.6) and (3. of the ML contour becomes part of the [/-contour. a plant whose NC point is always at the lower left corner of the template for all frequencies for which the templates are obtained (see Chap. 3.

the oval. the top portion.2. namely a constraint is placed on the damping ratio ^ of the dominant complex-pole pair of TD nearest the jco-axis. it is necessary to synthesize an L0(s) so that the disturbances are properly attenuated. Although this large peak does not appear in TR due to the design of the filter F(s).19) where ML is a constant for all to and over the whole range of £? parameter values.13 This damping ratio is related to the peak value of \T(ja>)\ = (3.2. This results in a constraint on £of the dominant complex-pole pair of TD. A value of M. and 2. becomes part of the Ucontour. in Fig.46 Chapters T ID. only one aspect of this disturbance-response problem is considered. of the M-contour on the NC. which corresponds to the value of the selected value of ML.18). can be selected to correspond to the maximum value of TR. For example. (3. it is reasonable to add the requirement (3." and it may be difficult to achieve the restriction on the peak overshoot ap of the time response: Therefore. consider the case for which the dominant complex-pole pair of Eq. For the present.7 dB for £= 0.4. as a> -> OR. a peak of TD. For a large class of problems.18) 1 + L(jco) For example. 3. the limiting value of the plant transfer function approaches . agefa. Therefore. efa. due to £ = 0. = • and Thus. can be "very large. for Lm Mm = 2 dB. (3. This constraint can therefore be translated into a constraint on the maximum value 7^ of Eq. If d(t) is very small.8 corresponds to the 2 dB M-contour on the NC. etc. it does affect the response for TD. This results in limiting the peak of the disturbance response.18) results in a peak of Lm T = 8 dB for £ = 0.

where the poles are real. (2.7). 3. ZP(jaJ) for a plant shown in Fig. The high-frequency boundary Bh.19) and SR(JO}I>. .20) yields V = ALmP = lim {{Lm^Kd). not necessarily straight lines. For the simple plant of Eq. is obtained by measuring down FdB from the ega portion of the M-contour as illustrated in this figure. 3. 3.elements of the plant matrix.^ .7).Lm(jco) }] (3. (3. (3.21) which. 3-9. the templates may not approach a vertical line if the Ay are not the same for all pv. With G(s) = A = 1 and F(s) = 1 in Fig.20) = LmK max- Lm Kmin = V dB If the nominal plant is chosen at K' = K'min.^ ] (3.Lm P. where K e{7. When the Ay are different.8.21) = Lm(Ka)max . then the constraint ML gives a boundary which approaches the [/-contour abode fa of Fig.7. the bed portion of the [/-contour in Fig. for this example. approaches a vertical line of length equal to A = lim Lm P^^ . (Note: For a MIMO plant P = {pt]}. The plant template.1. (2. plant of Eq.70} and a e {1.Lm(Ka)^n = LmlOO . The 3P[/fflj) is used to determine the corresponding tracking bounds BR(J(DJ) on the NC in the manner described in Sec. The remaining portions of the [/-contour. as a> -> ao.Lm(jco) } at—>co 2 {Lm(Ka)^n . is 40 dB. 3-9 TRACKING BOUNDS Bpfjca) ON THE NC As an introduction to this section. Fis determined by Eq.Lml = 40 dB For the m. then the widths of the templates are a multiple of 90°. 3. portions ab and de.p. applying the limiting condition. the procedure for adjusting the gain of a unityfeedback system to achieve a desired value of Mm by use of the NC is reviewed Consider the plot of Lm P(jco) vs. L = P. (3.1. are determined by satisfying the requirement of Eq.MISO Analog Control System 47 K lim {P(jto)} = —— where A represents the excess of poles over zeros of P(s). ^L(jw) is tangent to theM = 1 dB curve with a resonant frequency 0^ = 1. for this problem class. The plot of Lm L(ja>) vs.10} and where K = Ka.9 (the solid curve). 3. to Eq.8. the plant templates have the typical shape of Fig. CD -> <x>.

on L(/«i). It is assumed for Eq. It is desired to find a cascade compensator G(s). it is necessary to determine the resulting constraint. raising Lm L(ja>).3. they have negligible uncertainty. such that the specification 1 dB <Lm Mm <2 dB is always maintained for this plant gain variation while the resonant frequency &„.679) and the resonant frequency is a^ = 2.679 and the resultant resonant frequency satisfies the requirement 03m = 2. (3. or bound BR(J(H).9 Log magnitude-angle diagram.679. The procedure is to pick a nominal . 3. The manner of achieving this and other time-response specifications is the subject of the remaining portion of this chapter. This requires that the loop transmission L(faj) = G(jai)P(jaJ) be synthesized so that it is tangent to an M-contour in the range of 1 <Lm M <2 dB for the entire range of 1 <A <1. until it is tangent to the 2-dB M-curve.09 + Awm. the gainyl is increased.4b. 3. For this example the curve is raised by Lm A = 4. Thus. If Lm Mm = 2 dB is specified for Lm TR.48 Chapters 300° -28fl" -2W° -240° -220° -200° -189* -160° -Mfl° -120° -100° -80° ^0° -W° -10" 0' Fig. in Fig. The solution requires that the actual ALmTpfjcOi) < SptjfOi) dB in Fig. (3. only the uncertainty in P contributes to the change in TR given by Eq. that is. Thus. remains constant.1. Now consider that the plant uncertainty involves only the variation in gain^4 between the values of I and 1.17).5 dB (G = A = 1.09.17) that the compensators F and G are fixed (LTI).

then \Lm G(j2)\ = \\Lrn L(j2)\^ . if a trial design of L(j2) requires sliding 3P(j2) to the position.23) C' Lm L -180° -153.13J = 1 IdB (3. . The nominal plant transfer function.Lm P(j2)\A = \\. Note. As an illustration.2 . the plant's region of uncertainty 3P(j2) is given by the contour ABCD. (2.22) and is represented in Fig.4°}.4b of &(/«*) for each value of «?/. with K0 = 1 and aa = I . once a nominal plant is chosen.) on = 2. Lm P(j2) lies on or within the boundary of this contour.4 $ 'CD'in Fig.3° Fig.10.6). then 3P(j2) is translated on the Nichols chart vertically by the value of Lm G(J2) and horizontally by the angle ZG(j2). The templates 3P(ja^) are relocated to find the position of L0(jco) which satisfies the specifications in Fig.10. i. ZP(j2) for the plant of Eq. 3. consider the plot of Lm P(J2) vs. For example..4° -120° -101. it must be used for determining all the bounds BpHjoOi).. is 1 s(s (3. 3.10 by point ^4 for m = 2 [-13.7) (see Fig.10 Derivation of bounds Brfja. As shown in Fig. 3.0 dB. Since Lm L(j2) = Lm G(j2) + Lm PQ2). -153.MISO Analog Control System 49 plant P0(s) and to derive the bounds on the resulting nominal loop transfer function L0(s) = G(s)P0(s). 3. 3. 3.e.

5 dB at a> = 2. on the M-contour = Lm fl. as shown in Fig.) passing under the f/-contour. From Fig. over the specified bandwidth. The above procedure is repeated at other frequencies.4°) = 93. (2) Line up side A-B of J/ty «*) on the -90° line. respectively) is equal to the value of &(/c$) obtained from Fig.) on the NC containing the (/-contour to determine the bound Bj^jco. For other In the low frequency range (desired tracking bandwidth) generally an octave apart will provide a reasonable separation of the bounds on the N.) for a range of values of coj (cat. 3.2 dB is the smallest or minimum value of Lm L0(j2) which satisfies the 6. where the extreme values of Lm [L/(l + L)] are at C"(-0.25) When Eq.) as follows: (a) Use major angle divisions of the NC for lining up the 3P(ja>.2) = 6. 3. in Fig.49 dB) and the minimum at point A '(M = -6 dB) so that the maximum change in Lm T is. determine the locations of the template on the -90° line where jo)i) = Lma . then points. 10. C. «%. 1 1 as P0.) of permissible L0(joj).).5 dB specifications for SRQOJJ). The manipulation of the co = 2 template.50 Chapters Z.4b. in this case.v .5 dB = &(/2). 2. 10. Sufficient points are obtained in this manner to permit drawing a continuous curve of the bound BR(/2) on L0(j2). (-0. f Mark this point on the NC. (b) Select P0 to represent. 3.4° (3. 3. is repeated along a new angle (vertical) line.4b obtain values of S^jco. preferably* an octave apart.7 .11. as shown in Fig.11. . the lowest point of 3P(jo)i). . (The selection of (Oh results in the bound BR(jtoi.(-153.ZP(y2)A = -60° .2 dB). 7 dB) and A '{-7. Place the template 3P(j(o. Lowering the template on the NC to A "B "C "D ". if ZL0(j2) = -6(f. Any smaller magnitude is satisfactory but represents over design at that frequency. If the specifications tolerate a change of 6. . .51 dB. The procedure for determining the boundaries ER(jo^ is summarized as follows: 1 . 3. and a corresponding new rninimum of L0Q2) is found.25) is satisfied. then -4.(-7. 3. Move the template up or down until the difference ALm Tdjajfr between the values of two adjacent M-contours (Lm a and Lm (3. yields Lm L<j2)c» -Lm L(j2)A« = -0. (3. 3. in general. for ease of the design process.G(j2) = ZL(y7). resulting in a family of boundaries BR(jco.24) Using the contours of constant Lm M = Lm [L/(l + L)] on the NC in Fig..Lmft =SR(jajj) (3. the above trial position of Lm L0Q2) is well within the permissible tolerance. (Oh). Thus. For the design example shown in this chapter: (1) Select poinU in Fig.49) (-6) = 5. the maximum occurs at point C'(M = -0. lies on the bound BR(ja>i). Thus.

8). points 'on the <t> = <4 line becomes the left boundary or terminating point for the B^Qo^) contour and is a point on the [/-contour (a point on ab of Fig. In order to satisfy the requirement of Eq.11.25). 3.MISO Analog Control System 51 Lm adB LmpdB Fig. 3. -llff. plants the shape of the template may be such that if point A represents P0.19). another point of the template may be the lowest point that satisfies Eq.11 Graphical determination ofB^jco).. 3. No intersection of the A/L-contour by a template is permissible. If the template is moved further to the left it will intersect the ML contour and permit a peak of T(jw) greater than ML. (3. Note that obtaining the bounds BR(j(Qi) as described in step 2(6)(2) only guarantees that the difference 4(/ftV between the upper bound Lm Tv and the lower bound Lm TL for LmT = Lm [L/(l + L)\ will satisfy . 3. For example. up to -18CP or until a point of the template becomes tangent to the ML-contour. For the plant of this example the [/-contour is symmetrical about the -18CT axis. etc. Repeat step 2 on the lines -J0(f. When this equation is satisfied point A still yields points of £«(/<«/). in moving the template from the -9CP line to the left.3 Draw a curve through all the points to obtain the contour for BR(ja^). (3. the template may eventually become tangent to the A4-contour at some angle ^ as illustrated in Fig.

Over the specified BW it is desired that \TD{j(o)\ « J.e. (3.) usually decrease as w increases.27) yields TD(s) = —— l +l (3.. generally octaves apart. as discussed in Sec. but not necessary. For this latter situation it is desirable. 3-4.) is not observed. the magnitudes of the bounds BR(ja). i. from Eq. 3. 3-10 DISTURBANCE BOUNDS B^jcot): CASE 1 Two disturbance inputs are shown hi Fig. for this type of plant.)\ for (Oj > (Oi. 3.6). Thus. = //rtntoEq. > \BdJo). Both cases are analyzed. it is possible to have \BR(JO^)\. Note.LmTL (jto.1. When this characteristic of Sptjo). that \L(Jco)\ » 1 (or \l(jca)\ « 1).28). . avoid selecting an underdamped nominal plant P0(s).28) this equation has the mathematical format required to use the NC. (3. it is desirable to have S^ja^) increasing as cOj increases. It is assumed that only one disturbance input exists at a time. to synthesize TKu and TRL based upon an overdamped P0(s) model. ) 4. d. < cot.(t). until the highest bound BK(jo^ and lowest bound BRfjo)h) on the NC clear the [/-contour. over some portion of the region of plant uncertainty. For reasonably damped plants (£> 0. over the entire region of plant uncertainty. From Fig.. Repeat steps 2 and 3 over the range of (a* < CD. For a plant that is highly underdamped (£ <0.(t) = 0] CONTROL RATIO. 1 the disturbance control ratio for input d2(t) is (3. if y is specified instead of ML it dictates the side a-b of the [/-contour.52 Chapters (3.27) 1+ L Substituting!. CASE 1 [rfXO = Dou_.26) (jo)j).6). which results in the requirement.

In the frequency domain this specification may be approximated by \MR(jco)\= TR(jco) Y(jco) <Mm * (3.and frequency-domain response characteristics. for K = 1. 3.12.30) The corresponding time. resulting in the plot of Lm L(jaj) vs. ZHjeo) in Fig. co.11 contains data for two points on the Nichols plot of Fig. The disturbance transfer function: TD = 1/(1 + L) = el(l + e) has the desired BW 0 <co <a>2 for which IL(jaj)\ » 1 and I K j c o ) \ « 1.30) applies within the BW region.13. 3. 3. The plot of Lm L '(/«>) vs. Sincel =KL'= 7/^then (3. y(t) d(t) and \MD(Ja>)=\TD(ja. ZL '(Jco). then a negative value of Lm e yields a positive value for Lm L as shown in Fig. The plot of Lm Kjco)vs.32) APPLICATION. respectively. Since Lm 'j&) = Lm [l/L(jco)] = -Lm LQai).31) = (3. on the NC is tangent to Lm M = 1 dB contour. and the specification on the system performance is Mm = 1.) <a ==. Zeiwc these 2 points is also shown in this figure.29) DISTURBANCE RESPONSE CHARACTERISTIC. (3. The NC of Fig. 3.12 can be rotated 18Cf and is redrawn in Fig. A time-domain tracking response characteristic based upon Hf) = w_/(0 often specifies a maximum allowable peak overshoot Mp.MISO Analog Control System 53 TD(jO) * (3.33) .26 (2 dB). By means of a NC determine the value of K required to achieve this value of Mm and obtain the data to plot \M(jca)\ vs. o. Intersections of Lm L(jaj) with the M-contours provide the data to plot the tracking control ratio \M(a))\ vs. are. 3. Let L = KL' in the tracking ratio TL = L/(l + L). Thus Eq. Table 3.l x (3. where K is an unspecified gain.13.121 for K = 1. based upon the step disturbance forcing function d^t) = u_i(f). Now consider the corresponding disturbance control ratio for the same control system.

£e '(/'<») must be raised or lowered until it is tangent to the Lm Op-contour (\TD\max = On).e. then the plot Lm e'Qoi) vs. i. Note that this is the same procedure used for the tracking example of Fig. 3. For a given plant P having uncertain parameters. except that the adjustment in Lm l(jo)) is K'1.1 Data points for a Nichols Plot CO 0).12 Regular Nichols chart. TABLE 3. 3. consider that its template 3PQco.27).54 Chapters 3Lm / /I -270° Fig. If f(j(o) is given and it is required to determine K~! to satisfy Eq. Lm K'1 = A. for a given .. i.) for a given at has equal dB differences along it's A-B and C-D boundaries. (3.e.12. Lm L /.L -96° -98° Lm I £t 96° (or -264°) 21 15 -21 -15 C02 98° (or -262°) TEMPLATES. The amount A by which this plot is raised or lowered yields the value of K..

Lm [l/PijcD. "flipping it over" vertically. note that: (a) PAB = 180° + ZP = 180° .2 Data points for the templates of Fig.12 for the same frequency as for the template of Lm P(jcoJ and for the angles of Table 3.MISO Analog Control System 55 .12. the disturbance boundary BD(jo}i) for LQ&d = l/e(j(Oi) is best determined .LmPA) = . the corresponding angles are . 3. TABLE 3.280 (b) Templates of Lm PQw. Eq.60° =-240° Z(7/PCD) = -180 -pa>=-180 -100 =. it is located by first reflecting the template ofLm P(j(Qi) about the -180° axis. 3. (3.80° = 100° For 1/P. Data corresponding to the template location shown in Fig. ROTATED NC.)] is the same as the template of Lm P(jo).IIP A.) are used for the tracker case T = L/(l + L) and the templates of Lm [l/P(jco^] are used for the disturbance rejection case of TD = 1/(1 + L) = ?/(! + t).27). 3.12 are given in Table 3.LmPD) = 10 dB This template is arbitrarily set on the NC as shown in Fig. is arbitrarily set on NC in Fig. For the arbitrary location of the template of Lm [l/PQa^j\.12 Points Z/» /.B C.) but is rotated by 180°. Since Lm L is desired for the disturbance rejection case.PAB = ° .D -120° -80° 120° (or -240°) 80° (or -280°) The template of the reciprocal.2 Note that the template of Lm [l/P(ja>.120° = 60° PCD = 180° + ZP = 180° . Thus. 3.2. and then moving it up or down so that it lies between -5 and -20 dB.2.)].

The lowest point of the template must be used to determine the bounds and. where the rotation of the Lm [l/P(j(o)} template ABCD is reflected in Fig. 3.12 is rotated clockwise (cw) by 180°.-28 Fig. The rotated NC is used to determine directly the boundaries BD(jco. (3.-16 -12. 3.) for LD(jco.-20 24.13.28) in Fig. 3.13 Rotated Nichols chait .34) (Note: Use the negative angle for/j^since n > w) .LmTD = Lm\l + L] > .Lm TD or BD(J<BJ) evaluation 1° -100° -120° -140° -160° -180° -200° -226° -120° -240° -260° -280° Lm [ L m L 24 dB -24. This point is again used to determine the disturbance bounds BD(Ja>. Based upon Eqs.13.5 .-20 ISdB -16. 3. 3. deg 28.4 fe 0 --o | 3 4 --"4 & 8 ---8 12.-16 20. may or may not be the point corresponding to the nominal plant parameters.-12 16. o)..56 Chapters on the rotated NC of Fig.).-12 12dB -8.).For .-24 -20.-24 Phase angle. Thus. the NC of Fig. Point A for the simple plant of this design example corresponds to the nominal plant parameters and is the lowest point of the template 3P(j(n).8 -4- .Lm a(ja>i) > 0 dB (3. in general.27) and (3.

) = Lm am [see Eq.).(f) is P(j(0 ) (3. then L(ja^) = 8.32)] for the same values of frequency as for the tracking boundary BR(ja)j).4 dB. Repeat steps 3 and 4 over the desired frequency range c^ < o.36) . < ev 6.12. \L(Ja)i)\ » «(/». From Fig.) = -Lm M for LQtOj). up to the -180° line or the t/-contour. from Eq. 4.). For example. 2. 3. 3.33 and thus Lm a(ja>i) = -J8.5 the obtain values of Sod®. d2(f) = 0] CONTROL RATIO. for example.13. then .Lineup sideA-B of 3P(M). 3.).1 the disturbance control ratio for the input d. (3. Use major angle divisions of the NC for lining up the -3P(/®.. Note that when \L\ » 1.Lm a(jo)i ) = -SD (jiVj ) (3. 3-11 DISTURBANCE BOUNDS BD(jcOj): CASE 2 CASE 2 [d.. 3. Repeat step 3 on the vertical lines for -100°. Move the template up or down until point A lies on the M-contour that represents Sodo).Lm TD = Lm L>.) to represent the nominal plant P0 in Fig.13. BOUNDS BD(jca. 5. Since \L\ » I in the BW.) . The procedure for determining the boundaries BD(jo}j) is: 1.4dB andLm L(ja>d = 18.MISO Analog Control System 57 where a(ja$ < 0. The same nominal point must be used in obtaining the tracking and disturbance bounds. etc.(t) = D0u. From Fig.(f). on the -280° line for /(or the -80° line for L). (3. Mark this point on the NC.13 as P0. the constant M-contours of the NC can be used to obtain the disturbance performance TD. Transcribe these BD(fa)i) onto the NC that contains the bounds BR(ja}. Select the lowest point of -DP(/a>. For the design example used in this chapter select point A in Fig. 3. 3.35) In terms of L{jco. then.) becomes the boundary BD(ja).).). This requires the change of sign of the vertical axis in dB and the Af-contours. Thus the M- contour corresponding to TD(jco.34).). as shown in Fig. if a(ja)d = 0. Draw a curve through all the points to obtain the contour for BD^JO). -120°.

38) DISTURBANCE RESPONSE CHARACTERISTICS. the time. yields (3. 2. the following items in the order given: .40) d(t) and MD(jco)=\TD(ja.36) is multiplied by P. (3. for a unit-step disturbance forcing function. Evaluate in tabular form for each value of o.and frequency-domain response characteristics.) is: 1. (3. 3. Eq.The procedure for determining the boundaries BD(ja). From Fig. BOUNDS BD(ja>d. respectively. Eq./P0 and rearranged as follows: rfi ID - O 1 1 P PjL Po PO P PO =^ W (3.39) (3. (3. by MD(t) = (3. Based on Eq.5 obtain the value of &(/«*) representing the desired model specification TD = TD[r= ccp for the same values of frequency as for the tracker boundaries BR(jo)^. (3.58 Chapters Assuming point A of the template represents the nominal plant P0.37) with LOT TD = <£.30).41) where tp is the peak time. are given.37) Po P + GP P ° +L where W=(PJP)+L0 Thus.)\ = <am =a.

) for P e fP and |ff(/eBO| are known.MISO Analog Control System 59 LmP0(jtot) The ratio P0iPl is evaluated at each of the four points of Fig.42). It may be necessary to evaluate this ratio at additional points around the perimeter of the A8CD contour as shown in Fig. 3. 3.42) For arbitrary values of P0(jo). 3. Equation (3. with L0 replaced by its bound BA is rearranged to the form —— P B — (3. Before presenting the procedure for the graphical determination ofB^jcOj).): -w (-W) . Fig. PJP. it is necessary to first review.) and W(jc^). 3. Since the values of P0(j<&)/¥(}co. (3.38) with L0 = B0 .180") =/Bd Fig. 3.14 Phasor relationship of Eq.38).6 for each value of cot. where BD = Lm Bd. and W for to = 0%. graphically.14 presents the phasor relationship of Eq./P(jco. (3. the phasor relationship between Bd. the following procedure can be used to evaluate BjQco.).6.

. = <|> .180° P0o>i) Fig. 3. 3. 3.16.W is plotted and-Bj = \-Bd \Z<j> is obtained. 3. where point ^ is the nominal point.15 Template in polar coordinates. Draw a curve which is tangent to Fig. Use a compass to mark off arcs with a radius equal to the distance \W(jco. (b) Based upon Fig. A simple graphical evaluation yields a more restrictive bound (the worst case).)\ at a number of points on the perimeter of 3[P0(jcob/P(ja$j\.43) For one value of P</P shown in Fig. the solution for -Bd is obtained from W (3.14 and the location of the phasor -W(jai).16 Graphical evaluation o.18CT) is performed for various points around the perimeter of 3[Po/P\ in Fig. This graphical evaluation of Bj = \Bj[Z(</>.15. 3.60 Chapters (a) On polar or rectangular graph paper draw 3[P</P\ for each a* as shown in Fig. 3. the value of .15.

For the situation of Fig. 3. 3.17b.MISO Analog Control System 61 these arcs to form the first quadrant portion of the g-contour shown in Fig. Depending on the plant type desired for L it may be necessary to extend this contour into the second and fourth quadrants.3 for each value of en.15.3 Data points for the boundary BD(j(o) -180° -170° -160° 3-12 THE COMPOSITE BOUNDARY B0(jco.).) and BoQaoi) that are the most restrictive. the phasor from the origin of Fig. must he on or just above the bound J50(/o/). for each value of 0%. 3.)\ = \Bd(j(o.)\.) The composite bound B0(ja).3 for each value of 0%.) that have the largest values. 3.)/P(jct>i)]. 3. then the bound with the largest value or with the outermost boundary dominates. The composite bound B0(jco. Assuming the partial g-contour of Fig. (3.) and BD(jco. 3.16 is sufficient.). The synthesized L0(jo).17. In the frequency range ca.17 occur when the two bounds have one or more intersections.) that is used to synthesize the desired loop transmission transfer function L0(s) is obtained in the manner shown in Fig. 3. 16 to the g-contour represents -BJ(jo).17a. This contour includes the plant uncertainty as represented by 3[P0(ja>.43) and Fig. Plot the values of fib(/<aO from Table 3. measure from the graph the length BjQco. 3. is composed of those portions of each respective bound BR(j(o. 3. on the same TABLE 3. Based upon Eq. if | W(jco^\ » \P0(jcoiyP(jcoi')l then the g-contour is essentially a circle about the origin with radius \W(jco. 16.) for every 10° of Bjija). Forthe .17a the bound B0(jcn) is composed of those portions of each respective boundB R {ja)j) andB^co. the outermost of the two boundaries BR(jo). for the situation of Fig. The situations of Fig. < a>i < (Oh. 3.) and create Table 3. If there are no intersections.) becomes the perimeter ofB0(jo)i).). For the case shown in Fig.

17b the synthesized L0(jo}.„(/£»). only the gain K has a significant effect on the high-frequency response.). the importance of minimizing the high-frequency loop gain is to minimize the effect of sensor noise whose spectrum. Also.) must not lie in the interior of the B0(Jo)i) contour.17 Composite BJJco.18 A realistic definition of optimum6'55 in an LTI system is the minimization of the high-frequency loop gain K while satisfying the performance bounds. 3-13 SHAPING OF £. 3. lies in the high-frequency range (see Chaps. It has been shown that the . 6 and 9). This gain affects the high-frequency response since l\ma^J[L(/co)] = K(jco)~z where 1 is the excess of poles over zeros assigned to L(/a>). in general.3. situation of Fig.) Fig. and the effect of the other parameter uncertainty is negligible.62 Chapters BR(JCD.5.) B0(jco. 3. Thus.2.

.44) and where a = Tma. in combination with plant uncertainty templates. /? = Tmin.5 2/ Previous sections describe how tolerances on the closed-loop system frequency response. and it is unique.18. The control ratio -MO" -280' -260' ^40° ^20' -2W 180* -1*0" 140° 110° Phase angle.18 Graphical determination of Lm T^'co.MISO Analog Control System 63 optimum L0(/a>) exists.) at all en. In Fig. and &(/». are translated into bounds on a nominal loop transmission function LQco). it h'es on the boundary B0(ja>. (j>. For the set of plant parameters that correspond to point E within 3 and for a synthesized L0.18 the template 3P(i(Q.) forP<=£P. 3. as shown in Fig.) is located on the corresponding bound B0(jco. 3.) where point A is on the constant M-curve Lm {}.Lm ft = 4dB (3. the open-loop transfer function is LE = GPE. 3. deg Fig. and point C on the constant M-curve Lm a such that = Lm a .) = 4 dB.

3.20). Additional specifications are /I = 4. 3 . there are 4 poles in excess of zeros. 3. ) (3. A point such as Lm L0(j2) must be on or above the curve labeled B0(j2).4b. 3. 3.7). It involves building up the function L0 (jco) = Lok (jco) = P0 (jo)) Yl[KkGk (ja>)} k=0 (3. i. 3.19. (2.18. (3. and K= In order to minimize the order of the compensator a good starting point for "building up" the loop transmission function is to initially assume that LJ^ai) = P0(j a) as indicated in Eq.5) between points A and C in Fig.G0 = lZ(f. For example.19.19) and that at a> = 1 the required Lm L^l) is approximately 27 dB.6 for the example of this chapter) yields a maxi- mum variation in KJCQ) that satisfies the requirement Lm Tmax . For the plant of Eq. with ZL0 < -180° (see Fig. any value of P that lies within the region of uncertainty (see Fig. 3. 3 . Since the ML stability contour must be satisfied for all J plants then it is required that L0(/(o) have a phase margin angle y of 45° over the entire frequency . in order to satisfy the specifications. In order for Lm L0(ja>) to decrease from 27 to about 0 dB in two octaves. any value of P within 3PQ'cOj) yields a value of Lm T(= Lm T.Lm Tmin < 6R (Ja>.46) Thus.. suppose Lod(/4) = KJ"0(j4) = 0 dB Z-1350 (points in Fig. proper design of the prefilter F (see Fig. the shaping of L0(ja>) is shown by the dashed curve in Fig.e. Therefore. and that it also must be Type 1 (one pole at the origin).47) where for k = 0. LJjoi) is built up term-by-term in order to stay just outside the (/-contour in the NC of Fig. Further.47). L0(j<o) cannot violate the (/-contour.19.64 Chapter 3 7> = ———— (3. the slope of LJijai) must be about -14 dB/octave.) of Eq. Thus.45) obtained from the constant M-contours on the NC has the value Lm TE = -4 dB. The first step is to find the B0(jcOj) which dominates L0(/(o). A representative procedure for choosing a rational function L0(s) which satisfies the above specifications is now described.1) yields a tracking control ratio TR that lies between Lm Tv and Lm TL in Fig. (3. In this example a reasonable L0(ja>) closely follows the (/-contour up to co = 40 rad/sec and stays below it above co = 40 as shown in Fig 3.

(2. the value Lm LJjl) on the straight line approximation is selected at 30 dB (to allow for the -3 dB correction at the comer frequency). 3. In the same way it is seen that B0(J1) dominates all other B0(jaj) in Fig.e.19 Shaping ofLJja) on the Nichols chart for the plant of Eq.7). Hence B0(jl) dominates L0(jaJ) more than does B0(j4).). a pole at -1) which maintains -735° for co > /.19. The function L0(s) determined so far is . By selectingL^s) = kP'0(s) = k/[s(s + 7)] the first denominator factor of LM has a comer frequency at a> = I (i.. 3.19 because it is assumed that a slope of -6 dB/octave for a> < 1 suffices (additional values are 33 dB at co = 0.25. range for which L0(jco) follows the vertical right-hand side of the [/-contour and not just at the 0 dB crossover. 3. Thus. The B0(jco) curves for co < 1 are not shown in Fig. etc.39 dB at co = 0.MISO Analog Control System 65 24db 16db 8db Odb -- -8db -16db -24db -180° -140° -100° -60° Fig.5.

20).e.19. This is achieved by selecting T2 = 0. Since a second denominator term (7 +ja>T3) will be needed.20. Lo0(s)= 31.2. 3. In the NC of Fig.e. co > 5 is the region where the maximum phase lag allowed is -135° (i.2K(s+5)/s(s+l) (2)3K(s+5)/s(s+l)(s+15) (3) (3K/40)(s+5)(s+40) s(s+lXs+15) Fig. allow an additional 15° for this factor. i.20 Shaping of LO(J'(Q) on the Bode plot. ZL0(joi) must be > -135°).6 (1)0.66 Chapters . Mca = 10.6(7s(J + s) whose phase angle ZLoI(ja>) is sketched in Fig. and is provided with an additional denominator term (7 + ja>T3). whose angle ZLo0(j(o) is sketched in Fig.. since Z_Lol(jlO) = -772°. then 735° . and therefore a lead angle of 29° is needed at this frequency. a zero at -5. 3. giving a total of 75° + 29° = 45° lead angle required at co = 5. K=31. This results in the composite value 37.I IT = 23° more lag angle is needed. this is to be followed by an additional numerator term (7 + . 3. £0o(jo>) violates the -735° bound at co > 1. 3..20. ZL0<jj5) =-769° (see Fig. hence a numerator term (7 + jooT2) must be added At co = 5.2. However. 3.

e.. (7 + jT4) and (7 + jT5). 5) = -29 dB. allow a 2 dB safety margin. 3.). i. so no correction is needed for them. The resulting phase angle. so allow about 10° for it.2s/100+ s2/104)2 * ' ' = KP0 (s) n G.6. This locates the comer frequency at co = 100. giving 31. A damping ratio of £ = 0.5 + 2+ 3 + 1. but an additional 75° mar- gin is allowed (this is a matter of judgment which depends on the problem. the last corner frequency is at a> = 60. is -66. Since Lm Lo3(ja>) achieves this at co = 60. Thus 31. This requires selecting the comer frequency at co = 15 (or T3 = 1/15). 19) is at -22. the final comer frequency for the straight-line curve of Lm LtfQco) is at -29 dB. a final zero obtained from (7 + jo)T5) is needed Since the bottom of the t/-contour (see Fig. before the final -24 dB/octave slope is achieved. An angle of 180° could be selected at this point. A zero is assigned at a> = 40 with T4 = 1/40. £=0. giving . s/40) L 3(s) = —————————————— s(l + s)(l + s/15) A sketch of -/Zo3(/<y) is shown in Fig. which may include the presence of higher-order modes. Thus.66° . a different value of damping ratio.5 dB for the effect of (1 + ja>T4).5 dB. In order to achieve (an asymptotic) horizontal segment for Lm L0(jco). a 3 dB correction due to (1 +jcoTs). This means that 700° phase lag is permitted: 50° due to each complex-pole pair (180° .6(1+5/5) Lo2 (s) = —————————— A sketch of ZLo2(jcd) is shown in Fig.15° & 100°). 3.(s) K . i.5 is selected for the two complex-pole pairs.e.20. etc. thus L0(Jco) can make its asymptotic left turn under the [/-contour. with the appropriate dB correction applied to the log magnitude plot is chosen. in order to have an excess 2 = 4 of poles over zeros and to minimize the BW.MISO Analog Control System 67 jcoT4). and 7.. due to Lo3 and (1 +jw/60). Lm Lo2(j40) = -20 dB.20. giving 23° + 10° = 33° more lag allowable from (7 +jo)T3). 6(1 + s/5)(l + s/40)(l + s/60) ° 4 1. Looking ahead at co = 40. and finally two complex-pole pairs. The plan is to add two more numerator factors. T5 = 1/60. 3. giving a total of -(22. Thus.

The final L0(Jca) is good in this respect since it is close to these boundaries. Although there is a "slight" infringement of the ^/-contour. i. Even with no more savings. The vertical line at -14(f in Fig. then the price in achieving this is a larger cutoff frequency. A well designed. that reduction (i. without any reshaping of L0(jco) required. then it will be 13 dB at co = 2 (instead of the present 18 dB). 2. which is co = 100. B. for CD > 75. this 5-dB difference allows a cutoff frequency of about co = 70 instead of 700.. 3. . It is possible to economize significantly by allowing more phase lag in the low-frequency range. but not by much.19 is the dominating vertical boundary for L0(jco) for CD < 5 = co^. = 5 < co< coy &30.19 the phase margin angle y = -45° must be maintained for the frequency range co. 3. There is some phase to spare between L0(faj) and the boundaries. 3. if it is desired to reduce the number of poles and zeros of L0(s). and the right side of the ry-contour line at -135° is the vertical boundary effectively for a^ & 5 < co < 30 is ty. The design technique is thus highly "transparent" in revealing the tradeoffs between performance tolerances. For the beginner it is best not to use a CAD program (see Fig. On the graph paper for the Bode diagram plot the points representing Lm Bjija). easing) of the specifications at co = 1 to about 21 dB (instead of about 26 dB) has the same effect as the above. There is a tradeoff between the complexity of L0(s) (the number of its poles and zeros) and its final cutoff comer frequency.I in App.) and the angles corresponding to the right side of the [/-contour (the desired phase margin angle y) for the frequency range of ca^ < co < CDy. If 750° is permitted at co = 7. How badly the specifications are compromised by such easing can easily be checked. = 7. complexity of the compensation.19 and is shown by the solid curves in Fig. Use the straight-line approximations on the Bode diagram for the log magnitude at the start of the design problem.e.. On the other hand. no reshaping of L0(s) is done unless the simula- tion reveals that the specifications are not met. and the "cost of feedback" in bandwidth. an "economical" L0(jco) is close to the B0(ja>) boundary at each a>i. e. For the example illustrated in Fig..20.19 reveals immediately.g. B) until step 6. because of the inherent over design.e. Figure 3. 3. and the phase margin frequency co^. 3-14 GUIDELINES FOR SHAPING L0(jco) Some general guidelines for the shaping ofL0(Jco) are: 1. then a decrease of Lm L0(jco) at a rate of 72 dB/octave can be achieved. stability margins. with Lm L0(fl) = 25 dB. Use of more poles and zeros in L0(s) permits this cutoff frequency to be reduced a bit below 700.68 Chapters The optimal loop transfer function L0(jco) is sketched in Fig.

5. Once L0(jai) has been shaped. < 0.) be on the corresponding bound In practice. Generally it is desirable to first find the bounds BD(ja). 2. and employ the shaping discussion that follows Eq. Use frequencies an octave above and below and a decade above and below a corner frequency for both first. If the B W is too large.3'5 4. If it is desired that y(oo) = 0 for d(t) = u. \hefinalform ofL0(s) must have an excess of poles over zeros of at least eA+t where / > 1. 5. If the synthesized L0(jw) yields the desired performance. Generally the BW of L0(s)l[l + L0(s)] is larger than required for an acceptable rise time tK for the tracking of r(t) by y(f).) as close as possible to the bound B0Q&i). %>'s . In . any RHP poles and/or zeros of P0(s) should be included in L0(s). The last two poles that are added to L0(jco) are generally a complex pair (the nominal range is 0. A good starting L0(s) is L0d(s) = KgP0(s). For this situation a possible starting point is L^s) = KoP0(s)ls.. it is necessary to insure that TD(s) has a zero at the origin.).) are completely dominant compared to BR(ja).).) and then the bounds BR(J(OI). place L0(jco. then. If P(s) has an excess of poles over zeros.7) which tends to minimize the BW. with the plotted information of step 2. Experience shows that a value of /I + i of 3 or more for L(s) yields satisfactory results.(f). 3-15). An acceptable rise time can be achieved by the proper design of the prefilter F(s) (see Sec.48) for the frequency range ca < co* as guidelines in achieving Eq. then the required compensator is given by Specific guidelines for shaping L0(j(a) are: 1 . (3. Since exact cancellation of a pole by a zero is rarely possible. After finding the first Bj^ja. then increase the value of /".48). but above it in order to keep the BW of LJ[1 + L0] to a minimum.#z's = A. 3-15) and then verify that L0(Jco) does meet the design objectives by use of a CAD program. determine F(s) (see Sec. 3. it may be evident that all or some of the B^co. which is denoted by eA. 6. Do the shaping of Lm L0(jco) on the Bode plot using straight-line approximations for Lm L0(jaj). 4.and second-order terms.13'19 while maintaining the phase-margin angle corresponding to the right side of the [/-contour in shaping L0(jw). (3. in general.5 < t.MISO Analog Control System 69 3. An optimum design of L0(jaj) requires that L0(ja>.

i. A method for determining the bounds on F(s) is as follows: Place the nominal point A of the a)j plant template on the LO(JCQ. 7.22). The ability to shape the nominal loop transmission L0(s) is an art developed by the designer only after much practice and patience.) point of the L0(jai) curve on the NC (see Fig..?)1855 Design of a proper L0(s) guarantees only that the variation in \TR(Jco)\. .) 6.21 Requirements onF(s). 3. 3. 3.. 3..) boundaries are the optimal boundaries. The purpose of the prefilter is to position Lm T(ja>) within the frequency domain specifications.. then it will be necessary to utilize complex poles and/or zeros in G(s) in order to achieve an optimal L0(s). determine the maximum Lm T^ and minimum Lm T^. Arfi. 3.21.4b. increases. values of LmT LmT LmT. BD(ja>.70 Chapter3 that case the BD(ja>.e. which are redrawn in Fig. Traversing the template.e. dB Fig. This assumes that the tracking model is not designed to yield this increasing characteristic for &(/(%). i. For the example of this chapter the magnitude of the frequency response must he within the bounds Bv and BL shown in Fig.15 DESIGN OF THE PREETLTERF(. is less than or equal to that allowed. If ty(/a)i) is not continuously increasing as <w.

49) and (3.) = Lm F^'co.23 the difference between the Lm TRu-Lm Tmax and the Lm TR-lm Tmin curves yields the requirement for Lm F(jai). based upon Eq. (3. lie exactly on the tracking bounds BR(jo). from Eq.51) If values of L0(/o>..).Therefore.). thus $L (M) = Lm Tmax .BL (3.50) The variations in Eqs.MISO Analog Control System 71 dB LmT Desired ML . The tracking control ratio is TR = FL/[1 + L] and Lm TR O>. The process is repeated for each frequency corresponding to the templates used in the design of L0(jaJ).49) obtained from the A/-contours. in Fig.). 3. for each value of a). LmT(ja>. (3.) = - (3. (3.50) . Therefore. 3.contour Phase U .contour Fig.21.22 Prefilter determination.) + Lm T(jcoi) (3.e.Lm Tmin < SR = Bv .50).. These values are plotted as shown in Fig. it is necessary to determine the range in dB by which Lm T(jo$ must be raised or lowered to fit within the bounds of the specifications by use of the prefilter F{jo). then 4 = SR. 3.50) are both due to the variation inP. i..

t(f) with no steady-state error and to satisfy the performance specifications of Fig. 3. For step forcing functions the resulting F(s) must satisfy' lim \F(sy\ = 1 (3. 3. The system configuration shown in Fig.22). From the values obtained in steps 1 and 2.23 Frequency bounds on the prefilterF(s). Parameter uncertainty for the plant of Eq. 3. This is done by placing 3P(j(o^ with its nominal point on the point Lm L0{/a)i). The first objec- tives are to track a step input r(t) = u. Then use the M-contours to determine Tma^jco^ and TrtJjeoi) (see Fig.LmT dB Fig.46.1 contains three inputs. (2. from Fig.7) is shown in Fig. 3. 3.52) 1. 4.Lm T(jco) The procedure for designing F(s) is summarized as follows: (3. Use straight-line approximations to synthesize an F(s) so that Lm F(jco) lies within the plots of step 3.72 Lm Chapters . Use templates in conjunction with the L0(ja>) plot on the NC to determine 7^ and Tmin for each G>J. 3. 3. 3. plot T ~ \-Lm TR ~ vs co as shown in Fig.3. 2. An additional objective is to attenuate . Lm F(jco) = Lm TR (jco) . Obtain the values of Lm TRu and Lm TRl for various values of a.6.23.53) 3-16 BASIC DESIGN PROCEDURE FOR A MISO SYSTEM The basic concepts of the QFT technique are explained by means of a design example.

MISO Analog Control System 73 the system response caused by external step disturbance inputs d^t) and d2(t). Use the data of steps 2 and 3 and the values of 8$an) (see Fig. Plot curves ofLm B^jco. 3-10. using the tracking model (step 1).1) and denote it as P0(s). 1.) (step 3). Note that \L0(jeOj)\ > \B0<j(Oi)\ represents the loop transfer function that satisfies the most severe boundary BR and BD.) on the same NC.46). based upon the disturbance-rejection specifications. Synthesize an L0(jaj) = G(f(o)P0(/co) using the Lm BJjo^) boundaries and {/-contour so that Lm L0(ja)^ is on or above the curve for Lm B0(Ja>i) on the Nichols diagram. 2. Draw a curve through these points.19)]. based upon the desired tracking specifications (see Figs. and ML [see Eq. the values of S^jco. 4. = B0).) on the Nichols diagram (see Fig. 3. 6.). The resulting plot defines the overall boundary Lm B0(jco. Synthesize the tracking model control ratio TK(s) in the manner described in Sec. plant.) = G(Jcn)P(/cai).) must be on or above the curve for Lm BD(ja>.p. is as follows: 1.2. For m. Repeat this procedure for sufficient values of (Ot. Obtain templates of PQoji) that pictorially describe the plant uncertainty on the Nichols chart for the desired pass-band frequency range. as described in Sec. This procedure achieves the lowest possible value of the loop transmission frequency (phase margin frequency G^).) vs.p. Select a nominal plant from the set of Eq.) for tracking.3 and 3. and V for the universal highfrequency boundary (UHFB) Bh in conjunction with steps 6 through 8. (3. ML for disturbance rejection.).5) to determine the disturbance bound BD(Jco. the templates P(ja). For m. systems this requires that the synthesized loop transmission Lm LD(ja>.p. Determine the {/-contour based upon the specified values of <5?(/a>. 3-4. 3. (3. 9. 8.) to be as close as possible to the boundary value B0(ja).) on the loop transmission LD(Jco. </>. Determine the tracking bound BR(/cn) on the nominal transmission L0(/(»i) = G(j (o.) versus fo = ^BR{/CH) and Lm BB(/&i) versus $> = ^Bodfo. as applied to a m. .) (see Fig. For a given value of to/ at various values of the angle $ select the value of Lm BD(ja)j) or Lm BR(jca. Synthesize the disturbance-rejection model control ratios TD(s) in the manner described in Sec. systems this requires that the synthesized loop transmission satisfy the requirement that Lm L0(jcai) is on or above the curve for Lm Bdjcoi) on the Nichols diagram.19 assumingB^.) by selecting an appropriate compensator transfer function G(jco). whichever is the largest value (termed the "worst" or "most severe" boundary). Design LJijca.4b). 3. 3. 5.)P0(i o). 3-4. An outline of the basic design procedure for the QFT technique. 3.

For the L0(jco) obtained.4b.6)..(t) and d(f) = 0 for sufficient points around the parameter space describing the plant uncertainty (see Fig. may be too high resulting in saturation and/or sensor noise effects.5)] vs. by ignoring the disturbance rejection specification. Obtain the time-response data for y(f): (a) with d(t) = u.4b. 3. 3. but 8$ col) is satisfied By the proper design of the input filter F(s\ Lm TR = Lm FL/[1 + L] will lie within the bounds In problems with very large uncertainty and in disturbance rejection requiring a very large \L(ja>)\ over a "large" BW. Based upon the information available from steps 1 and 9. the plot of Lm L<jca)/[l + L(jw)] may be larger or smaller than Lm TRf. The analysis of the equation Y(jta) = D(jto) for the condition \G(ja))P(jco)\ » 1.54) l+L and T &F. 3. 3-4 and Ref. over the desired BW results in Y(jco) Thus. A situation may occur in which it may be impossible to satisfy all the desired performance specifications and a design trade-off decision needs to be made.(t) and r(t) = 0 and (b) with r(/) = w_. then L i (3. 1.74 Chapter 3 10. i. the gain that is required to satisfy the dominating Bfl(/o. (3.. For example. ca that lies between Bv and BL of Fig.). or Lm TRf of Fig. due to the gain in G{jco). synthesize an F(s) that results in a Lm TR [Eq. The possibility exists that the gain required in G(s) in order to satisfy the tracking performance specifications and the overdesign characteristic of the QFT technique may result in satisfying the performance requirement .e. B0 = BD. 11. For these situations design F(s) in the same manner as for the tracking models of Sec. synthesize the loop transmission fimction L0(s) by satisfying the bounds B0(ja)i) = BR(jco.) bounds.

Thus 0.ts = 2s (a) A tracking model for the upper bound. (3.y + 21/3.84 s.MISO Analog Control System 75 The simple plant of Eq.2078. B) was used to obtain the data and to execute the design procedures.56) A zero is inserted in Eq. is tentatively identified by V5> = ——————— (s + 3)(s + 4)(j + 10) L (3 58) - . 3-17 DESIGN EXAMPLE 1 This design example is for the control system of Fig. = 1.2.766 s.7) is used in the following sections to illustrate the details in applying this QFT design procedure. t.(f) and dj(t) = 0. tR = 0. based upon the desired performance specifications.. based upon the given desired performance specifications.55) as the nominal values: a = K = /.969 (3.969 The figures of merit for this transfer function with a step input are Mp = 1. (2.1 with r(i) = d2(t) = u. The CAD package TOTALPC1 (see App. Modeling the tracking control ratio T^s) = Y(s)/R(s): TRu :Mp=l.95 dR and a^ = 3. tp = 0. LmMm = 1.753 TR u (s) = ———————— s + 2± J3.342 s. 3-4) for the lower bound. (b) A tracking model (see Sec. 3. The plant transfer function P(s)= ——-— s(s + a) 1<K<10 1<«<10 (3.4 rad/sec.6584(5 + 30) .ts =2s TRi : Overdamped .56) which does not affect the desired performance specifications but widens fy between TR[/ and TKl in the high-frequency range. is tentatively identified by 19. Step 1.

these synthesized models should yield FOM within 1 percent of the specified values.(0 the output response has the formXO = e'&cos ht.02 s.76 Chapters and is modified with the addition of a pole to yield 3520 (s + 4)(s + 4)(s + 4. From the data for the log magnitude plots of Eqs.844 s.tK = 1. Step 2.59) are: Mp = l.59) to widen 4/ further between TRu and TRl at high frequencies.735 12." i. Sufficient values of a> up to cy. The desired specifications (see Fig.)\ CO 0. (3.59) The pole at s = -50 is inserted in Eq. Modeling the disturbance control ratio TaJ(s) = Y(s)/D2(s).5) are satisfied by choosing .851 1. the following values are obtained for SR(JCO. \y(f)\ < 0.57) and (3.749 11.224 46. should be obtained in order to ensure that this desired characteristic has been achieved. In practice. The FOM for Eq.594 15. (3. ^(fco) has the desired characteristic of an increasing value with increased frequency.194 12. Let the model disturbance control ratio be of the form Y(s) D2(s) (3.01 for t>t.60) For d2(t) = H_. (c) Determination of <^(/fflj). Since y(0) = 7..538 11. = 60 ms. 3.59). and ts = 1.01 3.e.81 As seen from the above table. it is desired that y(t) decay "as fast as possible. (3.5 1 2 5 7 9 10 15 100 <5K/«*X dB 0.4)(s + 50) (3.

3.95 -116.6° 25 -116.MISO Analog Control System 77 s + 70 ± >18 1 + LD For Eq.1 -168.2° LmP ZP -168.01 at f.7° -2.1 -135° -22.2° -21.65 -104.6° -41.67 -111.4° -33.7° -0. (a) Determination of V [see Eq.3 -111.6 -158.17 -101.6° 25.8 -104.5 -161.7° -8.1° -146. (3.3° 20 -95.4° -13.1 -171.5 -116.8 -101.3° 6.64 A B BBC BC C D AD ADD -116. Determination of the [/-contour.1 -168.6° 19. Analysis of P(s) is based upon the nominal values K = a = 1 (point A of Fig.6° -8.6 -158.6° -14.5° -34.3° 0.4° Note: Values of Lm are in dB StepS.3° -6.6° -135° 13. The template for each frequency is based on the data listed in Table 3.4 Data points for 3P(ja).5° -3 -13.7° 26 -93° 6 -93° 6 -95.25.8° -9 -135° -168.7° -7 -153.5° 11 5 -28. The values in Table 3.7° 5.8 -101.21)] for the plant of Eq.6° -17 10 -40 -174. 3.2 LmP ZP LmP ZP LmP ZP LmP ZP 26 -95.7° -172.5° 7 -153.5° -23 -135° -27 -153.24). (3. TDv(s) = s(s + 70) 1 (3.04 -95.3° -14.4° -28. ~ ft 0565 5.1° LmP ZP LmP ZP -116.3° -20 -174.55).6 -146.) Point a>i LmP ZP 0.2 15 -47 -176.2° 3 -135° 5 -116.4 which is computed using Eq.2 -101.2° -27 -176.97 -116. (3. TABLE 3.5 5 1 -3 -135° 17 -135° 19 2 -13 -153.4 are used to draw the templates ^PQcOj) shown in Fig.61) Steps 3 and 4.8° 13.7° -0. Forming the templates of P(jo)i).55) yields .61) the output decays rapidly so that [y(0| < 0. (3.

3.25 Construction of the plant templates.BC 10 A 1 >AAD 1 2 5 10 ->• a Fig. 3.BBC . •3 3 200 phase (deg) Fig.24 Template points. .78 Chapters K 1i B . .

) are: Frequency at 15 16 20 40 max bLm P(ja). Step 6.63) . For other values of a* the maximum values of ALm P(j(o.4 38.6 39.) dB 38.0 39.65 Therefore it can be seen. Determination of bounds BD(ja)i) for ^ Y ^ I ~~~ where e= 1/L = 1/PG and Y = DTD[r For the disturbance-rejection case: (3.m[P]max =Lm[Ka] = 40dB To determine the value of <% where ALm P «• V = 40 dB. (b) Determination of the Bh boundary. for this example.8./w PQcOi) occurs between points A and C. 3-8). select various values of ®.62) AZ. 3. For a> & 100: Point Lm Pdcot) -80 -60 -40 -60 A B C D From this data it is seen that the maximum value of AZ. as shown below. Select the M-contour that represents the desired value of ML for T (see Sec.MISO Analog Control System 79 lim [P(s)] = Ka and (3. that Vis achieved essentially at a^ « 40. At various points around the lower half of this contour measure down 40 dB and draw the Bh boundary in the manner shown in Fig.

26. ———T- '..•. -320.. -140. (3..-"-'-i S- T"'"" i N^ \ •••••" tea.) and the [/-contour are determined and are drawn in Fig. . -^ 50. -260..26) become the contours for BD. Lm TD = Lm BD. Point A of the templates of step 3 yields the maximum value for /= 1/L. From this figure it is seen that y~ 58*. the ML contour. -3 0. Determination of the bounds B^jca.. zi 'st. "":"" 20.. -ISO. -120.) to satisfy Lm — I = Lm ID <Lm[TDu} (3. _§^_ 0.-•' ^ iP^*—'^——""a i^r- ^——- -r* "•. For this example.e.''"vV *^se. -20.) given in step 2. -280. (3. the values of S^ja). £ ^ 2 4a I1 &7§- " B 30. By use of the templates. 0 phase (deg) Fig. Thus. \ T" ".5 are obtained from Eq..64) Note that there are no lower bounds on the response y(f). / '' 1• 1 -.. choose L > BD (jco. obtained by use of Eq.61). -24 -220. -1:60. BD = BR as shown in Fig. the BD bounds completely dominate over the BR bounds. -340.•.80 Chapters A Lm Y = A Lm = A Lm (3.26 Construction of the B^ca) boundaries. For 0 <co < 15 the contours of Lm TD(ja)i) (see Fig. -20. The data of Table 3. -«0.. .61)..:-". (. Thus. SO.c.65) for each frequency wt of TD(jco. -80. . Si?-' fe ^-^ . are points ofBD(ja)i) on and in the vicinity of the [/-contour.•. 3. i. •--..). -40... and the Bh boundary. 3. the bounds BR(ja>. the values of Lm LD in Table 3. 3. 3.27... 8iiil ss . -1 0. ••""?"" •• .... as shown in the next step. For ca > 20..5. -200. Step 7. -—V 10..) over the frequency range 0 < a> < 15 rad/sec. '•&^^ P -i^" : ^^- ^> : ^--j -10.

26 are more severe and become the B0(ja)i) contours for the overall system of Fig. 3.) contours. 3.28. An analysis of Figs. .27 reveals that the bounds BD(ja). Thus. 3. 3. 0 0 phase (deg) Fig. the BD contours of Fig.27 Construction of the B^jw.) for this example all he above the tracking contours BR(/a>j). then L0(s) must be at least a -340 -320 -3 )0 -280 -260 -240 -220 -200 -IBO -MO -ttto -{20 -1:00 -80 5 J) ^ 0 ~'.28 MISO system. 3. Step 9. (a) Minimum structure.M1SO Analog Control System 81 Step 8. Synthesizing or shaping of L0(s). Determination of the composite bounds 50(/'#0.26 and 3. R •^f •\ J 1 » > P V L A D k Fig. Since P(s) is Type 1.

0737 -0.5 -37.31 75.005 1.84 -1.67 -0.0029954 -0.5 -17. (3.85 79.4 23.01845 -0.1 11.6 86. This requirement may place a severe restriction on trying to synthesize an L0. that the desired system type is achieved. Ljdca.8 36.5 -31.1 -21.65 89. select Lo0(s) = P(s).05 -17.99 4.5 Frequency data for Eq.5 1 2 5 10 15 20 50 100 200 500 1000 2000 4000 Lm TD.) must be as close as possible to B0(joa^ but never below it.degrees 89.001154 <W zrD.6.2 where m = 1 for this example.1 -11.81 56.003 LmLo 43. For the approximate range o^ « 75 < CD.4 82.3 88.) For the case where TD = P/(l + L). it is desired that TD have a zero at the origin so thatX00) = 0 for di(t) = u_i(f). construct or shape L0(s) so that it comes as close as possible to the Ucontour.5 37.17 19.5 17.007 2.453 -0.5 14. For this example the frequency coy is the value of L0(ja>y) which results in Lm L0(j(»i) «• -36 dB. In addition. Thus.91 -9.46 -2. the initial L0(s) has the form ^W = 5 % TABLE 3. Using four cycle semilog graph paper and the data in Table 3.52 -14.61) a> 0.6 -31.67 -4. These restrictions can be satisfied by assuming the format of the optimal transfer function as (3.5 31.82 Chapters Type 1 in order to maintain the Type 1 tracking characteristic for L/(l + L).55 7. the angle ZL0{jco^ > -123° must be satisfied.44 -23. (b) Synthesis of L0(s). (Note: In general. dB -43.67) 0) -P 4 >' — ® CO .0 -35. Therefore G(s) must have a pole at the origin. < coy.4 4.

51 -89.40 -0.30. Since.59 -88.4° -108.29.08 89. the following F(s) is synthesized to lie between Bu and BL as shown in Fig.01 65. 3. Lm Tmin.33.01 -16.43 -82. 3-15.9)(s + 6.71 0 -0.68) where £= 0. and co^ = 200 rad/sec.34 12. 3. The resulting loop transfer function is TABLE 3. and Lm T^ are plotted in Fig.6 is used to assist in obtaining the desired L0(s).26 1.7° -108. (c) Determination of G{s).80 -1. For this example a table having the format shown in Table 3. The input filter F(s) is synthesized to yield the desired tracking of the input by the output y(t) in the manner described in Sec.27 6.0° -112.77 77.90 89. n = w + /t .0 72.69) (s + 2.7. 18.6 Data for loop shaping Angle contribution of poles (p) and zero(s) (z) in degrees ZJL0 Zl P2 ?2 P3.98 -62.53 -75. The curves of Bu.8° -110. Use L0(s) and P0(s). BL.42 88. a^ = 1000 rad/sec. 3.2 real poles. 3. L0(jaf) is drawn on the Bode plot of Fig.20 -10. and a pair of complex-conjugate poles.04 -29.30.17 89.26 -90.26 80. for this example |£0(/fi>)| » 1 over the desired BW.43 48.32 87. Step 10.95 -2.MISO Analog Control System 83 which has w real zeros. Thus.90 83.89 -21.4° -108.4 ^Loi CO 1 5 10 15 50 100 200 -180° -180° -180° -180° -180° -180° -180° 1000 2000 -180° -180° 32.2° 227.91 88.0 -136.98 -149.54 89. Since L0 is Type 2.85 (3. where P0(s) = l/s(s + 1) is the nominal plant. to obtain the transfer function G(s) = L0(syP0(s).53 18.6° + l)(s + 200)(s +14700 + yl5000) (3.20 -4. then F(s) = Tg(s). It is also plotted on the NC of Fig. then TD = 1/(1 + L) has two zeros at the origin.99° -181.5) .6° -111.

10. Fig. 3. Lower iSouad! "TOO.84 Chapters -180. -3 a I -20.30 Requirements onF(s) for Design Example 1. (3.68). . Fig. 3.j Upper Bpurid I -.29 Bode plot ofLJj'a) of Eq. .

C. (3.7 Time response characteristics for Eq. substitute the corresponding data into 1 -D(s) (3. The disturbance response plots.3Xs + 1.01 0. andLm L0(/aj) given by Eq. thus Lm TDl = Lm Mmj> = -20 dB.93x10 (5 + l. For Step 2.i(f).007 Point A B y«(0 0.wdOf = 0.6)(.00998 0.06 J>(°°) 0 0 0 0 . 3.058 0. The synthesized L0(s) is 9. 01 for tx > 60 ms is achieved.(t) = u. are shown in Fig. (3.06 C D 0.00004 0. d2(t) = 0.002 and 4 = 1.71) s2(s 26)(s + 900 ± y975) TABLE 3.06 0.31 for the entire region of plant parameter uncertainty. Time responses for a disturbance input: d(f) = u. 3. A QFT CAD (see App.71). 01. For each point of Fig. the value of lxo(OI ^ 0.0090 0. Steps 1. As shown in the figure.70) 1 + P(s)G(s) and determine yD(f) for each point.1. MpaMm = 0. B0 bounds. the disturbance time response characteristics for points A. 3.00226 0. where MP = 1. and D satisfy the specification that lXk)| ^ 0. 01 for tx > 60 ms. B. for the 8 cases.000399 Specified tx 0. Step 11. 3-5.606s. A) is used to obtain Fig.33 which shows the Ucontour. with r(t) = d.? + 45) (3.06 0. 1.t(t).7.70) Actual tx for Wt)| = 0. and 7 are the same as for example 1.0074 0. although the initial peak overshoots exceed the magnitude of 0. As indicated in Table 3.MISO Analog Control System 85 The output responses for this example with a unit step input are all identical. 3-18 DESIGN EXAMPLE 2 The plant and tracking specifications of Design Example 1 are used for this example.32. as shown in Fig 3.24.

3 0.86 Chapter 3 0.1 25 t (sec) Fig. 3.32 Disturbance responses plots for allJ plants.8 04 0. 3.31 Time responses to unit step input for all J plants.9 0. . Fig.2 0.

-'0. and Lm L0(ja>. B0 bounds. -3HO.-2. (b) All of the L. -80. -180. -120. $/?40/-220. -340. -% . -200. -100. -200. -40. -240. -260.33 (a) ^/-contour. -60. 0 phase (deg) (a) 0.M1SO Analog Control System 87 . -320. -140. -20. -320. -2 SO. 3. -80.-340. -20.-180. -280. 0 phase(deg) (b) Fig. . -3 0. -120. -1 0.) for obtaining L0(s). -60. -140.(ja>) plotted on one Nichols plot along with the stability margin requirement. -160. -160.

57) and (3.34.34 Requirements and resulting prefilter for design examples.022 and ts < 1. The simulation results are shown in Figs.03 thus satisfy the specification [y^O™*! ^ 0. 3.35 and 3. (3.62 s.33 are used to obtain the data for plotting BL. proper care must be exercised in analyzing the angular variation of f\jaj) as the frequency is varied from zero to infinity.Zj) j&(j(0 . Thus. These figures reveal that all disturbance responses are below 0. Lm T^ and Lm Tmai in Fig. 3. As an example. 3.72) where pt andp2 are a complex-conjugate pair and i Upper Bburid ! I I M I Lower Bound i M l ! Fig. Also.36. the tracking responses all lie in the range of 1 < Mp < 1. consider the plant K(jco .P2)(j(0 - (3.P^jO) .88 Chapters Equations (3.59) and Fig. the desired performance specifications have ah1 been met and a robust design has been achieved. Since the tracking specifications and models are identical for both examples. . 3.. 3.69) and is plotted in Fig.1.34. 3-19 TEMPLATE GENERATION FOR UNSTABLE PLANTS In the generation of templates for unstable plants. BL. the prefilter for this example is the same as given by Eq.

2 1.4 0.36 Tracking responses for Design Example 2.M1SO Analog Control System 0 02 0. . 3.4 1.8 1 1.8 2 Fig.35 Disturbance responses for Design Example 2.8 2 t(sec) Fig.2 1. 0 02 0.4 0.6 1.6 1.6 0.4 1.6 0.8 1 1. 3.

(3. and in Table 3. > <Odr All other angles are positive (ccw) for all 0 <co<°o as illustrated in Fig.72). in the limit as a) -> oo.-) are positive (ccw).73) In Eq. (3.72). 3. (b) the angular directions of all RHP poles and zeros are all taken ccw. The following guidelines should be used in the angular determination of </P(jcOi): TABLE 3. 3. as illustrated in Fig.8. 3. (3. The poles and zeros of Eq.) is negative clockwise (cw) for 0% < cod. because all angles are measured ccw. all angles of first-order factors (/«.5 +90 +135 +116. all angular directions are taken so that their values always he within the range of-90° to +90°.8 Angular variations of the minimum-phase stable and unstable plants PQ'co) [see Eq.5 +90 0 +63. 0 +45 +90 0 +45 zf(M) -90 -172 -270 -450 -405 -270 1 0 1 oo 2 0 1 oo +90 +90 +90 +90 +90 +90 -45 0 +90 +225 +180 +90 +45 +63. The poles and zeros of Eq.37 Case a>t Zp0 Angle in Degrees 4>i Z/>2 Z/>5 Zz. In Eq. (3. 3. for an unstable plant are plotted in Fig.72) the angle of (/«. (3.90 Chapter3 Z lim 01—»°0 =-270° (3.37a. Thus.37a. For stable plants. -/?. and is positive counterclockwise (ccw) for o>. . (3.72) as follows: Case 1.72) for a stable plant are plotted in Fig.8.73) the convention used is that.5 +90 +90 1. For Eq. 3. Two cases are analyzed for Eq. (3.37b and in Table 3. This feature is very important in obtaining a template where P contains both stable and unstable plants. For plants with RHP poles and zeros: (a) the angular directions as for the stable plant left-hand plane (LHP) poles and zeros of P(s) are taken in the same manner. -/?.5 +90 0 +63. Case 2. Note that the difference in the angle ofPQ'O) between the two cases is 360° and is necessary to account for the right-hand plane (RHP) poles.37b. eachjco term contributes an angle of +9CP. the angle of ftyo) varies continuously in a given direction as the frequency is varied between zero and infinity.70)] shown in Fig. 2.

1. 3. 2. The robust design technique of this chapter permits the design of an analog control system that satisfies the desired performance specifications within the specified range of structured plant parameter variation.MISO Analog Control System 91 (a) (b) Fig. 5 a MTMO system can be represented by an equivalent set of MISO systems. This design is based upon 1.37 Poles and zeros in the s-plane for (a) a stable plant. Finding the resulting bounds on the loop transfer functions Lt and input filter transfer functions F of Fig.m. (b) an unstable plant. the QFT technique can be applied to n. it is referred to as a MISO design technique.p. 3. plants and to digital control systems (see Chapter 4)14-17'27-52 . Two QFT MIMO system design approaches are available in which the equivalent MISO loops are designed according to the MISO design method presented in this chapter. respectively.27. plants. It is a frequency-response design method applied to the design of a MISO control system with an uncertain MISO plant P. Since the QFT technique is based upon the design of the loop transmission function L(s) of the MISO control system of Fig. Although this chapter deals with m.p. P ={P(j'&)} and ^jco) = {T(jaj)}. As will be shown in Chap. 3. 3-20 SUMMARY A general introduction to the MISO QFT technique is presented in this chapter. Specifying the tolerance in the oxlomain by means of the sets of plant transfer functions and closed-loop control ratios.

The QFT technique requires.m.13 That is. The advantage of this approach. is another DIG technique that allows the QFT design of the D(z) controller to be done in the s-domain. as discussed in Sec. in the manner described in Chap.or s-domain of the S-D MISO or MIMO control 92 .. when the plant is rap. the difference being that the design must take into account the right-half-plane (RHP) zero(s) that result in the w'plant transfer function due to the bilinear transformation. plant and the problem associated in satisfying the stability bounds.p.m. and the sampling time T (or sampling frequency to. the determination of the minimum sampling frequency (<%)„». plant when transformed into the w -domain. a bilinear transformation. The use of the z. the performance tolerances. it becomes a n. Thus. 3. The larger the plant uncertainty and the narrower the system performance tolerances. 3. when a plant P(s) is m.14 The QFT sampled-data (S-D) system design process is tuned to the bounds of uncertainty. the transformation either into the w. 4-5. the larger must be the value of (a)s)min.p.p. to obtain D(z). which is discussed in Sec. Once the ^-domain controller has been synthesized.DISCRETE QUANTITATIVE FEEDBACK TECHNIQUE14 4-1 INTRODUCTION1 This chapter focuses on the application of the QFT technique to MISO sampleddata control systems.to the w' domain bilinear transformation13 permits the analysis and design of sampled-data systems by the use of the digitization (DIG) technique. Thus. 48. it is transformed into the z-domain by use of the Tustin transformation. the w-plane detailed QFT design procedure essentially parallels very closely that for continuous-time systems of Chap. bandwidth (BW) that is needed for a satisfactory design. proper care must be exercised in satisfying the stability bounds prescribed for the QFT design The pseudo-continuous-time (PCT) approach. = 2n/T). is that it eliminates dealing with a n. Note.

1 w.AND w'. 4-2.or s-domain simulations satisfy the desired performance specifications then by use of the bihnear transformation the z-domain controller G(z) is obtained With this z-domain controller a discrete-time domain simulation is obtained to verify the goodness of the design. with minor exceptions.w = —————^—t—i———— T T 2 + sT+(sT) I2\ + — (4.to the z-domain transformations.1) (4. each require the use of a bilinear transformation. w should approach s.DOMAIN TRANSFORMATION The QFT design of the sampled-data control system requires the use of the bihnear transformations z = ———— -w +1 \v +1 (4. w . 4-2 BILINEAR TRANSFORMATIONS13 The two DIG techniques.Discrete QFT 93 system enables the use of the MISO QFT analog design technique to be readily used. in the w 'plane the desirable property that w' -> s as T -> 0 is achieved This w-plane property establishes the conceptual basis for defining a quantity in the w' . to perform the QFT design for the controller D(wl) or D(s). w = lim 2 + sT (4. If the w'.or ^-domain design.e. This section presents the z-to the w-domain to the z-domain and the s.3) and where This situation is overcome by defining 2 2 sT + (sT) /2! + --w' = ..4) Thus.2) 1 Note that the w domain lacks the desirable property that as the sampling time T approaches zero. i.

(4.6) -Tw' + 2 Equation (4.5).1) and (4. From now on. w' = w) except where noted in Table 4.94 Chapter 4 domain.to zplane transformations as follows: . Substituting w = Tw 72 into Eqs.1.6) represents an approximation of z = esT. the w' transformation is used throughout the text and the prime designator is omitted (i. 2z-l w = ———— (4. 4.. respectively. .e. (4. which is analogous to a quantity in the 5 domain.1.5) into the w-plane is shown in Fig. the z to w'plane and the w. The mapping of the z-plane via Eq. 4. JCD s plane Stable Unstable Region Region j 1m z Jfflw w' plane Stable Region Unstable Region Unstable region I Fig.1 Mapping of the j-plane into the z-plane by means of z = #T of the z-plane into the vr-plane by means of Eq.2) yields. (4.5) T z +l z.2 s-PLANE and t^PLANE RELATIONSHIP The relationship between the s and w plane can be found by examination of Eq.• Tw' + 2 (4. 4-2.

8) to obtain 2tanh(<r r / 2 ) —————-—————— (4. If WspT/2 is small ((DspT/2 < 0. substitute s = a^ into Eq.Discrete QFT 95 (4. substitute s =70^ into Eq. (4. 4.12) +a Fora 2 «2.8) to obtain 2tanhQ-a> ff> r/2) T 12) (4. (4.e. (4..7) sT sT - esTI2 _ g-sT/2 esTI2 + g-sT/2 which yields 2tanh(s7V2) (4./V = awp + ja>wp (4. the imaginary axis in the primary strip of the s plane is mapped onto the entire imaginary axis of the w plane (see Fig. tanh a can be expressed in the expanded form tanha = a+a 2 (4.297). i.1).13) . For the real axis.9) or 2 ten(a}spT 12) v = °>wp = (4. then from Eq.8) T For the s-plane imaginary axis. w = M + .11) By letting a = o-^T/2. (4. (4.6) in terms of z = e"T. 12).10) Thus. fromEqs.10) obtain v = a)wp sun^. 11) and (4.

7320 3. one whose order of the numerator wz and order of its denominator nz are unequal.0018 5.and w -domain values of v.26795 0.11) must be used to locate the s-plane poles and zeros properly in the w-plane.55 ——— ___ _ — 0.013 0. The mapping of the s-plane poles and zeros into the w-plane by use of these equations is referred to as prewarping of the s-plane poles and zeros.3032 3.. Table 4.0 2. J°>sp =s (4.619 76.0 70. when the approximations are valid. (4.63 44. into one for which the order of the numerator (ww) is equal to the order of its denominator («„).14) is very good and how it is dependent upon the sampling time T. (4. A characteristic of a bilinear transformation is: in general.0 60.76733 1.10) and (4. The relationships of Eq.26795 0. it transforms an unequal-order z-domain transfer function.5 1. the value of the gain constant Kv may be positive or negative.14) If the approximations are not valid.values Wi values <»s=120 0.0002 1. 0.13165 0.057 20.559 285. for cos = 120 rad/sec and cos = 240 rad/sec s-plane w-plane v.values Valid ^=120 .0 1.053 38.13 The sign of Kv is determined by the coefficients of G(z).0 40.0524 0. Further. (4.47 31.0 *-«.5).14) are the basis of a QFT design method (DIG technique) in the w-plane.0 1.41420 50.13165 0.0 5.0 100.108 — 4 4 i —— —— .1 illustrates the frequency range for which the degree of accuracy of the relationships of Eq.0 0. »z & w2 in the z domain and «„ = ww in the w domain. (4..11 58.5000 1 1. This characteristic must be kept in mind when synthesizing G(w) and F(w).0071 10.0 30..57735 0. That is.73205 ___ ___ w -plane v.2 66.394 99.0 120.0287 10.57735 1.0 10.01309 0.0004 5.5000 0.00655 0.06554 0..732 — — 2.15 142.02618 0.0 20. TABLE 4.02619 0. note that in transforming G(z) = KxG'(z) to G(w) = KwG'(w) by means of Eq.0000 1 2. then Eqs.= 240 approximation Range 0. which in turn are functions of T.96 Chapter 4 Thus.1 w.235 22.

for q = 1 Eq.to z.17) Utilizing the mathematical expression 1-a and letting s = jwsp in Eq. (4. Eq. 17) the following expression is obtained:13 . both equations are from now on referred to as a Tustin transformation. (4. (4.1 (4.to z-plane transformations is the Tustin algorithm.15) An advantage of the Tustin algorithm is that it is comparatively easy to implement. The following discussion is useful therefore in understanding the mapping result for the vector model.16) which is a bilinear transformation and can be equated to the trapezoidal integration (s~!) method Note.e.Discrete QFT 97 4-2. Eq. The Tustin transformation forq = l is defined as 2 1-z"1 2 z. (4. the accuracy of the response of the Tustin z-domain transfer function is good compared with the response of the exact z-domain transfer function.domain transformation of Eq. the accuracy increases as the frequency increases.56'51 The function of z that is substituted for sq in implementing the Tustin transformation is (4.to z-plane mapping. Also. (4.3 s. i.16).16) is rearranged to yield l + sT/2 z = ————— 1-57/2 (4. 16) is identical to the w.. Thus. To represent functionally the s.TO z-PLANE TRANSFORMATION: TUSTEV TRANSFORMATION13 One of the most popular s. The operator equation (4. 16) is as applicable to matrix equations as it is to scalar differential equations. 16) can be derived by approximating z = esT as a finite series. Also.

98 Chapter 4 The exact JT-transfbrm yields z = e}a>sp . Warping Direction of ffi^ jffl/2 Warpuig Direction of 8^ (b) Fig.19) (4.2 Map of = 2(tan co^WyT. (4.20) (4.298 A 0.5774(30°) 0. 4. .21) it/2 j-piane values | 0.3057(17°) -x/2 Tustin value. (a) Plot of Eq.18): e Equating the exponents yields =exra j2tan . Thus. the following equation is obtained from Eq. A -t tan or tan- 2 (4. where cosp is an equivalent s-plane frequency.. rad (deg) (a) .789 ^ 0.19) and (b) warping effect.523 (I 0.. (4.

23) results in the expression (4. in this approximation. there is again a warping penalty. the entire imaginary axis s-plane is mapped once and only once onto the UC. it is easy to realize that the imaginary axis of the s plane is mapped into the unit circle (UC) in the z plane as shown in Fig. 4. 4. The same stability regions exist for the exact 2-transform and the Tustin approximation.21). its numeri- cal coefficients adjusted) to finalize the design since approximations have been employed.2.21) generates d)sp. overshoot.30 rad. Proceeding in the same manner as used in deriving Eq.Discrete QFT 99 When 0)^712 < IT. (4.e.22) substitute z = e°sPT and s = asp into Eq. (4. (4. and settling time. The digital compensator (controller) must be tuned (i. Eq. or « 0. The left-half (LH) * plane is mapped into the inside of the UC. (4.T/2. As seen from Fig.22) Which means that in the frequency domain the Tustin approximation is good for small values of o>^. 4. The spectrum of the input must also be taken into consideration when selecting an approximation procedure with or without prewarping. Thus. 4. (4.2. consideration of the warping of the real pole component is now analyzed as a fine-tuning approach. To compensate for the warping.3 rad.1..19). However. Also.24) . (4. The continuous controller is mapped into the z plane by means of Eq.22) to yield T _ \ + aspT 12 \-6spT/2 Replacing e '" by its exponential series and dividing the numerator by the denominator in Eq. The real part of the s-plane pole influences such parameters as rise time. (4. Returning to Eq.16) using the prewarped frequency &sp.13 Compensation can be accomplished by using Eq. The prewarping approach for the Tustin approximation takes the imaginary axis s plane and folds it back to n/2 to -itfl as seen from Fig. prewarping of co^ by using Eq.2. which is depicted in Fig.22) is a good approximation when t»vT/2 and a>spT 12 are both less than 0. (4. It should be noted that in the previous discussion only the frequency has been prewarped due to the interest in the controller frequency response. then cosp*d)sp (4. The Tustin approximation prevents pole and zero aliasing since the folding phenomenon does not occur with this method.

|.1 "T s—mmmmm.. the zeros and the d-c gain are usually different! A characteristic of a bilinear transformation.3.25) satisfied..?. with Eqs. 4. A matched ^-transform can be defined as a direct mapping of each s-domain root to a z-domain root.14 5.6114 -2 -0. Fig..22) and (4. 4-2. The poles o£G(z) using this approach are identical to those resulting from the exact Z transformation of a given G(s). it transforms an unequal-order transfer function (ns * ws) in the sdomain into one for which the order of the numerator is equal to the order of its denominator (nz = wz) in the z-domain. 4. « > .-:T:-T.. -0.then (4. the Tustin transformation is employed for the DIG technique in this text.j Q.25) Thus.—-.:.100 Chapter 4 (or 1 » \trvT/2\) and 1 Tsp . is that. s + a -> 1 -e'aTz'1. 4. the Tustin approximation in the s domain is good for small magnitudes of the real and imaginary components of the variable s.. (4.3 represents the allowable location of the poles and zeros in the s plane for a good Tustin approximation.2. in general. JO) s-plane . The shaded area in Fig.. as pointed out in Sec.3 Allowable location (shaded area) of dominant poles and zeros in s plane for a good Tustin approximation. However.6114 T T -jlli JT Fig. Because of the mapping properties and its ease of use. . = . This characteristic must be kept in mind when synthesizing G(s) and F(s).2b illustrates the warping effect of a pole (or zero) when the approximations are not satisfied._ .

p.m.p.p.m.p.26) where L'm (5) is a m. (4. The frequency domain characteristics of A '(s) are determined by substituting s =. L(s) = (l-Ts)L'm(s) (4. (b) denominator contribution (j>D > 0°. (4.27) TS where 1-rs 1 + TS (4. plant has only one RHP zero is considered in this section. 4.26) is modified to l-TS Ts)L'(s)]A'(s)Ln(s) (4. 4.m. Thus. 90° (b) Fig.29) is rap.28).4 Analysis of the angular contribution of Eq. For this case.28) is n. .p. transfer function whose gain constant has a positive value./ft>intoEq. The case where the n.Discrete QFT 101 4-3 NONMINIMUM PHASE ANALOG PLANT The analog QFT design technique for m.31: (a) numerator contribution <j>N<CP. and (4. Eq. plants can be modified when the plant is n.

lag characteristic has essentially no effect on the determination of the optimal loop transmission function L0(s) or Lmo(s).32) reveal that^'f^ is an all-pass-filter (a. where the magnitude of the angle of Eq. In the low frequency range.p. 4. 4.p.Irm^ + (TO)2 + (rw)2 (4.33) is very small.p. (4. Whether the a.33) contributes an "extra" phase lag to the m. function Lm(s) as shown in Fig.f. 4.tan (<0r II) .#D (4. the a. Eqs.30) U' ZA'(jca) = tan Jl 4. \\ \ Fig.4). Thus.f) where its angular contribution $A' ~ &N ~ 0D ~ 20N < 0 (4.5.102 Chapter 4 A'(jco) = A' ZA'(jco) where (4.31) {—COT/ 1) .p.31) and (4. (4. 0 < cot < COL.5 An analog all-pass filter angle characteristic.f.32) and fa=-fo< Op(see Fig. .'s angular contribution is detrimental in achieving a \ \ Jj -90° a.

6 shows the B'h .ZA'(jco) < 0° .34) is rearranged to which yields ZLmo (jo.ZL0 (y'fli) . (4. For the nominal plant P0. i.(jco) < 0° then Eq.) + [-fA. 3.-|. plant having a RHP zero at z. from Eq. (4.m.m.19 are now relabeled as B'h -contours and B'0 bounds..) .1 ANALOG QFT DESIGN PROCEDURE FOR A n.contour) . (yo») = ZL0 (jo. (4.e. for each value of frequency corresponding to the template frequencies.p. analog plant: 1./. Note that the 5/. Obtain the (/-contour (B'h . is ®# = 0. and the BD.36) Since 0A. This is accomplished as follows: Eq. then ZL mo This results in the restriction that ^ > -/Stf5 -(-73(7) = -5(f. If at co = co^ <j>A'< -SOP. Figure 4.34) 2.p. 3 is modified as follows for a n. = co. BR and the B'0 bounds corresponding to the template frequencies cat for L0(s) in the same manner as for the analog plant.^. The limit on an achievable ea^ for a n.Discrete QFT 103 satisfactory Lmo(s) can be determined by the following approach: if the QFT design requires a phase margin angle 7 = 4(f at <».36) implies that the B'0 bounds and the (/-contour.52 4-3. for which -90°. are shifted to the right by \(/>A' to yield the corresponding .p. y < 4(f at co = ftty resulting in a larger value of Mm (or Mp).contour and the B'0 (jca^ ) bound It is necessary to modify the angle characteristic of LnoQco) in order to compensate for $A.(ja>)] (4.-contour and B0 bounds of Fig.m. PLANT The analog design procedure of Chap.27) the nominal loop transmission function is denoted as (s) (4.5|z. respec- tively. then the actual phase margin angle will be less than 40f...

104 Chapter 4 -288° -260° -2«° -210° -200° -180° -160" -M0° .p. 4. B^-contour ——' / -180° r B^forco. L0(s). . 4.2 °.0 ° 1 0 1 0 *0° 40° -20" 0* Fig. Lmo(jo)) needs to be here at co = o^ before the U-contour reaches the 0° axis Fig.7 The shifted U-contours for an n.p.) L(s).m.6 Nichols chart characteristic for a nonminimum phase (n.m.

one needs to exercise judgement as to where to try to locate the initial zero(s) that is (are) inserted into the synthesized function.6. for the case where only the stability ML requirement is satisfied. That is. Before proceeding with the design. and the corresponding 5/.7. For example.37) 6. the synthesized L^s) lies on or above each Bmodtoi) bound and is to the "right" or to the "right bottom" of the shifted {/-contour (^-contour).) to Lm L^jco^P. in order to achieve this decrease. the slope of Lm Lmo(j(o) must be 60 dB/dec.+i) bounds let alone yield a stable system. In synthesizing Lmo (s) .. Design the filter and perform a simulation in the same manner as described in Chap.-contours. as shown in Fig.e. locate the zero(s) as far left from the imaginary-axis as possible. If this occurs. 5. Form L0 (s) = A'(s)Lmo (s) in order to obtain G(s) = -^ (4.Discrete QFT 105 Bmo bounds. . satisfy only the ML specification. in one plus decade. as shown in Fig. 3. 4.+/ = 100. i. Thus. In order to minimize the value of t. Accepting a larger value for ML results in a "shrunken" {/-contour and may allow a QFT design that satisfies all the specifications. then a QFT design can be achieved to only meet the stability requirement. AdB = 66 dB. 3. 4. 4. This value for the slope results in a large "phase lag" characteristic which can force Lm L^JJai) to decrease so fast that it can not satisfy all the Bmo(ja>. let AdB = [B(ja>j) = 16dB] -[B(jcoi+J) = -50dB] = 66dB where a>j = 3 and £». the question that needs to be addressed is: can a realizable L0(s) be synthesized that satisfies all the bounds (Bmo bounds and B'h -contours) and yet be able to go from a dB value of Lm LnJjo).

4.Y*(s) (s) Note. Y*(s) represents the ideal impulse transform of Y(s).38) L(z) 1 + L(z) where = GzoP(z)Gl(z) (4.8 A MISO sampled-data control system.106 Chapter 4 4-4 DISCRETE MISO MODEL WITH PLANT UNCERTAINTY As the state-of-the-art of digital computers makes great strides.e'sT )/s. e. In this figure the ZOH unit represents a zero-order-hold device whose transfer function is G20(s) = (1 .39) (4. Thus..40) d(t). it is important to extend the QFT continuous-time system design technique to an S-D control system such as represented by Fig.g. D(s) f _^> : Filter i(kT) R*(s) v(kT) N e(kT)^ Controller ^^ V*(s) tSj^E*(s) 1 x(kT)^ ^^ m(t) J-T^ w(t) ^ ^^ M(s)'+<-' W(sf PI ant iV ' ' Y*(s) F*(s) ^*' X*(s) s) -"—— y(t) Y(s) Y(kT) ^ Y*(s) Fig. 4. such as digital flight control systems.8. .8 are Y(s) = P(s)W(s) X\s) = G*(s)E*(s) W(s) = M(s) + D(s) E*(s) = F*(s) . These equations are manipulated and transformed to yield the input/output z-domain relationships: Y(z) = L(z)F(z) PD(z) *R(z) ————— = YR (z) + YD (z) = TR (z) + YD (z) (4. digital control systems are playing a more important and greater role today. the starred functions represent impulse transfer functions. The basic equations that describe the MISO system of Fig. 4.

for a unit step disturbance input function. is readily determined.Discrete QFT 107 P(s) (4. 4-1 the design techniques that have been highly developed for the s-domain can readily be applied in the w-domain. (4. that (4. (4.46) Substituting Eq.48) and thus (4.47) (4.z -i )pe (z) = Gj (z)Pz (z) Note.42) (4.45) (4. (4. D(s) = 1/s.40) into Eq. Once L(z) has been synthesized then the controller G/(z).41).41) P(s) (4. This feature enables the . which is to be implemented.44) \ + L(z) 1 + L(z) _ PD(?) 1 + L(z) (4. 4-5 QFT ^DOMAIN DIG DESIGN As discussed in Sec.43) PD(z)=Z[P(s)D(S)} TR(z)= F(z)L(z) (4.49) The QFT design is based upon the uncertain plant being defined by Eq.39) yields L(z) = G!(z)(l .

54) =. then the approximation 3P(s) « 3P(w) (4.1 For the plant .tan —— = ^t \T) ^2) n 2n — T z (4.53) (4. both conditions of Eq. 4-2 are repeated below: 5 = J°>sp = (4-50> Tw + 2 -Tw + 2 z-1 (4. If for the range of parameter uncertainty and for a specified value of T.52) (2} ( coT\ co.56) are both satisfied then s » w. (4. v = . (4.297 co T (4.108 Chapter 4 QFT design technique for analog systems to be utilized for the design of discrete systems in the w domain if certain conditions related to T. The w-domain QFT design procedure presented in this chapter is based upon a stable uncertain plant. Example 4. z-. hold.51) = awp + j<ow =U+jv = I wp w (4. the sampling time. and vc-plane relationships from Sec.55) If -W— a T «2 and -^— < 0. The pertinent s-.56) are satisfied in the low frequency range.57) is valid in the low frequency range.

Example 4.88. Thus. As the value of a approaches 20. and the w = u +jv notation is used for the w-plane representation. (4. the double subscripts in Eq. Remember the 1/s in where Px is the LTI portion and Py is the plant uncertainty portion of P(s). (4. for 0 < a < 10 and 0 < Osp < 11.0625«2 -» 1«32 For both ranges of a: co^T/2 < 0. (4.297 -> co^ < 11.59) . Note.05 s. the "s" in the denominator of Eq.58) s(s + 2) s +b obtained by using either P(s) or P^s) will be identical. (4.88 Eq. then the QFT design can proceed in the w domain precisely in the same manner as for the continuous system. from the low side. Note. Consider two cases: (a) 0 < a < 20 and (b) 0 < a < 10. Thus. then Eq.14 For an achievable QFT design it is required that in the \im[PZl (jv)] (4.2 Given the plant P(s) = = Px Py X V (4. (4. for S-D control systems for which the minimum value of sampling time has not been specified.57) is a "fair" approximation with some warping. for simplification purposes.57) is valid.56) which yields: (b) « 2 = 0.Discrete QFT 109 K P(s) =————— s(s + a) where 1<K<10 and l<a<4 let T = 0. The range of values for which Eq.50) are dropped for the s-plane representation.41) comes from the zero-order-hold (ZOH) unit and is not part of the plant P(s).1'214 The design can be successful only if o)s is large enough for the specific plant uncertainty. If (os is fixed a priori.57) is valid is determined by utilizing Eq. it is best to assume initially that cos is not known a priori and the derivation of the minimum cos needed becomes one of the principal design problems. (4.

. Equation (4..J.p.1 CLOSED-LOOP SYSTEM SPECIFICATIONS Figure 3.63) S*R(jto)= Bfr(ja>) .62) for all P in 9> then for a discrete QFT designed system Eqs.60) is valid for co <0. (4.60) where H(jaj) is the analog transfer function. 4.. for ideal impulse sampling. do not change 4-5.4b represents the upper and lower bounds for the desired tracking responses and Fig.62) are modified to \BL (j B (4.B*L(jw)\> AT£( (4. 13. The reason for this is that for a> > 10. H*(jaj) is its corresponding impulsed sampled function. The specified BL and Bv bounds1 for the design example to be used in this chapter is shown in Fig.9)..63) to (4.65) are valid up to where BL is approximately Lm ah = -24 dB (o)h = 10 rad/sec in Fig. and co « cos. 4. so the allowable \TR(jco)\ uncertainty (i.110 Chapter 4 the resulting values of gain for all J plants.64) BL(JCD) Bu (jco)\ (4.9. Thus. For the QFT technique it suffices if Eqs. where for the analog QFT design it is desired that <LmTR(ja>)< (4.BL(ja>) (4.19 The latter case applies for the w domain QFT design for a system whose plant P(s) is m. BL decreases rapidly.e.. that (4. 3. The S^jco!) specification suffices to control the time domain response for m.5 represents the upper bound for the desired disturbance response.61) and 6R<Ja>) = Bu(jco) . As derived in Ref.65) which is valid for a "low enough" part of the frequency spectrum. (4.61) and (4.p.1cos where 0.2. systems and those for which the RHP zeros are known. BV-BL) soon exceeds the plant uncertainty. .1 is a good engineering rule-of-thumb value. where i = 1.

65) for the sampled model which is based upon the continuous model. then the following additional requirement is imposed: ZL >-180 (4. .45).i dB Note: Curves drawn approximately to scale. If conditionally stability is not allowed.67) o. it is very reasonable to simply use Eqs.44) and (4.66) l +L is a constant for all co and for all Pe in 9>e. Thus.40 This tends to force to. 4. (4.9 Bounds on LOT T^jat) used for Tj^jco) up to the -24 dB point. as indicated in Fig. A loop stability requirement similar to that done for the continuous system must be added to the disturbance response performance requirements. in feedback systems.63) to (4.5. to be significantly larger than what is required by the sampling theorem. Fig. (4. For example: is the range 0 to 10 rad/sec "low enough?" In general.Discrete QFT 111 The requirement for a reasonable stability margin independent of BL and Bv is to maximize the value of co at which the -24 dB value occurs on the BL curve. the price paid for the benefits of feedback is in the BW of the loop transmission L of Eqs. 3. For the example in this chapter a value of Lm ML = 6 dB is used.' This requirement (a damping constraint) is: L (4.

coc is the crossover frequency [|L*(/wc)| = 1]. and 10 rad/sec calculated for co. 3) from Eqs.LmF = Lm = LmT'R (4.40—— [ --- -180° -120° Fig.44) and (4. 5. and \L*(/ri)\ < lforat> a>c. 2.68). the designer finds the minimum L(z) [orL(/v)] to satisfy Eqs.68) l +L As it is seen from Eq.5 NN \ -20 ~\ --t10 -. For the QFT technique. = 60 rad/sec. can be obtained (see Chap. 1. in order to proceed with a QFT design. = 0. it is necessary at the outset to determine the bounds and the values of ML and y.(4. dB 20 \ \ 0.14 .10 Plant templates at &. (4. The change in TR due to parameter variation.63) to (4. A\Lm Tp(z)] can be made arbitrarily small by choosing L(z) sufficiently large..45) in the logarithmic domain and is expressed as follows: A(LmTfi) = LmTR . (4.5. 4.67). but almost identical for cos = 120 and 240 rad/sec and for the continuous-time system.112 Chapter 4 where y > 0 for co < coc. Thus.

a correlation is made between the values of co w = j tan(c». The appropriate values of v.2 PLANT TEMPLATES As is required for continuous-time control systems. Since QFT requires the shaping of L to satisfy certain bounds..8 in the w =jv domain. once again the prime designator is omitted (i. it is also necessary for S-D control systems to determine the plant templates 3P<. The number of plants J within the 9 = {Pe). K = 1 and a = / are selected as the nominal plant parameter values. are essentially the same as the plant templates 3Pe(jv$ = {Pe(jVi)} for the S-D system. = (2/T)tan(o)iT/2)} are shown in Fig. 4-5. Since at this point of the design process the desired value of cas has yet to be determined. A number of plant templates [for v.e.6) to be used for the templates are available from Eqs.. The template 3Pe(jvi). in order to stress the relationship between (cas)min.. 4-2).25. the plant templates -3P(/<w. The choice is intentionally made low. Hence. 4-5. where i = 1.2. 4-5.1 calculated at CDS = 60 rad/sec.(/ej. these "plant templates" are functions of the sampling frequency cos. (4. henceforth. Note. Therefore. To accentuate the value of doing a w-domain design [via Eq. (see Sees. For the values of cot < 0)0.)} at a "sufficient" number of o. After an individual achieves a sufficient understanding of the QFT technique. and to the primary strip 0 < jcos/2 in the s plane (see Sec. a good estimate can be made a priori for the value of (cos)min.r / o)s) and w' = jv = j ——tan(ft«r I <os) n for various values of o. is chosen to adequately describe the contour of the template which represents the region of plant uncertainty for each «. < cos/2.5)]. but this becomes apparent only for co > 0. These templates are very similar to the templates for the continuous-time system and those for o)s = 720 and 240 rad/sec.. (4. in Fig. values.25cos = ctoCv. As a consequence. where 0 < v < co.50) through (4.. and the performance tolerances.J. corresponds to the upper half of the unit circle in the z plane. only after cas is chosen. In order to determine the bounds.3 BOUNDS B(jv) ON L0(Jv) Recall that the^v-axis of the w plane. a nominal plant Peo(w) must be selected Thus.) = {Pe(j(o. a w -domain design [via Eq. the tentative value cas = 120 is selected to initiate the design process. is shifted on the NC to de- .2)] vs. w' = w). then such shaping is difficult to do within the unit circle in the r-domain. with its corresponding Srfjv. the plant uncertainty.1. 4.6). in the frequency range 0 < CD. 4.5 and 45. > vft2j) (an approximate engineering value. in Table 4. the same design techniques that are used to perform loop shaping in the s =jo) domain can be used.54). it is therefore reasonable to tentatively use the continuous 3P(jco.) templates in order to proceed with the design and eventually to be able to determine the required value of a>s.)}> for the continuous system. it is best to design Fand G. For S-D systems. (4..) = {P(/fi>.) value.Discrete QFT 113 4-5. see Sec.10 for the plant of Example 4.

' .5 1 0.70) w + 2/T dB 0 -20 -40 -90° Note: Curves drawn approximately to scale (0 0° p V 0. respectively: (4.11 Unsuccessful design at cos = 120 rad/sec.03 10 0.5 0.00 5 0.47) the corresponding nominal L0(w) and Pzo(w) are. 4.69) w + 2/T 2w (4.1 142.05 30 1.00 38.23 20 0.73 50 3. Based upon Eqs.2 40 1.6 Fig.131 5.40) and (4. (4.026 1..73 66.) for each value of v.27 10.58 22.114 Chapter 4 termine the bounds B(jv.013 0.

(4. the design technique. (4. 4-5.S) (B0 shifted to the right as discussed in Sec.Discrete QFT 115 . A good rule in selecting the nominal plant Peo from <P. 4. must be on or above B^'O.) 2/T -w A'(w) = ————— 2/T + w (4. For example. for a stable uncertain plant is modified3 to incorporate the all-pass filter (a. in Fig. although not plotted. in &„.) with a zero at 2/T [see Eqs.73) as follows: Let the nominal loop transmission be defined as L0 (w) = .70)].. 3. This selection criterion for Peo (P^) minimizes the number of plant templates needed to determine the bounds.A'(w) w+2/T (4. 4-5.69) and (4.5).9998) (4.Lmo WA(w) = LmoA'(w) (4.2) Peo (w) = ———————————————————————————— w2(w+0.38.1)].75) and where \A'(jv)\ = l f o r O < v < a?and Lmo is an rap.72) is reflected on the NC by the bounds B0(jv. The constraint imposed on Lm L0(fv) by Eq.72) for all Pe in £Pe.4 NONMDVIMUM-PHASE L0(w) It is important to note that in the w domain any practical L(\v) is nonminimumphase (n.m. (4. This result is due to the fact that any practical L(z) has an excess of at least one pole over zeros which leads via Eq.65) imposes the requirement that &TR (jv) < Bu (jv)\ .f.p.BL (jv)\ (4. Since \L0Q'v) = ^mo{jv)\ and .11. Lm L0(j0.99 x 10-6 (w . (4.72) for all P. (4. [Note the zero in the "true"1 w-domain is at 1 [Eq. function considered to have a positive value of gain. for all values of v.p.2)(w + 4377)(w + 38. The NC is convenient because it contains loci of constant \L/(1 + L)\.4) in order to satisfy Eq. as Peo. (4.52) to the zero at 2/T. is to select the Pei in &e that has the smallest magnitude and the largest phase lag over all other sPe in fPe.) on Lm L0(jv^ as discussed in Chap.74) where w-2/T A(w) = ————— = . Thus.71) Thus Eq.2.

65) tend to dominate.93° at p = 4 (v = 8/T). system are readily revealed as follows. and it approaches 18CP asp(v) approaches infinity. (or co. (4.76) or (4. leading in m. 4.77) a bound B'(jv) on L0(jv) becomes the bound B(jv) on £„»(/'v) by shifting B'(jv) positively (to the right on the NC) by the angle . The inherent limitations on the feedback capabilities of an n.p.m. (4. systems to a single universal \\ \ \ Fig.). The amount of shift is 9(f at p = 1 (v = 2/7). 151.78) is shown in Fig. the templates tend to be vertical lines (due to the bilinear transformation characteristic) and Eqs.12.ZA'(jv) = 2tan -1| — | = 2tan -1 /» 0 (4. (4.p. 4. .78). At larger values of v.116 Chapter 4 (4.63) through (4.78) A nondimensionalized p = vT/2 plot of Eq.12 Angle characteristic of Eq.

amongst the J plants being considered.11. to be used in the QFT design must include all RHP zeros ofPzo. 4. then 2tan -1 pK = 180° . 4.73 (VK = 142. at a value of v <VK. the Bh contours are denoted as a function ofp in order for these contours to be independent of the sampling frequency. For the situation where the uncertain plant parameter characteristics can yield an unstable plant(s). if the nominal plant chosen is one of these unstable plants then the ap. system. although not done in this chapter.p. In general.. where y = 3(f. i. Thus. 4-5.m. ->• v II III Fig. 4. .13 Regions of the frequency spectrum for synthesizing Lmo(jv). However. Lnotjvg) < (f in order that there exist a practical L^ which satisfies the bounds B(jv) and provides the required stability. approximately point K in Fig.contour is shifted in the NC by 2tan~.79) results inpK = 3.contours for the frequency range of 0 <VJ<VK.or Bh(/v). This situation is not discussed in this text.5. the B'h.Discrete QFT 117 high-frequency contour (bound) B'h basically effective for all a> (or v) greater than some a>h (or vh). to become Bh(jp) a function of p whose right boundary shifts to the vertical line ZLmJjv) = (f at the value p = pK.11. These requirements are very useful for determining the minimum cos needed for any specific uncertainty problem as discussed in Sec. The Nyquist stability criterion dictates that the Lmo(jv) plot is on the "right side" or the "bottom right side" of the Brfjp). in the n. and 2.f. for Fig. LmoQv) must reach the right-hand bottom of Bk(JvK).30° = 150" (4. a necessary requirement for a stable system. Requirement (2) is necessary in order to obtain an Lmo(jv) whose angle is always negative for all frequencies. For example.v7/2. 4. It has been shown that:3'19'40 1.11B 'h is shown with a width of 2(3(f) for ML = 6 dB.55 for cos = 120) for this example.e. in Fig.

It should be noted that the discussion dealing with the discrete QFT design procedure is based upon the gain value of Peo(w ) being a negative quantity. Thus. If this occurs.(W) = ^^(4.5 SYNTHESIZING Lmo(w) Based upon the analysis made in the previous sections. in obtaining L^^w) by use of a CAD package. This results in the gain of G. The sign of the gain value of Peo(w) is a function of the sampling time. < v02S where Eq. For example. These regions are: Region I. < <x> into four general regions. (4. fromEq.) to satisfy both the B^. in order to formalize the design procedure. A good engineering rule-of-thumb. 4. the value of vh may be less than that of vfl 2. < VK satisfy only the BA-contours (stability bounds). bounds and the B 'h. Region IV. For the frequency range 0 < v. 13.13 If the gain of Peo(w) is a positive . Depending on the value of T. < vh use the w-domain templates 3Pe(jv. (4.5 Region III. upon which the final synthesized L^w) is constructed. with the appropriate gain adjustment. Region II.. i. depending on the value of T. as shown in Fig.(\v) being a positive quantity..-contours.80) assuming perfect cancellation.81) Note: when starting with a Type m P0(s). i. results in a minimal order controller. is to let Lmol & P0\s=w be the starting loop transmission function..e. G. L . For the frequency range vftZ5 < v. it is possible to divide the frequency spectrum 0 < v.and B/.).9998) (4. Note. assuming T is small enough.69). the poles of P0(s) = l/[s(s + ])] are used for the starting L^j. vh may be less than vft25. = ———————— """ w(w + 0. i.e. < v. For the frequency range V* > VK use the w-domain templates to satisfy only the ML-contours and obtain the corresponding jR^-contours.56) is essentially satisfied use the analog templates.118 Chapter 4 4-5. from Eq. If T is not small enough and can not be decreased use these templates to satisfy the stability requirement. For the frequency range vj. (4. Only the w-domain location of the s-domain poles of P0 should be used in setting L^^vv) = P0(s)\s=w.) &3Pe(jv. the value of these m poles must be corrected in the CAD program to reflect their true value being exactly at the origin. It should be noted that for a tracking control system Imo(w) must be a Type-1 or higher transfer function. based upon the definition of Region I. satisfy only the B'hand ^-contours.e.68). 3P(/<y. the CAD data may not result in m poles of Lmol(w) being exactly at the origin.

are tried.2.05. and the corresponding right-side phase values ofBh(Jv) to LJijv) for v < VK: see Fig.3. The resulting Lm L^fjv).).58) are violated If \Lmo(jv)\ is shifted upward so that it satisfies all the bounds. and an algorithm due to Bode.59 is especially efficient in calculating numerically IL^Qv)! for values obtained from Eq. v fast enough on the right side of the Bh(jv) contour in order to reach the "bottom" at v < vK7 The difficulty is that d[Lm Lmo(jv)]/d[Lm v] is a function of ZL^fjv) resulting in an average of 20a dB/decade decrease for ZLmo(/v) = -9(fa (a = 1. v. Then for v > VK.0065 <p< 0. then it is above the critical point K at VK = 142. before the value of v reaches the value of v0.Discrete QFT 119 quantity the same design procedure is followed. Hence. v.11.73). Is there a practical L^jjv) that can satisfy these bounds? In other words.1.g. 0. the faster IL^Qv)! can decrease vs.05.2. they yield a positive angle and a positive dB value for L^. 4. Hence.40 used in Gera and Horowitz. the only difference being that the gain of the resulting G/(w) is a negative quantity. Thus. as seen in Fig.6 (p = 3. (4. of -135. e. ZL^ > 0 in order to satisfy the Bh(jv) contours for v > VK.2° occurs at v = 4. and similarly the corresponding angles for v = 1..J must monotonically increase.. based upon loop-shaping experience.11.. However. in Fig. 4-5. assign approximately -ICf to ^Lmo(jvK).1 L^fjv) determines IL^Qv^ within an arbitrary multiplier.11. because of the n.15 (0.p..986.497 can be determined. the right-side phase angles of Bh(Jv) of -720° occurs at v = 10.m. 4.97. is shifted vertically on the NC in Fig. (4. where VK < v0.993 and 0. That is.11 for -30° at v = 66.. For v < 22.78)]. "reasonable" values of -ZA'fjv) [see Eq.6 cos= 120 IS TOO SMALL The bounds on L^v) are shown in Fig.497 <v<22. so d^L^ldv > 0 and |L. so obtained. -45° at v = 38. that is. It is impossible to do so in this example.11 such that Lm /^(/v^) is at point K and checked as to whether it satisfies the other Bh(jVj) bounds. 4. until at v0 there is no more phase lag for a decrease of |Z/mo(/v)|. the Bh(jv) contour steadily moves to the right vs. The following numerical procedure is convenient for determining the minimum o)s that is required Work backwards. 4. 4. . since the zeros of L^ dominate. is there an L^'v) that lies on or outside the Bmo(jv) bounds at small values of v but then decreases vs. and -90° at 22. the more phase lag available for L^. A(w). it is necessary that Lmodv) reach the bottom of this moving barrier (contour). since the bounds in the range of 0.77) for specified values of v. cos = 120 and in turn. As is well known. a>s = 200 are too small a sampling frequency for this uncertainty problem. which is impractical.

e. where the plot of Lm L^fy'v) satisfies all the bounds.070 10 0.64 40 0.14. are identical in shape and differ only in the frequency value associated for a given contour. The Bh(/v.14 A satisfactory design: Lmol at a>s = 240. 4.132 10.4142 31.00 76. > 20.p. i.11 50 1.11 and 4.. The mechanism whereby a larger value of a>s does this is seen in Figs. because the sampled system approaches the continuous system as <DS approaches infinity (T approaching zero).0655 5.5773 44. . exists. 4. as seen in Fig. it is guaranteed that a satisfactory co.47 30 0.2679 20.39 Fig.500 1 0. 4. the value ws = 240 is satisfactory for this example. for v.000 5 0.14. The challenge is to use the minimum ws value that is achievable.) high-frequency contours in these figures. CO p V 0.82) '——ML dB 0r BS(J1528) Note: Curves drawn approximately to scale Figure 4..14 A satisfactory design: L mol at o>s = 240.120 Chapter 4 If the original plant is m.06 20 0. the corresponding contours are related as follows: =120 = Bh 0'2vi (4.065 0.5 0.0131 1.

In order to use this rule it is assumed that the value of the analog gain-margin frequency coc is a practical or realistic value. One needs to estimate and select the dominant low-frequency bound Bmo(/V.497 < v.84) 90° Therefore the estimated decrease in dB (est.03 because the low-frequency (ca) bounds are basically invariant with respect to co. then a good nde-of-thumb in selecting a suitable sampling frequency is as follows: Select the value of a>s to be two to three times the value of coc that results in ZL0{j(o^ = -180°. Assuming an average phase change of Qm = 75.5 . In Fig. < 5. to a>K. 4.3 in Fig.14 the synthesized L^s) has to satisfy all the bounds through BnoQ'cOj+j) and yet achieve the necessary dB decrease (A dB)in Lm in the frequency range <y. and VK. in Fig. < 142.) and a starting point for Lm LrJdv. 4. Examples of this situation are illustrated in Figs. An analysis of these two figures reveals that the same decrease in log magnitude from about 20 dB to -30 dB is achieved in the range 0. dB) is given by est.14 instead of in the range 0.e. For a S-D control system satisfying both requirements may not be obtainable for a given value of ML and/or T. 4. 4. Applying Eq.85) If the est.Discrete QFT 121 The bounds /^(M). dB > A dB then the value of cas that has been chosen is satisfactory.83) to Fig. dB = 90° 6dB Oct (4. labeled Lm L^. for cas = 120 and 240.497 < v. 6dB Oct (4.. i. (4.11.) on this bound. In the event a preliminary analog QFT design is accomplished for the plant P(s).14. up to vx. 4.11 and4. For this example.83) where d represents the number of octaves between v. 4.° then the average negative slope is given by 0. an extra octave is available to achieve this decrease in log magnitude.14. Increasing the value of ML and/or decreasing the value of T widens the frequency bandwidth from v.11 results in 1426 0. are the same for v. the Bode algorithm40 provides a satisfactory Lm Lmo(jv). < 285 in Fig. In order to determine if all requirements can be satisfied for a specified frequency bandwidth let V K (4.

153 and an est.m.40 which can however.5 which indicates that the value of cos = 240 is satisfactory.4}. for a unit-step disturbance input d(t).^transform for a unit-step forcing function. The function numerically obtained in this manner. When the n.1 0. function L0(w) is formed the proper sign on its gain term to assure a stable system can also be ascertained from a polar plot analysis.14. . 4. 4. Equation (4.76 (> 43 dB = A dB) then from Fig. Thus.0.p.14: 285.76 < 43 dB = A dB.10} and a e {1. is y(oo) = 0. can be achieved as accurately as desired by a rational function /^(w). dB = 45.Lmol WA(w) = Lmol (w)[A'(wy\ which is sketched in Fig.76.87) is obtained by dividing the resulting YD(w) by the bilinear transformation of the . for a nominal Type 1 Peo plant.87) has a zero at the origin. the value of ws = 120 is too small.p. This ensures. the Nyquist stability criterion is applied to L0(w) in the polar plot domain.122 Chapter 4 which yields d = 8. the resultingpseudo control ratio (4.39) (4.056)(w + 76. That is. which can be made at this numerical stage by allowing some over design. there is some trade-off between cos and the degree of Lmo(w). If a>s is increased to 240 it results in d = 9. a Type 1 system characteristic for TR and that YD(W) will not have a pole at the origin.153 and an est. Thus. If the disturbance response specification. K e {1. be of high degree. The numerator and denominator of the synthesized L»»/(/v) w-plane transfer function must be of the same degree. Stability is assured when loop shaping for the m.987)(w + 45.39) w(w + 3.84)(w . For the uncertain plant of Example 2. then L^w) and in-turn. the synthesized loop transmission is _ .059559(w + 12.86) = . L0(w) loop transmission functions must have at least two poles at the origin. For a Type m Peo then L0(\v) must have m + 1 poles precisely at the origin. Hence.7. dB = 40. Lmo(w) is performed on the NC.

7 ERROR IN THE DESIGN There is always the problem of choosing sufficient CD. enough to cause instability. These bounds.h. as shown in Fig. 16.h.9998 = 20/[the pole -0.70). The bound is moderately violated at v = 15.280 and violated by approximately » 12 dB at v = 152.71).Discrete QFT 123 4-5. are obtained by shifting the ML contour to the right corresponding to the template frequencies. because the plant templates do not remain at almost vertical lines (giving the Bh type of bounds for all co > some coh\ until a frequency several times larger than the largest pole or zero of Pe.. because the poles of P(s) map directly into Pe(w). for the discrete system. 4.. To determine the required pole-zero pair(s) the very-highfrequency (v. 14 for all v > 20. For v > VK.71)]. 4. Since the templates 3PJjv. 1528000. .) broaden out again for v > 219. > VK.9998 is the smallest magnitude and 4377 is the largest magnitude of a pole or zero in Eq.f.280. up to its maximum shift of 18CP. 12.71)] to v = 4377/20 = 219 #VK [the zero -4377 of Eq. it is necessary to obtain the more stringent bounds Bs in Fig. The poles and zeros of P(s) or Pj(w) are important.15b. in the w plane. However. for v. approximately.) templates of Fig. 4. there may be large shifts in the zeros of P£w).9998 of Eq.800. (allowing a factor of 20 for the effect of a pole or zero on the phase). 4. The polezero pair(s) that are selected must result in an Lmo2(w) whose corresponding v. v = vh = 20/0. That is. the templates are translated to the right so that they are always tangent to their corresponding ML contour in order to obtain sufficient points to determine the corresponding Bs bounds. as seen in Fig..15b for v = 1528. 1 5b. values. not the shifted Bh contour. points he on or below the corresponding Bs bounds.f. 4. 6. especially at large values of CD or v. These high frequency ML bounds are overlooked for the first trial design. 15280.15a are utilized to determine the Bs bounds shown hi Fig. the maximum shift occurs when -ZA'ijv) = 180° as shown in Fig. Thus l + w/140566 which satisfies the B^jv) boundaries. (4. with a far-off zero at w = -4377 for the nominal plant Peo(vi') of Eq. 4. Note that 0. (4. as well as the largest pole in Pe(w). definitely violating the bounds for v > 15. (4. with the "universal" type Bk assumed in Fig. for the analog system. 4. The result is almost vertical templates and hence B*-type of bounds from. Note. Thus the final loop transmission function. values. is .15a. The first design stops at point W in Fig. 4. the templates are measured from the shifted ML contours. The L^i is augmented by a high frequency pole-zero pair to correct the situation. as was the case for this example. 15b. It is usually easy to detect the largest pole or zero in the continuous plant. or v. these templates are shifted around the corresponding ML contours to obtain the corresponding Bs contours. 4. As illustrated in Fig. (4.

124 Chapter 4 BS(J15.7476 x 10-6O .000) v = 15.279 v = 38.39)(w + 17. (b) reason for unstable Lmol design and successful L^ = Lmo2(W)A'(W) Once L0(w) is determined then the vc-plane controller transfer function is determined from Eq..280) 'B s (j 152.800) dB 80 v .e.91) .197 120 v = 76.9999) (4.76. for cos = 240. G.89).0.510) Pzo (w) = ————————— i—————— ^———— :——'w(w + 0.394 J___ 60° 120° Phase 45° 90° (a) Fig.15 (a) Neglected plant templates at higher v. (4.3820 v = 7639 XI 100 A 1 0 8 ° -90° -45° 0° Bs(j 1.(w)=L°2(w) (4. i.90) where.528. 4. .

94) z(z .0.25)(z + 0.16 Determination of the v. utilizing Eqs.9231X2 + 0.84)(>v + 140566) 1 W (w + 3.98276) The resulting loop shaping function is Q.0.056)(w + 76.987)(w + 45.10634(z .88) yields • ^ = 0.99891) (4. (4.999891) ——————————————————————————— (z .9999)(w + 12.f.98276) . Bs bounds.0.89). (4.h.923 l)(z + 0.10634(z . (4.39Xw + 8785 A)(w + 17510) Substituting Eq.IXz . (4.86).25Xz + 0.7094)(z . 90° Phase Fig.Discrete QFT 125 ML contour Lmo(jv) L mo(iv) cannot lie to the right of the 0° axis.0. 19(w + 0. 4.7094)(z . (4.l)(z .90) the controller transfer function is 4979.51) into Eq.88).0. and (4.0.

) = 0 dB then the permitted §R(JV. In order to prevent this from occurring. that enough nondominant w-domain zeros are inserted into the simple F(w) in order to create a prefilter transfer function whose numerator and denominator are of the same order. The reason for this is: in applying the bilinear transformation in order to transform this simple F(w) to the zdomain results in an F(z) whose numerator and denominator orders are the same. (4.8182 Utilizing Eqs. In other words. The loop-shaping function L0(w) = AfyJL^fy) only guarantees that the variation <^(/v.63) through (4.97) yields .5) F(z) = ———-————z . (4. extremes of —/ dB and 1 dB with 4{/v.0. (4. (4. „_ (4.639 .) does not exceed that permitted by Eqs. 4-5.99131) Peo = Pn (z) = ———————i——————L (z -1)2 (z-0.) < -1 dB The 1 dB over design of \T'rfjv.95) whose zero at -229.) = -6 dB and BuQv.2 dB < F(jv.47) and (4.97416) . For this example. if at some vh BL(jv. it is necessary after the simpler F(w) is synthesized..) = 5 dB. An acceptable L0(jv) design may result in \T'R(jv^\. This was done in obtaining Eq.639) satisfies the wdomain but not the z-domain performance specifications.126 Chapter 4 where A(z) = 1/z. the prefilter in the z-domain is 0.) = 6 dB. be 10 times the largest magnitude of a pole or zero of Pe(w) e 9. let the highest frequency of concern for v.2 is nondominant.65).639/(\v + 7. By use of Eq.)\..94). For example..2) F(w) = —————-——————w + 7. 3.)\ allows I dB tolerances for \F(JVj)\.4 z(z+ 0. according to the QFT design procedure.397 xlO.8 DESIGN OF THE PREFILTER F(w) The design of F(w) is done exactly in the same manner as for continuous systems (see Sec.96) (4. It is then necessary that .15). As a general rule-of-thumb. _. due to the transformation the zero of F(z) will alter the desired QFT prefilter frequency domain characteristics.033329(w + 229.95) is obtained. where 3.51. where TR = FT'R. the satisfactory prefilter 0. Note that the prefilter F(w) = 7.1212(z + 0. (4. F(w) is chosen to satisfy the resulting tolerances on \F(fv.

100) (4.99131) Since the zero at -0.25)(z + 0.(z) at -0. the essence of the QFT technique is the synthesis or loop shaping of the optimal loop transmission functioa In applying the bilinear transformation to G.98276 and -0.3. as discussed in Ref. A computer synthesis and simulation flow chart is presented in App. The controller Bode plots of G/(w) and G...).0. and Bs bounds in the zdomain. The same situation can occur when the controller of Eq. One must also ascertain that the Dahlin effect.(z). then Eq.99131..(w) lying outside the good Tustin (bilinear transformation) region of Fig.J cases.2.0.Jplants necessary to obtain the tracking control ratio TRi(w). since they are close to the unit (U.(z) = 1 + L.e.25)(z .0.Discrete QFT 127 G. C. for this example. to obtain G. . are: P 8k (*) = P l (5)D(a) ) (4.49).98276)(z + 0.0.2.(z).(z) = —————————————————————————————————————— 313.99891 and the pole at -0.39) through (4. This "warping" is due to some pole and zeros of G.9231)(z + 0. Q. can result in significant warping of some of the poles and zeros of G. (z) = ————-——————-————-——————z(z.. 4-6 SIMULATION As stressed in the previous sections. The pertinent w-domain transfer functions for the i = 1. is not a problem.0.99131 are very close to one another.97416) G. Whether warping is significant or not can be determined by obtaining the Bode plots of G/(w) and G. 4. (4. the violation of the B0.9 lie essentially on top of one another. within the BW determined by where the -24 dB value occurs on BL in Fig.(z).05(z .C.97416) (4.99) The poles of G.. If the Bode plots.0.7094)(z .(w).0.. (4.99891)(z .. There is excellent correlation between these plots up to approximately « 40 rad/sec.(z)PZi(z) is formed for all i = 1. i. then warping is insignificant with respect to satisfying the loop-shaping requirements in the z domain. 4.923 l)(z + 0.98) is utilized The characteristic equation.101) . Bh. This warping may result in a significant degradation of the desired loop shaping characteristics. (4. on the negative real-axis. within the BW of 0 < (o < 10 lie essentially on top of one another indicating insignificant warping. and (4. .7094)(z .05(z .98276) (4. This is illustrated in the next section. can present a stability problem when L.. based upon Eqs.87).98) z(z .0. 13.98) can be approximated by 313.(z) should be obtained in order to ascertain that a stable system exists for all J cases. (z) = G.

Eq. (4.105) F(w)L z (w) (4.128 Chapter 4 2wP' (w) ———*—— w + 2/T (4. based upon these equations the pseudo disturbance control ratio is 1 P. zl(z -1).110) Since. (4.87).104) into Eq.109). from Eqs.109) Note that Eq.103) 2/r . (4.102) and (4. PtD(w) 2w 1 w + 2/T = PZl (w) (4.102) (4.105) yields TR (w) = (4.111) . is obtained based upon the unit-step disturbance function d{f) = u-j(t) and the following additional vcdomain transfer functions: (4.D(W) (4.108) w + 2/T 2w (4. (4. (w) = (w) = (4.107) The pseudo disturbance control ratio 7i> (w).109) is the bilinear transformation of the ^-transfer function for a unit-step function.104) (4. Thus. (4.106) Substituting Eq.

9657. (4. for all five cases.923 l)(z+ 0. A plot of the poles and zeros of L.(z) controller polepg at -0.C.92) to obtain the z-domain controller transfer function of Eq.88 to yield the modified controller 313.W _l + ^ i (z)G 1 (z)_ _ l + J P_(z)G 1 (z)_ where z-1 The w-domain tracking step responses for five extreme cases are shown in Fig.3 satisfy the specifications as represented by the boundaries in Fig.. see Fig. 4. Based upon the root-locus analysis of Lfa) = —I.D(z) _l + J P 2 (z)G.Discrete QFT 129 then Eq.(z).(z) = P2i(z)G. (4.(z).(z) are inside the U.25)(z .47). see Fig. and D(z) = z/(z -1). 4.7094)(z . (4.0.46). this controller pole is moved by an incremental amount toward the origin until all factors of Q. based upon Eqs. this controller pole is replaced by a pole at -0. it is determined for case 5 that one of the closed-loop poles is at -1.40). is a result of the G.98). 4. the bilinear transformation of Eq.19c. This unstable pole. 4.0.2 and 4.113) The z-domain pseudo-disturbance control ratio.115) z(z -0.008. (4. (4.51) is applied to Eq.0.14.114) D(z) _ p z [(z-l)/z]P e (z) P. for all five cases are shown in Table 4. (4.4. can be expressed as ~YDl(z)~ Tn z-1 P.88) .19. (4. For this example.9. the z-domain tracking control ratio can be expressed as F(z)Lt(z) F(z)Gl(z)PZ{(z) (4. By trial and error.05(z .98276 and the plant zero zz/z) at -0. The Bode plots for these cases all lie within the area bounded by the Bn and BL bounds of Fig. The poles and zeros of L.18.47).(s) are shown in Fig. 4. The w-domain figures of merit (FOM) shown in Tables 4.97416) (4. and (4.109) can be expressed as follows: (4.44) and (4.112) From Eqs.17. and the resulting characteristic equations Q. Since the specifications are met. 4.(z) (4.

Controller poles • .18 A root-locus pole-zero location resulting in an unstable system pole. 4.00 0.Controller zeros Unit circle z-Plane Region of plant zero variation Region of plant pole variation -1+jO Relocation of a controller pole Fig.Plant poles O.75 4.00 Case a k 1 1 1 2 4 1 3 2 3 4 2 10 5 4 10 Fig.00 0.130 Chapter 4 0.37 5. s 3.62 1.17 Step responses for: X .12 Time.Plant zeros A . .25 1.87 2. 4.50 3.

3403 0.0820 0.016 1.2 and 4.4711 0.14 tps 0. the w-domain simulation results are essentially duplicated by the z-domain simulation utilizing the modified controller Gi(z)m.44506 — — — — — .0820 Case (Plant) 3 4 0.340 0.000 1.086196 0.086189 0. only final values are shown.08618 0.2 Tracking FOM Case (plants) 1 2 Domain w z w z Controller G.08665 0.000 3 w z Gi(z)m 4 w z w z G.38 0.6021 TABLE 4. Various CAD packages (App.3 Disturbance FOM# Domain w z Controller GKw) G.000 1.013 1.353 0.000 tss 1.445 0.08648 5 0.000 1. TABLE 4.000 Unstable 1.0820 2 0.000 1.017 1. .000 1.0820 All responses are overdamped.6283 — — — 0.451 0.08619 0.(w) Gi(z) G!(z)m Gj(w) Gi(z) Mp 1.893 2.Discrete QFT 131 As Tables 4.(w) Gi(z) Gi(z)m G.6280 0.3 indicate. 1.471 0.0820 unstable 0.(z) G!(z)m 1 0.335 2.(w) Gi(z) G!(z)m 5 1.664 0.14 1.97 1.08648 0.478 0.609 — 0.906 2.139 L_ 1.8648 0. .801 0.08648 0..(w) GKz) G!(z)m G. A through C) include both analog and the discrete MISO system QFT routines which expedite the design process.

.132 Chapter 4 Im z Unit Circle z-Plane -Pol6ofG.9827 0.9657 0.9657 3 -0.7094 ^I -0. 4.9490 0 1 0. TABLE 4. location yielding: (b) a stable system and (c) an unstable system. * Root o -1+jO ->• Rez Fig.97416 | 0.(z) 'i Rez 1+jO (a) Q Unit circle — Root .9 9131 5 -0.98276 0.9006 ..(z) Case i Item Zeros 1 0.2500 Poles 0.9231 -0.locus branch -1+jO *• Rez Root of (^(z) (b) Unit circle Root .19 Analysis of root-locus plot of L(z) = Gj(z)Pei(z) = -I: (a) pole-zero locations determining system stability.4 The poles and zeros (all real) of L.97416 0.(z)-P.9'i'416 0.99131 2 0.locus branch Im z D i<\.9006 0.

for the value of T that is chosen and the model that is synthesized Eq. (4. (4.5). it is necessary to modify TRL(S) to have an excess of four poles over zeros in order to maintain the requirement that as o)i increases in value that <%(/<».Discrete QFT 133 4-7 BASIC DESIGN PROCEDURE FOR A MISO SAMPLED-DATA CONTROL SYSTEM The design procedure of Chap. Synthesize the-TRu and the !RL models in the s domain. its ^-transform TRL(Z) can readily be obtained. Since no ZOH device is involved then the ^transform of TRjJ(s) can not be taken. (4. or (2) synthesize TRj(s) that has an excess of poles over zeros of two in order to be able to obtain the . as discussed in the previous section.•) also continually increases. the smallest value of sampling time allowable should be selected The following step numbers. (b) If Eq.117). In order to minimize the warping effect. . (4. then the w-domain model is obtained as follows: T u (w)=T u (S)\s=w (4. 2. If Eq.116) (4.J transform of TRlj(s).116) and (4.16) is not applicable one can (1) select a lower value of T. The reader must ascertain that of « 2 and co^T/2 « 0. if this is permissible. Since the -^transform of the modified TRj(s) and TR[(s) can be obtained TRlj(w) and TRL(W) can be obtained by use of Eq. (a) The TRj(s) model usually contains one zero and two poles thus it has an excess of poles over zeros of one.14) will be satisfied. Since the TRL(S) model. 3. 1. has at least an excess of two poles over zeros.14) is applicable. correspond to those of Chap.117) TR^(W) = TRL(S)\S=W In general. in general. where necessary. (4. to be applicable to S-D MISO systems. In doing the latter. by the QFT-imposed requirement. The TD(W) model can be obtained from TD(S) as follows: (a) For the s-domain model that has only an excess of poles over zeros of one then the procedure outlined in Step l(a) can be repeated to obtain TD(w).297 in order to utilize Eqs. 3 is repeated in this section but amplified and modified.

< VK. < v0js). and if the low frequency range (0 < v. all performance specifications (Bmo.6 for some additional guidelines.15a] broaden out again due to the "far out" pole or zero of Pe(w) and then again approach a straight line. . (2) Whether or not Eq. the data for the templates. the templates in the w domain have the same characteristic as that in the s domain. The number of templates that are necessary for the QFT design can be determined as follows: (a) For the frequency range 0 < v.134 Chapter 4 (b) If Eq. (b) For the frequency range vh < v. 4-5) as follows: Region I: Where Eq. (4. they approach a straight vertical line of KdB height. (4. 4.. only the specification associated with the Bs (ML) stability contours need to be satisfied.56) is satisfied for all J plants the analog QFT templates 3Pe(j(o) can be used to determine the appropriate bounds. (c) For a discrete QFT design. The frequency intervals for a QFT design may be divided (see Sec.56) is applicable then 3Pe(ja>i) *3Pe(jvj.56) is not satisfied then TD(s) can be modified in the same manner as done for TRl£s) in Step l(b). the templates [see Fig.e. can be determined in the w domain in the same manner as is done for the analog design. Region II: For the frequency range of 0 to vh assuming the value of T is small enough. Eq. (4.-. Sufficient Bs contours need to be obtained in order to ensure a stable system.. As a general rule. 4-4. i. only the specifications associated with the Bk contours need to be satisfied Region IV: For the frequency range of VK < v. <VK. 3. This can be modified depending on the evolving shaping characteristic of L^. 4.56) is applicable. See Sec. templates below VK should be approximately an octave apart. < vh: (1) If an analog QFT design has been accomplished for a given system. (4. Above VK. B'h and BL) can be satisfied Region III: For the frequency range of vh < v. it is best to select TD(s) = a constant value. (c) For the frequency range VK < v. Bh. the templates should be between two octaves to a decade apart.

e. (b) For frequencies greater than vp. values of the templates. characteristic forvp< v. 1 1 . <VK.)-contours corresponding to the v. and Bmo bounds in the same manner of the analog QFT design.118) to satisfy the high-frequency (v. (a) First trial design: Synthesize L^.Discrete QFT 135 5. (b) Second trial design: synthesize (4.p.)-contours are determined by the appropriate angular shift to the right due to the a. BD. For v. Synthesize Lmo(w) in the same manner as done for the analog QFT design.15a]. values. 6.f.p. 10. The optimal bounds B^ can be determined as follows: (a) For 0 <Vj< 20/(smallest magnitude of a pole or zero.119) or -L(w)A(\v) ——mo^ ' ^ ) (4. the stabihty bounds Bs are determined by using the corresponding templates and applying the appropriate shift to the right of the ML contour due to the a.. i. (4. Once the B'h -contour has been constructed. > VK) Bs-contour bounds on their left or bottom sides [see Fig. The template is translated right or left while it is kept tangent to the corresponding ML contour. other than for the origin of Peo) = vp determine the BR.. 7. through item 9. < VK.69). the J?A(/v. characteristic. Once a satisfactory L^w) has been synthesized then the controller transfer function is obtained from Eq. > VK.120) . construct the BA(/v. Gj (w) = (w + 2 / 7')—^—2wPgo(w) (4. Select the plant from the set of plants Pti that has the smallest dB value and the largest (most negative) phase lag characteristics as the nominal plant Peo.f. 4. based upon the ML and v. the nominal point on the template yields points for the corresponding Bs bound. For the selected tangent points. = P0(s)\s=w for the frequency range 0 < V.

in order to apply the QFT technique. are essentially the same as the Bode plots of Lm G(z. plant to satisfy these requirements. 3 can be applied to the m. 4-6. The resulting modification of the m. it will be necessary to modify G(z) until the z-domain Bode plots of G(z) are essentially the same as those for G(w).6) to obtain G(z). (4.p. Once a satisfactory G(z) has been achieved. If these plots are not "reasonably" close to one another. This PCT design approach is also referred to as a DIG design method. a discrete-time domain simulation is performed in order to validate that the desired S-D control system performance specifications have been achieved. in order to take into account the use of this filter.p. 4-2. Simulation is first accomplished in the w domain in order to ensure that all system performance specifications have been satisfied. When the requirements in Sec. it is necessary to validate that the Bode plots Lm G(jv. analog QFT design procedure of Chap. If the wdomain design is not satisfactory then do another w-domain design to try to achieve the desired performance. the factors of the characteristic equation Q. Synthesize F(w) in the same manner as for the analog QFT design employing the templates and the plot of L0(jv). Remember F(w) must be equal order over equal order. 3. 4-S QFT TECHNIQUE APPLIED TO THE PSUEDO-CONTINTJOUSTIME (PCT) SYSTEM As noted in the preceeding sections.) vs. v. then the simpler analog QFT design procedure of Chap. .136 Chapter 4 where A(\v) =-A'(w). in the frequency range of 0 < cat < cos /2 where z = eJ(aiT .) and ZG(jv.p.) and </G(z.. results in a more involved wdomain QFT design procedure.f.3 for a psuedocontimtous-time (PCT) representation of a S-D system are satisfied and if P(s) is m. transform G(w) into the z-domain by using the bilinear transformation of Eq.m. Using the minimum practical sampling time allowable enhances the possibility for a given m. then warping has been sufficient enough to degrade the desired z-domain loop shaping characteristics.p. In order to validate that the desired loop shaping that has been achieved in the w domain has been maintained in the z domain. Before proceeding with the z-domain simulation. the n. z.{z) should be determined to ascertain that stable responses are achievable for all of the J cases as discussed in Sec. characteristic of a w-domain plant transfer function requires the use of an a. Once a satisfactory w-domain design has been achieved. 13. PCT system.p.) vs. 12. If this occurs.p.

21c. Therefore. whereas the second- order approximation is good for <DC < cosl3 (Ref. is approximated as follows: c -«~ "^i=c"w <"21) Thus. 4.122) Since Eq. 4. where all its sidebands are multiplied by 1/T [see Eq. note that in the frequency domain as cas -> oo (T -> 0).2la and (b). only the primary component needs to be considered in the analysis of the system. 4. Therefore. Using the first-order Fade' approximation (see Ref. 12). 4. then the primary strip becomes the entire frequency-spectrum domain.20.9) of Ref. which is the representation for the continuous-time system. To illustrate the effect of the value of T on the validity of the results obtained by the DIG method. first consider the frequency component E*(jco) representing the continuous-time signal E(Jco).Discrete QFT 137 4-8. This multiplier of 1/T attenuates the fundamental frequency of the sampled signal and all its harmonics are attenuated. Because of the low-pass filtering characteristics of a S-D system. requires a satisfactory pseudo-continuous-time (PCT) model of the S-D system. 60). the Pade' approximation Gpa(s) is used to replace Gzo(s) as shown in Fig. the sampler and ZOH units of a S-D system are approximated in the PCT system of Fig. To determine GA(s). (6. for the S-D system of Fig. as shown in Fig. the sampler and the ZOH units must be approximated by a linear continuous-time unit GA(s).13 Note that in obtaining PCT systems for a S-D system the factor 1/T replaces only the sampler that is sampling the continuous-time signal.20 where .122) satisfies the condition \\mT^fjA(s) = 1 it is an accurate PCT representation of the sampler and ZOH units. the PCT approximation of the sampler of Fig. 13].21c by the transfer function GA(s) = -Gpa(s) = -^— T Ts + 2 (4. In other words. because it satisfies the requirement that as T -» 0 the output of GA(s) must equal its input. 4. 4.20 is shown in Fig. for a high level of correlation with the S-D system.21b.3. in the complex-frequency s plane. (4. This approximation is good for coc < a>s/10.1 INTRODUCTION TO PSEUDO-CONTINUOUS-TIME SYSTEM DIG TECHNIQUE13 The DIG method of designing a S-D system. 4. The DIG method requires that the dominant poles and zeros of the PCT model should lie in the shaded area of Fig. when the value of T is small enough. Further. the transfer function of the ZOH device. 4. consider the S-D closed-loop control system of Fig.

-^ E(s) ^ J————*" ^ l GA(s) M(s)^ Gx(s) i C(s) ^ Gpc(s) (c) Fig."^'c*(s) C(z) C(s)^ R(s) ^X^ w ' T E( zf V 7f\fJ M(s)^ Gx(s) j G(s) Fig. (b) approximations of the sampler and ZOH. 4.20. (c) the approximate continuous-time control system equivalent of Fig. Sarnpler approximatia ZOH approximatioo !! M(s) E(s) -*• 1 T -T>- :: |! G (s) -> ^paW (b) M(s) _iL .138 Chapter 4 Gz(z) 1 Plant t _/.20 The uncompensated sampled-data control system. .21 (a) Sampler and ZOH. 4. 4.

This model is then used as a guide for selecting an acceptable C(z)IR(z). Thus.2-4. (The designer needs to specify.s 0.124) is used to illustrate the approaches for improving the performance of a basic system.7071.200 TP>* 4. utilizing the plant of the S-D control system.18 Tas 9.01 0.and z-domains.y)]T for the PCT unity-feedback control system of Fig. For T < 1 s there is still a relatively good correlation.1 1 Domain z s K. 4.215 3.45 Method DIR DIG DIR DIG DIR DIG T. One approach for designing a S-D unity-feedback control system is first to obtain a suitable closed-loop model [C(s)/R(.3 4.53 9.8+ 9.11 4.") TABLE 4.203 1.5 presents the required value of K* and timeresponse characteristics for each value of T. i. respectively.206 1.496 MP 1.22a yields. what is considered to be "good correlation.4393 2.906 2. (4. for £ = 0.478 9.4767.202 1.e. Note that for T < 0.90+ 13-14 13.16 4.892 3.2 s(s + l)(s + 5) (4. Table 4.Discrete QFT 139 4..336 6.21. The root-locus plot G^s) = -1 shown in Fig.2 SIMPLE PCT EXAMPLE Figure 4. Kx = 0. 4. the DIG technique and the direct (DIR) technique (the zanalysis). 4.123) The closed-loop system performance for three values of 7* are determined in both the 5.202 1.199 1.1 s there is a high level of correlation between the DIG and DIR models.76 z s z s 4-8.5 Analysis of a PCT system representing a sampled-data control system for £ = 0.147 4.20 represents a basic or uncompensated S-D control system where Casel s(s +1) (4.124) . for the plant of Eq. for a given application.

which transformed a completely stable system into a conditionally stable system.22b.5 (a) Fig. 4. (4. (4.e.jO.5 -0. the lag characteristic of Gzo(s) reduces the degree of system stability..7071 \ 'If s-plane s-plane [I . the ZOH unit degrades the degree of system stability.22a These figures illustrate the effect of inserting a lag network in cascade in a feedback control system.125). For the model it is assumed that the desired value of the damping ratio Cfor the dominant roots is ft 7071.. for a given value of £ the values of tp and t..22 Root locus for (a) Case L Eq.126) Gpc (s) 2b + 205 The root locus for GPc(s) = -1 is shown in Fig.140 Chapter 4 5-0. i. For comparison purposes the root-locus plot for Gj(s) ^ -7 is shown in Fig.22.125) and for T = 0. (b) Case 2. for a unit-step input. Eq. 4. 20/T R(s) (4.1 s. GPC (s) = GA (s)Gx(s) = 2KX/T s(s+l)(s+2/T) (4.124). (and Ts) are increased. 4. . as stated previously. Thus.-jO.5 KX . 4. as illustrated by Fig. Therefore. Thus.

KX(T + \)e~T (4.127).126) to obtain the approximate expression for C(z).TKX .534 5(53 + 215 +20^ + 9. 4.4883)(5 + 20.9672) (z . 4. (4.3 THE SAMPLED-DATA CONTROL SYSTEM EXAMPLE The determination of the time-domain performance of the S-D control system of Fig. The real and imaginary parts of the desired roots of Eq. Note that the pole at -20.20 may be achieved by either obtaining the exact expression for C(z) or applying the Tustin transformation to Eq.Discrete QFT 141 Case 2 [C(5)]T = 9. 4-8.4875 + .03 is due to the Fade' approximation for Gzo(s) [see Eq.4767. lie hi the acceptable region of Fig. (4.03) where Kx = 0.1 s and Kx = 0. for T = 0.534) (4. for T = 0. Case 3 0.4767.9048) (4. For the plant transfer function of Eq.121)]. 4. Proceeding with the exact approach first requires the ^transfer function of the forward loop of Fig. (4.1 s.19.3 for a good Tustin approximation.534_________ 5(5 + 0.128) Thus.124).l)(z .Kxe~T)z + e~T + Kx .0. (4.129) or . :(!-• z2 .0.(1 + e-T +KX.127) 2 _________9.002306(z + 0.

132) [C(z)]TU =[F(z)]TU[R(z)]TU As stated in Sec. results in a Tustin model of the control ratio as follows: . (4.131) _R(s)_T yields 1+ \C(z)}TU TU = [F(z)\TU [R(z)]TU This equation is rearranged to (4.134) COO = Z\c * (t)\ = — [C(z)]TU T 4.1 R(z) = S\r * (t)] . based upon Eq.142 Chapter 4 C(z) R(z) 0.002306(z + 0.0.136).133) and rearranging yields 1 1 C(z) = —[F(z)]w [R(z)}w = —[Tustin of FT(s)R(s)} J T (4.4767.130) The DIG technique requires that the s-domain model control ratio be transformed into a z-domain model. Applying the Tustin transformation to C(s) Gpc(s) = FT(s) (4. (4./0.126).9672) (z. (4.137) Thus.134) and (4.— [R(z)]TU T 1 (4. 4-8.135) into Eq. (4. (4. (4.133) (4.136) Substituting from Eq.132) yields ——— = [F(z)]nf C(z) (4.135) into Eq.9513 + .135) Substituting from Eqs. the Tustin transformation of Eq. with Kx = 0.04649) (4.

In converting a continuous-time system into a sampled-data system the timeresponse characteristics are degraded. Thus.0. Table 4. Therefore. 2.138) are essentially the same as those of Eq. (4. agrees favorably with those of the continuous-time model. 4.7 summarizes the time-response FOM for a unit-step forcing function of (1) the continuous-time system of Fig.04651)(z + 6.. (4. there is some variation in the values obtained when utilizing the exact C(z) and [C(z)]m.(s) and (2) the S-D system of Fig.7 s.21 for the two cases of G(s) = Gx(s) [with GA(s) removed] and G(s) = GA(s)G. (4. .138)].3.6 and 8.138) 5.5 s and the value of 4 occurs between 8. The time-response characteristics of the S-D system. [C(Z)]TZ/ results in a value of c*(/) * 0 at t = 0.20 based upon the exact and Tustin expressions for C(z). The table reveals that: 1. Eq. 4.(z +I) 4 3 (z . 4. using the values of gain obtained from the continuous-time model.6 illustrates the effect of this characteristic of the Tustin transformation on the time response due to a unit-step forcing function.130) due to the value of T used which resulted in the dominant roots lying in the good Tustin region of Fig.3. the resulting values of the FOM are in close agreement to those obtained by exact £ transformation. When using the Tustin transformation. the Tustin transformation is a valid design tool when the dominant zeros and poles of [C(s)/R(s)]Tu lie in the acceptable Tustin region of Fig. the order of the numerator polynomial of the resulting [C(z)//?(z)]ro is in general equal to the order of its corresponding denominator polynomial [see Eq.9513 ± y0.e.252 x 10~4) Note that the dominant poles of Eq.672 x lO. Note that the value of Mp occurs between 6. As may be expected. In using the exact 2 transformation the order of the numerator polynomial of C(z)/R(z). 4. i.4 and 6. The nondominant pole is due to the Fade' approximation of G20(s).Discrete QFT 143 Case 4 C(z) _ C(z) tf(z) _#(z)_ T O (4. which is in error based upon zero initial conditions. Table 4. (4.130) is one less than the order of its denominator polynomial. The degradation of the time response by use of the Tustin transformation is minimal.

1701 0.9844 1.6 Comparison of lime responses between C(z) and [C(z)]m for a unit-step input and T = 0.2894 0.029 1.022 1.3340E-01 0.7064E-01 0.5364 0.4153 0.043 1.6485 0.7461 0.3403E-01 0.022 1.7453 0.019 Case 4 (Tustin).1701 0.5672E-03 0.043 1.7024E-01 0.011 1.9823E-02 0.043 1.043 1. 0.4148 0.8944 0.043 1.8924E-02 0. 0.042 1.042 1.2284 0.1 s c(kT) k 0 Case 3 (exact).043 1.041 1.039 1.144 Chapter 4 TABLE 4.011 1.019 2 4 6 8 10 12 14 18 22 26 30 34 38 42 46 50 54 58 60 61 62 63 64 65 66 67 68 85 86 87 .043 1.043 1.1166 0.9460 0.043 1.021 1.9452 0.8281 0.043 1.8273 0.2897 0.038 1.6478 0.043 1.021 1.028 1.042 1.9836 1.042 1.2283 0.5370 0.042 1.1168 0.8936 0.043 1. 0.

4-9 APPLICABILITY OF DESIGN TECHNIQUE TO OTHER PLANTS As stated in Chap.4 THE PCT SYSTEM OF FIG.04821 1.69 8.8 is shown in Fig. and Dc(z) \ . 5).45 6. Each MISO system has two inputs. 3 and 4 can be applied to the QFT .45 8.4767 Case 1 2 3 4 Continuous-time system G(s) = Gx(s) GM(S) = GA(s)Gx(s) C(z) [C(z)]ru 6. 4.40 to 8. 4-8. whose order of the numerator and denominator are.75 K. the MISO loop QFT design techniques of Chaps.043 1. 4.5 PCT DESIGN SUMMARY Once a satisfactory controller Dc(s) (see Fig..75 8.48 6. Note that the sampler sampling the forcing function and the sampler in the system's output y(f) are replaced by a factor ofl/T.23b).13 A final check should be made before simulating the discrete design. ws and ns.7 Time-response characteristics of the uncompensated system i M. 4.45 Sampled-data system 0. The solution of these MISO systems are guaranteed to satisfy the original MEVIO problem.ffln should be essentially the same within the desired BW in order to ascertain that the discrete-time system response characteristics will be essentially the same as those obtained for the analog PCT system response.3 6. This diagram is simplified to the one shown in Fig. 1. 4. s1 0. and one output. then (a) if ns is greater than ws + 2 use the exact 2-transform to obtain the discrete controller Dc(z). or (b) if \vs < ns < \vs + 2 use the Tustin transformation to obtain tain [Dc(z)]ru. has been achieved.043 tp. 4.4765 4-8. If the plots differ appreciably it implies that warping has occurred and the desired discrete-time response characteristics may not be achieved (depending on the degree of the warping).e. i.23b and is the structure that is used for the QFT design. one desired (tracking) and one unwanted (cross-coupling effects) inputs. 1. the QFT design techniques can be applied to an mxm MTMO control system by transforming this system into an nf equivalent MISO systems (see Chap.. S 4s 8. the Bode plots of A{s)|j = jo.04342 1.8 The PCT system representing the digital control system of Fig. respectively.Discrete QFT 145 TABLE 4. Thus.23a.

p.1 RHP zero atw = 2/T. The reason is that the range of the s plane over which the mapping is almost directly so. it is necessary to verify all the J characteristic equations hi order to determine if the sign of the controller gain is correct for achieving a stable system. L(w) by the method of Chap. the minimum required sampling frequency cos is very strongly a function of the uncertainty problem. do not map directly into the zeros ofPe(w). This approach is simpler than using the approach of Sec. Since the QFT design is done on the NC.p. 4-11 SUMMARY MINIMUM-PHASE. design of the m equivalent MISO systems as discussed in the remaining chapters of the text.8. to the extent of the plant uncertainty and of the system's performance tolerances.23 The PCT equivalents of Fig.. 4-10 DESIGNING L(w) DIRECTLY 2 It is possible to perform the loop shaping of the n.m. AND UNSTABLE P® As pointed out earlier. that is. unlike the poles. then the sampled loop transmission function L(w) need have only one 4-11. 4-5. D(s) | m(t) +Y w(t) pa ""' X(s) M(s) + W(s) (s) Plant y(t) i——*• Y(s) *> 1 T (a) (b) Fig. If the continuous plant P(s) is m. 4. This can be asserted even through the zeros of'P(s). A. NONMINIMUM PHASE. 3 with the aid of the MIMO QFT CAD package of App.146 Chapter 4 Controller Kt) R(s) 1 T * F(s) V(s) +-SrE(s) B(s) v(t) e(t) x(t) d(t). 4. can be made as large .

the desired bandwidth may be achievable in the former by making cos large enough. because the bandwidth of importance recedes into lower v values (compare the low v bounds of Figs.14 4-11. 4-11. 13. is transformed into the zdomain to yield Dj(z).m. Ref. then a QFT discrete system design is achievable by use of a large enough value for (os.Discrete QFT 147 as desired by increasing co. 9. the RHP poles are arbitrarily close to the origin in the w-plane. plant.p.p.12.p. 4. The reason is that with large enough o)s. ws will have to be very large. even though the sampled L(w) has RHP poles and the RHP zero at 2/T. it is certainly not solvable as a discrete system design. Any practical sampled-data system is n. In designing the digital controller to achieve the desired control system performance requirements. The QFT design procedure of Chap. see Fig. in the w domain. The price paid is in the loop bandwidth for both. The main difference in quantitative design is the need for some experimentation to find (e^)/™. 4.p. how is maximum computational accuracy achieved? This particular aspect is discussed in Chap. 1. and the resulting GI(S) = D^s).m. but the harmful effect of the RHP zero at 2/T can be made arbitrarily small. then it may be impossible to achieve the desired performance for all P in ff. e.3 CONCLUSIONS The basic feedback concepts and quantitative design procedures are the same in the w domain for the sampled-data systems as in the s domain for the continuous systems. For example. by making cas large enough. as determined by the plant characteristics and the minimum practical sampling that can be used. If so.g. A(w) = (—w + 2/T)/(w + 2/7}.? The larger ws the smaller the importance of the surplus phase lag of the a.p.e.f. Unstable but m.11 and 4. i. The reader is referred to the technical literature that discuss the other factors involved in controller implementation..m.29 but it has been shown how time-varying compensation may be used for such plants to achieve stability (but not small system sensitivity) over large plant parameter uncertainity. the engineer must be aware of the factors that play an important role in the implementation of the controller. P(s) do pose a great problem. However.p. precisely as for the continuous systems. the QFT technique can be applied to the PCT system representation of the sampled-data control system. For an m. by means of the Tustin transformation.p.3 One should then first check to verify that the problem is solvable as a QFT continuous system design.2 DIGITAL CONTROLLER IMPLEMENTATION This chapter can not be concluded without mentioning the importance of the implementation issue as stressed in Fig. if P(s) is n.12. plants pose no problem. 3 can be applied to the PCT system.p.f.14 Unstable n... if the design is executed in the v domain (w = jv) for the former. When applicable.1. Some of the characteristics involved in the use of a bilinear . eliminating the need for the use of an a. If is is not solvable for a continuous system design. the resulting PCT system is also m. If it is barely solvable in the former.

The bilinear transformation characteristic: transforms an unequal order transform into an equal order transform by inserting zero(s) in the z-domain at -1 + JO To preserve dynamics of the PCT or w-domain QFT controller insert v-u nondominant zeros before applying the bilinear transformation to obtain D(z) Fig. As pointed out in Sec.m. 3 directly to obtain the n. . A it is possible to apply the method of Chap. 4. with the aid of the MIMO QFT CAD package of App. are stressed in Fig.24. 4-10. 4. that must be kept in mind by a control system designer.148 Chapter 4 transformation. L(w) with the requirement to verify all J characteristic equations to assure that the sign of the controller gain is correct for a stable robust system design.24 Bilinear transformation characteristics.p.

This extension is based upon the mathematical development in Sec. in149 . The design is tuned to the extent of the uncertainty and the performance tolerances. The extent of parameter uncertainly is defined a priori and so are the tolerances on the system responses to the crosscoupling interaction cy and to external disturbance functions (d^ 2. The solutions to these single-loop problems are guaranteed to work for the MIMO plant. It is quantitative in nature. Any technique may be used for the single-loop design problems . as stated previously. in which structured parameter uncertainty.state space. 3. 5-2 which results in the representation of a MTMO control system by nf MISO equivalent control systems. This QFT MISO design technique is extended in this present and following chapters to enable the design of MTMO control systems. The design for a MTMO system. The highly structured uncertain linear-time-invariant (LTT) MIMO plant has the following features:4 1. The synthesis problem is converted into a number of single-loop problems. or even cut and try. frequency response. It is not necessary to consider the system characteristic equation. and performance tolerances are derived from the original MTMO problem.MULTIPLE-INPUT MULTIPLEOUTPUT (MIMO) PLANTS: STRUCTURED PLANT PARAMETER UNCERTAINTY 5-1 INTRODUCTION The previous chapters present the development and design procedures of the quantitative feedback theory (QFT) synthesis technique for MISO control systems. external disturbance.

5-2 THE MIMO PLANT The P matrix may be formed from either the system state space matrix representation or from the system linear differential equations. The design method requires that the uncertainty in P be known or is at least bounded. which are each designed as outlined in Chap. respectively. 5-6) that the MIMO problem for an rnxrn system can be separated into m equivalent single-loop MISO systems and nf prefilter/cross-coupling problems. In any MIMO system with m inputs there are at most m outputs which can be independently controlled.}... the transfer function matrix of the plant may be represented by the associated set {pt.1 AMMO plant. 3 and 4 for analog and digital control systems. . (see Sec.19 Therefore. the first step is to modify the model so that the dimensions of the input and output are the same. The design process for these individual loops is the same as the design of a MISO system described in Chaps. 3. This set of transfer function matrices is contained in #*. which are known or unknown. the set of all plant transfer functions relating each input to each output.1 for which P is a member of the plant set &(P e &). In general. The system is then defined as being of order mxm. an mxe open-loop MIMO plant can be represented in a matrix notation as y(t) = Pu(t) where ( . By using fixed point theory7'21 Horowitz has shown. The state space representation for a LTI MIMO system is: Fig. are used for both the input and output vectors in the design procedure presented here.1. 5.150 Chapter 5 volves the design of an equivalent set of MISO system feedback loops. p2. . 3. the same dimensions. /= m. If the model defines an unequal number of inputs and outputs. ) 51 y(t) = /^-dimensional plant output vector u(t) = ^-dimensional plant input vector P = mxl plant transfer function matrix relating u(t) to XO Consider the MIMO plant of Fig 5.4>ls The cross-coupling is akin to the disturbance DI of Fig.7'52 When a system has variable parameters.

only finite set of P matrices are formed. ) 5 6 The solution of Eq. Let the matrix premultiplying the output vector Y(s) be D(s) and the matrix premultiplying the input vector U(s) be N. yi(s) andy£s) are the outputs. . yields: Y(s) = D-'i(s)NU(s) = P(s)U(s) Thus. a satisfactory control system design based on linear techniques. The plant transfer function matrix Pis evaluated as IB outputs has the form: a(s)y1(s) + b(s)y2(s) = fu1(s) + gu2(s) c(s) y. ^55^ This is defined as a 2x2 system. d(s) "/ j h ~ „. and w/(s) and u£s) are the inputs. . In the general case with m inputs and m outputs. the general plant model for a MDvIO system with two inputs and two 15 4) where a(s) through d(s) are polynomials in s. B.5) may then be written as: D(s)Y(s) = NU(s) ( . Only LTI systems are considered in this text. representing the extreme boundaries of the plant uncertainty under varying conditions. where D must be nonsingular. (5. ) 53 If the plant model consists of m coupled linear-time-invariant differential equations. ) 57 This plant matrix P(s) = \py(s)] is a member of the set ff= {P(s)} of possible plant matrices which are functions of the structured uncertainty in the plant parameters. the mxm plant transfer function matrix P(s) is: (5.MEMO Plants: Structured Parameter Uncertainty 151 (5. the system is defined as mxm.f through / are constant coefficients. In practice.2) = Gc(t) where the A. (5. for many real world nonlinear problems. . . Eq. In matrix notation the system is represented by: a(s) c(s) b(s) . and C are constant matrices and x is an /w-dimensional vector..8) ( . By obtaining these LTI plants that describe the extremities of the uncertainty (the nonlinearity of the system) the QFT technique can achieve.6) for the output Y(s). (s) + d(s) y2 (s) = huj (s) + iu2 (s) ( . Thus.

12) are of the form: .10) or (R.11) -(R4+Ls)j. m Outputs Thus.9) (5. ft) = (R.11) and (5. (s) + (R2 + R3+R4 + Ls) j2 (s) =Ez (s) (5.(R4 + Ls)h (s) =E! (s) (5.1 The Kirchoff s Voltage law is applied to the electrical network of Fig.12. 5. m Inputs and . (5.+R4 + LD) i.ft)-(R4+ LD) J2 ft) e2 ft) = -(R4+LD)ij ft) + (R2 + R3+R4 + LD)h ft) (5.2 to yield the following differential equations (the "Z>" operator notation is used): e. 5. Let U2. Eqs. Example 5.2 An electrical network.12) assuming zero initial conditions.152 Chapter 5 R Fig. +R4+ Ls) u (s) .

m dn D ml du dim (5.... based on Eq.13) and (5.MEMO Plants: Structured Parameter Uncertainty 153 dn(s)Y.6). For this example.7(s)U72(s) 12 (5..2 (S)U2 (s) d21 (s)Y. (5. Eqs..2..7) and where m = 2.18) Um_ mxl mxl which is of the form of Eq.13) (5.+ nim (s)Um (s) (5.19) .14) where. (s) + n22 (s)U 2 (s) (5.m (s)Ym (s) = nn . for this example.15) where/ =l. (s) + d22 (s)Y2 (. the expression for Fis: p. (5.14) can be expressed as follows: ' Yi Y2 D mxm 'Ui U2 = N mxm '• -Ym_ '• (5.v) = a2.2 (s)Y3 (s) = n11VI (s) + n. (s) + d. n!2 = n2] = 0.17) ••• Hm Hm2 Thus.16) N= Jlml "22 (5. (5. (sJU. These equations are of the general form: d d.

20) P21 P22 Thus.3b is the SFG of the compensated MIMO closed-loop control system where the compenstor and prefilter matrices are. when Pis singular the system is uncontrollable. From Eq. Figure 5.18) yields: (5. 5.3a is obtained which represents a plant with structured plant parameter uncertainty with no crosscoupling effects. (5.. i. and Y2 are not independent of each other.e.relates the /'* output to the f input and the tolerance matrix is given by 3(5) = . resulting in _ Pll Pl2 _ (5.154 Chapter 5 Suppose that pup22 =Pnp2i. Equation (5. \P\ is singular. an uncontrollable system. Thus. they can not be controlled independently. Therefore. for this example.21) reveals that Y.21) Pu Equation (5.18) the signal flow graph (SFG) of Fig. respectively: J 11 J 12 /22_ G(s) = S22 F(s) = f2l The control ratio matrix Tis: til ti2 t22 T= t21 (5.22) where the closed-loop system control ratio % = y/r}.

5. There are m2 closed-loop system transfer functions (transmissions) ty{s) relating the outputs y.5 the following equations can be written: . P.3b may be represented by the simplified SFG of Fig. 5-3 INTRODUCTION TO MIMO COMPENSATION Figure 5. Figure 5. giving m2 sets of acceptable regions TV(S) which are to be specified in the design. From Fig..{s) to the inputs r}(s). there are tolerance bounds on each t^s). G.5 in which F. (5. 5. The application of QFT to 2x2 and 3x3 systems has been highly developed and is illustrated in later sections of this chapter. 5. and (b) the compensated MTMO control system. In a quantitative problem statement. and d>= {P} is a set of matrices due to plant uncertainty. and where its elements of the J(s) matrix are given by ra(s) -> by > ty > ap. thus t£s) eTij{s) and 3(s) = { Ty(s)}.e.36 has the mxm closed-loop MMO feedback control structure of Fig.3 The signal flow graph of: (a) Eq.MEMO Plants: Structured Parameter Uncertainty 155 c> Fig. i. y. 5. Tare each mxm matrices.14).4.(s) = /j(s)r/s).

{s)}.23) where the system control ratio relating r to y is: .3b. G(s) = diag {g. and F(s) = {^(s)} is the matrix of prefilter transfer functions which may also be a diagonal matrix. 5. The first two expressions yield: y=PGu which is utilized with the remaining two expressions to obtain y =PG[v -y] =PG[Fr . that is. w = Gu u =v .y v = Fr In these equations P(s) = [py(s)] is the matrix of plant transfer functions. 5. G(s) is the matrix of compensator transfer functions and is often simplified so that it is diagonal.y] This equation is rearranged to yield: y = [I + PG]'PGFr (5.4 The simplified SFG of Fig.5 MIMO feedback structure.156 Chapter 5 Fig. 5. Fig.

Most design approaches treat stability for fixed parameter set. 5. Method I and Method 2.MEMO Plants: Structured Parameter Uncertainty 157 T^V + PGl^PGF (5. However. 3. The design objective is a system which behaves as desired for the entire range of uncertainty. the QFT design procedure systematizes and simplifies the manner of achieving a satisfactory system design. ensuring that the characteristic polynomial [the denominator of Eq. neglecting uncertainty./*) may vary. the diagonal compensation .25). 5.24) To appreciate the difficulty of the design problem. and attempting to cope with the plant uncertainty by trying to design the system to have conservative stability margins.6.25) There are m2 = 9 such ti}(s) expressions (all have the same denominator). The structure for a three-by-three MTMO system is shown in Fig." is an outgrowth of Method 1 in which the designed components of the previously designed loop that is are used in the design of the succeeding loops. Two highly developed QFT design techniques. . (5.\-P2l fllSl + P22 f2lS2 + P23 fil 83} IP 12 82 (^ + P3383) ' P 32 Pl3 S2 g 3\ P32 fll g2 + P33 /31 83! {P23 Pl2 82 83'(1 + P 22 8 J Pl3 83]V (5. and there may be considerable uncertainty in the nine plant transfer functions Pi. (5. They consist of the uncertain plant matrix P.24)] has no factors in the RHP for all possible py(s). Clearly. 5-4 MIMO COMPENSATION The basic MIMO compensation structure for a two-by-two MIMO system is shown in Fig.(s). for the case m = 3 with a diagonal G matrix. is extremely difficult. Methods 1 and 2 utilize the MISO design method of Chap. Method 2. "the improved method. this is a very difficult problem. Even the stability problem alone. exist for the design of such systems and are presented in this chapter. This requires finding nine//*) and three gt(s) such that each ti}(s) stays within its acceptable region Ty(s) no matter how the/?.7. note the very complex expression for tu given by Eq. In both approaches the MIMO system is converted into an equivalent set of single-loop systems.

gm_ Jml Jm2 ••• Jmm_ _Pml Pm2 (5. and the prefilter matrix F..26) The dashes in Eq. 0 G= 0|. F.8 is obtained The SFG of Fig. (5.9 for a 3x3 system is obtained in a similar manner. g2\. Substituting these matrices into Eq.26) denote the G.158 Chapter 5 matrix G. it is noted that if \P(j:aj)G(ja>)\ » I then T « F and the system becomes insensitive to the parameter variations in P. . (5. (5. 5-5 INTRODUCTION TO MISO EQUIVALENTS From Eq..0 0 . the denominator of this equation becomes P2lSl I + Pssgs which is a very "messy" equation for the purpose of analysis and synthesis in achieving a satisfactory design of the control system.23) yields the ti. (5.. In analyzing Eq. 5..(s) control ratios relating the /'* ' output to the/* input.. From these t^s) expressions the SFG of Fig.23). This chapter considers only a diagonal G matrix. though a nondiagonal G matrix allows the designer much more design flexibility. 5.23) obtain1 PG])PGF det I + PGl For a 3x3 plant and a diagonal G matrix.24 These matrices are defined as follows: 'g. and P matrices for a 2x2 system. 0 0" fll /21 fl2 •-• flm /22 ••• /2m Pll P21 Pl2\ P22\ Pi F= P= ..

5.^ + k . . 5. ^-J* G 1 nx3 ^j -i -i -i P y(t) Fig. 5.MIMO Plants: Structured Parameter Uncertainty 159 J r(t) -\-f —v ' *u *• G -i -1 P y(t) r Fig. f"A w v * ^H VJ +i Lj^V^.8 Two-by-two (2x2) MIMO signal flow graph.6 MIMO control structure two-by-two system (2x2) r(t) •TV.7 MIMO control structure three-by-three system (3x3) Fig.

each with two inputs and one output. Eq.23) is premultipliedby [/ + PG\ to obtain [/ + PGF=PGF (5. .1 EFFECTIVE MISO EQUIVALENTS The objective of this section is to find a suitable mapping that permits the analysis and synthesis of a MIMO control system by a set of equivalent MISO control systems. One input is designated as a "desired" input and the other as an "unwanted" input (cross-coupling effects and/or external system disturbances). Let Fig.27) When P is nonsingular. (5.9 Three-by-three (3x3) MIMO signal flow graph. This mapping results in m2 equivalent systems.28) which puts the constrained or the structured parametric uncertainty in one place in the equation.160 Chapters 5-5. then premultiplying both sides of this equation by P'1 yields [P~l + G]T = GF (5. 5. First.

.29) _* ml P . (5. rewrite Eq.l/P]m -l/P*2n.28) using Eq.32) 1 where A is the diagonal part and B is the balance of P1.30) V Adji}P 1/P]2 1/P*22 The Q matrix is then formed as ll <Jl2 -<J 1/Pn 1/P21 .MEMO Plants: Structured Parameter Uncertainty 161 Pll P*21 Pl2 P*22 -Pin -P*2n (5. thus AH = l/qu = P*i. * ••• r m..l/P*mm_ where P*j\ The matrix P is partitioned to the form ]= A + (5. Next. *P m2 -. (5.andbij = l/qij = P*jfori*j. This yields [A + G\ T = GF -BT which produces ..bn = 0. q21 Q= (5.32) and with G diagonal..31) J/P*ml l/P*m2 . P The m2 effective plant transfer functions are based upon defining: det[P] (5.

5. for a unit impulse input.11 and is the control ratio that relates the /'* output to the/ input.10 Schauder fixed point mapping. The general transformation result of m2 MISO system loops is shown in Fig. Tn T.33) are both diagonal.34) for the 3x3 case. Recalling that A and G in Eq. (5. where / =j = 1 in Eq.e.7. (5. i. 5. Figure 5. Similarly each of the nine structures in Fig.11 shows the four effective MISO loops (in the . If this mapping has a fixed point.35) This corresponds precisely to the first structure in Fig. 5.34) where each 7} and 7} is from the acceptable set 3. (5.) = 7} (see Fig.162 Chapter 5 Enclosure of all acceptable T&3 Fig. 5.7 This theorem is described by defining a mapping on 3 as follows: T (5. l [GF-BT] (5. <= 3 such that Y(T. yields the output gtfn-\ (5. where each of the m2 matrix elements on the right side of Eq. (5. based upon unit impulse functions.35).12. (5.33).. 5.33) This expression is used to define the desired fixed point mapping. Proof of the fact that design of the individual MISO feedback loops will yield a satisfactory MIMO design is based on the Schauder fixed point theorem. then a solution to the robust control problem has been achieved yielding a solution in the acceptable set . the 1.10).33) can be interpreted as a MISO problem.1 element on the right side of Eq.11 corresponds to one of the elements of Y(T) of Eq.

as y =(v ) +(V-) \SyJ \JljJcij rj = V + V Cij Sri S (5.12 obtained from Eq..38) and k = l.m (5.39) represents the interaction (cross-coupling) between the loops. (5.36) can thus be expressed..2.MEMO Plants: Structured Parameter Uncertainty 163 boxed area) resulting from a 2x2 system and the nine effective MISO loops . and the transfer function w^Cy relates the /'* output to they* "cross-coupling effect" input c#. .. Thus. 5. Eq.36) represents the control ratio of the /* MISO loop where the transfer function wuVy relates the /'* output to . for the mxm system of Fig.37) (5. The transfer function of Eq.4 The control ratios.36) where (5.11 Effective MISO loops two-by-two (boxed in loops) and three-by-three (all 9 loops).40) or Fig.34) have the form (5. for unit impulse inputs. (5.. with r^s) = 1 (a unit impulse function).^resulting from a 3x3 system.'* "desired" input /-.39) Equation (5. 5. (5.

(5. In each of the nine structures of Fig.10).{s) must be chosen to ensure that this condition is satisfied. 5.44) Thus (5. All the gi(s). 5.42) (5. .45) represents the maximum portion of by allocated towards cross-coupling effect rejection and by represents the upper bound for the tracking portion of /*. with r(s) = 1.-.11 it is necessary that the control ratio yv(s). must be a member of the acceptable set t(i e 3. The object of the design is to have each loop track its desired input while minimizing the outputs due to the cross-coupling effects. in the low frequency range is expressed as (5.43) and where now the upper bound. For any particular loop there is a cross-coupling effect input which is a function of all the other loop outputs.41) where (5.J(s) (see Fig. 5.164 Chapter 5 Fig.12 Effective MSO loops (in general).fi.

5. i..34) may be replaced by a % and al 1 the elements of T on the right side of the same equation by tig. TKu or TD in Fig. For a controllable and observable plant. These are obtained from the design specifications. b^ is a function of the response requirements on the output^ due to the input /}.9 it. The lower bound % needs defining only when there is a command input. and the second subscript. 5.5).p. The cross-coupling effect ctj(s) expressed by Eq.4 The first subscript k refers to the output variable. Therefore. the transmission zeros can be computed from the determinant of the system matrix which is defined1>75 by si-A B C 0 where A. The fi.m.{s) are the compensator elements of the F(s) and G(s) matrices described previously. which is a solution to Fig. each in its acceptable set.e./ refers to the input variable.MEMO Plants: Structured Parameter Uncertainty 165 thus constituting nine MISO design problems. (5. 5-6 EFFECTIVE MISO LOOPS OF THE MIMO SYSTEM There are two design methods for designing MIMO systems.{s) and g. The number of transmission zeros26 for a system having the same number of inputs as outputs is equal to d where d is the rank deficiency of the matrix product of CB. and then ^{5) on the left side of Eq. there exists a fixed point.12 is treated as an individual MISO design problem.2. k = l.11 and 5. (5. for the respective output/input relationship.39) represents the interaction between the loops. Note that if the plant has transmission zeros in the right-half-plane (RHP) it only indicates that qti may be n. or the del P may have zeros in the RHP. and / are mxm matrices. then this solution must be unique. C. If each element is 1:1.4b or 3.. 3 and 4. It should be noted that if the phase margin frequencies co$ of the loops are not -widely separated there is considerable interaction between the loops5" 5-6. (5.. 7. A more formal and detailed treatment is given in Ref. 5. ^-.m . If all of these MISO problems are solved..46) where the numerator b^ is the upper response bound. 3. This means that there exist nine tik and %. B. Using a diagonal matrix results in restricting the design freedom available to achieve the desired performance . which is solved using the procedures explained in Chaps. In the first method each MISO loop in Figs..1 EXAMPLE: THE 2x2 PLANT For this example a diagonal G matrix is utilized.

(5.50) -A Pll A />H Substituting Eq.49) 1 A P22 -A />«' (5.51).30): P!= where tfll 1 $12 (5.47) P22 •P21 . obtained from Eq. gt.51) 1 421 J21 t22^ S2f22_ ~°h2 + g2 The responses due to input 1.82/21 8lfl2 (5. (5. The P and Fl matrices are..FromEq.e.Pl2 Pll (5.48) where A = . This is offset by the resulting simplified design process.49) into Eq. The elements of a diagonal G are denoted with a single subscript. respectively: Pu P21 Pn P22. (5. are: . (5. (5.P12P21.28) yields: 7 —— (In + 81 1 ' —— In I til tn SlJ 11 . i.166 Chapter 5 specifications.

54) For input ——— g.51).52) and (5. (5. )ields the equations shown in Fig. 1 (5.MEMO Plants: Structured Parameter Uncertainty 167 In ' ! 1i2 (5. .53) HL 121 + (-L + + —— <l22 = £2/22 These equations are rearranged into a format that readily permits the synthesis of the gi's and them's that will result in the MIMO control system achieving the desired system performance specifications. = f The responses caused by input 2.13.52) tlL + (_L + 121 122 ).53) are manipulated to achieve the following format: — For inputn : tn = ~ = ~~ ——~ .54).+ §2 (5. respectively. Equations (5. (5. obtained from Eq. 5. are: . Associated with each equation in this figure is its corresponding SFG. 22 qn ——— +§2 q22 Multiplying the tn and t12 equations by qn and the t2i and t22 equations by q22 in Eq.

57) L2f22 '22" l+U (5. These terms are Junctions of the other tk/s and the structured parameter uncertainty of the plant.58) are of the format of Eqs. In a similar iashion.{/ffl)| "large enough.2 by interchanging 1 -> 2 and 2 -> 1 in the equations for tn and tI2. Equations (5.. Theoretically. by making |Z." so that Cy & 0 a "decoupled system" is achieved. .55) (5. 5. 2. the % expressions and their corresponding SFG may be obtained for any mxm control system.55) through (5. t22 and t2i can automatically be obtained from the expressions for tn and t. Note: 1. (5.45). The Cy terms in these equations represent the cross-coupling effect from the other loops.f» 1+L.36) through (5.13 2x2 MISO structures and their respective ^equations. (5.168 Chapter 5 (5. 3.58) Fig.= 12 L..56) t.

5. from Eq.MIMO Plants: Structured Parameter Uncertainty 169 5-6. For all P etFand t2i e T2!.59) Let (j> . be the actual value of tc. (5. (5.2 PERFORMANCE BOUNDS Based upon unit impulse inputs.14 Upper and lower tracking bounds. For LTI system. the linear superposition theorem is utilized in the development of the performance bounds. The "actual value" expression corresponding to the tn expression of Eq. and cu = -(21/112. -co Fig.. be the actual value of ttj f be the actual value of tr TC..41) obtain: ty = tnj + tcij (5.55) is where the rr term represents the transmission due to the command input r/ and the TC term represents the transmission due to the cross-coupling effects. the output $/ must satisfy the TR. A 2x2 control system is used to illustrate the concept of performance bounds. .

as illustrated in Fig. Only magnitudes need be considered for the rap. 5. thus. systems. From this figure the following expression is obtained: (5. it is necessary to specify a priori the closed-loop transfer functions %.60) is expressed as follows: an(ia>)<\\ Trn -\TCII\\ -.p. is shown in Fig. in terms of the actual values: an < | <t>nG<*>)\ ^bn(jca) -\Trn + Tcll\^ lower bound ft upper bound fr (5. Thus. As is discussed later.64) . (5. a^ alj . system. Note that the I's in this figure and in the remaining discussion in this section represent only magnitudes. for a 2x2 system. for CD = 0%.44) and (5. 14 [see Eqs. for the high frequency range only an upper bound is of concern. 15.2 Consider the bound determinations % by . system since the magnitude determines the phase. AT is split up based upon the desired performance specifications.62).6o) Since the relative phases of the t's are not known and not required for m.. (5.63) In the "low frequency range" the bandwidth (BW) of concern.45)] for an m. Thus. Consider the specifications bu and an on tu.the smallest bound (5.p. based upon <*=<Tril)L*tru\*(Tru)v = lti (5.. 5.62) These represent an overdesign since they result in a more restrictive performance.61) and | T |+1T l^bn (ja>) . on LI = giqn." Example 5. there is no need to be concerned with a "split. (5. bril . Thus. and ba.170 Chapter 5 performance specifications on tn. A pictorial representation of Eqs.61) and (5. then to ensure the achievement of the desired performance specifications Eq. 0 < co < cot.the largest bound (5.

rCll = 0.05 T —l (5.15. can be determined in the same manner as for a MISO system (see Chap.13 and to Eq.65) where for the cross-coupling effect the upper bound for tC]J is given by \t Cij \= t* cjj ^b Cij= h en where i= •>i = 1 ' i 1 u u Substitutuig the expression for cn [see Eq. for illustrative purposes only.15 Allocation for tracking and cross-coupling specifications for 4 responses. the bound on the crosscoupling effect bCif is determined as follows: I ten I = <T cij = i en =0.05 and Arry/ = 0.55). (5. Referring to Fig. (5. 5.65) and then ma- . 5.MEMO Plants: Structured Parameter Uncertainty 171 CO Fig. (5.46)] into Eq. 5. 3). and Fig. for /. In Fig..15. 5.2 at a) = (Oj.

. P..70) "High bounds" on the NC require h.172 Chapter 5 nipulating this equation the following constraint on /-.The following magnitudes.67) ii Inverting this equation yields: ti 9l2 (5./plant models. (5. for each of the .69) In determining these bounds.68) 11 20 which is of the mathematical format that allows the use of the NC for the graphical determination of the cross-coupling bound bcn (see Sec. thus. (i = 1.2 (5.\>- .71) .g. where bn is the upper bound on t21.66) where the upper bound b2J is inserted for t21. insert the corresponding plant parameters into these equations and determine the magnitudes^..= 20 9l2 = 20 11 (5. This is accomplished by making the substitution of L = lie. B. respectively.68) are. the optimum choice of the (5.. are used from these J sets of values to determine the cross-coupling bound: (5. it is necessary to insert the actual plant parameters into these equations. of the following mathematical format: \A\>\B\ (a) \C\<\D\ (b) (5. That is. for each value of a>.. j_ ii >20 11 9. is obtained: •L.).. Thus. 3-10). Equations (5..J). CK and A.(/».2. in order to minimize the required compensator gain..67) and Eq. into this equation. It is necessary to manipulate this equation in a manner that permits the utilization of the Nichols Chart (NC).

Method 2 may involve a tradeoff in the design parameters. This method is an improvement over Method 1 in that it utilizes the resulting designed g/s andfy's of the first MISO equivalent loop that is designed in the design of the succeeding loops. adjust tcll until the condition is satisfied The MDvlO/QFT CAD discussed in App.MIMO Plants: Structured Parameter Uncertainty 173 specifications are those that result in achieving essentially the same tracking and cross-coupling bounds. the optimal bound b0. Depending on the starting quantities. 4 5-6.n is and then. 16.4 QFT DESIGN METHOD 2 The QFT design Method 2. for many problems. by decreasing (increasing) one quantity and increasing (decreasing) the other quantity expedites achieving the condition of Eq. The final MISO equivalent loop to be designed uses the exact transfer functions of the previously designed g/s andfy's thus this loop has the least amount of overdesign.4 The order in which the MISO loops are designed is important.42) through (5.e. (5. determine the other bounds required for the other L/s. . in loop shaping Lh based on Eqs. 5-6. (5. A automates this procedure.2.45) and (5. it is necessary to obtain the bounds for L0. by trial and error if a CAD package is not available. and ta.57). (5. is given by Eq.e.72).72). This may require trial and error.3 QFT DESIGN METHOD 1 hi the manner described in Sec.. i.71) is to do a design initially based on the following assumption: With this design. i. determine how big r. This procedure is illustrated in Fig.. which satisfy the specifications for both t.72) A recommended method for determining an appropriate set of constraints (specifications) on Eq. bcn»brll (5. (5. By the procedure just described. This feature of Method 2 reduces the overdesign in the early part of the design process.. 5. Any order may be chosen but some orders may produce less overdesign (lower bandwidth) than others. The general rule hi the choice of the design order of the loops is that . Once these specifications are satisfied then proceed to do the looping shaping to determine L0. etc. for determining the bounds of the first loop to be design. may yield a better control system design. 5-6.

The h.) uncertainty of qa may be very much greater than for the other loop(s). That is.174 Chapter 5 Trial 1 Result tc dominates Trial 2 4db \/ 13 db U Easier on Tc but harder on AtR but T_ still dominates. choose the "starting" loop / for which it is most important to minimize the BW requirements. 5.g. Some of the factors involved in determining the BW requirements are: 1. v/XTLTLO 6db Thus iterate ATR and Tc values between Trial 2 and 3 until BR ~ Bc within the range 0<COi<0)h Fig.g. The "high frequency gain" (h.f. uncertainty may effect the size of the templates. 4db Trial 3 6db 9db AtR now dominates. 3.f. Loop / has severe bending mode problems that other loops do not have. Sensor noise 2. . the most constrained loop is chosen as the starting loop.16 Procedure pictorial representation to achieve Br a Bc.

the BW of the succeeding designed loops are higher than the B W of the previously designed loops. 9. are also roughly similar for the different loops. If. then try a different starring loop selection.73) is manipulated to yield: . If Method 1 can not be used then it is necessary to reevaluate the performance specifications. up to the frequency range in which the templates first narrow. If Method 2 can not satisfy the B W requirements for all loops then Method 1 must be used. that in general. For the discrete case. Thus. 5-6. The reason for this choice is that there is a tendency towards BW propagation as the design proceeds from the first loop to the final loop that is designed." That is.MEMO Plants: Structured Parameter Uncertainty 175 Professor Horowitz provides the following insight: (1) Analyze the various qa templates over a reasonable range of frequencies: almost vertical at low and high (for the analog case) frequencies. This is based on the knowledge. and the % . the responses must be less than some bound. 4). (2) If the feedback requirements per loop are roughly the same.5 SUMMARY For the cross-coupling effect rejections problem. Further insight into satisfying the B W requirement is given in Chap. */. see the discussion in Chap. 5-7) is satisfied. which indicates the 4 specifications are about the same. 9 All of these factors emerge from the transparency of QFT which helps to reduce the trial and error that is involved in achieving a satisfactory design. Also. Equation (5. assuming the diagonal dominance condition (see Sec. before widening again for co. the second loop to be designed should be the one having the second smallest feedback requirement.. the loop with the smallest amount of feedback should be chosen as the starting loop.qti = Lt. > co. then the loop with the smallest qa templates should be the "starting loop. rather than the final frequency range (o}t > cosl2) in which the templates approach a vertical line (see Chap. in choosing the loop with the most severe BW limitation causes a design problem in the succeeding designed loops./2 (a>s sampling frequency).73) where g. etc. that is: Thus the loop equations become (5.

77) can be expressed as (5.76) Remember to use only the magnitude in the cross-coupling calculations.74) Substituting Eq.46). Eqs. Therefore. This assumes the worst case.36). with impulse input functions. I/.74) yields: 7 +L > (5. and in the first term k = 2 and in the second term k = 3.75) For example.46) into Eq. (5. Eqs. (5.78) yield.78) Consider first the 2x2 plant.176 Chapter 5 1 + L.37) and (5. (5.41) can be approximated by ta*ycu (5-77> Thus from Eqs.40) and (5. Eq.. 5-7 CONSTRAINTS ON THE PLANT MATRIX21 In order to use the QFT technique the following critical condition must be satisfied Condition 1: Pmust be nonsingular for any combination of possible plant parameters.j = 2. (5.75) becomes: (5.46) and (5. i. where / = l. Eq. (5. (5.\>^ (5. to ensure that Pl exists. for i =j = 1: . (5. m = 2. Specifying that \ycn\ < ba (the given upper bound in the high frequency range). (5. in a 3x3 system for the first loop.e. In the high frequency range y (jo) approaches zero as co -» oo.

81) '12 / \>bn 2 922 (5.80) are rearranged to 21 11 '11 (5. where it is specified that (yc)2J < b2l (the given cross-couphng up- per bound). results in 7> (5. (5. thus Eqs.79) yield: -bn 121 122 (5.82) ' A 021 921 Multiplying Eqs.82).81) by Eq.MIMO Plants: Structured Parameter Uncertainty 177 bn> -b2i (5. = 2.46) and (5.j = 1.83) Since Pll _P2l Pi 2 P22_ .Pl2 Pll_ Qll $12 and then .80) I +L2 Equations (5. = L2 « 0 in the high frequency range.79) For. where L. (5.-UH Pll _~ ?21 . Henceforth. (5. . the cross-couphng bound notation bcij is simplified to by for / */.79) and (5.

This condition may also be obtained by using the right column of the MISO loops in Fig 5. Condition 2 (2x2 plant): As co -» oo Pyj^H/^^l or /'/7/'22 -|/'/2/ 2. then Eq (5.46) and (5.l 7 >0 (a) (5.178 Chapter 5 1 * £ * <?/? ~~ 1 *~~ _ £ * Pn Pn 1 _ £ 121 122 />/2 1 Pn (5.6.88) .11.84) into Eq. 7> n n ?/y (5.11 since the loop transmissions LI and L2 are again involved in the derivation.85) is the diagonal dominance condition for the 2x2 plant. 5. for / = 7.e.86) (5. For / =j = 1 (the left column of Fig.3 respectively.2. m = 3. (5.85) Since pn andp22 are elements of the principal diagonal of P.84) _ £ P22 P21 P21 P22 Substituting Eqs. (5.85) is not satisfied then refer to Item 3 in the summary of Example 6. This condition is obtained considering only the left column of the MISO loops for the 2x2 plant of Fig 5. (5.83) yields Condition 2. Next consider the 3x3 plant i. where L1 = L2 = L3 &0 as co -> <x> and where it is specified that Eqs.78) yield.87) 23 21 1> Letting 133 n *2J «32 (5. (5. If Eq.. Equation (5.11).85(a)) must also be satisfied for all frequencies.

A.91) and (5. (5. 1> 9l2 (5.94) and (5. (5. (5.93) the following expressions for 2/ and A? are obtained (5.86) through (5.94) (5.89) bn Equations (5.MEMO Plants: Structured Parameter Uncertainty 179 (5.92) and using _ 922 9 33 =2 923932 (5. respectively.88) become.90) I <l23 (5.91) 1 .95) Substitute Eqs. (5-92) From Eqs.95) into Eq.90) to obtain .

P 3 2 1-- Pn Multiplying Eq. into Eq. (5.96) yields: Pl2 Pzi P22 P23P31 P 22 P 33 Pll (5.180 Chapter 5 <\22 22 (5. applies only for QFT design Method 1) As co —> oo I P*21 P33 I+ 1 P*23 P31 I J + (5. Condition 3 (3x3 plant.97) by Ps yields Condition 3.97) * * P 2 . .98) PllP23P 3 2 . (5.96) 111 113 <l33 133 ^22 I 32 < 1- 131 Substituting q^ = I / p .

In some design problems it may be necessary or desired to determine a Pn upon which the QFT design is accomplished Doing this minimizes the effort required to achieve the desired BW and minimizes the cross-coupling effects. With plant uncertainty the off-diagonal terms of Pn will not be zero but "very small" in comparison to P. resulting from the original ordering of the elements of the input and output vectors does not satisfy Condition 2. = 0. 7. This desired result can be achieved by post multiplying Pby a matrix Wto yield: P =PW = [p.e.. i. is specified in order to achieve the performance specification: | "c. ] where p.(J(o). Plant uncertainty and loop interaction (cross-coupling) makes the ideal response unachievable. In general. Thus. i *j. having no parameter uncertainty. This condition is necessary only if Eq.5. for .{/®)| < bi.MIMO Plants: Structured Parameter Uncertainty 181 See Ref. then a reordering of the input and output vectors may result in satisfying these conditions. it is clearly best to let _//. Condition 1 ensures controllability of the plant since the inverse of P produces the effective transfer functions used in the design. Since |£.. an upper bound % / *j. For an LTI plant. . the system performance specifications describe a range of acceptable responses for the commanded output and a maximum tolerable response for the uncommanded outputs. the output yk ~ 0 due to c^. for all Pe 9 A system designed to this specification is called BNIC. This was done on an AFTI-16 design by Horowitz30 as shown in Fig. If the P matrix. <bi.. (5. 7 for higher order plants. #j. for the nonnominal plant cases in &. The uncommanded outputs are treated as cross-coupling effects (akin to disturbances). it is possible to essentially achieve zero cross-coupling effects.8) is used to generate the plant. Thus. f „ .. 5-8 BASICALLY NONEVTERACTING (BNIQ LOOPS A basically noninteracting (BNIC) loop4 is one in which the output y^s) due to the input rj(s) is ideally zero. for all o. 17. =0 for i * j resulting in a diagonal Pn matrix for P representing the nominal plant case in the set 9.

e .elevator deflection. 3 and 4. . If it is possible to find the//. satisfy the design specifications of the original MIMO system.17 Output time response sketch for 2x2 plant: a . within the BW. are available in which the equivalent MISO loops are designed according to the MISO design method outlined in Chaps. Guidelines for finding the P matrix. The method of representing a MIMO system by an equivalent set of MISO systems is presented using P~'. 6 and 7. the compensator must be "good enough" (large enough gain over the desired BW) to handle the "feedback needs" of the worst of the m systems for each value of ®.12. Since the gt's are the same for all m MISO systems in each row of Fig.pitch rate.roll rate. Two design approaches. are given.9.182 Chapter 5 -*-t Fig.aileron deflection. 5-9 SUMMARY This chapter describes the multiple-input multiple-output closed-loop system and the plant matrix. Thus. which relates the input vector to the output vector. the mxm MIMO has been converted into m design problems. q . 5. 5.? and the g/s which satisfy the performance specifications for the m2 systems then it is guaranteed that these prefilters and compensators. respectively. 5. when used as elements in the MMO systems of Fig.8 or 5. andp . which are discussed in detail in Chaps.

so the plant set P~' generates sets of ^ = {</. to the design of the m2 MISO equivalent systems.3. = xC.. k = 2.} .>}. and design performance trade-offs.Z — L (6. and one output. the QFT design technique developed in Chapters 3 and 4 can now be applied.11.11. P* = [//%!. with the appropriate modifications indicated in Chapter 5.?) in Fig. This chapter presents the details of the QFT Design Method 1. ft.11 for m = 3).}.. 5.11 is any member of a set ^generated by the tu in TU and the q^ in <^. sensitivity analysis. 5. Let the plant inverse matrix P~' = \l/qv\. Replace the mxm MIMO problem by m single loops and m2 prefilters (see Fig.{s)/rj{s) [outputy.5]. The cross-coupling effect in Fig 5..DESIGN METHOD 1 -THE SINGLE-LOOP (MISO) EQUIVALENTS18 6-1 INTRODUCTION The QFT technique developed in Chapters 2 through 4 is based upon a MISO system structure having two or three external inputs and one output. and the uncertainty in P generates sets Q = }.2) '' 183 . let T^S) be the set of acceptable transfer functions ta(s) = y. one cross-coupling effect input signal. { c =. The following chapter presents the details of the QFT Design Method 2. In Fig 5. In Chapter 5 it is shown that an mxm MIMO system can be represented by m3 MISO equivalent systems having a desired tracking input command. Thus. 5.{s) in response to the command input /•/. As indicated in Chapter 5. In general in Fig.

4] B33e[-l..W] ^e[15. the uncertainty problem (due to the <ph c.2. in the words of Professor Horowitz. 20] (6. 5. three cross-coupling bounds and one tracking bound.2\ 2?. and/« so that its output is in TI2 for all qu in c^\ and all cn in c. etc. 4] A3I e [-0.m. the MISO design problem is to find L. is the same for all the MISO structures in the first row of Fig. System stability (and much more than that) for all P in ffis automatically guaranteed. thus. for part of the structured parameter uncertainty range. (5. Thus. 3-12.. such that the output is a member of the set T. = g.3 e [10.. for the middle MISO loop in the top row of Fig. "was done as a Master's thesis by a typical graduate student who like nearly all control graduates.1 1 find L. It is easier to present the important ideas by means of a design example. Also. 5.4] 5.5.3e[\.1] all other Ay = 0 By* e [-1." The plant and uncertainties are A. the toughest portions of these four bounds are combined to form the optimal bound to be used in loop shaping L/.15. e [-1. 4. the optimal bound bou is derived and for which the toughest of these bounds must be satisfied by Z/. for each row there will exist.i and/..m. It is not necessary to consider the highly complex system characteristic equation [denominator of Eq. satisfy the MIMO uncertainty problem.25)] with its uncertainty py plant parameters.5. 1] B2le[l.q. for all qu'va <^n and for all Cu in c-. 2] A22 e [5..p. has had no courses dealing with uncertainty. 6-2 DESIGN EXAMPLE The power of the technique is illustrated by presenting the results of Ref.1 1.2 e [-0. the plant is unstable for part of .3) Note thatp/2 is always nonminium phase (n. = g..q.) and/?/. in the manner described in Sec. which.8] B. -1. etc. is also n. In each of these three structures. then it is guaranteed that these same fa g. If the designer designs these MISO systems to satisfy their above stated specifications. assuming a diagonal prefilter matrix F.11.2] B22e[5.8. 1. needs. That is. 101] B. 2] A./) gives bounds which in-turn result in the optimal bound boll the level of feedback L.25] F e [0. 5.184 Chapter 6 for the set of acceptable ?$ transfer functions.2] E € [-0. For the top row of MISO loops in Fig. Similarly. e [-0. Note that L. e [2.p..

nor in columns 1. (see Sec. and (c) allocation.0 2. Hence..0 10. just as in the single-loop system (see Chap.1 Tolerances (a) on |fc.0 5. 6.2 0.{/ffl)|. as previously noted62. The four remaining specifications are |*/X/<0)| < 0. (b) on \t]2 (ja>)\.2\.2. 0.<*)]. it follows that if det P(s) has RHP zeros.2 0. 4 and App. 6. At small a> there is no diagonal dominance in any row. \t32. . 7-10). but success can be guaranteed only if the performance tolerances are compatible with the n.0 5. thus.1a for ^u(fco)\.1 0.p. 6.0 0. D-2).11 are qn = [det P\l\Adjlt P\.3.m.p. and in Fig. in the frequency domain \tij(jco)\ suffices when they are m.0 (a) (b) Fig.5 1.m. time-domain specifications on the system output and on as many of its derivatives as desired may then be achieved by means of bounds on fe{/'o)|. Since the MISO plants in Fig.1 0. The technique can still be used. Rosenbrock's technique61 can not be used even if there was no parameter uncertainty. They are shown in Fig. character. In iact.p. 5. 1 for all a>. 6-2. i = 1.0 10.5 1. det P(s) has all its zeros in the interior of the LHP.0 20. so this problem does not arise.2. \t}j(ja>)\. the equivalent single-loop transmissions are n.0 20.0 2.37 For this design example.Single Loop Equivalents 185 the parameter range.1 PERFORMANCE TOLERANCES Performance tolerances are on the magnitudes of the elements of the 3x3 closed loop system transmission matrix T = [//.1b for \t.

cross-coupling effects. where Y =Y + and T R = Y_ R R IN. because ideally zero t.m. But this is impossible because of parameter uncertainty. so upper bounds must be assigned.p.4) where £ represents the variable parameter in T. t23.2.5) (c) R kn Fig. The functions which satisfy the assigned tolerances on ty constitute a set %. Figure 6.186 Chapter 6 The first five are called interacting and the latter four basically noninteracting (BNIC). 8). N FL 1+ L -L 1+ L P 1+ L (a) TN = (b) (6. The effect of these items on system performance can be expressed in terms of the sensitivity Junction1 _ T_dS (6. Using the linear superposition theorem. The matter of m. or n. t}2 are deliberately chosen interacting in order to obtain a design problem with considerable variety. Y . character of the BNIC elements is of no concern.3. t2l. 6-2. t. and external system disturbances (see Chap.2 An example of system sensitivity analysis. parameter uncertainty.2 is used for the purpose of analyzing the sensitivity of a system to three of these items.2 SENSITIVITY ANALYSIS" An important aspect in the design of a control system is the insensitivity of the system outputs to items such as: sensor noise.p. 6. t3l are desired for all a>.

5b) tends to be greater than the BW of Eq.7) P(f ) *• CO Fig. 6. the following transfer functions and sensitivity functions (where 5 = P) are obtained. then: For the low frequency range. (6. (6.3. it is necessary to achieve a slope for Lm L0(ja>) that minimizes the effect on the system due to sensor noise. This is the most important case.6) ^ Since sensitivity is a function of frequency. . since the "minimum BW" of Eq. (6. where \L(ja))\» 1.2. 6. Based on the magnitude characteristic of L0 for low and high frequency ranges.6b) (6. respectively: TR = _FG_ -TN P ~° l+L 1 l +L (b) (c) (6.3 Frequency response characteristics for the system of Fig. 6. from Eq.Single Loop Equivalents 187 and L = GP.5a). as illustrated in Fig.

where \L(ja>)\«1. . 6.4 Bandwidth characteristics of Fig. (6. 6.1 Ob) in the "high fre- -LmP dB Sensor noise at plant input Cut-off frequency Fig. 6. (6.6b) (6. the designer must try to locate the condition of Eq. 6.9) -» log[S£7V]<0 (a) ] > 0 (b) (6.188 Chapter 6 For the high frequency range. with respect to sensitivity. (6. from Eq.4. the low frequency sensitivity given by Eq.8) is unsatisfactory since it can present a serious noise rejection problem.2. small enough in order to minimize the sensor noise effect on the system's output For most practical systems n > w + 2 and (6. it is necessary to try to make the phase margin frequency <»+ (the loop transmission BW). As seen from this figure.4. Based upon the analysis of Fig.7) is satisfactory but the high frequency sensitivity given by Eq.8) The BW characteristics of the open-loop function L(j(o).10) Thus. (6. are illustrated in Fig.

Also shown in this figure are the corresponding lower bounds aa and ah. There are two kinds of performance tolerances.0(i * j) in Eq.. la and b. in Fig.i = 1. the noise effect on the output is negligible. .11 y. respectively. i. (c) 1 + L.1.3 SIMPLIFICATION OF THE SINGLE-LOOP STRUCTURES For the second row of Fig. 2. The analysis for external disturbance effect (see Chap. 2. That is.q.e. respectively.11) At a fixed P e & (and hence nf %) \y . V • = (b) where L.]. Let some fixedPbe chosen as the nominal plant matrix P0 = (#.\max occurs at I Cji 'max {I **' « The relative phases of the talnik are not known so the extreme magnitude of bk (6.. 8) on the system output is identical to that for cross-coupling effects.3: 3. 6-2.3.. i.11) for (7.) it is necessary that (6. For the BNIC (k. 5. low sensitivity is conducive to their rejection. 2. Their respective outputs are: J f-L-i i (a) ' 1 + L. so it is clearly best to force % = Oby setting/.3: etc.) range where the system performance is not of concern.. For either case.13) x — "ij^-J ) because the relative phase of the two terms is not known.e. = g. generating a nominal P0~' = [l/qijo] and the .j(s) has the two components due to rj(s) = I and to Cjj(s). (6. 6.f. tolerances on 4 and on tv for/ ^y(BNIC).Single Loop Equivalents 189 quency" (h.7). where ba and bh represent the upper bounds of the control ratio tlt and tu. (6.3. 2.12) resulting from among the J plants must be used in the design.1..

15) is denoted as a D-type. In this example.0qjqiio. it is necessary that: | > ay (6.11) the above may be summarized by Eq. Suppose that due to the uncertainty in qih \ %| of Eq.11).11) is in the interval [a1.190 Chapter 6 nominal loop transmissions L. The specifications on the interacting tu terms (all 4 and ta.2 A22 A3! Bn =0-1 B13=4 B22=5 A n = 1.8 32i = 2 B23 =4 = 20 (6. qik b.1 = -1 = 5 = 1. b'^ ] [see Fig.. Icj. by being the designated bounds shown in Fig. Tc are not known a priori.lib) and (6. specifically A/.3) are assumed as: An = 2 3. 6. 16) and \ < 1 + L. = L. so L. the nominal plant parameters of Eq. (jco) \ < by (co) with ay.0 = &</„„. (6.. using Eq.12) the condition given by Eq.13) can now be written (for the BNIC tu) to yield the following design specification: 1 1 + liia ' '" „ h v _ z h v (6.q **> max X k*i Cij 1 (b) (6. (6.16) From Eq.- an 1 + Li ~ c . (6.15) ~cvqa qn quo k*. (6. Specification type described by Eq.17) bki q —*Iik max . t}2) are. (6.14) B3l=l E =2 B32 333 = 10 F =2 Using Eqs.1. ay (to) < I Tij (joa) + TCU. (6. (6. Since the relative phases of of TtJ. (6. 6.

B B .2 2.21 q .12 3l3 .Table 6.2 < a2 -i-ll 111 31 2. D31 32 33 33 f33_ B3 .3 a (7 + .22 . 1 1 1 3.12 22 B22 D23 < ^22 i-22 ^23 23 3. B33 L 22 1 + Z3 q — 33 + 31 q —33 ^32 q ± 32 q .22 23 a .1 Constraints on the equivalent single-loop structures An .

1.29 12121 PllP Hi'22 . Thus. and based upon the location of "noisy sensors" might require a trade-off in the phase margin frequency specification for each loop. (6. Notes: templates. (6. 6-3 HIGH FREQUENCY RANGE ANALYSIS The analysis in the h. (6. in Eqs.18) In a similar manner.20) 1. for L2. (6. 6. the designer selects the value of c\f above which the crosscoupling effects have no effect on system performance. range.82). fromEq.19) Remember that these equations apply in the h. Based upon the knowledge of the control system to be designed. Eq.17b) is almost the same as Du of Eq. (5. utilizing the J plants. by subject to Eq. range where the lower bounds are neglected.16). Note that Eq.81): qu where A = in K- (6. (6. The results of this section are summarized in Table 6. when designed to yield the desired response has the value \fu{jco)\ » 0..5. obtain: 922 921 922 921 (6. for i =j = I . tri.18) by Eq. can be based upon the fact that the filter /J.41) and (5.19). (6. for the case of / =j.192 Chapter 6 Equation (6.17a) is denoted as Aa. as shown hi Fig. (2) the A ratio can be anything. (6. (5. is "dropped" to zero and only the upper bound bn is used eliminating the need to determine b'.17b) as Ba. are obtained of the ratios of the q's in Eqs. from Eq.f.. consider the bounds shown in Fig. Multiplying Eq. 6.18) and (6.15) except that ic is not known and must be suitably chosen together with ay.42) can be neglected For a 2x2 system.f. the lower bound a..5. Thus. Thus.19) yields: (6. Note: (1) t21 still needs to be specified even when/7J = 0. (5.

the diagonal dominance condition.22) Remember that Pll P21 P= Pl2 Pl = P22 P22 ' Pl2 detP P2i Pu and .Single Loop Equivalents 193 1. For the 2x2 plant.f.) cut-off fora.0 CO llf Fig.21) PnP 22 which must be true over the entire range of P e ff.5 High-frequency (h. Let _ Pl2 P21 PtlP22 (6. 6. for ca -> oo (say w >o}h) is given by: PnP 21 <l (6.

/)(/ + L2) .24) yields die characteristic equation: L2)-y]2 = 0 (6.2 = 0. \rithfjj = Ofori #j. 6-4 STABILITY ANALYSIS A 2x2 plant is used to discuss the stability analysis for a MIMO system. resulting in an unstable system. q's. especially for n. (5.p.| are both » \yn\ and at h.25) At "low" frequencies |/.6 shows the polar plot of (1 + L. i. Since the diagonal dominance condition must be satisfied. For a LTI plant. the "locus of critical points" is the -1 +jO point. . for this 2x2 plant the polar plot in Fig. Figure 6.e.. in Eq./ into the equation for t. The Nyquist stability criterion1 should be applied utilizing a polar plot analysis. it is the zeros of die q's .194 Chapter 6 P22 P22 (6. as co —> x> |pw/>22| » \pl2p21\ then the magnitude \ya\ better be < 1 or the yn locus will be encircled.3 can not encircle the origin.)(l + L2) for all P <= ^(shaded area) and the locus of critical points represented by y12.55). (6.6 can not encircle the y12 locus for P e fPor (1 + /. not die zeros of die/>'s which determine if die system is m..f. whose characteristic equation is G(s)H(s) + 1 0. Thus.p.23) dciP Thus.\ = \L2\ « 0 -> 1 . yields: Thus. the denominator of Eq. with no parameter uncertainty. By substituting the equation for /./| and |L.y. \L.y.m. 6.

7.Single Loop Equivalents 195 -270° -180° -90° Fig.2).6 Polar plots of Eq. Consider the following two examples. (a) Consider Suppose.. i. 5-6. dominates L2 . 6-5 EQUILIBRIUM AND TRADE-OFFS There are interactions between the rows of the SFG in Fig.7. 5. 6.e. and let the <j>'s represent the actual values of the outputs (see Sec.e. i.25).: — $12 "dominates" in the first row of Fig. (6. dominates LI and — $21 "dominates" in the second row. Example 6. 6.1 Trade-off Example #1 The SFG for a 2x2 plant is shown in Fig.11 via the specifications by. For this example assume \til\mm = hi. 6. for a fixed value of ».

Thus.2 Trade-off Example #2 For Column 1.L is used here in doing a "trade-off1 on the specifications for one or more by in order to hopefully achieve the de- sired dominance through the analysis of: q22 bn b2i bn bsi <hi I I <h31 (6. it is possible to ease the design by being concerned with one or more bj/s. For example: Column 1 Column 2 c\2 - „ Row 1 C2] = -b!2 C22 = ———— ROW 2 If the 1.26) i Q33 1 bn . 5. Example 6. (b) Assume The terms that dominate in each row are in different columns. from the SFG of Fig.1 and 2. as done in Example 6. The same process. 3 12 si C21~ 132.2 terms dominate then it is possible to ease the design by decreasing b2i and bl2.196 Chapter 6 then if the value of the specification b32 is lowered it helps $n which decreases the demand on LI.11 for a 3x3 plant: Oil )21 .

t2. because now \y2!\ < 8instead of 0.\ and \c3!\ in Fig.Single Loop Equivalents 197 Column 1 Column 2 2. or y32 (or y33Yy3i will impose the same burden. 6. Any further decrease of b21 involves "trade-off. (Decrease of b2J is permissible because the BNIC \t. suppose b2t is decreased from 0. . For example. This does not matter ify22 dominates L2. 5../q12 Row 1 Row 2O—-» $22 Fig.11. But it makes it harder on L2. assuming that yn andy31 dominate Lj and L3. To accomplish this may require a "trade-off1 in one or more of the system's performance specifications. If so.. b21 can be decreased until say ylt imposes the same burden on L2 as does yn. which eases the burden on L. it is conceivable that before y2i~y22 occurs. a sacrifice of L2 for the sake of LI and/or L3.) This decreases \c. DEFINITION Design equilibrium exists when it is impossible to reduce the burden on any Lh without increasing it on some other L. There is a bewildering multitude of possibilities and options. As illustrated by these examples.3I\ < b2l is required.1. Is there a simple rule for determining when 'free' easing of burdens is no longer possible. denoted by yji'y^.1 to 8.. However. either y12 (or VnYyn.e. and only "trade-off' is available? This question is partly answered by the following definition and theorem. it may be possible to modify one or more by until the dominant terms lie in the same column." i.7 Equilibrium and trade-offs for a 2x2 plant. and L3 for satisfyingyu e rtl andy31 e T3. respectively.

TRADE-OFF After equilibrium has been reached. .2 then it is also necessary to reduce one or both of b22.. -> <x>. In serious situations.2. 4 for the proof... as bv -> atj.f ~yik~. it is still possible and may be desirable to do 'trade-off . the sacrifice of Z. the designer can then consider modifying the (%£>#) tolerances.for example..that is... See Ref..e. 6. but it is highly desirable to reduce L. say column / dominates all L. Also.g.j is small at low co (see Fig.8..e. for one or more i. Assuming thatyu =y. This variety offers the designer very useful design freedom. The trade-off situation is radically different in the high frequency range. Clearly. according to the circumstances .y3i..k. It is also emphasized that all the above discussion is at a fixed cot value. This is another of the valuable insights offered by the design technique.\ but making the specifications harder ony2i. e.. because bv > asj (in the interacting yv). 5. and the design technique offers this kind of flexibility. a number of equilibria can exist. i.11 one or both of b2!.5). as any column may be chosen. At low co values. Thus. ya. and the difference between bv and a. i.b31 may be reduced thereby decreasing \c. so it is important to reduce \Lx(j:a>)\ rapidly compared with (o.198 Chapter 6 THEOREM A necessary and sufficient condition for design equilibrium is when all the L.. This does not precludey. SR = bti -a. It may be desirable and it is certainly possible to have different columns dominating at different cof values. i = l. sacrifice one or more Lio for others. The above analysis led to the low and medium frequency bounds on Lio shown in Fig. the amount of such improvement is quite limited. In Fig. if the sensor on y. 6. The potential profit to be gained by exploiting column dominance is greater at high frequency because the ^ tolerances are greater in this range of co. for the sake of very small improvement of some other L. is noisier than the others. He may wish to economize on some Lx. or q^ has elastic modes in a rather low co range.m dominate the Lh withy any fixed column. is greater than in the "low frequency" range.b32. in the situations noted above. are dominated by the members of the same column.

Single Loop Equivalents



Fig. 6.8 Bounds in the Nichols Chart on Lio(ja>\ i=l,2,3, and L, designed.


At large co the a^aj) of the interacting tv of Figs.6. la and b become so small that they can be made zero. Hence. the^4ff in Table 6.1 can be ignored and each TCIJ = b,/. so all the constraints in the table are now of the D type, i.e., replace the B0 by Du. There is now no limit on the permissible reduction of the bif and the problems of 'equilibrium' and 'trade-off become so simplified, as to permit analytical derivation of the bounds Bi(ja>) on Lio(ja>). First the claim that the 6,y are optimally chosen so that D,,~Dij,..., Dmi~DmJ, for all ij is easily proven. Suppose, (with no loss of generality), that D,, dominates Lt. Reduce bt2 (recall it has replaced fcn) until Da~Dn.
This certainly doesn't hurt L/ and helps Da, i # 1. Hence, if any Da dominates L, this reduction of ba helps L,. If not, a single Da does so, then there is at least no harm in this reduction of bn. Similarly, reduce bn, until Aj-Aa-A/- Using exactly the same argument, one can deduce that D23~D22~D21, etc., for each row. The above result applies even if the |?//i/«|max are different in the elements of row L etc. However, if they are the same, as they are in this case then let v,y A Uax which appear in row / of Table 6.1 The above give

I2 b22 + V13 12 b23 + V13


Chapter 6

Consider the m equations










which constitute a linear homogeneous set in the m bk!. For a solution to exist the

determinant of the coefficient matrix must be zero. This matrix, denoted by (f> has v,, for its off-diagonal and A., as its diagonal elements. Exactly the same matrix results by taking any a value in the set of m equations

= lk .

k = l,2,...,m


The choice of the Ak is up to the designer but he or she can choose only m - 1 of them. The condition del fi = 0, determines the last 2. In this way. the designer can deliberately sacrifice some loops in order to help others. For any fixedy, the above set of m equations in bkj (k - /,..., m) is homogeneous, so the ratios by/by ( k = 2,... ,m) are determined by the choice of the /lk. Even if the bkj Ibij emerge very large, one can always make the b,/ small enough so that the b^ satisfy the tolerances. Hence, it is not necessary to solve the equations for the b^ /b,,-. Of course, this is true only because all a^ = 0. If the above approach is used at lower co, where some or all akj * 0, the bkj /by should be calculated, in order to be certain that the tolerances are not violated.
Example 6.3 Trade-off Example #3 Equations (5.39) - (5.56) for the 2x2 plant, where fu = 0 (i *j). in the high frequency range simplify to:

Single Loop Equivalents


Column 1 Rowl

Column 2


<b,, (a)

t 12 t


12 *n

1 +L


c a


c a


~ I +L


?7 (d)

Substitute into these equations the expressions for Cy's from Example 6.1 to

obtain the following set of equations: Column 1 Rowl

Column 2

-b22qnlqi2 l + L,


Row 2




-b 12922/I. 21


The equations in Eq. (6.32) are rearranged as follows, in order to perform a frequency domain analysis:

Column 1

Column 2

In -b2, \l+L\>^^- (a) 9!2 bn -b,j b2i

9n -b22



Row 2



9 22



Chapter 6

Suppose that





then the 7,7 term |Eq. (6.33a)J dominates over the 1,2 term [Eq. (6.33b)j and the 2,2 term [Eq. (6.33d)] dominates over the 2,1 term [Eq. (6.33c)]. Thus a diagonal dominance exists. The objective is to balance the dominance across all columns. For example, in this case. b23 is increased until both terms match, i.e..
= AI = A,

i \+Li\>

for the 1,1 and 1,2 terms of Eq. (6.33) and

\\+L \>

for the 2,1 and 2,2 terms of Eq. (6.33). Thus the dominance is now "balanced" across all columns for this 2x2 plant. Note that hi the high frequency range the problems of equilibrium and tradeoff become simplified. This simplification permits an analytical derivation of the bounds B,(jca) on Lio(jw).

For a 3x3 system the bounds Ufa to) on L3o{j(a) tend to be significantly larger (more stringent) than those on L,0, L2o in the low and middle & range, where the trade-off opportunities are limited because the interacting a,/ ^ 0. In the higher co range, when Eqs. (6.27) and (6.28) apply, there is much greater scope for such trade-off. This was utilized to help L3 in the higher a> range by setting A, = A2 = A, and solving del <P = 0 for A3. As the vv are functions of co, det fi = 0 gives a relation between the A/ which is also a function of a>. The result is shown in Fig. 6.9 for co > 100, for which the v,,(/aj) are fairly constant with a>. Given 17 + L,-f! <Af and /,, = L^q/q^ one can find

Single Loop Equivalents





1.2 dB


Fig. 6.9 Optimum relation between Ay = A2 and A3 for <o > 100.

the bounds on Lio. which of course depend on the set Q = {qu}. The results are shown in Fig. 6.10, where: Fig. 6.10a corresponds to A/ = A2 = 1. A5 = 2.3 dB; Fig.

6. lOb corresponds to A, = A? = ft 7, A5 = 2.7 dB; and Fig.6. lOc corresponds to A, = A? = 0.5, A3 ^ 3 dB. The q^a)) are almost constant with a> for «>, > 100. so these bounds apply for all o>, > 700. As expected, the larger A becomes, the easier it is on L,0L2o and the harder it is on L3o. The big difference in the bounds in Fig. 6.10 is due to the small A compared with the large A5 that were deliberately used. For this reason the actual L3o(fco) was able to reach its final asymptotic slope at co ~ 250 (see Fig. 6.11) compared with ~ 450.230 for Llg, L2o even though \L3o\ is much more than \Lla[. \L2o\ at low and medium w. The Lio chosen to satisfy the bounds are shown in Fig. 6.11. Design simulation results were highly satisfactory and are shown in Fig. 6.12 for extreme plant parameter combinations.

Example 6. 4 2x2 Plant

Consider the 1,1 MISO equivalent of Fig. 5.11,

O. Thus:


Chapter 6


Fig. 6.10 The resulting bounds onL/ 0 for various A values for a> > 100.


Fig. 6.11 Bode plots of Lio(ja>).

Single Loop Equivalents



-t2i In

The 1,1 cross-coupling expression for the bound determination becomes:


This expression is rearranged to the following format:

\i+L, \>
where \T2i\ = b2i ,

tn en



Next, consider the 2,1 MISO equivalent of Fig. 5.11. with f21 = 0. For this example, since
(21 ~ tc2l

Then the upper bound for tc

is ba, i.e.,

Tc2J = 02

This expression is rearranged to the following format:
11+1-2 |>


where the maximum magnitude bn, from Fig. 6. Ic is used for tn.

Equations (6.37) and (6.38) are inverted, respectively, to yield the following equations:









Chapter 6

Fig. 6.12 Simulation results for representative step responses.

By letting 4 = 1/L, and 4 = 1/L2,, these equations become

1+ f

< Mm,


1 +1-

< Mn,7



which are now of the mathematical format required in order to use the NC. Note that:

Single Loop Equivalents


Thus, based on Eqs. (6.40) and (6.41). the rotated or inverse NC must be used in conjunction with the templates to determine the cross-coupling effect bounds BCl(jco,) and BC2(fa>,) for // and (2. respectively (see Chap. 3). These bounds are transposed to the regular NC to become bounds for L, and L2, respectively.
Example 6.5 Boundary Determination for Example 6.4

Consider for Eqs. (6.39*) and (6.40*) that:

= Lm



= - 36 dB - Lm ^77 < Lm Mn
For this example, BC7 is determined in the high frequency range where the templates

are essentially a straight line. Assume for this example that q2i is independent of q22 (not true in flight control). The graphical technique for determining this boundary is illustrated in Fig. 6. 13 as follows:
Trial 1: From Eq. (6.42) First trial

Actual value

value for q22



Lm q22h = -36 dB - (-18 dB) = -18 dB * Lm q21li = +6 dB
Lm q22l = -36 dB - ( -/ dB) = -35 dB * Lm q,2l = -20 dB

Trial 2:



Lm q22li = -36 dB - (-42 dB) = +6 dB = Lm q22li = +6 dB
Lm qni = -36 dB - (-16 dB) = -20 dB = Lm qn, = -20 dB


Chapter 6

Locus of:

Fig. 6.13 Cross-coupling bound determination for LmL2 on the in verse Nichols Chart.

In Trial 2 the trial and actual values agree resulting in obtaining a point on Bcr In a similar manner other points for this boundary can be obtained. This procedure is automated in the MIMO QFT CAD package of App. A. Usually the dominating case(s) on a grid line are quickly found Thus, the optimal L2o must lie on or below the boundary'
-B C2 (j«?i)vs. Z£

Single Loop Equivalents


on the inverted NC where the values

for various values of -Zfa = ^L2, are plotted on the regular NC to obtain the boundary

then L2o must lie above this boundary. For Prob. 6.1, in trying to satisfy the requirements for both Lt and L-,. Eqs. (6.37) and (6.38), respectively, assume that the specifications are met for L, and not for L2. Thus, for this situation reducing bu allows a reduction in b2i proportionally to maintain the requirement of Eq. (6.37) and in turn satisfy Eq. (6.38) for a satisfactory L3 since for this example it is stipulated that the q's are independent. When the q's are not independent then it is necessary to obtain the templates

in order to determine the bounds. Example 6.6 Cross-coupling Boundary Determination for a 3x3 Plant

Consider the 3x3MMO control system of Fig. 5.11. From Eq. (5.28) obtain


1 111




1 22 t32

(23 t33_ -

+" S-, 02

t21 _t31


1 132


J 21 82

J 1281
J 22 82

J is8l


_J 31 83

J 32 83

J 33 8 3_


Chapter 6

The first row of Eq. (6.43) yields:

fill! til =
1 + /./


= /„

= /„ + tcil


fnLi l + Li


= trl

= trl + tcil


In a similar manner the expressions for the 1,2 and 1,3 expressions for row one of Fig. 5. 1 1 are obtained. Given that \tu\inax = by and assume, for \t,,\ma, = bn, that <f>,, (the actual control ratio) is satisfied. Thus, for this situation it might be difficult to satisfy fa, $33* etc. If this is the case, then the designer needs to do a "trade-off1 in trying to achieve the best performance possible. Performing the design in the high frequency range simplifies the task as discussed in Example 6.5. The equations for the determination of the cross-coupling bounds are:

For the first row of Fig. 5.11 :
1 In \






For the second row:





For the third row:



Single Loop Equivalents


L 3 |>

bn + 3,





Consider the interactions between the rows of Fig. 5.11 via the b,, specifications. For example, suppose b2! is decreased from 0.1 to a lvalue. The decrease is acceptable because for BMC %,\ < b2, is required This decrease in turn decreases \cu\ and \csl\ which eases the burden on LI and L3 for satisfying yn e 3n and y3! e 3u. This is based upon the feet that yn stndy3I dominate L{ andL3, respectively. Thus, the design for L2 is more difficult because ]y21 \ <£instead of O.I. This may not matter if y^ dominates L2. If so, b2! can be decreased until, for example, y21 imposes the same burden on L2 as does y23 (y^-^j). Any further decrease in b2! may involve a "tradeoff," i.e., requiring a sacrifice on the specifications on L2 for the sake of L, and/or L3. However, it is conceivable that before y2i~y22 occurs, either y12(or y13)~yji or y^or y33)~y3i. There are other trade-offs that may be possible to investigate. In summary, for the h.f. range, the main factors to keep in mind in trying to achieve column dominancy are: 1. Adjust the by's until one column dominates. 2. Once Item 1 is accomplished, try to make the nondominant elements in each row equal the dominant ones by reducing the b^'s in each denominator. This will make all three columns equal and allows one to use any column as the constraining column. This reduces the 3x3 matrix of constraints to a single constraint on each loop, i.e.,

3. From Sec. 5-7, with respect to Condition 2 (e.g. a 2x2 plant) if
12V PuP 21
PllP22 <1

is not true then renumber the output terminals. For example:

Thus, for the original 2x2 plant:
Ul .U2. ?21 Ul + P22 U2


Chapter 6

For the renumbered system:
P2lUl +
_Pl]tfl + P12U2
P2I P'll P22 III

P'l2 \ll2_

which results in Condition 2 being satisfied, i.e.,
P22PU >1


4. Items 1-3 apply to an mxm plant as well.

Up to now it is assumed that all pu elements of P have the same number of excess poles («) over zeros (w), that is, A, = (n - w) = A for all /plants. Thus, in taking

the Umiting angle for all plants is /l(-90°) and the templates become a straight line in the limit. In the event that A, is not the same for all LTI plant cases then in taking


the h'miting angle /L,(-9(f) for each LTI plant case do not all have the same value. Therefore, the template as a> -> <x> will not be a straight line.

The specifics of QFT Design Method 1 are presented in this chapter. In addition, system sensitivity, stability, and performance trade-off analyses are discussed in detail. The case of plants having RHP zeros and/or poles are also discussed in this chapter. It should be noted that if each ptt of P does not have the same value of excess of poles over zeros then as a> -> °o the templates may not be straight lines. The reader is urged, in performing a QFT design, to constantly have in mind the material given in Chap. 9.



The design technique of Chap. 6 inherently involves some overdesign, as seen from Eq. (6.1), in which t21, t31 can be any members of their acceptable sets T21> T31 and qi2,qi3 any member of their uncertainty sets. As noted in Sec. 6-2, it is therefore necessary to use the worst case values which leads to overdesign. Actually, in the real world, there is a correlation between the t2i,t31 and the qu,qi3, etc. For example, in Eq. (5.55) it is possible that q12 is large when t21 is large in the expression for cu. Such correlation a can only help make cu smaller. Thus, for Method 1, it is not possible to use this correlation, and so one must take the largest t21, the smallest q12, etc. This is the price paid for converting the MDVIO problem into the much simpler MISO problems, and avoiding having to work with the horrendous
denominator in Eq. (5.25).

Another disadvantage of Method 1 is that there emerges a certain inequality (see Sec. 5-7, the diagonal dominance condition) which must be satisfied by the plant elements; e.g., for m = 2, as o -> a? it is:




or vice versa. Rosenbrock's dominance condition is tougher. It requires this inequality to be satisfied over the entire frequency range, not just as s -> oo . His method can therefore not be used for the 3x3 example of Chap. 6. The equivalent of Eq. (7.1) for m = 3, as co -> oo. is

39) yield: + a > ^ = g.3) The objective is to choose fn(s). equations are used that are independent of the t.214 Chapter 7 Pl23 = (7. A) contains both design methods. (7.} = [I/ qji\ (e) (7.j(/a>)\.2.f22. (7. Method 1 is used when the diagonal dominance condition can be satisfied and when the BW constraints can not be satisfied by Method 2 (this BW constraint is discussed in a later section). qti (b) (c) t2J e T2J (d) P1 = [p].. the resulting equations are exact. in the 2x2 system. Vp e <?and Vt2j e T2J appearing in c^.n by simply finding t2j from T= [I + PG\1PGF.t.2). The final step. and involves less over design than Method 1. These are precisely the MISO design problems of Sec.. P e &. Method 2 is used when the diagonal dominance condition can not be satisfied and/or where overdesign needs to be minimized The MIMO QFT CAD package (App.2 or alternatively hi. Instead. When t. 1 and for an impulse input. The necessary constraints for Method 2 are discussed in Sec.36) through (5. 7. f. Hence. In summary. (7. and gi(s) such that the yij(s) has no RHP poles. 7-6 and in App.3). Thus. Thus. for Fig.t22 (as in Fig 5. if indeed^ E rtj for all t2J (of c. 5-6 and 5-7 the design equations are again Eq.]) e T2j. 1 1 for m = 2). it was necessary to seek a modification of the technique that succeeds in avoiding the condition of Eq. In Sees.3) with 2 replacing I and the c2v containing t!v.. 5-6 through 5-8. is to choose f2i. Eqs. Thus. This modification lead to the development of Method 220.j is substituted for ytj in Eq.f!2(s). (7. andg2 to ensure that the t2v(s) have no RHP poles and that t2v e T2v and P e &. then the design objectives for tv have been achieved by \hefu. (5. andgi but only if actually t2j e T2J. and satisfy the tolerances on \t. D. 7-2 DESIGN EQUATIONS FOR THE 2x2 SYSTEM In Method 2 the same design equations as before are used for t.2) was not satisfied by the FY-16CV 3x3 lateral system for any p e ff. . here.2) It turned out that Eq.

andf12 are known (Use Lt not L. 7.2 MSO structure for % (j=l. andfzs so that in Fig. 7.4). 5. D).2 the outputs^.3a-d). (6. (7. 7.17a) has been completed by means of Eqs. (7.4) P = [Pj.0 — see App. r=o——> •<\ >2 ^°—^—py2j -1 Fig. 7. and>^ are stable and satisfy the tolerances on \y2J\ and l^l. This design is done after the design of Eq.} (e) or the MISO structures of Fig. These are single-loop problems similar to Fig. (7. It is then necessary to find g2. so that Lt.1 MSO structure for ttj (j'=J.4a-e) are ./2i. Eq. t2J = ——— —— (a) L3. except that only the uncertainty P e ^need be considered. as the c2j in Eqs. respectively.2. (7.fn. = g2 q22e = e.(d) (0 (7.2).2\ Eq.1.3). o- ->———-C>———>———*>- -1 Fig.MEMO System Design Method 2 215 r.

Even if this requirement .s. design execution is that of a MISO single-loop system — which is what makes this design procedure so tractable. if the t2j appearing in Cij are in T2J (which they are).3a-d).. although the idea and approach are motivated by the fixed point method in Method 1. the design equations for t21 and t22 are of the same form as Eqs. Note: for nonsquare plants the use of a weighting matrix W = {w. (7. see Constraint 1 in Sec. Now Fig. 7. L2. There. with the c2J functions of ttj. The equations forj.The order in which the loops are to be designated and designed. respectively. so fixed point theory is required to justify Method 1.2. The theoretical justification of the above design procedure is as follows: 1. 2. etc.216 Chapter? not functions of the elements of rv. If Eq. (7. 7.. Thus. andy12 correspond precisely to those for tu and t12. 3. (even if the specifications for ta and tj. (73a-d).e.i = qn. ) 75 as Loop I -> q.p. L.. in order to achieve arbitrarily small (a. 4.1. by proper choice ofthejSvand&forthegiven/^and gt. is based upon the following: (a) Choose the loop that satisfies detP q =———— bem. are not satisfied). i. (AdjP). (7. 7-8.j} is required in order to achieve an effective square plant matrix Pe = PW since a square plant is required for a QFT design (see Sec. 1.5) is not satisfied. then an adjustment of the "weighting" factors wv may result in satisfying this requirement.1 has been designed [via Eqs. which they are in Eqs. (7. ( ..3a-d)] so Hhatyu andya are stable and satisfy the specifications on tn and t12 Vp e <?. 7-10).) sensitivity20. no fixed point theory is needed to rigorously justify this design procedure. Thus. At each step. Designation of the order of loop shaping .4a-b) are exactly the expressions for t21 and t22. 7-3 DESIGN GUIDELINES The following items are intended to provide the reader with a heuristic insight to the design process. The design specifications are satisfied for t2v of Fig. the design is satisfactory for t21 and t22 because Eqs.

7-8. This prevents a. But such overdesign does not exist in Fig. E). 4. if qn is m. respectively. 7-4 REDUCED OVERDESIGN18 Figures 7. See Constraint 2 in Sec. then q22e. for the second row of the 2x2 system.e.s. 1 and 7.. The uncertainties in t21 e T21 and fe e T22 are assumed independent of P e & in Figs. 3. 1 1. sensitivity if with gain k (a ± uncertainty value) is independent of the signs of zv and/or pu. 7. 5. (b) YJI is "large. In Fig. Numerator or denominator factors of the form (1 + Ts). If this condition is not satisfied then it is impossible to achieve a.s.11 (for 2x2 system)7 in which t21 and t22 appear in the cross-coupling cn and c2j . 9 and App. tolerances for co > coh. 2. 7. For the 2x2 plant if (a) Y is "small" then q22e = q22. must be excluded. (c) When applying Method 2.m. in general. then it is possible to achieve a. (b) When applying Method 2.7. will all be m.2. tolerances.s. This may be acceptable but may lead to a wide B W for L2. Transmission zeros ~ If there are transmission zeros in the RHP it only indicates that qv may be n. where T has a ± uncertainty value and is independent of other parameters. 7. 1. choose the loop that has the toughest or most stringent specifications as Loop 1.p. \ YJt = 1 then q22e can be unstable. where 7 is the desired phase margin angle.p. or the del P may have zeros in the RHP only.2 because c21 and c22 are not functions of the elements of any TV.p. This correlation is not being exploited.If 3a>h s for all co > coh so that the width of3qn(jaj) does not exceed 1 8(f .MEMO System Design Method 2 217 is not achievable a satisfactory design may still be possible depending upon the application (see Chap. 1 because in reality there is a correlation between the t21 and t22 and the q. etc.p. There is inherent overdesign in Fig.2 are the same as Fig." i. the design . if more than one loop can be m. 5.j ofP1 .1 and 7. Template width .

by Method 1 involved overdesign precisely as in the first row of this figure and in Fig. that is: Li = g. Note that y12 is defined by Eq. in Eqs.2.1. as detailed in Ref. they are otherwise identical hi form. = 2. Finally. Thus. LI and the three fv are chosen so thatj^/.7).8) L2e P21P13 . for ah1 P e ff. (7.7). but the t. (7. or from the third row of Fig. (7.j are replaced by the. (7. 5-6 and 5-7.5c) and for all P e £P.y/. 7. of course be reversed withy^ and y22 using Eq. Although the forms for c2J and L2. 5.2. are stable and satisfy the tolerances on ttj for all /$ e % appearing in ctj in Eq. (7.3 = '-+C* l + L2e (a) /^=-^^. of course. so the design techniques for both are basically the same.3) with 1 replaced by 2 and with tn and tI2 using Eqs. giving forj = 1. (7.218 Chapter 7 for t2l and t^. (7. of Eq.g/qlk ( .11 for / =3 and eliminating tu and t2t by means of Eqs. (7.7) and (7.8) the^ and L2.P23 -fijP*2i (e) again resulting in MISO problems. 5. (7. are chosen so that^ are stable for all t3J e T3j appearing in c2J and. 20. In Eqs. ) 77 l+Li giving MISO problems. LI and the^ are known from the designs of Eqs. The equations for^ are obtained from the second row of Fig. //' 12 1 — ————— (b) P'3l=^ (0 L2e=g2q22e ' (d) (7. MISO design equations for t3j are obtained by finding t3J from T = [I + PG\!PGF.12 in which tv and t3j appear.4) by exchanging numbers 1. be the same as in Sees. the resulting equations are: . (7.4c) wherey = / and.7). The above procedure can.8) are different from those in Eqs.8).3.qir cij = -^t.2. 7-5 3x3 DESIGN EQUATIONS tyijj = 7.

Finally.7a-c) because the tv e % (k = 2. they are chosen so that the t3j are stable and satisfy the tolerances T3J.8a-e) guarantees that the t2J tolerances are satisfied. 7-6 EXAMPLE . the design based on Eqs. design execution is that of MISO single-loop systems. for which the technique of Chaps. respectively. (c).2. In Eqs. providing the t3J appearing in c2j are e T3J which is the case here.9). (7. providing that the t2J and ty. (5.3x3 SYSTEM DESIGN EQUATIONS Consider the design of a 3x3 control system for which rt(t) * 0 and r2(f) = r}(t) = 0 =fa = 0). These equations constitute single-loop uncertainty problems. (II). p*32(i + LJ) (j) Note: SeeEq.7a-c) are used for the second or third channel.terms of Eq. Hence. (7. (7-9) 2 = _ C3j . etc. (7. aadfy are known. ? L3e= ? -A (b). in Eqs. Since Lh L2. This example entails four parts: part (1) — the set-up of the .———————————————————— _ _ TM . (7. Suppose the nine^ and three g.7a-c) guarantees that the ty tolerances are satisfied. the only unknowns in the Eq. there is some over design in Eqs. the order can be changed and equations of the form of Eqs.9a-j) are such that the t3J are stable and their tolerances are satisfied (which is so by definition here).fij.93) for the expressions for the 75. (7. Now the design based on Eqs. that at each step.3 andj = 1. (7. i. the % tolerances are also satisfied. 1 + L3e (a). Of course.r i 2 3 (l + ^) +/ / J ( l + I 2 s )-( / / 2 «).e. 3 and 4 apply. (7. (7. appearing in cv are e T2J and T3J. In the above.)(l + L 2 e ). while there is no such overdesign in Eqs. again. Cg).. (78a:/) there is less overdesign because only the ty appears.P*31 (1 + ^2e ) (0. The justification of the above design approach is the same as for the 2x2 case.3) appear as crosscoupling effects uncorrelated to the plant uncertainty. which has been established.MIMO System Design Method 2 219 f*L*. *32 . 3= 2+il3 3) (e).9) are the fij and gj. = (l + L. Note. L3=g3q33 (d).9a-j).

Part (1) The pertinent equations are (refer to Figs. ——— C33133 Part (2) Utilizing Method 1. .220 Chapter 7 pertinent equations. . 5.10) in order to yield t\\sfu and g1 that satisfy the design specifications.11. 5.10) __________ \(l21 VziJ t2l= ————————————————————— 1 + Z-2 .the design approach. the bjs are substituted into Eq. (7. l + L3 .11 and 5.13): Rowl I tl3~————— W I+LI _ cisQu 1+Li Row 2 (7. In gen- . and parts (2)-(4) .12) + L3 (a). (3X C22<l22 t22 = —————— 1 + ^2 (0 Row 3 (7. It is assumed that the loop to be designed first is row 1 of Fig.

If the design specifications cannot be achieved with a diagonal prefilter then it will be necessary to utilize a nondiagonal prefilter matrix.3.2 and Eqs.b) of Ref. giving the designer useful flexibility. Note: one may initially start the design with a diagonal prefilter F matrix in order to simplify the design process. etc. Part (3) Substitute tn obtained in Part (2) into Eqs. with w = 2. 7.7) type equations are used for the first two rows of Fig. Apply Method 2 to design^/ andg5 [see Eq. it is clear that a larger variety of mixtures is possible.b) with u = 1. B. design^. (7.12) having only two unknowns:^/ andg3. The theoretical justification is now as follows: the design for channel 3 is correct by definition then the fixed point theory. Thus..b) of Ref. [see Eq. Part (4) Substitute tn and t21 obtained in Parts (2) and (3) into Eq.9)]. for the 3x3 system. have been chosen so that these mappings map rff into themselves. 7-7 mxm SYSTEM: m > 3 The procedure for generating the design equations for mxm MIMO systems with m > 3. that is.2).12). 5. These equations can be derived or obtained from Ref. A and B.12). in terms of only the parameters. For the next chosen channel (say the second). in which all the ttj and t2j are eliminated by means of Eqs.8)]. the design starts with Eqs. and L2e so that the y2J's are stable for all t3J e T3J in c2J and for all P e &. (V. with u = 3. (7. The process continues until the end. but eliminate all the tv and %by means of Eqs.. The nine^ and the three g. This method was used for the 3x3 FY16-CCVlateral design modes. so a fixed point exists. One uses for any channel (say the first) design equations in which all the tu (i * 1) appear as cross-coupling effects. Apply Method 2 to design^. (7. (4a. that hopefully meets the specifications. Eqs. the ba's are not involved in the design off31 and g3. precisely as in Ref. andg. as in (Ref. start again with Eqs. Eq. 7). A Denote the resulting design equations as Eqs. (4a. in order to be able to exploit this flexibility to its fullest extent. For example. 7. .30 The two sets of design equations for t.9a-j) for channel 3. (7.'s are stable and satisfy the specifications on ttj for all 4. Denote these as Eqs. the designer must understand MISO design theory used in the design execution. and the theoretical justification is the same as given previously for m = 2. it is necessary that Lj andfv be designed so that the y. (7. 7. 7. (7. For the next chosen channel (say the third). Mixtures of the first and second techniques may also be used. etc. should be clear from the preceding sections.11) and (7. This is the concept of Method 2 which results in Eq.11 for both channels 1. is used to justify the designs for channels (1. A.MIMO System Design Method 2 221 eral. it is now possible to obtain t31. which reveals the cost of feedback and the available trade-offs among the loops. However.j and t2J are taken as the mappings on the acceptable sets TV and the third set of mappings is simply t}J e T3J. In general. For larger m. e % in ctj and for all P e fP. (4a.

it is required thatyn e TU andya e T!2 for all P e ff. Since P ranges over £P. 7. 7-8 CONDITIONS FOR EXISTENCE OF A SOLUTION This section considers the conditions required for the applicability of the QFT design technique. sensitivity also achieves attenuation over a.e. sensitivity of the tij over arbitrarily large (a. for example.222 Chapter 7 There are additional important advantages in Method 2 since the diagonal dominance conditions given by Eqs (7. sensitivity which is defined as the BW achievement of a.s. BW (i. In the general m case. if qn satisfies certain constraints. 1. BW.3).l. a RHP pole of (Adj P)lk is cancelled by an identical zero of p* and does not appear in the other terms of det P. t21 e T21 and t22 e T22. sensitivity") for tn. yn is: tkl (AdjP)!k/(AdjP) 11 1 + 1 detP (AdjP) n J (7. "a. This problem has also been studied in an abstract setting by Zames and Bensousan. irreducible conditions applicable for LTI compensators in general.1. for example.qn can be made a.l.s.3. Such as.s. of course.1) and (7. it is seen that a.2) are no longer necessary.s. over a. The reader is referred to Ref. and compares the two sets of conditions. Such cases IK are excluded. it considers the inherent. There may be exceptional cases when in det P. These have been detailed in App. 20 for details. tolerances over a. SENSITIVITY" IN SINGLE-LOOP DESIGN In Eqs.3 is the extended logarithmic complex plane (NC). normally cancelled by similar ones in the denominator.) sensitivity"] is that det P has no RHP zeros.l. (7.63 In Fig. 7-8.L B W of external disturbances acting on the plant. Also. RHP zeros of (Adj P)u are. 7 so are only qualitatively described here by means of Fig.1 CONDITIONS FOR "a. Instead the principal condition to be satisfied [to achieve "arbitrarily small (as. Indeed this is at least theoretically possible.) BW. 7. since (Adj P)ik is a term in the expansion of det P.13) RHP poles of (Adj P)lk are normally cancelled by similar poles in det P. This is done for as. the cross-coupling component in. the set {Lt = giQttyqud'co)} is not a single complex number (at any fixed . 1 of Ref. Figure 7. ti2 are achievable if Lt = g.

lying in any of the 3P[Lj(j(o)]. 7-3).. ZL^co) must be negative.RflMO System Design Method 2 223 co) in the NC but a region. Stability over the range of 9> requires that 3P[Li(jco)] move downward in between the vertical lines Vi and V2 without the points . The benefits which do arise may nevertheless suffice for the specific system being designed (see Refs. In a design with significant plant uncertainty 3p[£. This is so over the important co range of tu and ta (their bandwidths generally). Unstable qn are included in the above discussion and do not require separate treatment.p then q22e. If this constraint is not satisfied for Method 2 the theoretically attainable benefits of feedback are limited. are stated here for the case of "a. but two important constraints (also see Sec. above the zero dB line as shown in Fig. in order for \Li(jco}\ to decrease.14) Tij = ttj (jco) = set of acceptable ty (joo) (7. that in order to maintain 3p[Z. (Obviously. 02.e. 27. Note.0. For those co for which ^[Lfid)] is so located.. because L.p. and 29). on the average. in order to achieve the desired sensitivity reduction.. if qtl is m.(jco)}. never be a part of any 3P[L..02. sensitivity": Constraint 1 qn = det P/(Adj..p.(/w)] must lie relatively high up. transcendental compensation can be used for special countable cases. the area of3p[quQ'a>)].3 for co2. for m <3. It follows from the above that the manageable uncertainties depend on the assigned tv tolerances. the template of L. Such large values for 3p[Li(ja>)] can be maintained theoretically for any finite co range. with phase margin angle y. = giqn..2y) degrees phase width for 3P\qn(jco)]. P)n must be m. are also m. the uncertainty in the magnitude and phase o>iqn(jc6) (i. . As co increases and the 3P\Li(jco)\ descend lower on the chart below the zero dB line. O.(/<»)] above the zero dB line. In practice only the (780° ..l.j (jco). however.s. denoted as 3p[L(jca)}. but it is essential that the points ... This appears to allow (36(f. . etc.. Eventually LI(JOJ) must decrease and -* 0 as co -> oo. can be a.02. However.15) must be modified to permit such zeros of tv. any zeros of qn(jco) on lias jco axis must be known and finite in number in order for gj(s) to be assigned poles there.) If the range of such zeros on they® axis is uncertain.. Note: if qn is m.. 26. VPe£P (7. 7.y) phase width is tolerable in this range. which is the same as 3p[qn(ja))] but translated vertically by 20log\g1(jco)} dB and horizontally by </gi(joJ) degrees.p. clearly their width may increase again.. then the specifications av (jco) < tij (jco) | < b.

tolerances for a> >cah. with the uncertainty in T including a sign change which is independent of other parameters. 7.3 Templates of LI(JCO) on logarithmic complex plane (Nichols Chart). then it is impossible to achieve a.. for Method 2. sensitivity" if with the k uncertainty including a sign change which is independent of the signs of the zv and pu.s. This prevents "a. is to obtain: .3 for all co > co. A good way to apply this constraint.s. the width of 3p[qn(ja>)] exceeds (18(f -y). Also excluded is a fector (1 + TS) in the numerator or denominator of qu. Constraint 2 Suppose -?<Bk.224 Chapter 7 Fig.. y being a desired phase margin angle.

(1 + LI) and det Pare already required.4) are used for the second channel. condition of Constraint 1 therefore applies to q22(l + Lj). and then Eq. "a. As for Constraint 2.p.j]. For m = 2: Pll ?22 ' Pl2 P21 _^ kn k22 -kuk. P e &. used in the design technique of Sees.16) = PI kn^ Let K = [k. = det P/(Adj P)ti of the first channel i. The above. for which Eqs.p.s. for example. (7. This ratio must not change sign over the range of uncertainty. sensitivity" can be achieved by L. by the usual LTI design techniques. Equations (7. sensitivity" design. 7-8.e. Constraint 2 gives kltkn > k12k21 VP e 9>. To remove the ambiguity it is also assumed that the plant terminals are numbered so that for at least one P e £P. to achieve significant sensitivity reduction. or vice versa. applies to channel no. (7. BW « that of L2 [denoted by BW(Z. qa must be m. in more abstract form.3) are used. The m. i. If this condition is not satisfied then it is impossible. VP e ^(RHP poles are tolerable).p. so Constraint 2 yields kiik22>ki2k2i VPeff (7. 7-2 and 7-5.17) This is a diagonal dominance condition as s —> oo -which applies only to the first channel being designed when applying Method 2. So from Constraint 1.4b) implies Constraint 2 applies top22 of P.y -> ktjs^ as s -> <x>. 7-8. That is: U(S .. Zanies and Bensousan63 have defined a diagonal condition as s -> <x>.MIMO System Design Method 2 225 detP detp0 where P0 represents the nominal plant matrix. Constraint 2 states that there may be no change in the sign of det K/kj as P ranges over ff. 1. for "a. knk22 > k!2k2].. most of which is not new because m.2 APPLICATIONS OF SEC. This automatically takes care of all loops.1 TO DESIGN METHOD 220^ Constraints 1 and 2 therefore apply to the q.) « BW(L2)]. In this chapter it is assumed that all £ff > 0 for all P e £P.21 (7. Suppose P = Ipij] has each/?.s.

(7. L2 and LI over a. 7-8. When BW is a specification it is important that one template for each loop be obtained at their respective loop B W specification. sensitivity" is achieved (as it may be) by BW(L.The latter also attenuates c22. (7. and (Yl2 1^2 + Yl3 H3) It has not been ascertained whether these two sets of constraints are identical. sensitivity" of t22 be achieved despite large uncertainty in F? Clearly by large L2e.) « BW(L2) « BW(L3). i. and since a. (7..226 Chapter 7 For the condition assumed with Eqs. Application to the second. 17). with no sign changes in the k. Suppose it has them. has RHP zeros? For if not. (7. Constraints for m > 3 may be similarly developed. it follows that L.4) with . Suppose (1 + Lt) has RHP zeros.3). Hence. This same principle applies to all ta. If "a.3a). and proven by Eq.3a) are thereby unaffected. and is basically the same as that derived from examination of Eq.4b) and (7. . Large L2. L. so the constraints apply to qtl and q22. (7.8). 7-8. However. by simply asking whether a stable tu is possible if I + L..l. (7.3a) and its analog for t2J (by interchanging 1. = 7. is achieved by large g2q22 because large Lt (needed likewise for small tu sensitivity) gives L2. 7-8. Eqs. Since F is outside the feedback loops.8b)].s. sensitivity. How can "a. the result is that the constraints of Sec. the usual feedback method. 16) and (7. but t21 is affected [see Eqs.s. 7-8. (andl^) must satisfy the constraints. 7/2). BW in order to achieve "a. so neither should the zeros of the numerator of Eq. The zeros of 1 + gjqn in Eq.j. which is not done here.2). BW is needed. system stability is unaffected. for i = 1.e. which may not be small because of gi in its numerator.4a) for t22. gives the same results as for the m = 2 case. unless in Eq.l. (7. the results are the same.3 INHERENT CONSTRAINTS20 It is important to determine whether the constraints in Sec. 7-8. because L3e has the same form in both cases [compare to Eqs.3a). for . It may also be so in the general case but this would require consideration of simultaneous sign changes among the kv. The termfugjqn is unaffected. 1 apply to q22 and qnl This is indeed so.3 shows that the constraints of Sec. For this purpose examine Eq. there is need for a. For m = 3 the application of the constraints to the first channel makes it applicable to qn. the hypothesis (/ + Lj) has RHP zeros is untenable. and there is a small change mf21. 7-8. over an a. (7.l. (7.1 apply to qn.2 are due to the specific design technique or are inherent in the problem itself. These are RHP poles of//. 1 must always apply to each qlt i'• = 1 to m. (7." But do the constraints of Sec. Sec. = 1].l. = g2^22.s. the numerator of tn has these same zeros.

How is this systematically done in the case of significant P uncertainty? For m = 2.s. 7-9 NONDIAGONAL G The constraints on Pin Sec.3). P or sensor noise problems) may be achievable via H. H may be very helpful in reducing the amount of feedback needed to achieve specified tolerance sets given by Eqs. by making V = PH quasi-diagonal. Eq. sensitivity. (7. To the knowledge of the authors of Ref.p. Thus. 7.4a). (7. similarly Eqs. as s -> oo. one subdivides the plant set into subsets ZPj which are correlated with the subsets % of Ty.3) by maximum use of the correlation which exists between the ^uncertainty and the t2j e TZ in Eqs. The constraint det Pis m. 7 this approach has not as yet been attempted in any numerical problem. are now applied to these pairs 0>u. r.15).7) through (7. even if "a. 7. let H be a fixed LTI pre-compensator matrix inserted ahead of the plant and let V = PHbe the new effective plant in the set u = {PH. (5. VP e fPis not eased at all. P e 3}. However.p. and obviously cancellation of RHP zeros of det Pby det H cannot be done for many reasons. then it is necessary that the constraint violated by ff. because it applies to the sign of det V not changing.40) through (5. sensitivity" is not attempted In design Method 2 the constraints are in effect only for "a. Equations (7. so their reduction via H is desirable.8)." Hence. but is solvable by Method 2. The design techniques with diagonal G. That is. For example in Eq. 7-4. are now applied to set u instead of set 9. because of n.m. it is clear that the best method is achieved with Eqs. A suggestion for this purpose has been given in Ref. (7.ju separately for each u. The other important constraint involving diagonal dominance as s -> co. (7. also appear in the design technique of Ref. so that a design unachievable by diagonal G (say. L2e must handle the uncertainties due to L2e itself and attenuate the effective cross-coupling set {c2j}. sensitivity" are not eased by a nondiagonal G. This is the case in Ref. Are these constraints eased if a nondiagonal G is used? To answer this question. Off-diagonal plant elements appear in all the design Eqs. (7.46). even though P has large nondiagonal components.4). It may be possible to considerably reduce \Cy\max by means of H. because det V= (det P)(det H). For basically noninteracting tolerances on t^ (k * i). s.14) and (7.52 However. the constraints on £?for "a. they are always present there.7) and (7.9) in the 'cross-coupling' components. (7. If H is helpful in overcoming some constraint. for a given plant set ff.MEMO System Design Method 2 227 From Sec.s. and Eqs. 7-8 are deduced on the assumption that G is diagonal. It is worth noting that the constraints of the diagonal dominance type as s -> <z>.3). is also not eased. 7.ftj is made zero. it is possible that a specific synthesis problem with given % ^"sets may not be solvable by Method 1. so only the latter need exist. (7. is not violated by u. let the normalized H have 1 for its diago- .

2 = /4 h21 = u.j. one should check the effect on the resulting sets of vn = pn + vp12. 7-10 ACHIEVABILITY OF A m. because of the requirements on the loop transmissions due to their uncertainties.. giving H = Pla'JA. EFFECTIVE PLANT det Pe For some control systems (e..e. utilizes the Perron root interaction measure for decoupling plants (control authority allocation) in multivariable QFT robust control system design. P = P. because in practice H =Pa~'A with Pa' *P' exactly.2 and 4 for discussion relevant to this topic. one tries to diagonalize P/ by means ofPjH = A.l(s -p). i. by a ixm -weighting matrix W.j] or a frequency sensitivity matrix W = [w.p. Instead. a gain matrix W = [w. (See Ref. giving PH = PPa'A with RHP dipoles.V(](D).|v2y| at each a>. The use of the off-diagonal components in multivariable QFT can ease the diagonal controller design problem.e. The difficulty is how to design the off-diagonal elements. The expression for the output y(s) may be obtained by two different approaches: Case A From the differential equations describing the system of Fig.4. especially if one must consider engineering factors such as cost-benefit trade-offs of using cross-feeds. i. v22 = np2l + p22. Then v. 7.y2J in Eqs.3) and (7. system integrity and plant input signal levels. The objective is to minimize over ^max|vw|. are clearly very helpful in choosing H(J(O). Durban. If the constraint detP(s) be m. 7. The reader is referred to the literature for more details.p. The final choice depends on the relative importance of the two terms in the numerators ofy. strong structure in the plant uncertainty. 3.p. The inputs and outputs of an aircraft may be selected in such a manner as to yield the mxm aircraft plant matrix P(s).4).j(/a>)] as shown in Fig.4 obtain .. (7.2 = npn + pn and v2/ =p2t + u/?».228 Chapter 7 nal elements and h. Boje has developed some nondiagonal G decoupling design approaches81 based on the use of the Perron root interaction measure. 7. South Africa. then one should not try to diagonalize P by means of PH = A diagonal. flight control) there are often more control inputs (f) than outputs (m) available. and Boje of Natal.) If the elements of P have a RHP pole in common.g. However. det P. Sees. Sketches of the sets in the complex plane.(s) for an effective mxm plant Pe(s) by augmenting the basic mxl plant P/. is not satisfied then it may be possible to achieve a m.

ne(s) must not have any RHP zeros in order for Eq. V! .18) is manipulated to yield y(s) = D1 (S)N(s)u(S) = Pt(s)u(s) (7. de(s) = detD(s). and satisfy the constraint that det Pe(s) be m. Equation (7. 7. (7.23) to be m.18) D(s)y(s) = N(s)u(s) where D(s) and N(s) are mxm matrices of polynomials and the output y(s) and input u(s) are mxl vectors. Thus.p. (7. —^Y! Pb : ' >Y Fig.21) det D(s) det Pe (s) = del N(s) (7.20) Thus from Eq.22) detPe(s) = ne(s) d.20) the following expressions are obtained (7. .4 An mxm effective plant P&s).MEMO System Design Method 2 229 (7.19) where the effective plant matrix is given by (7.23) where ne(s) = detN(s).p. w j ve P. andne(s) and de(s) are polynomials.(s) (7.

Assume that this p. clearly det[Pt(s)W\ =p.19) and (7. by Pb and Wh. 7.29) Let the submatrices of Pb and W.. (7. (7. d(s) = dij(s). respectively.j{s) are considered to have a common denominator.j(s)ld(s\ W= [w6].(s) is m.p. (7.f n 30) where . of Eq..26) From Eqs. n.(s). The number of terms in the summation of Eq. Choose W so that the mxm submatrix formed by the rows indexed by >..25) y(s) = Ph(sWu(s) (7.j(s) = n.. az. + • • • + det ft.{s) and w. i. p.. and wv is a gain to be determined in order to try to achieve a m..{s). The detPe(s) obtained from Eq.p. ] (7. be represented.26) it is seen that the effective plant matrix can also be expressed by Pe(s) = Pb(s)W (7.4..24) (7.27) may be expressed as follows: det Pe (s) = det [ft (s)!T] = [p.2 (s) det Wj. A sufficient condition for det [Pb(s)W] to be m.28) is det Pe (s) = det H. in general thep/.p. respectively. det P. where w = 1. (s)w. Then.2 +••• + det Pbw (s) det Wb.. /„.. A short proof of this fact is as follows: let pt(s) be the determinant of the mxm submatrix of Pt(s) formed by the columns indexed by /.28) is given by (7. /„.28). .j(s) and d(s) are polynomials.. 2. is that at least one pt(s) be m... (7.e. is the identity matrix while all remaining rows are zero. (s) det Wb. elements of Pb(s) can have different denominators By use of the Binet-Cauchy formula" Eq..p. m < i.28) where p. (7. where v(s) is a Ixl vector and W is only a gain matrix the following expressions are obtained: y(s) = R(s)v(s) v(s) = Wu(s) (7..230 Chapter 7 Case B From the block diagram of Fig. (s) det Wbl + det Pt. are the determinants of appropriate mxm submatrices of Pt(s) and W(s). Note that although in this section all the p..27) where Pb(s) = \pifs)].

37) in Eq. \ k. (7. | det | wijt | = det | nv (s).d(s) and (7. where k. = det wijl is a scalar.36) Based upon Eqs. = —— det| nv (s)w \ det | m d (7. (7. 7.p.36) and let .23) and (7.30) can be expressed as det Pbn (s) det Wb.36) e?"~'(s) must be a factor ofn^(s)w.MEMO System Design Method 2 231 d(s) and (7.34) det | n.35) d.34) represent the same open-loop system of Fig. (7.30) to be written as follows: 1 det Pe (s) = —— Z(det | nv (s)i \ det | wyi \ + m d (s) (7. Since Eq.p. The following development illustrates that a sufficient condition for the existence of a m.(s) which results in de(s) .31) (7.4 then ne(s) (7.32) Thus each term in Eq. Assume det | ntj (s). (7. (7.34) through (7.33) which permits Eq. det Pe(s) is that at least one det Pt(s) must be m. (7. (7.z where «gis a polynomial. Factor out dm~'(s) from every term in Eq.p.y (5)2 | det | wiJ2 • • + det | «ijaz det | Wy.34) is m.

41) is manipulated to the mathematical format of Ni(s) _N2(s)_ = -1 (7.p.38). Equation (7.p.40).p. (7. (7.-.(s) V'*9) „ + det In/. 'I = kl Nl (5) '^ (7. for all Pb e P and be expressed by Eq.p.42) which permits a root-locus analysis of Eq. |det|w ftz |] where k2 is a scalar.. The values of w. (7. submatrices to be identified as Eq. This assumes that throughout the region of plant parameter uncertainty.39). the initially chosen m. (7. submatrix in Eq. (7.38) and (7. (7. then the zeros of the polynomial (7. Determine the number Oy submatrices of P*w of Eq. k2.41) lie in the LHP for all Pb e P.j of W^ associated with . Eq. Pe(s).232 Chapter 7 kl det ——'I «.38) Also.41) must all be in the LHP. let the summation of the remaining terms.38). (7. 2.. To enhance the achievability of m. Select one of the m.(».29) can be expressed as follows: det Pe (s) = —— [kiNi (s) + k2N2 (s)] (7. (7.40) In order for det Pe(s) to be m. Note: kj.34) is m.. det Pe(s) the following guidelines may be used: 1.34) that are rap.p. (7. and A^s) are now functions of w.which are to be selected in order to try to achieve a m. be expressed as k2N2 (s) = ——!-— [det | nt] (s)2 |det | wiJ2 | + • • • m d .42) are in the LHP then the weighting factors wff are selected in hopes that all roots of Eq. (5). Since the zeros of Eq. 3.p. after factoring out dm~'(s) from each term. Based upon Eqs.

e. terms of Eq. This problem has been studied by several researchers and the resulting techniques may be used for this purpose. in order to simphfy the root-locus computational effort. Hence (7.p. (7. non m.42).1 m. (7. By changing the values of the gains w.44) where the d. dv(s).p.p.38) and (7... be altered in such a manner as to increase the values of their corresponding det W^. it is necessary to factor out d(s) from the numerator polynomials of Eqs.39).p. perfect factoring may not exist. Thus. . Now consider the most general case. ww(s) and hw(s) are scalar polynomials. Once the n^s) and wv are specified. as possible.43) and det W. or failing that has them as relatively "far-off". = det (7.j in the manner described will result in the Oy . (b) The 01-0.p. it may enhance the achievability of a m. Pe(s) by the analysis of Eq.45) For each value of w. This dominance of the ay-l terms in Ni(s) may result in N2(s) being m.16^ Thus. for the general case some or all of the nj(s)ldj(s) may have RHP zeros and/or poles. the Binet-Cauchy formula permits the determination if an mp. Pbtt.j(s)/dj(s) has a range of uncertainty which may be correlated with that of the remaining a.j{s) and h^s) are not all the same and where «„(*). (7. n.MIMO System Design Method 2 233 (a) The remaining Oy -1 m. i. (7. The problem now becomes one of trying to choose fixed ww(s) and hw(s) polynomials so that the det Pe(s) has no RHP zeros over the entire range.39) to dominate in the resulting expression for N2(s). effective plant detPe(s) is achievable over the region of plant uncertainty.1 submatrices of P(s). Depending upon the CAD package that is utilized in determining d(s) and the numerator polynomials of these equations. Pbw be altered in such a manner as to decrease the value of their corresponding det W^. Also.

every LTI compensation technique has these constraints. sensitivity.2 through 7.22 The reader is urged.e.e. to constantly refer to Chap. the technique in Sees. (b) The MEV1O uncertainty problem is rigorously converted into a number of MISO uncertainty problems. (f) These techniques are applicable to the design of a reconfigurable aircraft with surface failures. Also. with the following features: (a) There is detailed control over the nf individual system transfer functions. They are present in the new techniques only for a. give constraints on the plant which are inherent and irreducible. (c) For "arbitrary small sensitivity" over arbitrary large bandwidth (BW).s. (e) The overdesign inherent in the fixed point techniques.s.7 has been reduced. 7.7 i.. (d) Part of the constraints (at infinite s) in (c) were always present in the previously developed MISO equivalent technique. sensitivity" was not required.234 Chapter 7 7-11 SUMMARY This chapter has presented synthesis techniques for highly uncertain mxm MDVIO LTI feedback systems with output feedback. Solutions of the latter are guaranteed to be satisfactory for the former. 9 and to Apps.. D and E. . fixed point theory is not required for justification of the new technique.5. i. but some overdesign is still present. even if "a. Relatively simple MISO single loop feedback techniques can be used to solve the MISO problems. in performing a QFT design.

ff{s) = {/%$•)} and ^C?) = {Pa(s)} are sets of matrices due to plant and disturbance uncertainties respectively. the first portion develops the necessary equations that take into account the MIMO system's external disturbance inputs. the following equations are written s) + Pd(s)dext(S) u(s) = G(s)v(s) v(s) = r(s)-y(s) ( . This chapter considers the analysis and design of a MIMO external disturbance rejection control system. From Fig. P(s).8 MIMO SYSTEM WITH EXTERNAL DISTURBANCE INPUTS12 8-1 INTRODUCTION Previous chapters have dealt with MIMO tracking control systems with no external disturbances being applied to the plant. and P^s) are mxm matrices. the design procedures can be applied to other MIMO external disturbance and tracking/external disturbance problems.1.12 Although this chapter addresses the specific aerial refueling problem. The following development quantifies external uncertain disturbances. 8-2 MIMO QFT WITH EXTERNAL (INPUT) DISTURBANCE Output disturbance rejection is the primary design criterion in this chapter. The objective is to find a suitable mapping that permits the analysis and synthesis of a MIMO control system by a set of equivalent MISO control systems.1 represents an mxm MIMO closedloop system in which G(s). Therefore. The remaining portion applies this development to a real world application problem. ) 81 235 . 8. Figure 8. Previous discussions of MIMO QFT did not consider external input disturbance in the calculation of cross-coupling rejection bounds.

the system control ratio relating d. Of (8.2) where henceforth the (s) is dropped in the continuing development.1 QFT Compensator with output external disturbance.7) . ) 83 which yields (8. 8.6) by [/ + PG] yields (8. (8.2) From Eqs.^ to y is Td = [l + PG]"1Prf ( . (8. v = -y u = -Gy ( . ) 86 Premultiply Eq.5) Based upon unit impulse disturbance inputs for <C. where for the regulator case with zero tracking input r(/) = [0.236 Chapter 8 r ext u I_________I Fig.1) and (8.4) is rearranged to yield: (8.4) Equation (8.0.

(8. The P1 matrix is partitioned as follows: (8. P' = [p*] = [IfqJ.12) where A is the diagonal part o£P~' andB is the balance of P1.9) P* ml Pm2 r\* ••• Pmo n* The m2 effective plant transfer functions are formed as = J_ detP (8.8) Pn = Pn "• '•• PL PSn P*21 P*22 (8.10) Pv the Q matrix is then formed as 111 1l2 '" Qlm 12m Q 321 <\22 ••• (8.11) with G diagonal. b.j = P*t for / *j../ = l/qa .11) Qml 1m2 '" 1mm where P = \pt). Substituting Eq.13) Rearranging Eq. (8.13) produces .MEMO System with External Disturbances 237 Premultiplying both sides of Eq. (8.12) into Eq. Thus /l. (8. and Q = [qj = [!//»*]. and b& = l/q.7) by P' results in [p-1 + G\Td=P -}Pd Let (8. = 0.p*t. results in (8.

e. 14) yields: Y(Td) = BPd .2.+ l+ (8.15) for the 3x3 case.2 3x3 MISO equivalent loops for external output disturbance. Fig..16) 8i< 111 Vl2 1l3 V 412 Equation (8.14). 8. Td e 3d.238 Chapters (8. then this Td is a solution of Eq. 8. (8.7 The theorem defines a mapping Y(Ta) where each member of Td is from the acceptable set 3d (see Fig. (8.2 shows the effective MISO loops resulting from a 3x3 system. Figure 8. for a unit impulse input. 5.15) This equation defines the desired fixed point mapping. (8. . for a unit impulse external disturbance inputs.1) element on the right side of Eq. Eq. Since A and G in Eq.10). (8. i.14) Rearranging. where each of the m2 matrix elements on the right side of Eq.16) corresponds precisely to the first structure in Fig. the (1. provides the output -**.BTd] (8.14) are diagonal. If this mapping has a fixed point. (8. Proof of the fact that the design of each MISO system yields a satisfactory MIMO design is based on the Schauder fixed point theorem.14) are interpreted as MISO problems.

15) have the form ya = wa (deiey. For this development..15). for each feedback loop of Eq. the interaction term.17) and for the 1-2 loop case. = q. From Eq. (8.21) 111 112 .18) k=\ ik . and the corresponding outputs^. each of the nine structures in this figure corresponds to one of the elements of Y(Td) of Eq. Additional equations. which is the output of loop 1 due to disturbance input 2.18). The control ratios for the external disturbance inputs d^.20) 1 + 1^2 111 Substituting in for td22 yields: "/2 i r 1+ Lj (8. yields.19) c«= where (O* represents the external disturbance effects and ^represents the crosscoupling effects. for unit impulse inputs. (8./(l + gg. i. the following control ratio: (8. Eq..MEMO System with External Disturbances 239 Similarly.17) x = number of disturbance inputs (8. the equations for the case of a 2x2 MDVIO system are presented.: (dext)ij= Z =i qik (8. including the cross-coupling terms from loop 2. (8. quantifying both the external disturbance (d^a and the internal cross-coupling effects Cy. where w.e. These equations are used to define the disturbance bounds for subsequent loops based on the completed design of the first loop. (8.) and (8. are derived to utilize the improved method (Method 2) QFT design technique. not only contains the cross-coupling interaction but also the external disturbances.X ' k*i m = dimension of square MIMO system Thus.

J2 (8.22) (8. (8.23) 112 Substituting in for ^ and rearranging yields qn)- .27) .25) where 7I2= Cl. (8.240 Chapter 8 d2 _ 9n ' l + L.-.flj(l2fll2- Solving for tdl2 yields L2)(p ^ q.26) -r12Pdi2+r12tdl2 (8.24) (l+L1)(\+L2)q12q2l (8.

MEMO System with External Disturbances 241 / rf =—————————————————— r " n+/ " ' * • (8. (8. in general.29) the effective plant is defined as: Substituting Eq.2 2 T * a P . representing the cross-coupling and the external disturbance effects. (8. are calculated at a given frequency to satisfy .28) Equation (8.32) The interaction bounds (the optimal bounds for a pure regulator control system). for the 2x2 case. (8. the improved method control ratio of the J* interaction input to the i* system output is: 1] dij qk(\+Lk) _^v——k>_——— where/ = 1.29)..30) into Eq. (8.2and/fc*/ = 1 .29) From Eq.28) is rearranged as follows: tda = ——2 2 . (8. yields: _q12(l+L2) " Thus.

The order in which loops are designed is important as pointed out in Chap. Then the design is continued through the remaining loop by use of Method 2. but some orders produce less overdesign (less bandwidth) than others..242 Chapter 8 where/= 1. the designer must then "tune" the interaction term specification accordingly. The intent is to present to the reader a design problem. At this point it is important to point out that when the interaction term specification is considered. the first loop is then redesigned using Method 2. 8-3 AN EXTERNAL DISTURBANCE PROBLEM The remaining portion of this chapter applies the QFT design technique to a realworld problem. The last loop designed has the least amount of overdesign. Since each loop may not exhibit the same interaction characteristics. 7. In other words. the interaction term specification would consider external disturbance effects only. . = 1. interaction term specification tuning provides flexibility in the QFT design process. Any order can be used. the designer must decide how much is to be allocated for the cross-coupling effects and how much for the external disturbance effects. But if the loop interaction term is a mix of cross-coupling and external disturbance.2 and**/ (8. As a second iteration.2 and fc * / (8. For example.33) or where . from the onset of the specification of the control problem to the verification of the design results by means of linear and nonlinear simulations. therefore the most constrained loop is done first by Method 1. if one loop is only affected by external disturbance.34) The improved QFT method uses these equations to reduce the overdesign inherent in the original design process. the designer can "tune" the external disturbance rejection specification depending on the nature of the interaction term for a particular loop. with as much detail as space permits.

MEMO System with External Disturbances 243 8-3.gravity and changes in the moments of inertia /„ and /„. Piloting a large. the receiver aircraft will change position relative to the tanker. By using MLVIO QFT. thus reducing the pilot workload and fatigue factor.. The pilot must maintain a very precise position relative to the tanker. An AFCS must be designed to regulate the receiver's position. and possibly impact the pilot's capability to perform his/her mission. an AFCS is designed that operates throughout the range of the changing aircraft dynamics and rejects disturbances including those at the output. This could endanger the flight crew. . He/she maintains position visually. The pilot must pay close attention and take corrective action to maintain position. This AFCS is designed to precisely regulate position relative to the tanker by applying the MMO QFT disturbance external rejection design method to address the rejection of the disturbances entering the system at the output. and extending recovery time between missions. the pilot must contend with maintaining position in the presence of wind gusts. One way to ease the pilot workload is to implement an automatic flight control system (AFCS) for air-to-air refueling. high inertia aircraft during air-to-air refueling requires intense concentration. movements in center-of. The pilot must compensate for changes in aircraft dynamics due to taking on fuel. The AFCS needs to be able to maintain a precise position of the receiving aircraft (receiver) relative to the tanker in the presence of such disturbances as wind gusts. Refueling these aircraft during flight provides unlimited range of operation for this fleet of aircraft. Hence. automatic control of the receiving aircraft during aerial refueling operations is most beneficial.3 ASSUMPTIONS The following assumptions are made: • Only the desired outputs are of interest for final performance.1 AERIAL REFUELING BACKGROUND The United States Air Force (USAF) maintains a fleet of large cargo/transport aircraft. 8-3. Besides dynamic changes. Since these aircraft can take on large amounts of fuel. applying the appropriate control inputs when changes in position occur. specifically. and the pilot's fatigue level increases and can reach an unsafe level. Cargo/transport aircraft are generally large and have high moments of inertia. long flights and multiple air-to-air refueling can seriously strain and fatigue the pilot.000 pounds. Excessive changes in position will disconnect the refueling boom from the receiver. air-to-air refueling can take up to 30 minutes. and in the changes of mass and moments of inertia. However. decreasing flight safety.2 PROBLEM STATEMENT During air-to-air refueling. up to 250. Compound this over long flights and multiple refueling.25 8-3.

8-3. The AFCS controls the receiver and is independent of the tanker in as much as the tanker is used as the point of reference.244 Chapters • Position of the receiver aircraft relative to the tanker during air-to-air refueling can be accurately measured • The CAD packages used. . The first assumption is required in applying MTMO QFT. The third assumption is concerned with the limits of CAD packages and their numerical robustness. for the QFT design process using a published document65 containing C-135 cargo aircraft stability derivatives tables and plots. • Implement a Mach-hold. and z outputs in order to keep the receiver aircraft in a volume specified as the area of boom operation. A full six-degree-of-freedom nonlinear simulation is performed in EASY5x. (4) to evaluate the new control law.65 Models are developed for disturbances due to wind gusts and received fuel. and (5) to validate the MIMO QFT design with disturbances at the output. altitude-hold. y. and z) of a receiver aircraft relative to a tanker. The second assumption is required because no sensors are currently in place to measure the position of the receiver relative to the tanker. wing leveler autopilot that operates for all aircraft models. (3) to simulate the design for linear and nonlinear performance on MATRIXx. wing-leveler autopilot. and nonlinear performance on EASY5x.4 DESIGN OBJECTIVES The design objectives are: (1) to utilize the aircraft models.6 METHODOLOGY The design approach requires six steps: • Generate Linear time-invariant (LTI) state-space models of the aircraft for different weights and center of gravity. 8-3. altitude-hold. The AFCS is designed to reject disturbances at the x. The control system is simulated for linear performance in MATRIXX. y. JVflMO/QFT. The MIMO/QFT CAD package (App. (2) to present a design for multi-channel control laws using MTMO QFT for several flight conditions with special emphasis on aircraft center-of-gravity and weight changes. QFT compensators control the reference signal of the autopilot to maintain "formation" during air-to-air refueling. developed in Ref. 8-3. EASYSx. The MIMO QFT plant is the bare-aircraft model augmented by a typical Mach-hold. MATRIX* and Mathematica are adequate for the design process. A) was utilized to achieve this design. 12.5 SCOPE The MIMO QFT external disturbance rejection technique is applied to the design of an AFCS regulator for the automatic maintenance of the three-dimensional separation (x.

• Simulate the design on MATRIXX and EASY5x to validate the AFCS design. MATRIXX is used to design the autopilot using root-locus design techniques. generated by EASYSx. 8-8 which presents the linear and nonlinear simulations and illustrate the "goodness of the design. and desired input/outputs. The user need only provide A/C stability derivatives. TheMach 0.3. the C-135B will have a CL between 0. 0. The 6 DOF state-space models. flight conditions. during air-to-air refueling. The AFCS QFT design is presented in Sec. 8-3.7 OVERVIEW OF THE AERIAL REFUELING DESIGN PROBLEM Section 84 discusses how the external disturbances are incorporated into the 6 degree of freedom (DOF) aircraft equations. and represent weights ranging from empty/low fuel to loaded/full fuel A/C.® Therefore. EASY5x is a computer aided design (CAD) tool written by Boeing Computer Services used to model. The autopilot is designed to control all 16 plant cases. simulate. 8-7 followed by Sec.MIMO System with External Disturbances 245 • Model the disturbance due to wind gusts and refueling.45.6.23 Wind gusts and fuel transfer disturbance models are developed as well as the AFCS concept. CL = 0.69 at 28. and analyze dynamic systems. altitude hold. are loaded into MATRTXX.65 A Mach-hold. for 8 different A/C weights. An autopilot is used for two reasons: 1. Autopilots reduce the high frequency cutoff of the A/C 2.1 C-135B MODELING EASYSx is used to develop the state-space 6 DOF bare (uncontrolled) A/C model.500 feet flight condition is considered These discrete values are selected based on the availability of data. The MIMO QFT mathematical expressions are reformulated based upon these modified aircraft equations.2. Typically. 8-5 and 8-6. This A/C is chosen because of the availability of the aerodynamic data. 8^. normal refueling speed and altitude. 8. the 16 plant models envelop the structured uncertainty of the C-135B during air-to-air refueling. All A/C have autopilots . Sixteen bare A/C plants are developed to account for the uncertainty of the C135B during air-to-air refueling. The bare aircraft and autopilot are shown in Fig.27 and 0. The 16 models are based on two different coefficients of lift. • Design the AFCS using QFT. The air-to-air refueling AFCS concept is discussed in Sees." 8-4 AIR-TO-AIR REFUELING FCS DESIGN CONCEPT The aircraft (A/C) modeled in this chapter is the cargo variant of the C-135 class aircraft (C-135B). and wing-leveler autopilot is included in the C-135B model.

2 DISTURBANCE MODELING Disturbance models are generated by developing augmented state-space models of the A/C in the presence of wind gusts and foel transfer inputs. aileron. Since autopilots are available. based upon external disturbance inputs represented by the vector d. and y-position in a local inertial reference frame where x is positive out the nose of the A/C. The three outputs frame of reference is translated from the A/C center of gravity (eg) to the approximate location of the air-to-air refueling receptacle on the top of the A/C. The Mach hold command input is used to control the x position.246 Chapter 8 Fig. The outputs are zposition (altitude). reducing cost and overhead The inputs to the autopilot are thrust. considers three disturbance components: pitch plane wind induced dis- .3 C-135B bare aircraft with autopilot. x-position. Lowering the cutoff frequency of the A/C reduces high frequency parameter uncertainty which in turn reduces the size of the QFT templates. and rudder commands. Mach and altitude are self evident. using it in the QFT design eliminates duplication of a control system to provide input to the bare A/C. By using the rudder for the QFT controller. 12. altitude hold controls altitude. and altitude is positive up. good performance is obtained while leaving the aileron controller to handle wing leveling. 3 of Ref. rudder is chosen over aileron because the rudder does not roll the A/C. y out the right wing. 8-4. elevator. 8. The development in Chap. and the rudder command is used to control the y position.

(8.36) sx = Ax+Bu+ftd (8. 8.A] ~l ffd (8. 8-2 is applicable for this problem. as the disturbances are applied to the bare A/C. Total disturbance modeled is *d = t pitchd pitch + * latdlat + *• rfdrf (8-35) where rpiteh and /7/are additional inputs that identify pitch plane flight behavior. and refueling disturbance Frf. ..MDVIO System with External Disturbances 247 turbance /^.1. Thus. In the same manner.A]A Bu + [si . lateral channel wind induced disturbance /I.39) is represented in Fig.A] ABu + C\sl . The state-space equation now takes on the form x = Ax + Bu + ffd = to ~ ~ (8 36) ' Note. they enter into the inner most loop. P(s) = P^s) for a square plant matrix P(s).40) and where the plant model PF is partitioned into the two matrices P(s) and Pd(s). If P(s) is not a square matrix then a weighting matrix W(s) must be used to yield Pe(s) = P(s)W.A]'1 f!d (8. then from Eq. identifies the lateral plane flight behavior. Equation (8. the QFT loops are closed 8-5 PLANT AND DISTURBANCE MATRICES Based on zero initial conditions.38) y = Cx = C\sl . around which the autopilots. and the matrix P^s) models the transmission from the external disturbance inputs to the output of PF.37) x = [sl. and later. the input /I.39) where (8. the QFT formulation of Sec.

(6) nominal boom length is 477.5ft up or down. (3) it can move as much as 70° up and down from normal position and maintain its connection to the receiver.41) where . the relationship in Fig. These perturbations are viewed as disturbances by the receiver. The control problem's goal is to minimize the perturbations to be within a specified volume of space where the refueling boom can operate. (4) horizontal movement is limited to Iff" left and right while maintaining fuel flow.4 can be used to develop the equations required to define the regulation control problem. In order to maintain connection. altitude.5ft left or right.8ft). R is the nominal boom length measured from the boom hinge point on the tanker. (8) it can expand or constrict as much as 73. and velocity while the receiver compensates for these changes and maintains relative position.248 Chapter 8 8-6 CONTROL PROBLEM APPROACH The tanker's position is assumed fixed and hence the receiver aircraft's position is measured from this frame of reference. In this approach the control problem can be viewed as a formation flying problem. Using the tanker as the point of reference. X is the horizontal distance between these same points. unmodeled. The C-135 tanker refueling boom has the following operational constraints: (1) nominal boom operation position is 30° down from horizontal. (5) the disconnect limit horizontally is 15° left and right. in order to maintain fuel flow. Y measures the distance between the center line of the boom hinge point and receiver receptacle. Normal boom operating position and length defines this volume. The perturbations are caused by wind gusts and disturbances due to refueling.65 These dimensions provide a maximum perturbation from nominal boom position of approximately 2. (7) it can expand or constrict 13.85 ft up or down. The receiver maintains the total obligation of regulating its position. (2) the boom can move as much as 4° up and down from normal position and continue delivering fuel. and 11.5 inches (39.5 inches and maintain refueling. the maximum perturbation can be 7. Equations are developed that identify perturbations from the set position. Z is the vertical distance between the boom hinge point and receiver aircraft's refueling receptacle. Other. The tanker is free to change course. disturbances may include the tanker changing course. 8. The following equations are derived R2 = X2 + Y2 + Z2 (8. but cannot deliver fuel. 7ft left or right.5 inches and maintain contact but not refueling.

Eq.45) dt dt dt dt . 8.41) and squaring the terms.4 Control problem geometry.. X . (8. Substituting Eq. yields (8. z « R. (8. x. (8.42) into Eq. y.43) is approximated as +2zZ (8.43) Since r. (8. Y.42) Y = Y+y Z=Z+z which are the sums of the nominal positions (overbar terms) and perturbations (lower case terms).44) Taking the derivative with respect to time where the overbar terms are constant yields — dr _ — dx — dy — dz (8.MEMO System with External Disturbances Hinge Point (Tanker) 249 R Y . Into page Receptacle (Receiver) Fig. Z respectively.

47). and setting r = 0 yields: _X_ £_ Z_ R (8.x=y = z = 0. First the disturbance rejection specification is identified Next.1 DISTURBANCE REJECTION SPECIFICATION The primary goal in designing the AFCS system is to regulate the position of the A/C receiving fuel relative to the tanker.47) R . Therefore. rearranging.46) and (8. (8. . 8-2. the control problem is to design the compensator G of Fig. 8-7.250 Chapter 8 Integrating. any deviation z cmd Fig.Receiver Thus. r = 0 if and only i£.46) R R ' defining x= XT -XR T .Tanker (8. 8. Previous chapters have provided the detailed step-bystep guide for the QFT design process.5 Control problem. As discussed in Sec. 8. 8-7 THE QFT DESIGN The details for the QFT AFCS design are now presented. design of the loop transmissions for all three channels is described.5 that will satisfy Eqs.

reveals12 that MISO loops {2.6 Disturbance rejection model response to 10 ft/sec impulse. the system response will deviate no more than 2 ft.MEMO System with External Disturbances 251 from the nominal set position is considered a disturbance. {3.1}.3}.3} are below the disturbance specification before compensation is applied i 111 1 1 1 1 1 1 1 1 1 1111 i 111 1111 i 111 11111 1111 1 1 1 1 0 0. Therefore the following disturbance specification is derived. and {2.5 Time (sec) 3 3. Given an impulse input of magnitude 10 ft/sec. Equation (8. {3. which is superimposed over the Lm P(s) MISO loop plots (see Fig.5 1 1.7).5 5 Fig.6 shows the disturbance rejection model response to a 10ft/sec impulse input. and Fig. Additionally.48) identifies the transfer function for the disturbance rejection specification. Generic disturbance rejection model = Y(s) Dea(s) 400s (8.5 4 4.1}. A maximum deviation from the nominal set position of 2 ft in any direction is specified which will confine the receiving A/C to a volume that permits continued fuel delivery.2}. 8.48) An analysis of the Lm plot of the disturbance rejection specification. the system will attenuate to half the maximum deviation in less than I sec. {1.5 2 2. 8. Hence a disturbance rejection specification is determined based on modeled disturbance inputs and the basic QFT design based upon unit impulse inputs. Since the most severe disturbance is due to wind. the disturbance specification is "tuned" to the wind input of 10 ft/sec. . 8.

8.3 CHANNEL 2 LOOP DESIGN (METHOD 1 "LOOP 1" DESIGN PROCEDURE) For channel 2 plant case 2 is chosen to be the nominal loop. Since channel 2 couples strongly into channel 1 it is designed first. Plant 2 is chosen because through initial design attempts it proved to be the most difficult to shape around the stability contour. The bandpass of the plants are relatively low. In shaping the loops the overall system bandpass is designed to remain approximately equal to the plant bandpass. A successful shaping of plant case 2 guarantees sta- . 8-7.7 P(s) Log magnitude plot. This requirement may require trade-offs on meeting certain higher frequency bounds. Channel 1 is then designed. The unproved method is then applied to utilize the known g2 to recalculate the disturbance bounds for channel 1. 8-7. a benefit of using an autopilot. Note: in practice the loops would have been renumbered in order to be in sequence and "simplify" the performance of the design process.252 Chapter 8 Fig. Channel 3 is designed last since it is completely decoupled and thus a 1x1 SISO system.2 LOOP SHAPING The order of loop shaping is determined by the amount of cross-coupling each MISO loop exerts on each other.

= _ ———— ———— ——— . and also from the difference in the effect of wind disturbance between the 2 classes of A/C plants based on CL.2 and CL = 0. but also larger cross coupling disturbance. Though there is a phase difference.6.6. This type of behavior makes it difficult to shape a loop that is stable.98 ±jl)(j + 10)(s + 20)(5 + 120) 8-7. thus overdesign is reduced For the same reason as in channel 2. satisfies the composite bound criteria. 8-2. Channel 2 is relatively easy to shape and proved to have the lowest order compensator g2. 8. as in channel 2 the templates show a 360° phase difference between the two plant cases of CL = 0. each plant correctly goes around the stability contour indicating a stable design for all plant cases. the nominal loop for channel 1 is given by plant case 2. Therefore. The curling causes a large change in phase with little or no change in magnitude. guaranteeing a stable design satisfying the disturbance rejection specification.9 shows the loop shapes on the NC for all 16 plants. The channel 2 plants are 36(f out of phase between the plants derived from the aircraft plant with CL = 0. see App. Utilizing the known structure ofg2 a more accurate calculation of the cross-coupled disturbance from the compensated channel 2 to the uncompensated channel I is achieved.MIMO System with External Disturbances 253 bility for all plant cases. The loop tends to curl at certain frequencies as shown in Fig. .2 and 0. The loop shaping is more difficult for channel 1. This is evident from the 360° wide templates and the stretching of the bounds over 36(f. E in Ref. 12. ———_ s(s + 0. Figure 8. The following compensator is designed for this channel: . The magnitude uncertainty arises due to the strong coupling from channel 2 into channel 1. stability and disturbance boundaries are calculated and composite bounds are formed in the MIMO QFT CAD Package.4 CHANNEL 1 LOOP DESIGN (METHOD 2 "LOOP 2" DESIGN PROCEDURE) After g2 is designed the improved method (Method 2) is applied using the equations derived in Sec. Again. D of Ref. 8. These have smaller magnitude compared to those obtained by Method 7.30). But unlike channel 2 there is a magnitude uncertainty evident in the channel 1 templates. these plants have not only more external disturbance. . Templates.10. (8. The plants of CL = 0. From this figure the 360° degree phase difference hi some plants is evident. based upon Eq. and . The disturbance and hence the composite bounds are generated. Compensator g2 poles and zeros are added to shape the loop.6 have a larger wind induced disturbance as shown in App. the channel 2 loop easily satisfies all QFT loop shaping requirements for composite bound and stability contours. The phase difference does not present a problem as the MIMO CAD Package is able to accommodate this scenario.8. As shown in Fig. 12.

9 Channel 2 Nichols plot for all plant cases.254 Open Loop Tnoimiitkn for Quancl 2 Chapter 8 Fig. . P0 = plant case 2. 8. Fig. 8.8 Channel 2 loop shaping.

8.25 + J0. A lag-lead compensator is used to "stretch" the low frequency curl.3)(5 + 0. A loop shape is finally achieved that satisfies the lower frequency bounds. P0 = plant case 2. The loop for channel 1 is shaped with a compromise on the bandpass. and slightly increases the system bandpass.+ 135 ±J65.38)(* + 1100) 8-7.747)(s + 200) g.32 ±J3. 184)0 + 90)(s. The channel / compensator g.10 Channel 1 loop shaping. The channel 3 templates have relatively small phase and magnitude uncertainty. ————————————————————————————————————— (8. 8.MEMO System with External Disturbances 255 Fig. Additional lag-lead compensators are tried to further "stretch" the curl but caused the loop to increase in magnitude as the frequency increased. 1 1 shows the uncertainly in the low frequency range of the plants. This is an indicator of the difficulties in achieving a loop shape that satisfies design criteria. The following compensator is designed for channel 1 : (s + 0. The Nichols plot of all 16 plants in Fig. There is no . stability.433)0 + 3)(s + 9)(s + 1. has a higher order than the channel 2 compensator. 14 ± J3 .5 CHANNEL 3 LOOP DESIGN (A SISO DESIGN) Channel 3 exhibits none of the channel 1 or channel 2 characteristics.50) s(s + 2)(s + 0. maintains a low system bandpass. Though there is large phase and magnitude differences between the plants the QFT method is able to achieve a design that satisfies stability and disturbance rejection for all plant cases.

8. further evidence of relatively small uncertainty in channel 3. Notice the large change in phase with no decrease in magnitude. This is deemed acceptable since it occurs below the 0 dB line at frequencies below the cutoff. The channel 3 loop has a tendency to curl up as the frequency increases. as detailed in Sec. The lack of cross-coupling disturbance and relatively certain external disturbance is evident in Fig. The external disturbances have similar effects on channel 3 for all plant cases. y position has the largest margin of disturbance allowed. 8. coupling from channels 7 or 2 into channel 3. The Nichols plot of Fig.12 where the bounds collapse around the stability contour. Again. To achieve stability the very low frequency bounds are penetrated This trade-off is considered acceptable since channel 3.Q5)(s + 0.51) . 8.5 ft. The transfer function for the channel 3 compensator is: (s + O.2)0 + 0- + 5)Q + 30) (8. 83' . 8-6. 7.256 Chapters Fig. The main difficulty is to add compensation to shape the loop around the bounds and stability contour at +180° and then add further compensation to keep the loop from penetrating the stability region at -180°.11 Channel 1 Nichols plot for all plant cases.13 shows a very tight grouping of all plant cases.

80JOOI2014016 Fig. .-4fl-i) 0 2 [41. -Domaia Open Loop Trammisskes for Channel 3 Fig. 8. 8.61.12 Channel 3 loop shaping.13 Channel 3 Nichols plot for all plant cases. /*„ = plant case 2.MIMO System with External Disturbances 257 1 'SsS ffl ® ffi 82ffl[!«2«a»18«l(ie]«i:«lCI)»)^«.

2}.2}. 8. 8.258 Chapter 8 8-7. except as noted. Due to the conservatism of the QFT technique it is left up to the nonlinear simulation to determine if the specifications for the loops {1.2}.2}.3}. 8. {2. >1 6. The closed loop MISO plots of Fig. 328. and {3. are an excellent indicator of success in meeting the design specification.1) MISO loop (14) MISO loop (U) TOO. and {3. MISO loop (1.14. From these plots one can easily see that the disturbance rejection specification is met for all MISO loops except in the low frequency portion of the MISO loops {1.14 MISO equivalent system Lm plots.1 J. .6 CLOSED LOOP Lm PLOTS The overall equivalent MISO system closed loop Lm plots are shown in Fig.14. {2.3} are actual met and to attest the goodness of the design for all loops. Fig.

6 deflected to a maximum value of approximately ft 008ft with slower dampening. Linear simulations are run for all plant cases in MATRIXX. The results of the linear separation for channel 1 (Z separation) demonstrate excellent results with very little perturbation from the set point. Figure 8. . are performed in EASY5x. The plots demonstrate two distinct responses corresponding to the A/C lift coefficient CL.6. one for each CL = 0.2 and 0. The channel 2 (X separation) linear simulation demonstrates similar characteristics for response based on CL.MEMO System with External Disturbances 259 8-8 AIR-TO-AIR REFUELING SIMULATIONS In this section the compensators designed in the previous sections are installed in the AFCS and simulations are run to analyze their performance. The CL = 0. The A/C with CL = 0. 8-8. 8. Nonlinear simulations are for two plant cases.Z separation deflections for all plant cases. 8.16. The aircraft with CL = 0.2 A/C has a maximum de- 1 I -.1 LINEAR SIMULATIONS Linear simulation are performed in MATRIXX with the modeled external disturbances forcing the system to deflect from the set point.15 Linear simulation . Also the response dampens faster for the A/C modeled with CL = 0. The simulations are executed in the presence of all external disturbances simultaneously. Again.01 ~"°12 u 2 4 5 Time (see) 6 7 B 10 Fig.003ft.2 show a maximum perturbation of approximately 0.15 presents the channel 1 response. excellent rejection of external disturbance is achieved as shown in Fig.2.

20. time limitation prevented performing a nonlinear simulation for each plant case. The nonlinear results are consistent with the linear results.025ft. while the CL = 0. namely very small perturbations for channels I and 2. Though considerably larger than channels 1 and 2.260 Chapter 8 Fig. EASY5x has a Dryden wind gust model preprogrammed in the CAD package. 8. 8-8.425ft from the set point. The nonlinear simulations require considerable time to setup and perform. are recorded as shown in Figs. with a larger deflection in channel 3. As in the linear simulations. 8. The maximum perturbation in channel 3 is approximately 1.16 Linear simulation -A'position deflection for all plant cases. Recall that the A/C with d = 0.18 and 8. the nonlinear simulations are within the design specifications.2 and 8-6. .6 has a larger uncompensated perturbation due to external wind disturbance.9 ft. the channel 3 perturbation remains within the design specification.2 NONLINEAR SIMULATIONS The nonlinear simulation are performed in EASY5x. The Dryden wind gust model is used in the nonlinear simulations versus the disturbance model developed in Sees.6 A/C deflects approximately 0. Two nonlinear simulations are run: one for an A/C with CL = 0. see Fig. Channel 3 (Y separation) has the largest perturbation from the set point in the linear simulation. The nonlinear simulations demonstrate the same excellent results that are achieved in the linear simulation.2 and one for a CL = 0.17. therefore.6. 8-4. flection of approximately 0. 8.

Y position deflection for all plant cases. Y. plant 1 Q. position deflection.2. . 8. 8.18 Nonlinear simulation -X.MEMO System with External Disturbances 261 0 1 2 3 4 6 8 7 8 9 10 Fig.17 Linear simulation . Fig. = 0. Z.

The optimal bounds are now a combination of not only the most stringent portion of each of the tracking and cross-coupling effect bounds but also that of the bounds due to the external disturbance (d^y of Eq. are the control surface and thrust response of the autopilot. 6 or Method 2 of Chap.. .2 1. The aileron. 8. This modified figure now represents the tracking/regulator MIMO control system.control surface and throttle response. 1 1 the Cy terms are replaced by the d. 10 and 5. 7. * 0. Figs.. that is Fig. -80 0 2 4 6 Time <MC) 8 10 12 14 16 Fig. terms. 8.262 Chapter 8 Also presented in the nonlinear simulation plots. 8. plant 1.19 Nonlinear simulation . The autopilot is a "text book" design and is not very sophisticated A QFT design using the actual C-135B autopilot can probably achieve similar results without extreme engine response requirements. 8-9 TRACKING/REGULATOR MIMO SYSTEM A MIMO control system that involves both tracking and external disturbance inputs (see Fig.2 is modified by including the tracking inputs r. The engine response is most likely due to the autopilot design. 8. In Figs 5. rudder. and elevator responses are well within the physical capability of these devices.19). On the other hand the thrust requirements are probably beyond engine response capability. 19 and 8. (8.1) can now be designed by either Method 1 of Chap.

8 . 8.MEMO System with External Disturbances .20 Nonlinear simulation .X. Z position deflection plant 2 CL = 0. Y.21 Nonlinear simulation . 8.2 -i S x j . .2 r i •: il 0 \ V 0 « 4 6 .6.4 1 *° 263 •* 0 I .. ttme (BBC) 10 1Z 14 16 -A Fig. 12 14 16 Fig.control surface and throttle position.

are presented. A) was modified to also handle the pure regulation problem. which is the interaction bound for a pure regulator system for the improved method.10'73 . The system response is within design specification. the inherent nature of QFT's ability in handling large plant uncertainties is discussed Finally the Lm plots of the closed loop MISO system is shown.12 This modified CAD package has been used to design an PCS that involved satisfying both tracking and regulation specifications. The tracking MIMO QFT CAD package (see App. indicating a successful design based upon the trade-offs that are required The compensators designed in this chapter are integrated into the air-to-air refueling AFCS. covering the particular difficulties in shaping the loops for each channel. Equations to calculate the new optimal bounds. Also. Each loop shaping is detailed. The results of this development are applied to the design of a real-world problem: the AFCS. Linear and nonlinear simulations are performed with excellent results.264 Chapter 8 8-10 SUMMARY This chapter presents the development of the improved QFT method to include the effects of external input disturbances on the system's outputs. The QFT design process worked extremely well in designing the AFCS in the presence of an external output disturbance.

The scientific method uses mathematical methods: (1) to gain insights into. the researcher (1) is not concerned with whether his or her efforts will result in the solution of "real world problems. the engineer must take over and apply the new results to real world problems. corollaries. The scientific method is necessary in order for the researcher to be able "to see the trees from the forest. (2) to generalize. Generally. although most of the real world problems are nonlinear. 265 . It is rather an engineering endeavor. Once the scientific approach has successfully advanced the state-of-the-art. the mathematical solution of the control problem needs to be implemented and the validity and applicability of the modeling assumptions need to be verified by the engineer. in many areas of science and technology. at this stage. The engineer is at the "interface" of the real world." and (2) applies linear analysis and synthesis techniques most of the time. in this chapter the process of applying the QFT robust control system design method is presented from this broad perspective and it is shown that it is uniquely suited to address and solve engineering control system design problems.NOW THE "PRACTICING ENGINEER TAKES OVER"23437 9-1 INTRODUCTION Control system design is an interdisciplinary and multi-stage process and is not a sub-area of mathematics. and (3) to expand the state-of-the-art. it behooves the engineer to reduce the real-world control task to a tractable mathematical problem. and the body of knowledge and theoretical results available in the technical literature. and lemmas. and where applicable. While applied mathematics plays a crucial role in facilitating the solution of robust control system design problems. Often this requires the proof of theorems." and thus to be able to achieve positive results. The required simplifying assumptions and the attendant mathematical modeling effort require sound engineering knowledge and judgement. Furthermore. Hence.

I W. corollaries. Horowitz applied the scientific method in the development of his Quantitative Feedback Theory (QFT) approach to the engineering design of robust control systems. I. when dealing with nonlinear systems and real world problems.6^72 et al (see Ref. and simplifications that are needed for him or her to proceed from A to O. and O. Merino. and the design is validated. Holton.38 F. It however behooves the engineer to make the required modeling assumptions. given the required performance specifications. This approach will expedite the engineer's robust control system design and enhance the quality of this design. In PRACTICE (engineer) There is a difference between practice and theory. This requires that the engineer have a good understanding of the physical characteristics of the plant to be controlled. steps T to Z entail extensive simulations and/or experimental tests where the validity of the model is verified. from P to S. 2) have used the scientific method to further enhance the mathematical rigor of the QFT technique. It however behooves the engineer to make the required modeling assumptions. the control system design task is a multi-stage process which entails many steps. the implementation issues are addressed. an engineer who has a firm understanding of the results of the "scientific method" and has a firm understanding of the nature and characteristics of the plant to be controlled must develop an "engineering method" by developing appropriate "Engineering Rules" (E. Mathematics is most helpful in taking the engineer through some of these steps. D. say from A to Z. and in the face of parametric uncertainty. and simplifications that are needed for him or her to proceed from A to O. In conclusion. in order to help the engineer to bridge the "interface" gap. the control system design task is a multi-stage process which entails many steps.39 O. Finally.R. Nwokah. is a QFT solution feasible.) that will assist him or her in "bridging the gap" Therefore. This body of knowledge must be coupled with "the body of engineering knowledge" pertaining to the application. a successful practical control system design process can be described as an approach that utilizes existing and/or developing new E.Rs by an experienced control system designer in order to bridge the gap between the scientific and the engineering methods. say from P to S. The goal of these researchers is to establish theorems. so that the mathemati- ." Thus. Mathematics is most helpful in taking the engineer through some of these steps. hypotheses. say from A to Z. This "bridging the gap" (the "interface" gap) can be highlighted by the following anonymous quote: "In THEORY (scientist) There is no difference between theory and practice. Bailey. and lemmas that can tell the designer at the onset whether. hypotheses. for example.266 Chapter 9 Thus. so that the mathematical problem is tractable and the existing theory can be applied.

3 of Sec. . 9. In this situation the designer needs to reduce the size of the templates (see E. 9-3). or by grouping failure cases so that individual designs can be accomplished and then scheduled. This can be done by either. tells the engineer at the onset of the design process whether a fixed compensator G can be synthesized that will yield the desired system performance in the face of the prevailing structured uncertainty. see Fig. reevaluating the design requirements and eliminating individual effector failure cases. and the design is validated An example of this multi-stage process is presented in Ref. When control effector failures need to be accommodated. the width the of template can become excessive and a single successful design might not be possible. Finally. If only the template's height is the problem then the engineer needs to employ straight gain scheduling. 9.1.1 A template representing J LTI plants at co = 30 rad/sec. 15 e (deg) Fig. 9-2 TRANSPARENCY OF QFT The elements resulting from the application of the scientific method which provide the "transparency" of the QFT design technique and that enhance its abih'ty to solve real world problems are: (a) Template The size of the template (width and height).Practicing Engineer 267 cal problem is tractable and the existing theory can be applied.R. 33. steps T to Z entail extensive simulations where the validity of the model is verified. the implementation issues are addressed.

altitude. . 9.variationin parameter #2 •" ' expanded template l|ir -initial template Phase Fig. flight conditions. and has a nonlinear model of the aircraft for that flight envelope. the minimum set of LTT plants needed for a QFT design are those that form the boundaries of the templates over the frequency bandwidth of interest. Since plants on the interior of the template are guaranteed to meet specifications. angle of attack. Nonlinear aircraft models are linearized at operating states.2 Template expansion procedure. . the designer reduces the computation requirements and simplifies validation and verification of the design.2 for two independent parameters. By choosing the minimum number of LTI plant cases required to fully describe the templates of interest. flight envelope (operating scenario). This process is depicted in Fig. 10).nominal variation cases .variation in parameter #1 "**.. The problem is to find a minimum set of LTI plant cases that describe the template boundaries utilizing all permutations of the independent parameters that describe the flight envelope.268 Chapter 9 One of the key problems associated with QFT design is choosing LTT plants cases that represent the nonlinear plant in the operational region of interest. hi aircraft design the designer knows the proposed operating region of the aircraft. One method used to choose LTI plants is through the use of template expansion. and side-slip angle as well as other parameters such as center of gravity and weight.extended cases 00 s^ . For example (see Chap. 9. These flight conditions include states such as velocity.

The selection of values. 9. Thus. a plant p. select the nominal plant p. a template boundary can be developed. At this point the limits of variation of the parameters describe the boundary of the template. at the top of the templates. (d) Minimum order compensator (controller) (MOC) G Why are high-order compensators unacceptable? The implementation of a w-order compensator in a digital flight control system (DFCS) yields an w-order discrete-time dynamical system. . while keeping the remaining parameters fixed at the nominal flight condition. Furthermore. the plant whose unstable pole lies furthest to the right in the s-plane) as the "worst case plant. when all J plants in ^ are stable. (c) Signal flow graph (SFG) Forj = Peu = PWu. during the first m-1 time instants the input has a somewhat limited effect on the output.1). are further enhanced by the engineer's firm understanding of the interrelationship of the plant outputs with the inputs to the W matrix.Practicing Engineer 269 The first step in the template expansion process is to use engineering knowledge of the aircraft to find the limits of variation of the individual parameters in the flight envelope. it is required that this plant always lies. the points making up the expanded template boundary are used to expand the boundary again.3 represents a MIMO QFT control system structure. to be the plant lying at the "top" vertex of the template (see Fig. for the output is partially determined by the m initial conditions. that has the highest degree of instability. for each loop. Hence. Variations about these nominal points describe an expanded template boundary. in turn.) frequency.y and modifying some of these values during the simulation phase of the design process. be obtained at the respective specified BW(Z-. 9. so that all variations are taken into account. The next step is to choose a nominal flight condition as a starting point on the template. That is. (b) Phase margin frequency In order to ensure that the value of the specified phase margin frequency e^ is not exceeded by any of the plants in the set then. The next step in the process is to expand the boundaries of the template by choosing points on the existing template boundary to be nominal points and then plotting new lines describing parametric variations at each of these nominal points.3 that represents Pe can be helpful in the initial selection of the values of vc. graphically. for all template frequencies." This rule for the selection of the nominal plant will facilitate the achievement of the specified value of o)^. In this fashion. Fig. 9. Once the original template boundary has been expanded. etc. Having the SFG for the portion of Fig. The latter is equivalent to a first order dynamical system with /w-time delays. the control action is delayed. Next plot lines on the templates describing the effect of varying each parameter through its limits. the design engineer must then select. as the nominal plant. It is advisable that one of the templates. In the event that one or more plant in the set ff'is unstable.

Another factor that must be considered in maintaining low-order controllers is that on board flight control computers have limited capacity due to other non-control related computing requirements. in a relatively "short design time. the poles and zeros of the nominal plant qiio are put to good use in synthesizing a satisfactory loop shaping transfer function Liio. have recourse to plant P. In QFT. In general. these methods yield high-order compensators. Hence. or design methods. the time delay caused by an w-order controller is m/60 s. = Some or all of these elements. 9. "doctoring" or "padding" by inserting additional poles and/or zeros into P. The FCS's time delay is 20/60 = 1/3 s which is unacceptable. during loop shaping. and in-turn Pe.3 MIMO QFT control structure block diagram. Example . This results in an "augmented" plant matrix P^ which these designers base their design on in order to achieve a compensator G whose order equals the order of the augmented plant. about 30% of the computer capacity is allocated to the DFCS and 70% for non PCS requirements. are not available in other optimization based. As a rule." a design that meets all the performance specifications that are specified at the onset of the initial design effort. or comparable ones.Consider an 60 Hz PCS sampling rate and a 20tt-order Hm compensator. This minimizes their ability to achieve. multivariable control system design techniques. Doing so yields the MOC g. in order to achieve the MOC G. . To achieve a "low-order" compensator (controller) some designers.270 Chapter 9 R Pcmd Y Hcrnd W J L Fig.

R. it may be desirable to obtain complete decoupling for the nominal plant case.y elements. - 0 0 = PW (9.. for less attention needs to be given to cross-coupling effects (Cy) rejection. Also. qa plants are most desirable for they allow the full exploitation of the "benefits of feedback..e. with the additional benefit of reduced closed-loop B W. i. in general..p.e. Now.e.p.1) and where and (9. the degree of decoupling will have been enhanced This greatly facilitates the QFT design process... .p. It turns out that one can apply the Binet-Cauchy theorem64 (see Chap." i. m.Practicing Engineer 271 9-3 BODY OF ENGINEERING QFT KNOWLEDGE Through the many years of applying the QFT robust control design technique to many real world nonlinear problems..2) Thus. it is desired to know at the onset if it is possible to achieve m. the following Engineering Rules have evolved: 9-3.1 Weighting Matrix An Ixm weighting matrix W = {w^} is required to achieve a square mxm equivalent plant matrix Pe whenever the plant matrix Pis mxl. Method 1 is then more readily applicable. will not occur.™ 0 P22 .«.1 E.3) 0 o . high gain. P>= (9. Pmm Although for the nonnominal plants complete decoupling. qa's are possible. qt's by the proper selection of the vc.. i. Pit 0 * #. 8) to determine if m.

and/or lemmas pertaining to these bounds. it is outside the bandwidth of concern in manual control and it does not present a problem. If theorems. For manual flight control systems.3 E. corollaries. either some specifications need to be relaxed in order to achieve a solution.p.272 Chapter 9 9-3. 9.Out" "Close .4). external disturbance and cross-coupling effects rejection. reveal that no loop shaping solution exists then one must be attuned to stepping back and doing a "trade-off. obtained by the scientific method.4 Right-half-plane analysis. If this RHP zero is "far out" to the right. For these cases a satisfactory QFT design may be achievable. 9. this is not necessarily deleterious since the pilot inputs a new command before its effect is noticeable.R2 n.In"- Fig. . one must determine if the location of the RHP zero(s) is in a region which will not present a problem for the real-world design problem being considered. Thus. it is assumed that the unstable pole is outside the closed loop system's lower bandwidth. if a RHP zero happens to be "close" to the origin (see Fig. or one must be willing to live with a degree of gain scheduling.p. 9-3.2 E. in other words. a graphical analysis of the template shown in Fig. bearing in mind the need to satisfy tracking specifications. and satisfying the stability bounds.m.ni.1 can reveal the following: jco S-plane / "Far .R3 Templates The adage "a picture is worth a thousand words" applies to the preliminary task of determining if a robust control solution exists. qu's For qu's that are n." In other words. 9.

performance specifications must be realistic and commensurate with the real world plant being controlled Situations have occurred where the conclusion was reached that no acceptable design was possi- . identified an expanding area of the flight envelope. Naturally. 80% failure still permits enough control authority! This degree of failure can only be determined by a person who is knowledgeable of the physical plant to be controlled. in order to achieve a solution one needs to relax the requirement that the "worst failure case" be accommodated. On these templates. that required more plants for a better definition. the worst failure case is the culprit in generating this large "angle width./s of Pdo not have the same value of A (excess of poles over zeros) then as o -> <x> the templates may not become straight lines. This is especially true for real-world control problems that involve control effector failures accommodation. it is necessary to stipulate for what failure case or cases a successful design is achievable. 9. generally. in flight control34 (see Chap.4 E. 25%." Thus. a nominal flight condition for an unmanned research vehicle was chosen to be 50 kts forward velocity. If all thep. 50%. when shown on the template. is too large thus not enabling the determination of the tracking bounds for a given w.2. through maximum and minimum values.9% of the mean aerodynamic chord From this nominal flight condition. (c) The effects of the structured uncertainty on the template's geometry is now discussed Thus. In general. In determining "reasonable failure cases" that can be accommodated by robust (not adaptive) control .1073 In these design problems. (b) The situation where the templates are too "wide" (the magnitude of phase angle width) thus prohibiting a QFT solution or a solution by any other multivariable design technique. These variations produced an initial set of templates. An attempt is made to choose flight conditions in such a way as to fully cover the flight envelope with the templates. linearized plants that represent different flight conditions in the flight envelope are extracted from a nonlinear truth model.Practicing Engineer 273 (a) The maximum template height. A possible method of reducing the size of the templates is given by E. variations caused by each parameter are identified Each variation.R. knowledge of the plant (application) "is king" when it comes to the design of a feedback compensator or controller for the said plant. see Fig. 10). each parameter was varied in steps.8. a weight of 205 pounds and center of gravity at 29. when this situation arises.one must consider if 10%..4 Design Techniques No matter what design method one uses. 1000ft altitude. To do this. in dB.R. One can then decide if gain scheduling is required and is feasible in order to yield a design that yields the tracking bounds. 9-3.

by Method 2 it is known that cot of the succeeding designed loop is larger than the previously designed loops. 9-3. it is desirable to minimize the amount of gain required in each loop /. . while at the same time meet the performance specifications in the face of the given structured uncertainty. then loop 2. To achieve this by an automatic loop shaping routine may be difficult. To achieve this goal. can use his "engineering talent" to make use of the "dips" in the composite B0i(j(oj). where <% < <% < e% < .7 E. 9-3. first close loop 1.6 E.274 Chapter 9 ble. and then zeros and poles are successively added in order to obtain the required loop shape. . This insures that the ensuing compensator/controller is of the lowest order. where feasible. etc.R5 QFT Method 2 Arbitrarily picking the wrong order of the loops to be designed (loop closures) by Method 2. Thus.5.. For these situations when one "stepped back" and asked the pertinent question "was something demanded that this plant physically cannot deliver regardless of the control design technique?". see Fig.5 E.R6 Minimum Order Compensator (Controller) (MOC) In order to ensure the smallest possible order compensator/controller. etc. with a good understanding of the Nichols Chart and a good interactive QFT CAD package. resulting in: (9.. entails picking the loops in the order of increasing values of the desired co^ i. Indeed. The proper order of the loops to be designed by Method 2. the nominal plant's poles and zeros are being used to shape the loop. it was determined that some or all of the prescribed performance specifications were unrealistic. which is highly desirable.9..R7 Minimum Compensator Gain To minimize the effects of noise. 9-3. The designer by shaping Loi to pass through these dips.e. can ensure achieving the minimum compensator gain that is realistically possible. a control system designer. can result in the nonexistence of a solution due to the chosen order of closures.. one starts the loop shaping process by using the loop's nominal plant Lol = qllo. This may occur if the solution process is based on satisfying an upper limit of the phase margin frequency co^ for each loop. saturation. the compensator is obtained from gt = LJqllo.4) Finally.

5 QFT stability and composite bounds. The templates 3PP(ja).9 E. 1). utilize unity feedback loops for the mxm MIMO plant P that will yield an mxm preconditioned plant matrix PP. for all templates. QFT affords the robust establishment of these FOM. In some flight control system design problems an inner loop controller such as that of an angle of attack command has been used to achieve a PP. 14. Thus.) 9-3. and the phase margin frequency CD^ of a feedback control system using the NC. BW(Lj) template ensures that a>^ < BW(L. 9.).2 of Ref.1.8 E. by choosing the nominal plant: (a) To correspond to the maximum dB plant. This template reduction size is predicted by performing a sensitivity analysis (see Sec. on the «ty. Indeed. the gain margin. 9-3. in general can be smaller in size than the templates 3Pdco.Practicing Engineer 275 Fig.11'12-65 (See Sec.R8 Basic mxm Plant P Preconditioning When appropriate.R9 Nominal Plant Determination It is easy to determine the phase margin angle y.)for all plants . 8-4.). This concept has been used in a number of MIMO QFT designs. The QFT design is performed utilizing the preconditioned matrix PP.

9. due to the conservatism of the QFT technique. For example. is utilized to obtained the algorithm for the software implementation of Gz(z).12 E." Transient performance is sometimes of more interest than the ultimate steady-state output. 9-3. shown in Fig. ensures the achievement of a robustly guaranteed gain margin is easily accomplished (c) Whichever is the "left-most" plant on all templates ensures that the desired Y is robustly achieved With respect to item (a).6c.13 E. by use of the bilinear transformation the controller Gz(z) of Fig.6b is obtained from the compensator Gj(s) or Gc(w) of Fig.6a.11 E. for a more realistic test in determining a control system's performance through simulation a ramped-up step function is used as shown in Fig.g. the smaller the value of T. Use of this type of a realistic forcing function has the tendency to minimize the saturating aspects of the system.R11 Asymptotic Results Asymptotic results provided by mathematical analysis are not as useful as they seem to be. will ensure that the phase margin fiequency for all /plant cases are as close as possible to co^ 9-3.13 For example. for all templates. .7. Consider the manual control disturbance rejection case where fast disturbance attenuation is more desirable than total disturbance rejection which entails a very long "settling time. 9-3. the goodness of the design is judged on a pre-specified planning time horizon beyond which the performance is less important since the human operator will inject new inputs to the system.RIO Optimization and Simulation Run Time In many real-world manual feedback linear or nonlinear control problems.10 E.276 Chapter 9 (b) To be the uniformly maximum dB plant. Thus. Unfortunately. the greater the degree of accuracy that is required to be maintained The numerical accuracy is enhanced by a factored representation of the controller and prefilter.. 9. if the maximum dB plant is not the same on all templates within the BW then select the maximum dB plant on the template for the desired «^.R12 Controller Implementation Tight performance specifications and a high degree of uncertainty require small sampling intervals T.R13 Nonideal Step Function for Simulation For mathematical analysis of LTI systems ideal step functions are utilized In the real world ideal step functions do not exist. The equivalent cascaded transfer function representation (factored representation) of Gz(z). 5 seconds and there is no interest in the long time intervals commensurate with the slow phugoid dynamics. 9. in manual flight control the time horizon is determined by the aircraft's short period dynamics. e. Selecting the nominal plant in this manner. 9-3. 9.

consider the extreme case of zero inputs to linear plants: the output will always be zero. 9. which.domain J cascaded controllers Fig. Both displacement and rate saturation significantly reduce the achievable benefits of (high gain) feedback Thus. which is afforded by linearity. 9. The intrinsic scalability property.domain compensator = 0(z) (c) z. including QFT.6 s. yield linear compensators for linear. unfortunately. breaks down in the face of nonlinearity.7 A ramped-up step function. The worst offenders are saturation type nonlinearities. plants.Practicing Engineer 277 -*0(s) I(w)- (a) The design compensators -0(z) (b) z. the achieved robust performance applies to "small signals" only. irrespective of the (linear) plant and so infinite robustness is . 9-4 NONLINEARTTIES . are invariably located at the plant inputs. but uncertain. Hence.THE ENGINEERING APPROACH All the current robust control design methods.or vc-domain to z-domain bilinear transformation: Formulation for implementation of the G(z) controller. time rise time Fig. The latter are encountered in actuators.

for it will cause instability and departure. 9. nonlinearity and saturation need to be addressed when feedback control is used to stabilize open-loop unstable plants. 9. This will have catastrophic consequences.8. However the error signal will be ramped into the compensator. from a "small signal" point of view saturation is equivalent to opening the feedback loop. This is also generally true in robust feedback control systems once significant slewing is attempted. The resulting plant matrix P (y = Pit) is (AdjD)N ^ =[pv} (s)] detD (9. actuator "anti-windup" schemes must be employed. One might want to relax the robustness requirements and subsequently employ a limited degree of gain scheduling.278 Chapter 9 achieved The lack of robustness becomes evident when the plant is being driven hard (with a large input signal .5) with D(s) = [dy(sj] and N(s) = [ntj(s)] being mxm polynomial matrices in s. when saturation is encountered during simulations of the designed feedback control system.6) where the elements of P are transfer functions.it is "slewed") and nonzero inputs are applied. This results in the tracking not being as tight as initially desired. 9-5 PLANT INVERSION Given an /n-output vectory. Another possibility is to somewhat relax the tracking performance specifications and reduce the BW of the prefilter. thus delaying or eliminating the onset of actuator saturation. .8 AMMO plant. Suppose that the inverse plant ym Fig. the LTI plant transform equations can be written in the form: D(s)y(s) = N(s)u(s) (9. Furthermore. Hence. and is due to nonlmearity and saturation. Indeed. an /w-input vector u for Fig.

. (9.8). . where the matrices A. Thus. and which. if F1 is obtained from Eq. } d\ Jj} d J det{ni}-}/ d1 det{nv}l J 2 ddet{n^} In general. For the numerical calculation of P~' it is much better to use Eq.7) is needed. 1 1): ._ A d j { n / d _ 2Adj{n. (Adj N)D Pl = N1D = ^-^—— detN (9. 11). (9.10) In other words.11) where «# and d are polynomials of degree /and n. then m-l cancellations of polynomials from the numerator in Eq. for the mxm control system: (9. from Eq.11) with the m-1 polynomials in its denominator is required. since u = P'y. i.8) or from the state equations i. (9.. (9. nxm.5). C are nxn. t<n and d is the characteristic polynomial of A. in s.9) that describe the w^-order plant.Practicing Engineer 279 (9. (9. yields the following expressions: y = C[sI -A]-lBu (9. and mxn.13) Thus. The numerical poles/zeros cancellations will of course not be exact. B. \^-\(9. (9.11).8) rather than Eq. The reason why is best illustrated by considering a 2x2 (m = 2) plant utilizing the second approach. rather than directly from Eq. The designer can obtain it either from Eq. in turn.e. respectively. respectively. (9. : x=Ax + Bu (a) y = Cx (b) (9.e. as in some of the QFT techniques. because of the inevitable computer round- . for example.

Then. Conclusion Obtain det P and P1 directly from Eq.0123456.p.5). 4 the QFT design method applied to MISO sampled-data (S-D) control systems is presented. if possible. (9. is determined by the value of the open- . Moreover. Figure 9.0123456} The factored form polynomial is expanded into coefficient form. Proof outline ~ If the plant is not controllable and observable then even a formal inversion of P won't be possible because some of the rows and/or columns of P will be linearly dependent over the real field.11) is invertible iff the system (A.. (9. which may not be readily available from the state-space form. 2. Note that m-1 such cancellations of each zero of the numerator with the m-l poles of the denominator must occur. As pointed out in this chapter.50.5). Therefore.p. C) is controllable and observable. the designer is very strongly advised at the very onset of the design process to obtain.. the following holds: Theorem . one needs the plant equations in the form of Eq.e. P is mxm. ^-domain plant becomes a n. in order to recover numerical accuracy one must factor all the numerator and denominator polynomials of Pl and then check out the inevitable inexact cancellations.936 illustrates the degree of accuracy of factoring this polynomial by various 1992 CAD packages. The roots of the expanded polynomial are then obtained and plotted. are entered into the CAD package to create the polynomial: Roots ={1.. Finally. (9. . bearing in mind that the entries of P are not real numbers but instead are (rational) functions in the dummy variable s. a m.0123456. The MATHEMATICA package yields the most accurate roots which expedites the cancellation of the common factors of Eq. in the formulation of Eq. The following set of roots.. To do this.. formal invertibility is almost guaranteed Indeed. 9-6 INVERTIBILITY The question is often asked: "How does one know that f\s) is invertible?" Obviously. the plant needs to be a square matrix. 9-7 PSEUDO-CONTINUOUS-TIME (PCT) SYSTEM In Chap. (9. i. (9.The mxm matrix P(s) in Eq. as is well known. Thus. each having at least 8 significant digits of accuracy.12) no "pole/zero" cancellations occur iff the above system is controllable and observable. controllability and observability of the control system are "a given" in the real world where over modelling should be avoided at all costs. in converting an analog system whose stability.5). B. (9.280 Chapter 9 off. in general. the data in the form of Eq.13). plant in the w-domain. Also.m.

——— ———— 40 -30 . .—— • r _. Because.__._.p. -— 40 -20 -10 j h--j--—— • -to ••> -n -« •« -«o o r** Roots of high precision polynomial (Mathematica). ——— . — — -H . Thus. to a S-D system.m. 4. ——— ———— ———— —— — — • —— —— . the degree of system stability is decreased. COK > 0% (or VK > v. | j -10 . Macsyma). of the w-domain n. Maflab. 9.9 Figures from Sating's AFTT Thesis. there is a "restricted" frequency band Aw = 0% . Control-C.36 loop gain K.__. in which loop shaping must be accomplished while satisfying all the bounds. for S-D systems this "restricted" frequency band makes loop shaping a little more difficult to accomplish. Roots of standard precision polynomial (MATRTXX. In general. characteristic. — — . Such a "restricted" frequency band does not exist for analog systems. its stability is now determined not only by the value of K but is also a function of the sampling-time T.„.). in converting an analog system to a S-D system. Fig.Practicing Engineer 281 J r——I ™- " ———— j .0%. as pointed out in Chap.

as amply reflected in the size of the templates. may in some cases prevent a successful design. Now.282 Chapter 9 Finally. 4. The PCT QFT design approach was used to design a MIMO digital flight control system for an unmanned research vehicle34 (see Chap. will shift the region of uncertainty from one of low frequency to a high frequency range. this situation can occur in the design of a flight control system (PCS) where one initially closes an aileron/rudder interconnect feedback loop. For example. QFT is good at handling structured (parametric) uncertainty. systems that are strictly proper in the s-domain are proper but not strictly proper in the w-domain As shown in Chap. The resulting s-domain controllers and prefilters are then transformed into the z-domain by use of the Tustin transformation. When QFT is applied to design a MEVIO PCS that already has incorporated an aileron/rudder interconnect. in general.. . a technique for "by-passing" the restricted frequency band problem for a w-domain QFT S-D system design is to convert this S-D system to a pseudo-continuous-time (PCT) system.35 The designs were successful and met all desired performance specifications..e. for in MIMO QFT the use of high gain will decouple the equivalent set of MISO plants. i. unmodelled (e. For example. Hence. However. Hence. in flight control the plant inputs are to the aileron and rudder deflection and the outputs are roll rate and side-slip. based upon Bode's theorem. 10). QFT. the Bode effect prevents a successful design because the uncertainty has been transferred to an unfavorable high frequency band. provided that the inputs and the outputs of the MISO controlled system are properly chosen. that putting a unity feedback loop around a plant P containing parametric uncertainty. essentially does the same job as the use of an aileron/rudder interconnect.g. for QFT automatically does the job. uncertainty at low frequencies and within the control bandwidth of interest—as opposed to other robust control design methods which address unstructured uncertainty. 9-8 BODE'S THEOREM40 It is well known. The purpose of using an aileron/rudder interconnect is to minimize the cross-coupling effect (also referred to in flight control as adverse yaw). high frequency) dynamics. The QFT design is then accomplished in the ^-domain for the PCT system. The criteria for converting a given S-D system to a PCT system must be satisfied in order to accomplish a satisfactory QFT design. in effect. applying QFT to an effective plant transfer function P that includes the unity-feedback loop. This approach was also used to design a MIMO digital robotic control system. namely. the direct application of QFT to the design of the lateral directional channel FCS obviates the need for an aileron/rudder interconnect.

Nonlinear (a) Nonlinear Functions (b) Nonlinear Dynamics 3. Warping Fig. 9. both from a theoretical viewpoint and the real world aspects of the control application. The major factors that play a vital role in the design process of a flight control system are discussed in the following chapter and are depicted in Fig. Visualization L Hardware-i n the-Loop StmulattoD-ImpiementaUoo 1. Computation Cycle Time/Sampling Rate 4. Noise Corrupted Measurement. Quantization Error.10 The QFT control system design process. 9. . implementation and flight test.there are many important factors that must be considered. This figure shows four major aspects of the control design process: control theory (design techniques) simulation."In Bridging the Gap" .10. during the design process. the following items: (a) The important considerations of control authority allocation (b) The imbedded desired performance specifications (c) An awareness of the imperfect real world aspects of the particular control system that is being designed ' Problem Inception 'Control Authority Allocation/Actuator Solution ••Imbedded Desired Performance Specifications At Onset of Design Process Awareness of "Real-World" Plant Sinnil 1. Throughout the text it is stressed that QFT has the ability from the inception of the control problem to the testing and proving of the final designed control system to holistically take into account. Real-Time Operation of control algorithm 2. Linear 2.Practicing Engineer 283 9-9 THE CONTROL DESIGN PROCESS34 As demonstrated throughout this chapter . Available for Feedback 3. during and upon the completion of a successful control system design.

As an example. However.284 Chapter 9 9-10 SUMMARY Engineers throughout the world are applying the results of the scientific method to achieving solutions for real world problems. In conclusion. it provides facilities to deal with uncertainty which are not available in traditional methods. the following: "The QFT approach has the obvious advantage that it is close to engineers' existing experience on classical design methods." In this chapter guidelines are provided to the control engineer on how to interface between the scientific and engineering methods. an attempt is made to bridge the often lamented gap between theory and practice. Grimble. an aerospace firm has stated to Professor M. More recent tools such as Hm design also show promise but are very different to the existing procedures used in parts of the aerospace industry. University of Strathclyde. The QFT approach therefore appears to have the attractive features of providing a link with existing techniques whilst at the same time providing many of the advanced features needed for the 90's high performance systems. . What might be needed are tools for the future which combine the attractive features of QFT and Hx approaches.

it is important for the designer to follow a design and implementation process that allows the testing of the assumptions as early in the process as possible so the control system can be redesigned. it is necessary that engineers of the future must be able to bridge the gap between the scientific and engineering methods.R. This chapter provides the next step in enhancing this goal: overcoming problems encountered during design and implementation of a QFT control system in the real world.. for example. perform nonlinear simulations. "assume linear behavior" that result in obtaining linear plant models. As described in Chapter 1 and detailed later in this chapter. the control design process should include simulation of the control system on increasingly realistic models which helps transition to implementation on real world applications.) as a first step towards achieving this goal. i. Chapter 9 presents a set of Engineering Rules (E. Thus. etc. to take into account unmodeled effects. As pointed out in Chapter 9. Control design problems generally involve real world nonlinear plants. In utilizing control system design techniques.10 THE DESIGN AND IMPLEMENTATION PROCESS FOR A ROBUST CONTROL SYSTEM 10-1 INTRODUCTION Chapters 2 through 8 have involved the theoretical development of the QFT technique and the associated control system design process with minimum concern with the real-world aspects of the control problem. which require linear plant models. Most of the real world implementation problems are the result of assumptions made during the design process. Thus. account for unmodeled effects.e. it is necessary that assumptions be made that allow simplification of these nonlinear plants. implement hardware. Included are testing of the assumptions that are made in order to design with linear models. 285 . in facing the technological problems of the 21st century.

1. That is. If the control system does not meet performance specifications at any stage of the process. Following is a discussion of issues the control designer faces in implementing control systems in the real world. hardware/software interface.2 PERFORMANCE SPECIFICATIONS Performance specifications are essentially mathematical models developed from the functional requirements and are utilized during the design process in order to achieve the desired system performance robustness. the designer must be aware of how the specifications and requirements relate and what tradeoffs need to be made. such as integrator wind-up.1 indicate the features of the QFT technique that assist in the design of control systems and can best meet performance specifications and be implemented on the real world system. the designer must follow a design process such as that shown in Fig. Note. 1. During the design process. 10. the designer must have a clear understanding of the problem that needs to be solved. it becomes clear that the functional requirements cannot be met. the designer might be called upon to use engineering judgement and the knowledge of the goals of the controlled system to modify these requirements.1 FUNCTIONAL REQUIREMENTS Before the design process can begin. 10-2. 10. i. the control system is redesigned and retested. The designer must also understand the environment in which the system is required to operate.e. The following sections describe the individual stages of the control design and implementation process. Therefore. In general. Since performance specifications are normally only interpretations of the functional requirements. the designer might need to apply . this is not a step that a control designer normally takes on his own. and bending modes. costly time can be wasted in the design-test-redesign cycle. Together these two requirements make up what is referred to as the functional requirements. 10-2. they also become more expensive both in cost and time.286 Chapter 10 This chapter begins with an in depth discussion of the control system design process. redrawn here as Fig. The ovals inside the circle in Fig. as the simulations become more realistic. it is very important to be able to find potential problems early in the design process for the control system. the environmental requirements. 10-2 CONTROL SYSTEM DESIGN PROCESS In order to design a control system for a real world control problem. If the designer does not start with a clear understanding of the functional requirements. the designer must understand what the controlled system is required to do and what are its operational requirements. If during the design process. This figure represents a design process that moves the designer from the problem definition stage to the successful control system implementation in steps of increasing reality.1..

Functional Requirements This is the problem statement. performan and disturbance boundaries a function of frequency Tryout of controlled system under actual operating conditions Ability to manipulate closed-loop response as a function of frequency Qperator-Jn-the-Loop Simulation •Visual cues •Moving base Engineering Visualization Used to better understand results of simulations and System tests Design can be brokei down by open or closedloop operating condition Hardware-in-tfac-Loop Simulation/Implementation Engineering Interactive Simulation •User supplies commands and then can react to resulting dynamic behavior •Gives a better understanding of control system operation •Real-time operation of control algorithm •Noise corrupted measurements available for feedback •Computation cycle time/Sampling Rate •Quantization Error. 287 . 10.Performance Specifications These are the mathematical expressions which represent the functional requirements. Warping Fig.1 The QFT control system design process: Bridging the Gap. Included are the operational goals of the controlled system and its operating environment Dyn nics Model Mathematical model of the system to be controlled Control Authority Allocation Algorithm to optimize the use of control effectors for efficiency and decoupling OFT Control System Design ' Produces control algorithms for test and implementation Concurrent Nichols Plots Linear Simulation Implement and test candidate control algorithm using Simplified Dynamics t Robust Stability.

For example.e. i. an oversimplified model can lead to trouble as in the case of bending modes as discussed in Sec. a simplified model of the dynamical system can be used to represent the system in the control design process. This model can be as simple as a linear-time-invariant (LTI) transfer function or a complicated set of nonlinear differential and algebraic equations with time varying parameters. In many cases. This control authority allocation is accomplished by the proper selection of the Wy elements of the weighting matrix W. The model is developed from a knowledge of the system and its operating requirements.. 10-2. The optimization of the control effectors' control authority allocation can be used to help decouple the system and assist in achieving the desired robust system performance. from an analysis of the LTI transfer functions a designer may be able to determine their nondominating poles and zeros. based upon knowledge of the real-world operating characteristics of the plant.288 Chapter 10 engineering judgement in order to make the necessary modifications to the specifications that.3 DYNAMICS MODEL The dynamics model is a mathematical model of the system to be controlled. there may be redundant control effectors. i. judgement must be exercised by the designer.5 QFT CONTROL SYSTEM DESIGN As described in detail in this text. by deleting the nondominating poles and zeros from these LTI transfer functions reduced order models are obtained. 10-2. That is. Thus. Not only does a reduced order model simplify the design process. enables achieving a robust control system design. in determining the percentage of the control authority that is allocated to the various controlled variables. the designer should try to use as simple a model as possible that represents the important system dynamics in the design process. the number of control effectors available to the controller may be greater than the number of controlled variables. while still meeting the requirements. In fact. 10-8. but also reduces the risk of introducing numerical inaccuracies in the de- sign process. Also. But remember. In these cases. the QFT design process is used to develop mathematical algorithms that can be implemented in order to achieve the desired . the control effectors available may induce cross-coupling hi the dynamical system and do not clearly control any one variable. those which have a negligible effect on the system's performance (those that he outside the system's bandwidth).4 CONTROL AUTHORITY ALLOCATION An important part of the design process is the control authority allocation assigned to each of the control effectors. 10-2.e. Depending on the dynamical system. a method for determining the percentage of control power available from each control effector to each controlled variable must be determined.

vertical velocity. pitch attitude. control authority allocation. by a computer simulation. If some or all of the specifications have not been met. the dynamics data to better understand exactly what occurs during the simulation.8 ENGINEERING VISUALIZATION After each of the simulations it is valuable to animate. A QFT design can be accomplished by hand using Nichols and Bode plots. but computer software such as Mathematica. the control system must be simulated with each LTI transfer function to check the result against the specifications. 10-2. . These systems are simulated and the results are compared to the specifications. in the case of an aircraft this means that the designer can view the angle of attack. the designer can either redesign the control system or reexamine the requirements. the designer can obtain a clearer understanding from watching the computer animation of the maneuver. and Matrixx greatly simplify the design process. the designer must ensure that the assumed percentage of effector damage is realistic with respect to its associated remaining control authority available to satisfy the control system performance requirements. to still be able to fly an aircraft. they are implemented along with linear representations of the dynamical system. and altitude simultaneously. for example. and/or the percentage of control effector failure). For designs that involve control effector damage. the initial specified requirements may not be realistic. forward velocity. For more specific details of the maneuver the designer can then return to the data plots. As with the linear simulations it may be necessary to accomplish a redesign or a revaluation of the specifications (performance specifications. the designer must ensure that the system performance is close enough to the specifications to meet the overall functional requirements. pitch rate. For example. Also. 10-2. Since QFT design involves linearizing nonlinear equations.6 LINEAR SIMULATION Once the control algorithms have been designed. Implementation issues and insights provided by the QFT process to the designer are discussed hi the following sections. In some cases. Matlab. Note that the three dimension engineering visualizations integrate all of the dynamics of the simulation. Instead of trying to decipher the position and attitude of the aircraft from six two dimensional plots. the simulation complexity is increased by adding nonlinear components and any other components that are removed to simplify the simulation.Design and Implementation 289 control system performance.7 NONLINEAR SIMULATION Once the control system has passed the linear simulation testing phase. 10-2.

10 HARDWARE-IN-THE-LOOP SIMULATION/IMPLEMENTATION At this stage of design and implementation the control system algorithms are implemented on the same type of hardware systems as those that control the dynamical system.12 SYSTEM TEST The final testing of the control system involves implementation on the dynamical system and operational testing. noise corrupted measurements for feedback. Such simulations. a pilot flying an aircraft. Many of these simulations surround the operator with visual cues and some inject motion into the simulation. i.9 ENGINEERING INTERACTIVE SIMULATION When there is an operator involved in the controlled system. for a specified aircraft. this type of control system is referred as a manual flight control systems. Once the controlled system has been shown to meet the performance specifications for the operating environment.290 Chapter 10 10-2. This allows simulation of real-time operation of control algorithm. 10-2.11 OPERATOR-EV-THE-LOOP SIMULATION In order to insure the controlled system meets the requirements of the human operator a simulation is set up to allow the operator to interact with a simulation of the system. Other hardware components such as actuators and sensors are also connected to the system. a successful control design has been achieved. In the case of a control system designed for an aircraft. . the interactive simulation involving a pilot gives the designer the ability to perform a simulated flight test before the design leaves his/her desk. It should be noted that in reality the pilot is a part of the overall flight control system. An interactive simulation provides the designer with the ability to implement the control system in the same fashion that it will be implemented on the dynamical system. and computation cycle time/sampling rate quantization errors. it is often useful for the designer to use an interactive simulation in order to obtain a better understanding of the operation of the system.e. These types of simulations are used to improve the handling qualities of the controlled system by giving the operator a chance to try out the controlled system and then using his or her responses to help shape a redesign. A hardware-in-the-loop-simulation is also useful to ensure that commands issued from the control system move the effectors in the correct directions and the outputs of the feedback sensors have the correct polarity. for example at the WrightPatterson AFB Lamars simulator. Thus. 10-2.. he forms the "outer loop" of the control system. 10-2. The interactive simulation also gives the designer the ability to test the system continuously throughout the operating environment. are often performed by a pilot. for example.

To illustrate some of the real-world problems that are encountered in performing the control system design process are shown in Fig. To design robust flight control systems using the QFT design technique 2.13 REDESIGN At every stage of the control system design and implementation process the designer makes a decision to move to the next stage or to redesign (modify) the control system. To flight test these designs 3. The objectives of the project described in this section are as follows: 1.Design and Implementation 291 10-2. To implement an inner loop PCS on the Lambda URV that would be part of an autonomous flight control system 4. 10.2 is a remotely piloted aircraft with a wingspan/ of 4 ft and is operated by the U. .1.S. Once the control system is modified the simulation testing is repeated 10-3 DESIGN PROCESS EXAMPLE The Lambda Unmanned Research Vehicle (URV) shown in Fig. 10. Air Force for research in flight control technology. A « Fig. 10.2 Lambda Unmanned Research Vehicle (URV).

and take the design through flight test.A refinement of the plant model was made in order to take into account a bending mode that was neglected in the previous designs.1 FIRST DESIGN CYCLE Requirements There were two major design requirements for this project. Overshoot and settling time were chosen to be 25% and 1 sec. Roll rate was an overdamped response that settled quickly.A redesign of the PCS was accomplished to satisfy requirements 1 through 3. Specifications The time response specifications were selected base on the open-loop response of Lambda. Cycle 4 .. Aircraft Model The aircraft model developmental process began with the use of Digital Datcom.Cycle 1 was repeated but involved the design of an improved integrator wind-up limiter. Yaw rate overshoot and settling tune were chosen to be 15% and 2 sees. see Sec 3-4.292 Chapter 10 In accomplishing this design project required four cycles around the control design process loop. for pitch rate response. These specifications were transformed into LTI transfer functions for use in the QFT design. Cycles 1 and 3 were unsuccessful and cycles 2 and 4 produced successful flight tests.This cycle involved the satisfaction of only the first two of the project objectives. These four design cycles are: Cycle 1 . and the settling time was chosen to be one second. The QFT digital flight control system design presented in this chapter is performed as a pseudo-continuous-time (PCT) desiga The resulting ^-domain compensators and prefilters are transformed into z-domain controllers and prefilters by use of theTustin transformatioa 10-3. a computer program which predicts stability and control derivatives for aerospace vehicles based on the physical characteristics of the vehicle. The pitch rate was an underdamped response that settled fairly quickly. This baseline model was refined by using system identification software to estimate the aerodynamic derivatives from actual . respectively. The second was a need for an inner loop PCS on Lambda that would interface with an autonomous waypoint directed autopilot. Cycle 2 .. Datcom information forms the baseline model of the aircraft. but it did not reach steady state as fast as the other two. Yaw rate was also underdamped. Cycle 3 . The first was a desire to develop a robust flight control system using QFT. respectively.

covering the entire proposed flight envelope.80 This simulation allowed the implemented FCS. nonlinearities such as control surface travel limits were introduced into the linear simulation. the FCS was flight tested (Flight Test #1). The second design was based on the DATCOM model with the refinements made with system identification. and control surface saturation. Nonlinear Simulations A nonlinear simulation was developed at the Air Force Research Laboratory (formerly the Wright Laboratory) that incorporated a six degree of freedom simulation. it was discovered that the angular rate sensors had high levels of noise.5 deg/sec. While this design produced the desired responses in the linear simulation. Thus. air vehicle kinematics. to accomplish the design and for linear simulations. which is programmed on a EPROM chip. developed by AFIT (see App. The noise was recorded and was incorporated into the nonlinear simulation. This remedy was an "engineering decision" in order to obtain a satisfactory design. FCSDesign There were two QFT designs accomplished at the Air Force Institute of Technology6579 (AFIT). When the FCS was implemented in this simulation. automatic trim calculation. for designing control systems allowed for a rapid redesign. the allowable gain was modified to achieve a redesigned controller. This information combined with the Datcom information provided a working model for the flight control system design. . with peak values on the order of 0. The first was based on the DATCOM model of Lambda alone. In the third design cycle a more satisfactory resolution of the noise problem was achieved Once simulations of the redesign were satisfactory. Hardware-in-the-Loop Simulation Software from the nonlinear simulation were used to develop a hardware-in-theloop simulation. After successful linear simulations. This second design used linearized transfer functions to represent Lambda in various flight conditions.Design and Implementation 293 flight test data. The FCS amplified this noise and this effectively masked any control command signal. when implemented in the nonlinear simulation the original control system exhibited undesirable behavior due to the initial assumptions about allowable gain being incorrect. Linear Simulations All PCS designs were simulated using Matrixx and LTI state space models representing the full flight envelope of Lambda.78 Maximum likelihood identification was used to identify the natural frequency and damping ratios of the short period and roll modes. A). to be tested in the aircraft. The MIMO QFT computer-aided-design program. The noise problem was minimized by lowering the loop transmission gain and then testing the resulting FCS in the nonlinear simulation.

when the aircraft was commanded to bank. the rudder was commanded to deflect in the wrong direction.12 of Chap. thus. An additional requirement was incorporated for the second design cycle that involved the design of an improved integrator wind-up limiter. . The PCS was thus turned off and the testing involving the lateral control channel was terminated. was used in the second flight test. The individual transfer functions were then implemented. there was no attempt to use a turn coordination algorithm.294 Chapter 10 Flight Test #1 Two major difficulties caused the first flight test to fail.R. 10-3.R. turn coordination logic was implemented around the inner loop PCS that relied on the roll angle. 12. the same QFT PCS designed for the first flight test. 9.2 SECOND DESIGN CYCLE Requirements The requirements for the second design cycle did not change from the original requirements. when the PCS pitch channel was turned on. In this cycle instead of each of the controllers being implemented by a single software algorithm relating their respective outputs to their respective inputs. These transfer functions were individually transformed into the discrete time domain. PCS Design Since the problems encountered in the first test had nothing to do with the QFT designed PCS. the aircraft developed a high pitch rate. the first was reversed polarity on an angle sensor and the second was an integrator wind-up limiter scheme that did not work. This implementation allowed limitations to be placed only on those pieces of the PCS that contained pure integrators and provided the controller accuracy that is indicated by E. Specifications and Aircraft Model The specifications and the aircraft model for the second design cycle did not change from the original requirements. Later. This test was also terminated and post analysis revealed that the scheme used to limit integrator wind-up had caused a numerical instability. the continuous time domain transfer functions were factored into poles and zeros in order to create first order cascaded blocks as illustrated in Fig.6. by their own respective software algorithm. they were implemented in the manner described by E. Post flight analysis of the flight test video and data showed that the polarity of the roll angle sensor was incorrect. during the same flight test. That is. Since the inner loop PCS was to be implemented as a part of an autonomous system. The insertion of an integrator wind-up limiter involved a different form of the controller implementation for the second design cycle. During the second flight test. 9.

3 Response to pitch-down command. Due to problems with the first flight test the PCS was engaged only during the test maneuvers.Design and Implementation 295 Linear. The maneuvers performed consisted of unit step commands in all three axes. setup. This set of maneuvers was first performed with the QFT PCS and then with the open loop aircraft. Nonlinear. 10. and landing. Hardware-in-the-Loop Simulation All simulations consisted of checking out the new implementation of the PCS. the temperature was in the 6CFF+ range with winds at 5 to 7 mph. 10. the QFT PCS performed as it was designed.3. . It is important to note that during this maneuver the aircraft covered a large portion of its dynamics envelope by varying in forward air speed from 75 kts to 110 lets. As shown in Fig.T (lower) 11 12 13 14 15 Time (see) Fig. Lambda was flown in manual mode for take-off. ——— TRuq (upper) ••••••. Flight Test #2 On 20 Nov 92. The figure shows the responses of Lambda to a step pitch down command The dotted lines in the plot represent the specified TRu and TRL. There were no problems encountered during any of these simulations.

was included in the nonlinear aircraft model by recording actual noise and inserting it as a block in the model. Also. by the addition of a hardware noise filter on the output of the sensors. a sideslip command was used to cause the aircraft to intrinsically fly coordinated turns. To do this a template expansion process was developed and is explained in Sec. to about 0. Thus. That is. Aircraft Model The sensor improvement. mentioned above. Therefore.296 Chapter 10 10-3. Changing to sideslip command allowed the use of the yaw rate sensor to implement a yaw damper to reduced the dutch roll mode oscillations. the roll rate response specification was change to match that of the pitch rate. After this change the roll specifications for overshoot and settling time were 25% and 1 sec. The yaw damper was designed and then incorporated in the aircraft model for a PCS design. this aliased noise had a relatively high BW. This caused some modeling errors. This yaw damper was implemented by adding a washout filter. PCS Design Matrixx was used to develop linearized plant models about flight conditions in the flight envelope. The remedy was to place a filter at the sensor output before the sampler. An attempt was made to choose flight conditions in such a way as to fully describe the flight envelope with the templates. the goal of turn coordination is to reduce the sideslip angle to zero during a turn by using the proper amount of rudder deflection during the turn. 10-4. respectively.057sec peak to peak. After the second flight test these errors were corrected through the use of system identification applied to flight test data that resulted in a refinement of the aircraft model. Since Lambda has a sideslip sensor. This cycle involved the design of an inner loop PCS that had intrinsic turn coordination. During the second flight test the pilot felt that the aircraft's roll rate response was too slow. During the system identification work for the second aircraft model. It was determined that the noise originated from a motor on the sensor was a high frequency being sampled at a lower frequency. . the sensor noise problem was reduced by an order of magnitude.3 THIRD DESIGN CYCLE Requirements The requirements for the third design cycle had not changed from the original requirements. Specifications For this iteration of the design a sideslip angle command was incorporated as part of the inner loop controller. designed through the use of a root locus plot. some of the parameters had been scaled incorrectly. This allowed a redesign of the PCS to improve the system performance.

the simulated responses were very similar to the flight test results. Using this procedure. there were violations of stability criteria in the frequency domain and. the porpoising behavior occurred. Nonlinear. Various models were incorporated into the nonlinear model and simulated This simulation used the identical flight test inputs as simulation inputs and compared the simulated outputs to the flight test data. when the PCS was engaged. or porpoising. The PCS was implemented in the cascaded method outlined previously. Instability caused by actuator rate limiting was ruled out by inserting severe rate limited actuator models in the nonlinear simula- tion. but involved a refinement in the aircraft model to incorporate effects of the bending mode discovered in Flight Test #3. was inserted in the model.Design and Implementation 297 Linear. . modeled as a lightly damped pair of poles. Aircraft Model A model of the porpoising behavior encountered in the third flight test was identified by assuming that the behavior was caused by an unmodeled effect. see Sec. a longitudinal bending mode at 73. using the aircraft model with the bending mode. All simulations produced the desired responses to given stimulus. behavior.4 FOURTH DESIGN CYCLE Requirements The requirements for the fourth flight test had not changed from the original requirements. as expected. Note. Specifications The specifications for the fourth design cycle were the same as those for the third cycle. and Hardware-in-the-Loop Simulations The refined Lambda model was implemented in all three simulations.2 rad/sec seemed to be the likely cause. While the post flight test analysis was inconclusive. Flight Test #3 During the third flight test. a violation of the gain margin was ruled out by increasing the inner loop gam in the model and observing the response. the aircraft exhibited an uncontrolled pitching. PCS Design Matrixx was used to develop linearized plant models about the given flight conditions and the PCS was redesigned based on the model containing the bending mode. when the PCS from design cycle three. 10-3. 10-7. When a bending mode.

This design was implemented and all three simulations were run and tested. during onset of the command. The intrinsic turn coordination scheme worked as designed The pilot was pleased with the handling qualities and felt comfortable flying with the PCS engaged at all times. and Hardware-in-the-Loop Simulation A fourth design cycle was accomplished using the new model. During the flight the PCS was engaged and then left engaged for the entire series of tests. the porpoising effect was eliminated. The problem is which operating conditions to choose. Choosing too many LTI plants may yield points that lie inside the template contours and can lead to computational problems during the design. When the data was examined. The PCS performed as designed. altitude. Gross limits were set for these values from knowledge of the . Only those operating conditions that yield points that he on the contour of the templates. More flight testing of this PCS will be required to answer this question. The field conditions were a little gusty. In the case of Lambda the parameters that were varied to set the operating conditions were airspeed. the computational workload and associated problems may be minimized by reducing the number of plants to be utilized in the design process to only those plants that lie on the template contours. Note by applying engineering insights it is readily determined that the template contours and not the LTI plants which lie within the template's contour determine the performance bounds that need to be satisfied by the synthesized functions. the problem was unavoidable. Through engineering knowledge of the problem the designer is able to determine the particular parameters that effect the operating conditions and the physical limits of these parameters. Flight Test #4 The fourth flight test occurred in September 1993. His one criticism was that the roll rate was too slow. These LTI plants determine the template contours. This PCS design simulation responded within specifications and. in some cases. it was found that all of the 60 Hz data had been lost. Lambda's response exhibited more overshoot and longer settling time than specified These problems could be attributable to the gusty conditions. for all frequencies of interest. but. Nonlinear. Since the roll rate was limited by the maximum roll rate detectable by the roll rate gyro. but much of the 10 Hz data had been captured Analysis of this data showed that the PCS did cause Lambda to respond within the specified envelope. and center of gravity. Thus. 10-4 SELECTION OF DESIGN ENVELOPE At the onset of a QFT design. as expected. are necessary. since no gust disturbance was specified during the design process. weight.298 Chapter 10 Linear. but within acceptable limits for the flight test. the designer must select a set of operating conditions in order to obtain the LTI transfer functions that represent the dynamical system that are required for the design.

4. At the end of the process.Design and Implementation 299 aircraft and the possible flight envelope. Identify the variations caused by each parameter. Determine the important parameters that describe the operating condition and their minimum.variation in parameter #2 k . shown in Fig.4 Template expansion process.variation in parameter #1 .extended cases . This can be accomplished by connecting the points on the template due to each parameter variation. 2. is a graphical process that tracks the effect of variations of the parameters which are involved in selecting the operating conditions and determine the resulting LTI plants.4. plot the results of varying each parameter through its maximum and minimum while holding the rest of the parameters at their nominal values. other template frequencies should be checked to insure that a complete set of operating conditions have been chosen. For the template frequency of step 2. The process is as follows: 1. . plot the dB vs. phase values of the nominal operating condition. On this same graph. Choose a template frequency for the expansion process.initial template Phase Fig. 10. • nominal variation cases . 10. . and nominal values. The template expansion process. maximum. the template expansion process was used to find the set of operating conditions that fully described the flight envelope. Next. 3. 5. This forms an initial template. This frequency should be representative of the dynamic system in the bandwidth of interest.

This is accomplished by holding the remaining parameters at their nominal values and plotting the four points of the templates resulting from the extremes of the two parameters. 8. 7. These variations produced an initial set of templates. assumptions were made as to the polarity of feedback and command signals. With this type of simulation the control algorithms can be implemented and the control effectors can be monitored during simulated operation. A situation like this occurs when a control effector has reach its limits.300 Chapter 10 Choose the two parameters that cause the largest variations and use these to expand the template. In order to prevent integrator wind-up. the commanded value must be "integrated" back down to the operational range before it becomes effective. identified an expanded template area of the flight envelope that required more plants for better definition. The other phenomena that a hardware-in-the-loop simulation can identify is . if possible. in steps. as nominal points for further expansion with other parameters. on this expanded template. through maximum and minimum values. each parameter was varied. 1. During implementation these assumptions must be tested.000ft altitude. From this nominal trim flight condition. on the template. On these templates. a weight of 205 Ibs. Each variation when translated. This is one of the reasons to use a hardware-inthe-loop simulation. 10-6 HARDWARE/SOFTWARE CONSIDERATION During the modeling and development of the control system. while holding the other parameters at their nominal trim values.R. The problem occurs when the controller tries to reverse the command. the variation corresponding to each parameter was identified. For Lambda. With the transfer functions separated in this manner individual integrators can be limited. Use the outside points. 10-5 CONTROL SYSTEM IMPLEMENTATION ISSUES 6. and center of gravity at 29. An implementation problem that can cause stability and performance problems is integrator wind-up.12 of Chap. a nominal flight condition was chosen to be 50 kts velocity. Choose other frequencies in the bandwidth of interest to ensure that the operating envelope is completely defined. This is the situation that occurs when the controlled system can not respond quickly enough to the commands from the controller and the commanded values keep increasing due to integrator action. these transfer functions can be separated into first and second order transfer functions (see E. The longer the system is in this state the more the commanded value increases. For implementation. anti-windup algorithms must be applied to integrators during implementation.9% of the mean aerodynamic chord. 9). During the QFT design process the controller is in the form of transfer functions that can be of any order. Feedback signals can be checked by moving sensors by hand.

5 1 1.5 Pitch resonance during flight. the existence of a bending mode was discovered during a flight test. Thus. The goal of turn coordination is to keep the sideslip angle at zero during turns by using the proper amount of rudder deflection. In aircraft.5 3 3. one source of higher frequency modes is structural bending. the controller may need to be redesigned to account for the noise. If the feedback noise is within the bandwidth of the control system. 10-7 BENDING MODES During the design of a control system. Sometimes it is possible to implement a hardware filter after the sensor.5 B Fig. .5 4 4. 10. A control system that excites a bending mode in a flying aircraft can produce disastrous consequences.Design and Implementation 301 the effects of feedback noise on the controlled system. but before the sampler too reduce the noise in the bandwidth of interest. and the noise has not been included in the modeling or simulation. 10-7. the effects of higher frequency modes on stability and performance must be considered. This might result in a trade off between performance and noise rejection. During the modeling process it is very important to include the effects of these higher frequency modes so they can be minimized during the design process. a sideslip angle command was incorporated as part of the inner loop controller.5 2 2. the aircraft operators decided that they would prefer a different feedback structure in the PCS that included turn compensation. to implement turn compensation. In the case of Lambda. Since 0 0.1 LAMBDA BENDING EXAMPLE Following the initial flights.

Pitch attitude data from this flight is shown in Fig. a porpoising behavior was observed. Using this procedure. the longitudinal portion of the QFT PCS was engaged and the sensor data was recorded for further analysis. 10-7. At this point Lambda began oscillations in the pitch axis and the QFT PCS was disengaged immediately. To ensure flight safety. designed through the use of a root locus plot. The yaw damper was designed and then incorporated in the aircraft model for a PCS design.302 Chapter 10 Lambda has a sideslip sensor. When this design was finally flight tested.5 whose high resolution data was at a 60 Hz sample rate.2 UNMODELED BEHAVIOR A model of the porpoising behavior was identified by assuming that the behavior was caused by an unmodeled effect.6 Lambda bending model structure. In order to collect sensor data on this behavior. This simulation used the actual flight test inputs as simulation inputs and compared the simulated outputs to the flight test data. The pilot had Lambda flying in level flight when the longitudinal portion of the QFT PCS was engaged. .626s+!76. Lambda was flown back to level flight. sideslip feedback was used to cause the aircraft to fly coordinated turns. Changing to sideslip command also allowed the use of the yaw rate sensor to implement a yaw damper to reduce the dutch roll mode oscillations. Lambda was flown to a safe altitude by the pilot before the QFT PCS was engaged. 10. Instability caused by actuator rate limiting was ruled out by inserting severe rate limited actuator models in the nonlinear simulation. a violation of the gain margin was ruled out by increasing the inner loop gain in the model and observing the response. This yaw damper was implemented by adding a washout filter.3 s^21.63 Rigid Aircraft Dynamics Plant Fig. Various proposed models were incorporated into a nonlinear model of Lambda and simulated. 10. Bending Mode -20 Elevator Actuator 136.

2 2 The simulated response was very similar to the flight test results. It was possible to excite and observe such a mode by tapping rhythmically on the tail of the aircraft. A bending mode modeled as a lightly damped pair of poles at 13.28* + 174. 10.Design and Implementation 303 Upon reviewing the video record of the flight it was suggested that the aircraft appeared to have a second-order bending mode in the longitudinal axis. Matrixx was used subsequently to develop new linearized plant models containing the r-r-mtt +-I--H-H-* i-i-i-i-ii LL. was inserted in the nonlinear simulation as shown in Fig. 10.i-i. .ii Fig.2 rad/sec.6.7 Lambda bending models. just within the bandwidth of the PCS. This model generated a pitch acceleration signal from elevator deflection which was passed through the second order filter: -20 s + 5.

5 ±y9. and disturbance rejection without exciting the bending mode. the resonance occurred with the PCS that was designed in Design Cycle #3. The new aircraft model was implemented in the nonlinear simulations and tested with both filters.7. . a nondominating s-domain zero is inserted into/w at -150. As expected. stability. The new PCS passed a hardware-in-the-loop simulation and was scheduled for a flight test. there were violations of stability criteria on the Nichols chart. The bending mode is close enough in frequency to the performance bandwidth of Lambda that care needs to be taken to design a controller that will be able to take advantage of the available bandwidth to deliver performance. The standard method of design would be to add a notch filter to keep the mode from becoming excited. With the MIMO QFT CAD program it was possible to shape the loop so that at 5 rad/sec the loop intersected a point on the Nichols Chart where the stability boundary and the performance boundary met. It was also possible to design a fourth-order compensator to replace the earlier fifth-order design.5)(s + ll)(s + 3. As an alternative. It would also increase the order of the compensator. The Bode plots of these models are shown in Fig. The new controller was determined to be: 125(s+ l)(s +2. The new plant models were entered into the MIMO QFT CAD software. Therefore./35. The PCS was redesigned based on the new models using the PCS from the previous design cycle as a baseline.{s).4) s(s + 10)(s + 35±. A standard notch filter would not take advantage of any beneficial dynamics at frequencies near the bending mode. lowering the complexity of the controller instead of increasing it. The PCS resulting from Design Cycle #4 responded within specifications.7) A characteristic of a bilinear transformation is that. in general. with the old controllers.9 + y2) s(s + 2)(s + 80)(s + 36±y48) The MIMO QFT CAD software showed that. it transforms an unequal-order transfer function (ns * ws) in the ^-domain into one for which the order of the numerator is equal to the order of its denominator (nz = wz) in the 2 domain.304 Chapter 10 bending mode about the given flight conditions. 10. The previous controller was: 1093(s+8. This was an optimal point for the loop to pass through given Lambda's performance bandwidth. the inner loop filter was revised to compensate for the new information. This characteristic must be kept in mind when synthesizing gf(s) and fi.

based upon these simplified models.5 5 Fig. 10-8 SUMMARY Control design and implementation in the real world is an iterative process. 10. At any point in the design process. The QFT PCS was engaged and there was no noticeable oscillation. once again. This redesign is followed. 10.5°) recorded at 10 Hz.Design and Implementation 305 I o -1 -2 -3 0 0.5 2 2. After successful testing of the designed control system.5 1 1. At every point of the design process . Pitch attitude during flight.5 4 4. Pitch response during this flight is shown in Fig. 10. the control system must be redesigned and retested on the simplified models. Initial steps are performed with linear models that have been formulated with simplifying assumptions. it is tested on increasingly realistic (nonlinear) models. if the control system does not meet performance and stability specifications.8. During the next flight test.1). Unfortunately.8. but within acceptable limits for the experiment. by testing on the nonlinear model (see Fig. The only problems encountered were some roll performance problems which could be attributed to the windy conditions. the field conditions were gusty. The pilot was very pleased with the handling qualities and felt comfortable flying with the PCS engaged for the entire series of tests.5 Time (sec) 3 3. the test data recording function failed during the flight so that the only data available is low resolution data (±0.


Chapter 10

the designer must be aware of test assumptions so engineering judgement can be used to help guide the design to a successful implementation and operation. The bottom line is that the controlled system must meet the requirements set out at the beginning of the process.



This appendix presents the MIMO QFT CAD package that entails the implementation of algorithms for the design of a robust multivariable control system that, in addition, rejects external disturbance signals. Both analog- and discrete-time MIMO tracking control systems are considered. The CAD package is capable of carrying through a robust control design from problem setup, through the design process, to a frequency domain analysis of the compensated MIMO system. For analog control problems, the design process is performed in the s-plane, while for discrete control problems the plants are discretized and the design process proceeds in the w'-plane. The package performs the following operations: 1. 2. 3. 4. 5. 6. 7. Automates the selection of weighting matrix Discretization of the plants Formation of the square effective plants The polynomial matrix inverse required to obtain the equivalent plants Generation of templates Selection of a nominal plant Generation of the stability, tracking, cross-coupling disturbance, external disturbance rejection, gamma, and composite bounds 8. Loop shaping 9. Design of the prefilter elements.

This is followed by a frequency-domain analysis of the completed design, and export of this design to the MATLAB SIMULINK® toolbox in order to validate the frequency domain design by time-domain simulations. The bound generation routines and graphics have been enhanced. The allocation of the degree of the cross-coupling effects rejection is automatically performed, while in the proc307


Appendix A

ess of generating the tracking bounds. Gain scheduling may be included in the weighting matrix. The improved (second) method (Method 2) may be applied for the general case of a mxm effective plant for both external disturbance rejection and tracking control problems. The MIMO QFT CAD package is implemented using Mathematica and is hosted on Sun Work-station platforms. The CAD package can also be used to perform a QFT design for the special case of a MISO control system. This CAD package is now maintained and continually improved by Professor Michael Grimble, Industrial Control Centre (ICC), University of Strathdyde, 50 George Street, Glasgow Gl 1QE, Scotland (e-mail: m.grimble@eee.strath.ac.uk). A PC version is being developed by ICC.


The CAD package, as outlined in Fig. A1, is a design tool for applying the Quantitative Feedback Theory (QFT) technique to analog and digital multivariable tracking control and external disturbance rejection design problems involving MTMO plants having structured plant parameter uncertainty. For tracking control problems, a MIMO square effective plant Pe with m inputs and m outputs is to be controlled by use of a diagonal compensator G and a diagonal prefilter F in the feedback structure shown in Fig. A2. For external disturbance rejection problems, see Chap. 8 and Fig. A3, a diagonal compensator G is designed such that the system rejects outside disturbances which are projected to the outputs of Pe through the disturbance plant model Pj, as shown in Fig. A.4. The system structure of Fig. A3 can be used for a control problem specifying both tracking and external disturbance rejection requirements. For both classes of control problems the closed-loop system is required to meet the appropriate stability and performance (tracking or external disturbance rejection) specifications.

The design for a continuous-time system is done in the s-domain by a thoroughly defined analog QFT design process. These same procedures, as shown in Fig. A.1, are utilized for discrete-time systems that are described in the w'-domain (referred to in this sequel as the w-domain). As indicated in Block 9, the user selects the analog or discrete design CAD package route. Once F(w) and G(w) are synthesized, they are transformed into the z-domain for implementation by a digital computer. Thus, the following sections, although they refer to the s-domain design, apply equally well to a w-domain design.




poles of A,


Fig. A.1 MIMO QFT flow chart for analog and discrete control systems.


Appendix A

Fig. A.2 MIMO QFT control system with no external disturbance.

Fig. A.3 MEMO QFT control system with external disturbance [/(?) = 0],


Fig. A.4 Partitioning of PF.




Using the QFT design technique, external disturbances applied to the uncertain MIMO plant PF are to be rejected by use of a diagonal compensator G in the feedback structure shown in Fig. A. 3 such that the closed loop system satisfies the stability and performance specifications. Obviously, a prefilter F is not required for pure external disturbance rejection problems in which it is assumed that the tracking command input in Fig. A.3 is zero. Thus, the mjon SISO equivalents, for the case in which md = m = 3, of the mjcm MIMO external disturbance rejection system of Fig. A.3 are shown in Fig. A.5. The plant model Pf is partitioned into two distinct plant models Pe and Pd as shown in Fig. A. 4 for the QFT design process. The plant Pj models the transmission from the external disturbance inputs to the outputs of PF and features only the external disturbance rejection problems. Pd does not affect the closed-loop stability of the m feedback loops in Fig. A.3. The plant Pe models the open-loop transmission of Pr in the feedback loop and takes the place of PF in pure tracking control problems.


When the plant matrix Pis not square, then a square mxm plant Pe is formed from

Fig. A.5 3x3 MISO equivalent loops for external disturbance inputs [r(i)= ff\.


Appendix A

the mxi plant P by use of the (xm weighting matrix W as shown in the block diagram in Fig. A.6 for analog designs, and Fig. A 7 for discrete designs. Even if m = /one may still use the weighting matrix Wfor the purpose of gain scheduling if needed Thus:
P, =PW

The open-loop plant P is, in general, constituted in four component parts. A block diagram showing the placement of the loaded plant model PL, the actuator dynamics TACT, the sensor dynamics TSEN, and the sensor gain matrix WHEN is shown in Fig. A 8. The expression for the plant matrix P of dimension mx£, which in general is not square, is:

The disturbance plant model Pd is in general constituted in three component parts. A block diagram showing the placement of the loaded model Pdl, the sensor

Fig. A.6 The formation of the analog plant Pe.


A 7 The formation of the discrete effective plant Pe.



dynamics TSEN, and the sensor gain matrix JfWis shown in Fig. A. 9. The plant Pd of dimension mxmd is, in general, not square. The expression for Pd is therefore:



According to the partitioning of the disturbance rejection system in Figs. A.3 through A. 9, the bare plant is composed of two transfer function models PL and PJL as shown in Fig. A. 10. These matrices, which may not be square, are loaded in by the designer and placed into the control system structures used by the CAD package shown in Figs. A.8 and A.9, respectively







P •



—3*—— x i

Fig. A.8 Components of the plant P.


Fig. A9 Components of the Plant Pj.

Fig. A. 10 Partitioning of the bare plant model.

sensor dynamics TSEN^S). i. the Binet-Cauchy theorem (see Sec. for these cases. A.6) The QFT design process then proceeds in the s-domain for a continuous design or in the w-domain for a discrete design using exactly the same design steps unless stated otherwise. for each plant case L Pn for plant case i is formed from P.1) For both analog and discrete control problems the plant model P to be controlled (see Fig.e. must have full rank. controllability and ./). each plant Pt(s) is discretized as each Pef is formed To discretize P.p. 1) is in general constituted by the four analog component parts: the loaded plant model PL.. 7-8) is applied to Pei(s) in order to determine whether a m. (minimal) gain scheduling of W may be required to affect a QFT design by allowing a different weighting matrix W.p.4) For a discrete-time control problem. viz.(s) an exact S-transform is performed. not square. The effective plant matrix P.t in the w-plane by an appropriate W.314 Appendix A A-6 FORMATION OF PLANT MODELS FOR TRACKING CONTROL PROBLEMS (SEE BLOCKS 4-13 IN FIG. A. The plant f\s) for the J plant cases i = 1.(w) (A. In some multivariable control problems..: P. (•*) = WSENS. If so.. Since QFT requires a square plant then the square mxm effective plant Pe is formed from the nonsquare mxt plant P by use of the exm weighting matrix W. the analog plant P is embedded in the digital control system by placing a zero-order-hold before the inputs of the plant P and by sampling the feedback signal of the plant outputs. The plant P of dimension mxfis.2.(s) in order to minimize the number of RHP zeros of det P. and the sensor gain matrix ffW(. being m.5) For discrete-time control problems. Tsms 0) PLl (*) TACT (*) (A.J is therefore formed as follows: P. the degree of uncertainty in the plant P may render impossible a successful robust design. in general.. and Wl as follows: P^ = \Pv\ = PiWi (A.s). det PSi(s) is achievable by an appropriate W. Thus. For continuous-time designs.to w-transformation resulting in w-plane transfer functions. followed by the z.(z) -> P.. For discrete-time designs one can apply the BinetCauchy theorem to P. actuator dynamics TACi(s). it will result in all (#.p. It is desirable to select elements of W such that the determinant of Pe(s) is m.

or n.MIMO QFT CAD 315 observability are assumed.p.m. PSi is recomputed. withy.8) The Q matrix elements become the equivalent plants of the MISO loops. character of the effective plants (qu\.7) The equivalent plants are then formed by inverting the elements />. When using a diagonal prefilter. the weighting matrix is revised. only the diagonal MISO loops have a transfer function component due to tracking: ta = tm + tcij (A. i. A nonzero determinant is indicative of full rank The numerator factors of the determinant.: . 1 1) consists of both a tracking and a cross-coupling component. A. = 0 and / *j.*• . one plant case at a time. these determinants determine the m. each MISO loop transmission (see Fig. Thus. 7-8). adi p. * Kl=-JJ-i = {pvy} detPe (A. that is. A. they are not unique to QFT. have a transfer function component due to cross-coupling only.. which are zeros of the qv. are examined as well.e. and have diagonal elements that have the same sign for all plant cases as w -> <x>. If any PSi is unacceptable based on the above criteria. These are conditions that any of the usual LTI design techniques must satisfy (see Sec. and the tests are applied again. A-7 INVERSE OF P. (SEE BLOCK 15 IN FIG. The CAD package also allows the designer to list the determinant of Pe. The CAD package therefore allows the sign of the m diagonal plants to be examined for the J plant cases as a> -> coin table form.1) The polynomial matrix inverse is performed using the Mathematica inverse function: . A-8 MISO LOOPS OF THE TRACKING CONTROL PROBLEM By the principle of superposition.9) Off-diagonal loops..p. (A.

= giqu...10) Expressions for tracking and cross-coupling transfer function components of the (if) MISO loop are explicitly given by: (L.11) eg .. and where Z. The cross-coupling effect input.316 Appendix A Fig.-. a function of all other controlled outputs...12) where the index z specifies one of the J LTI plants.e.?. where > (A.• ^ ' (A.. J. All 3x3 MISO equivalent loops containing both tracking and external disturbance inputs.2. z = 1. i. can be expressed by the equation: .\ (A.

.15) to Eqs. 18). . (8...2. 19)]. 14). is expressed [see Eqs.— -1 I ffl f.. r = 1. and where Z.13) A-9 MISO LOOPS OF EXTERNAL DISTURBANCE REJECTION PROBLEM For external disturbance rejection problems. (8. (8.e. (A.. a function of the external disturbance input and all other controlled outputs (cross-coupling). is given by: t J = . A..19)] by the equation: [P' where [p* pd\ 1 e Pdlij-L. and (8. (8.MEMO OFT CAD 317 (A.J. i.14) the following relationship holds: The SISO loop disturbance input.5) consists of only a disturbance component due to cross-coupling effects and the external disturbance forcing function: The transfer function Td relating the disturbance input of the (ij) SISO loop to its output. = g#s. [see Eqs.18) and (8.12) through (8. X"1 l f y is the (ij) element of the transfer function matrix product Pj'Pd- . ' " = ^ ^ " (A-15) where the index r specifies which LTI plant is being considered. each SISO loop transmission (see Fig. In comparing Eq.

The standard approach of QFT Method 1 assumes worst case conditions and does not take this design information into account. A. then the weighting matrix W can be modified.318 Appendix A A-10 Q MATRIX VALIDATION CHECKS (SEE BLOCK 16 IN FIG. the CAD package allows the Q matrix transfer function elements to be displayed in factored form for any selected plant case. and the Q matrix recomputed. For both tracking and external disturbance rejection problems. A-ll IMPROVED METHOD (SEE BLOCKS 19 AND 20 IN FIG. Also. That is. takes into account any correlation between the uncertainty in the designed MISO loops and the next row of MISO loops for which a design is to be performed. a QFT Method 2 (improved method) design must be used. then a QFT Method 1 design may be used.1) The improved QFT Method 2. the Q matrix elements are tested to verify that the condition of diagonal dominance is satisfied.1) For analog and discrete tracking and external disturbance rejection problems. it cancels nearly identical pole-zero pairs based on a user specified ratio of the distance between the pole-zero pair to the distance of the zero from the origin in both the right-half and left-half plane. The improved method requires the derivation of the q-equivalent plant transfer function for the next MISO loops yet to be designed. Additional tools for examining the equivalent plants qu of the Q matrix include a Bode plot function and a transfer function display subroutine. and where: L *= . The Bode plot for the set of J plant cases is useful for displaying variation in equivalent plant transmission as an aid in selecting representative template frequencies. If the results of this test are not satisfactory. If diagonal dominance holds for all plant cases.17) where the compensator gk for row k of the MISO loops has been designed (Lk is known). Otherwise. The Bode plot for a Q matrix element can be displayed for a specified set of plant cases or for all J plant cases together. A. the new set of transfer functions required by the improved method are generated using the equation: (A. To reduce the order of the Q matrix transfer functions the package performs automatic pole/zero cancellation.

A-12.. The notation k on Lk is not to be confused with the standard notation i on Lt. The notation k.1) A-12. A-12 SPECIFICATIONS (SEE BLOCK 17 IN FIG. The number in parentheses in the subscript of the improved method plants indicates the number of times this method has been applied to generate the plant transfer function. k = i-1 if loops are designed sequentially. i.1 STABILITY SPECIFICATIONS A stability margin is specified for each row of MISO loops..i for «=1. denotes the index of the loop most previously designed.) Improved method plants are generated for all plant cases and for each row of MISO loops for which the compensator has not yet been designed. The stability margin may be specified in terms of the gain margin gm. (A.MIMO QFT CAD 319 ( ilge(n)(like(n) and qua(o) = q& are generated during the matrix inversion of Pe. For example. 17) is obtained after applying the improved method n times. Only the ML contour stability specification is stored in memory..20) For the off-diagonal MISO loops the following upper bound is specified: . (i.e. The two remaining specifications are calculated from the one specification that has been entered. the phase margin angle 7 or the corresponding ML contour.. and qtje(n) in Eq. the plant q^a) is obtained by applying the improved method once.2..J (A. and the improved method was applied after each design is completed. A. as used here.2 TRACKING PERFORMANCE SPECIFICATIONS Frequency domain performance specifications are defined in the form of LTI transfer functions. For the diagonal MISO loops upper and lower bounds are specified as follows: aii <\tli !f<b.e. n compensator elements of the matrix G were designed..

m channels. i. A-12.. Eq. an allocated tracking bounds. i = l. A-13 BOUNDS ON THE NC (SEE BLOCK 23 EV FIG. cross-coupling bounds. 17). new RHP poles will not be introduced in qiie. the s-domain transfer functions yielding the desired performance specifications. and for a template frequency a> = ®. If the sampling rate is not sufficiently high for the above assumption to hold. and gamma bounds when using the improved method The allocated tracking and cross-coupling bounds are generated such that the proper reduction in overdesign is achieved when using the improved method This set of bounds can be replaced by a single composite bound before beginning a design. (8. This constraint is given by: (A. sign changes in the denominator are prevented and new RHP poles are not introduced.2.3 EXTERNAL DISTURBANCE REJECTION PERFORMANCE SPECIFICATIONS For all MISO loops the following upper bound is specified: | tv !„£ by for / *j. (A.23) A unique minimum value sk is specified by the designer for each of the k = 1.33) and (8. . A. several bounds may be included in the set plotted on the NC. By requiring the magnitude of the denominator of Eq. These bounds include stability bound. external disturbance bounds. 17) be larger than a small value 3t. are transformed to the z-domain and then transformed into the wdomain.34) in Chap. (Note that upper case letters for the bounds are used therein.. 8.. the performance specifications are approximated in the bandwidth of interest by making the substitution s -> w..1) For a given row / of MISO loops.) For analog design problems.e.J (A. By proper design of each compensator gk. In the case of a digital control problem...320 Appendix A A-12.. (A. all performance specifications are defined as sdomain transfer functions..4 GAMMA BOUND SPECIFICATIONS The improved method requires the derivation of the effective q plant transfer function.22) This upper bound is determined based upon Eqs.2. .

The stability bounds constrain the maximum closed-loop transmission of the MISO loop with unity gain prefilter to have a bounded magnitude of: (A. one for each off-diagonal MISO loop in the row of MSO loops for which the compensator is to be designed. Each bound is generated based on the constraint: Cij <bi for i (A.24) The bound is plotted for a given frequency by plotting the path of the nominal point while traversing the ML contour with the template generated for that frequency. overdesign is substantially reduced by modifying Eq.27) Lhfk. .25) which is a function of all other cross-coupling controlled outputs.MIMO QFT CAD 321 A-13.1 STABILITY BOUNDS A stability bound is generated for each template.26) When the improved method has been applied. (A. cross-coupling bounds are generated for each template.26) as follows: (I »I.2 CROSS-COUPLING BOUNDS For tracking problems. The specifications dictate that c$ is less than an upper bound for each plant case z: (A.unknown where f^ = 0 when k *j for the case of a diagonal prefilter F required by the MIMO QFT CAD package. A-13. *"'•«- C = \f Lhfkj unknown (A.L.

30) where Lt and Yv are given. A-13. each gamma bound is generated such that Eq. allocated tracking bounds are used to insure that the variation in closed loop frequency domain transmission 4 of the diagonal MISO loop . 17) for an improved method design.. A-13.-1 l + m (l c'i IJ.29) Equation (A. (A30) prevents a sign change (preventing the introduction of RHP poles) in the characteristic equation of the improved method plants' of Eq. (A 17). The bounds are generated such that the correlation between m and q» in Eq. (A. (A29) is properly taken into account over the range of the plant parameter uncertainty which is outlined by the template when the nominal loop transmission does not violate these bounds. (A.25) . respectively. It is desired that the magnitude of the denominator of the effective plant qa calculated using Eq. (A30) is satisfied when the nominal loop transmission Lh does not violate the gamma bound. Given the range of plant uncertainty defined by the templates and variation of yv among the plant cases.322 Appendix A Based on Eqs.4 ALLOCATED TRACKING BOUNDS For tracking problems. not be smaller than a specified minimum value despite plant uncertainty. by Eqs. (A28). i. a lower bound is placed on \1 + Lt as follows: 11 + L | > « ')max I * ] " (A.28) By substituting Z. the gamma bound is generated based upon satisfying: \l + Lk-iy\>sk (A. Thus. where the compensator for row/ is to be designed after the compensator for row / of the MISO loops is designed.(A27). (A.29) is the basis upon which the cross-coupling bounds on Lta are generated. This enhances the ability to design a stabilizing compensator for each successive feedback loop. m A*. (A 18) and (A 19). = 1/m into Eq. Satisfying the constraint in Eq. the latter is transformed such that the bound is plotted on the inverse NC.e.3 GAMMA BOUNDS Gamma bounds are generated for each template.

5 EXTERNAL DISTURBANCE REJECTION BOUNDS External disturbance bounds are plotted for each template.32) where the transmission TR. Variation in the closed loop transmission of the diagonal MISO loop results from uncertainty in the response due to tracking and from the presence of the crosscoupling effects: where t fii and /^ are given.W =l(C * ) ~" r *. by Eqs.Lm[TRmla .35) ' . (A. 1 1) and (A 12). A-13. The disturbance entering the ((/) MISO loop resulting from the external disturbance entering through Pd and from the cross-coupling transmissions is given by Eq. with unity gain prefilter. (A.(< (A. The constraint on Li used to determine a point on the tracking bound is: Lm[TRmiK + (tCii )max] . respectively. : I deij « < [ ? .(tcu )max] < SR (A. (A 16). (A. The specifications dictate that deij is less than an upper bound for each plant case t in the set of /plants. i. the actual variation in ta is less than (tcii)max ..e. 14) and (A 15)].MIMOQETCAD 323 does not exceed the variation §R permitted by the performance specifications. is: TR = -^(A-33) and the most extreme transmission due to cross-coupling effects is: (^.)mar in Eq.34). to be above the bound. By constraining £. only an allocated tracking bound is generated for diagonal MISO loops.34) Because points on the cross-coupling bound (if generated) are identical to those on the allocated tracking bound for the value of (c.1 Ift I (A. ' ? „ ] + I . one for each MISO loop in the row of MISO loops for which the compensator is to be designed [see Eqs.

deleting. = gqu. (say n times so far) the calculation of is modified as follows: by replacing the term |f#|/|<frfc. stability.6 COMPOSITE BOUNDS A set of composite bounds is formed based on any or all of the tracking. g. Since L*. external disturbance rejection.{w) is transformed by a bilinear transformation to g.1) The compensator for an analog system (a controller for a discrete system) is designed to satisfy design specifications for the entire row of MISO loops in which the compensator is used.{w) and g. The composite bound is formed. then one can proceed. or modifying the poles and zeros of the compensator and by allowing adjustment of the gain until an acceptable loop shape is obtained. A-14 COMPENSATOR DESIGN (SEE BLOCKS 24-25 IN FIG. A-13. since robustness will be maintained in the z- . and (b) with (A. The open-loop transmission is shaped by adding. Stability is checked during loop shaping by examining the nominal closed-loop MISO transmission in factored form. The compensator design may thus be performed for an entire row of MISO loops using a single design iteration based on the composite bounds plotted on the NC. in the summation above (a) with bkj^\/Hik(n) which is utilized when the improved method has not yet been applied. for a given frequency. If very close. by retaining the most restrictive portion of all the bounds. All closed-loop s. For a discrete design.324 Appendix A For a Method 2 design in which the improved method has already been applied. cross-coupling effects.{z) are compared for 0 < co < 2/3(cos/2). A.or w-domain poles should be in the left-half-plane.{z). As a validation check. the Bode plots ofg. bounds for all MISO loops are plotted together on the NC. is the same for all MISO loops in a given row.36) when the improved method has been applied to take into account the designed open loop transmission Lk. and gamma bounds.

the maximum and minimum closed loop transmission due to tracking TR with unity gain prefilter. If very close. . gt needs to be modified in order to achieve a stable system for all cases. the Bode plots of F(w) and F(z) are compared for 0 < a> < 2/3(a>s /2) (see Block 29 in Fig. As is the case for tracking bounds on the NC. are as follows: Lm(b'n)-Lm(TRmJ (A. The prefilter design begins with the determination of TRma and 7^ . a portion of the permitted range of variation of 4 is allocated to the cross-coupling effects. respectively.39) Lm(a. A1).1) The proper design of the compensator guarantees that the variation in closed loop transmission due to uncertainty for 4 is acceptable. Thus. lies between the two filter bounds and satisfies tn{s) = 1 in the limit as s -> 0.MEMO QFT CAD 325 domain. As a validation check. proceed with the formation of At(z) = Li(z)l[l + Lt(z)] in order to ensure that all the poles of At(z) are inside the unit circle (see Block 26 in Fig. if the plots are essentially on top of one another then one can proceed.37) (A.40) Once the filter bounds are generated. at each template frequency a>i using Eq. but does not guarantee that the transmission is within the upper and lower performance tolerances a-d and ba. (A 33).i)-Lm(T Rmin ) (A. A-15 PREFILTER DESIGN (SEE BLOCKS 28-29 IN FIG. Next. a prefilter is synthesized such that the Bode plot of the nominal tr. b'n = bn-\(t Cit )max\ On = Un + (tjmax \ (A. If not. since robustness will be maintained in the z-domain. A. A.1).e. The prefilter is therefore required to translate the closed loop transmission 4 such that it satisfies the upper and lower performance tolerances.. i. restricted tolerances are placed on t.38) The filter bounds on the nominal tr. F(w) is transformed by a bilinear transformation to F(z). For a discrete design.

and the mxm diagonal compensator matrix.. whose elements tv are the transmissions plotted on the Bode plots. and Pe..J are plotted on the NC along with the ML contour to validate that the stability requirements are satisfied for each feedback loop /'. and Pd.j.1) The CAD package provides a number of tests to validate that the completed MDVIO design meets the stability and performance specifications for the J plant cases. for all plant cases i = 1. to allow the designer to validate that performance specifications placed on the closed-loop system have been met over the frequency range of interest. the mxm diagonal prefilter matrix. P. The first (Block 27): poles of At for each feedback loop / are checked to validate that all poles of the characteristic equation are stable. The designer can then insert nonlinear elements such as saturation or rate limits. For each off-diagonal ty. based upon the disturbance rejection specifications by. and G are the mxm plant matrix. The model is then simulated to verify that the time domain figures of merit specifications are satisfied. a set of/ Bode magnitude plots are plotted for each ttj.(#„). J Bode magnitude plots are plotted along with the performance bound by.2. respectively. For a continuous-time control problem. an ana- . If some plants are n. an array of the J open-loop SISO loop transmissions (L. and ba..41) where / is the identity matrix. For each diagonal 4. respectively. If no open loop transmission violates the ML contour. one for each of the J plant cases. one cannot rely on the loop shapes on the NC. for the set of J plant cases along with the tracking performance specifications.326 Appendix A A-16 DESIGN VALIDATION (SEE BLOCKS 25-33 IN FIG. The mxm closed-loop transfer function matrix T. and G are the mxmd external disturbance plant matrix. Each Bode plot illustrates the frequency domain transmission t. A.. For the second test (Block 30). MATLAB SEVTULAB models are generated based on the completed design.. For the final validation. For external disturbance rejection problems. is formed for the J plant cases based on the equation: T = [I +P e G] 1 P d (A. J Bode magnitude plots are plotted along with the performance bounds a. For tracking control problems an mxm array of Bode magnitude plots is generated for the mxm matrix of elements 4 of the closed-loop transfer function matrix T. F.m. the mxm effective plant matrix..42) where / is the identity matrix. where: T = [I + PeG]~l PeGF (A.. and add anti-windup protection.p. then the stability specifications are satisfied. and the mxm diagonal compensator matrix.): = #.

a hybrid simulation is performed based on the s-domain effective plants Pn(s\ and the z-domain compensator G(z). . Bound generation routines have been optimized to reduce overdesign. and prefllter F(z). equivalent plant formation.and discretetime control system design based on Mathematica. This includes problem setup.MIMO QFT CAD 327 log simulation is performed For a discrete-time control problem. The design procedure is automated. compensator and prefllter design and design validation for mxm MIMO systems. A-17 SUMMARY The MIMO QFT CAD package has been developed for both analog. The package has been extended to handle in a unified way external disturbance rejection problems as well as tracking control problems.

This CAD package can also be used by students for their basic control theory courses. and a multitude of data displays as the MIMO QFT CAD package of App. Questions in the use of this package should be addressed to Dr. The computer programs called TOTAL-PC is discussed in this appendix as a specific example of a CAD package. which are divided according to gen328 . whereas others are not comprehensive in their capabilities. B-2 OVERVIEW OF TOTAL-PC TOTALai>a2 is an option-number package that reflects a hand calculator or lineterminal environment for interface speed and agility. very high-order plant modeling.wpafb.af. most packages are command-driven interactive programs through the keyboard or a file designation. A. Ewing for updated versions. During the early 1990s the improved version of TOTAL-PC has enhanced the analysis and design of control systems. Some software products include extensive system building capabilities.mil Users of this CAD package should periodically check with Dr. In general. Robert Ewing whose e-mail address is: ewingrl@sensors. and provides an extensive set of control system analysis and design CAD capabilities. Some of these packages require extensive knowledge to use their full capabilities. The disk attached to this text contains the TOTAL-PC CAD package and its associated USERS Manual. nonlinear system construction. This CAD package along with the MATLAB® and the MIMO QFT CAD packages are the ones employed in the text examples. 1).APPENDIX B TOTAL-PC CAD PACKAGE B-l INTRODUCTION TOTAL-PC is ideally suited as an educational tool while others are more suited for the practicing engineer. This FORTRAN package was originally developed at the Air Force Institute of Technology (AFIT) in the late 1970s. This CAD package contains over 150 commands (see Table B.

Air Force Institute of Technology.J.2 Ewing. B.3 Lamont. B. OH 45433-7765. developed at AFIT." M. It divides each of its options (commands) into various groups. Wright-Patterson AFB. 1. Wright-Patterson AFB. June 1992." School of Engineering. OH 45433-7765.S. G. Wright-Patterson AFB.: "Development of an Analog MIMO Quantitative Feedback Theory (QFT) CAD Package.R. Air Force Institute of Technology. B. The TOTAL QFT CAD package.4 Sating.33 for MISO systems was designed as an educational tool as illustrated in Fig. B. has accelerated the utilization of the QFT technique for the design of robust multivariable control systems. School of Engineering. A. It contains the conventional analog and discrete control system analysis and design options and the QFT design options. R. thesis.I Larimer. AFIT/GE/ENG/92J-04. Ohio. 1978. OH.TOTAL-PC CAD 329 eral functional categories.: "An Interactive Computer-Aided Design Program for Digital and Continuous System Analysis and Synthesis (TOTAL). B-3 QFT CAD PACKAGE Up to about 1986 there were essentially no Quantitative Feedback Theory (QFT) CAD packages designed specifically to assist in doing a complete QFT design for a control system. Wright-Patterson AFB. .B. Graduate School of Engineering.1986." MS Thesis. R. Air Force Institute of Technology.1992.: "TOTAL/PC. P Street. P Street. Air Force Institute of Technology.: "ICECAP/QFT. GE/GGC/EE/78-2. S." School of Engineering. In 1992 the MIMO QFT CAD package8A of App. REFERENCES B.

BD(jW).330 Appendix B 10 For each plant transfer P^Cs) into GTF G(s) into HTF Form TDi=CLTF GTF 1 + GTF* HTF 3 Denote nominal plant. B. F(s) Form G(S)=LO(S)/PO(S) Fig. obtain template data Print out J plant transfer functions Form Templates Plot time and frequency responses 11 ofYDi(s) 12 4 For each plant form OLTF=GTF*FiTF form TD|=CLTF = GTF 1 + GTF* HTF 6 Obtain bound data: BR(jo>). . 7 Gooff-line Loop Shaping to synthesize L0(s) synthesize F(s) 8 Go Back on-line enter Lo(s).B0CJa>).1 CAD flowchart forMISO analog QFT design.

TOTAL-PC CAD 331 Table B.s) from continuous state space model .I Some TOTAL-PC options No. Option Transfer-function input options 0 List options 1 2 3 4 5 6 Recover all data from file memory Polynomial form-GTF (forward transfer function) Polynomial form-HTF (feedback transfer function) Polynomial form-OLTF (open-loop transfer function) Polynomial form-CLTF (closed-loop transfer function) Factored form-GTF 7 8 9 Factored form-HTF Factored form-OLTF Factored form-CLTF Matrix input options for state equations 10 List options 11 12 13 14 15 16 17 18 19 AMAT-Continuous plant matrix BMAT-Continuous input matrix CMAT-Output matrix DMAT-Direct transmission matrix KMAT-State variable feedback matrix FMAT-Discrete plant matrix GMAT-Discrete input matrix Set up state space model of system Explain use of above matrices Block diagram manipulation and state-space options 20 21 22 23 List options Form OLTF=GTF*HTF (in cascade) FormCLTF=(GAIN*GTF)/(l+GAIN*GTF*HTF) FormCLTF=(GAIN*OLTF)/(l+GAIN*OLTF) 24 25 Form CLTF=GTF+HTF (in parallel) GTF(s) and HTF(.

impulse. continuous F(t) and discrete F(kT) 30 31 32 33 34 35 36 37 38 39 List options Tabular listing of F(t) or F(kT Plot F(t) or F(kT) at user's terminal Printer plot (written to file answer) Calcomp plot (written to file plot) Print time or difference equation [F(f) or F(kTj\ Partial fraction expansion of CLTF (or OLTF) LIST T-PEAJC. T-RISE. pulse. ramp. final value Quick sketch at user's terminal Select input: step. T-DUP. M-PEAK. T-SETTLING. sin off Root-locus options 40 41 42 43 44 45 46 47 48 49 List options General root locus Root locus with a gain of interest Root locus with a zeta (damping ratio) of interest List n points on a branch of interest List all points on a branch of interest List locus roots at a gain of interest List locus roots at a zeta of interest Plot root locus at user's terminal List current values of all root locus variables Frequency-response options 50 51 52 53 54 55 56 List options Tabular listing Two cycle scan of magnitude (or dB) Two cycle scan of phase (degrees or radians) Plot F(j'co) at user's terminal Create GNUPlot-Frequency plot Create GNUPlot-Root locus plot .332 Appendix B 26 27 28 29 GTF(z) and HTF(z) from discrete state space model Write adjoint (*sI-AMAT) to file answer Find HTF from CLTF & GTF for CLTF=GTF*HTF/(1+GTF*HTF) Find HTF from CLTF & GTF for CLTF=GTF/(1+GTF*HTF) Time-response options.

Create GNUPLOT-Nichols Log magnitude/angle plot Chalk Pitch Axis HQ Criterion Analysis Polynomial operations 60 61 62 63 64 65 66 67 68 69 List options Factor polynomial (POLYA) Add polynomials (POLYC=POLYA+POLYB) Subtract polynomials (POLYC=POLYA-POLYB) Multiply polys (POLYC=POLYA*POLYB) Divide polys (POLYC+REM=POLYA/POLYB) Store Polynomial (POLYJ into POLYD Expand roots into a polynomial (s+a)n Expansion into a polynomial Activate polynomial calculator Matrix operations 70 71 72 73 74 75 76 77 78 79 List options ROOTA=eigenvalues OF AMAT CMAT=AMAT+BMAT CMAT=AMAT-BMAT CMAT=AMAT*BMAT CMAT=AMAT inverse CMAT=AMAT transposed CMAT=identity matrix / DMAT=zero matrix 0 Copy one matrix to another Digitization options 80 82 83 84 85 86 87 89 List options CLTFO) to CLTF(z) by first difference approximation CLTF(s) to CLTF(z) by Tustin transformation CLTF(z) TO CLTF(s) by impulse invariance CLTF(z) TO CLTF(/) by inverse first difference CLTF(z) TO CLTF(s) by inverse Tustin Find FMAT and GMAT from AMAT and BMAT CLTF(X) to CLTF(Y) by X=ALPHA* (Y+A)/(Y+B) .TOTAL-PC CAD 333 57 58 59 Tabulate points of interest: peaks. etc. breaks.

FTP.) List special commands allowed in option mode List variable name directory List main options of total Print new features bulletin Augment AMAT:[AMAT]=[AMAT] J3AIN*[BMAT]*[KMAT] FomiOLTF<IX*aJSrajy/[<XDP^ (Integral of (CLTF SQUARED))/2PI= Double-precision discrete transform optionst 140 141 142 143 144 145 146 147 148 LIST OPTIONS CLTF(s) TO CLTF(z) (IMPULSE VARIANCE) CLTF(s) TO CLTF(w) W=(Z-1)I(Z+1) CLTF(s) TO CLTF(w') W =(2/T)(Z-l)l(Z+l) HI-RATE CLTF(z) TO LO-RATE CLTF(z) HI-RATE CLTF(w) TO LO-RATE CLTF(w) HI-RATE CLTF(w') TO LO-RATE CLTF(wf) OPTION 144 (AVOIDING INTERNAL FACTORING) OPTION 144 (ALL CALCULATIONS IN Z PLANE) tOption 37 is not available for the discrete domain. TDTF.334 Appendix B Miscellaneous options 90 91 93 96 97 98 99 119 121 129 LIST OPTIONS Rewind and update Memory file with current data List current switch settings (ECHO. TLTF. Quantitative Feedback Technique (QFT) 150 151 152 153 154 155 156 157 158 159 Define-Plant List options Define-TLTF: (Lower Tracking Ratio) Define-TUTF: (Upper Tracking Ratio) Defme-TDTF: (Disturbance) Select-Nominal plant: (1-10) Define-FTF: (Pre-filter function) Define-LOTF: (Nominal loop function) Display Plant. TUTF. etc. ANSWER. LOTF QFT design/report macro .

(4. Block 6: For this block the following steps are involved: 1. Boundary transfer functions: TRu (s). Block 1: The input s-domain data consists of: 1. (4. Transform Pei ( z ) .2 present a QFT CAD flowchart for the s-. Block3: Formes) = P^/s [see Eq. Variable plants: P. 1 and C. and TD (s). 2. to Pe..or w-domain design?" Blocks 2 and 27: For the analog s-domain design see App.. 4-6. Block 2: "Is this an s. This appendix. TRl (w). 335 . TRl (s). and z-domain designs.AND z-DOMAIN) Figures C. 4.J] Block 4: Obtain P.42) where r = 1.{z) [Eq. and TD(w) by replacing s by w in the boundary transfer functions of Block 1.(s).41)] Block 5: Form TRu (w). (w).2. which complements Sec.APPENDIX C TOTAL-PC: DISCRETE QFT DESIGN PROCESS C-l DISCRETE QFT DESIGN (w'.. The following descriptions deal with the w '-domain design procedure of Chap. via z = (Tw + 2)l(-Tw + 2). B.. w-. provides additional information for each of the blocks in the flowchart as follows.

w-. 1 QFT simulation flowchart for the s-.336 Appendix C Fig. and z-domains: Part 1. . C.

.(z) factors Relocate troublesome LHP poles of G((z) closer to origin in unit circle forming G. and z-domains: Part 2.Discrete QFT Design 337 Plot time response. YDi (KT). YB (t) Plot time response. w-. YD^ (t) Plot time response. C2 QFT simulation flowchart for the 5-. YRi (KT) and frequoicy response Edit and generate report or QFT design (Y/N?) Show unstable Q. and frequency response CLTF = GTF/Cl + GIF HTF>=T Plot time response.m(z) Fig.

3. and B^Q'v.118)] . 4.. (4. 4. (w) = L0 (w)/PZo (w) [see Eq. Obtain data to construct the templates. and F(z) for the frequency range 0 < vt < v24 corresponding to 0 < a> < o)24 where co24 is the value of the frequency at the -24 dB point on BL(y) plot (see Fig. 2. Form/. Block 8: Reenter TOTAL-PC 1. The templates are constructed and plots for BR(/VJ). (4.90)] Block 11: 3.ON obtain G. Block?: Exit from TOTAL-PC At this point of the design process the designer exits the CAD package. and the Bs contours are obtained.(\v\ F(w). With CANCEL. Denote the nominal plant P<.) bounds. 3.).70) and (4. Enter Lmo(w) and F(w).74)]. 2. Next the optimal loop transmission transfer function Lmo(w) and the prefilter F(w) are synthesized.9).(z). Via bilinear transformation obtain G. A(w) is formed automatically Block 9: 1.o(w).110)]. Block 12: 3.(z) and F(z). Bh. G. Block 13: [See Eq.. Form PZi (w) = Pei (w)[2w/(w + 2/T)] [see Eqs. Ensure the nominal plant PZl = PZa.338 Appendix C 2. Block 10: 1. = ~Lmo(w)A(\v) [seeEq. Obtain Bode plots and data for G. (4. BD(jv. (4. and the BJ. Based on the utilization of the QFT options redefine TSAMP=0.

(joi) | Block 15: With CANCEUON 1. (4. 3. 2. Block 14: 1.40) and (4. (z). Block 19: 1. Obtain the time response data to determine the figures of merit (FOM) and the plots for yDi (?) . (w)].ON and form Q. Enter PZi(w) asGTF 2. Obtain the factors for each Q. Set CANCEL. Form OLTF = GTFxHTF [= L.(z) P2. Block 16: 1. Obtain the frequency response data and the plots for | 7b. Set TSAMP = T and obtain Pz. Enter G.Discrete QFT Design 339 1. The determination if all J control ratios (tracking and disturbance) are stable. Form FLTF = F(w)[OLTF/(l + OLTF)] [= TRi (w)] [see Eq.106)]. Obtain the time response data to determine the FOM and the plots for .(4. . if not proceed to Block 25. Obtain the frequency-response data for the plots of | TR: (ju) | Block 17: 1.ON obtain TD. If alU plants yield stable responses proceed to Block 20. With CANCEL. (t) • 2.(z).(w) as HTF 3. Block 18: 1 2. (z) [see Eqs.102)].(w) = GTF/(1 + GTFxHTF) = CLTFfor each plant i.(z) = 1 + G. Set forcing functions /•(/) and d(t) to a unit step forcing function. 2.VR.

. 6.(z). Enter Pz. Obtain the frequency response data for TDi and T R i .ON obtain CLTF [= GTF/(1 + GTFxHTF)] [=TD. Plot time and frequency responses for YD. 7. 4. or 20-24? 8. With CANCEL. Note that if a printout is desired. Block 22 through 24: With CANCEL.(4. Block 26: Relocate the pole(s) on the negative real axis of Gj(z). (kT).6. FormFLTF = F(z)[OLTF/(1 + OLTF)] [= TRa(z)] [seeEq. the user is allowed to edit data before printing. 3. that result in the closedloop unstable pole(s). 5. 2. Set forcing functions r(t) and d(t) to a unit step forcing function. Query: Do you want a printout of input and calculated data of Blocks 1 to 24 or 1-19. (4. Repeat Blocks 19 through 24 and exit fromTOTAL-PC. 4-5.112)]. If not. Block 25: From Block 18 show all unstable factors of Q. Form OLTF = GTFxHTF 2.ON 1. Block 21: 1.(z)} [seeEq.340 Appendix C Block 20: 1. the user is allowed to delete zero(s) and pole(s) that justify cancellation. Enter G/(z) as HTF. 25.(z) asGTF.(kT). A printout of OLTF is given in order to determine if "proper cancellation" has occurred. 26. Exit from TOTAL-PC.113)]. 2. Obtain discrete domain time response data in order to determine the FOM and the plots for }>R. closer to the origin within the unit circle as discussed in Sec.

The nominal plant is: P0(s) = Vs D-2 PROBLEM Design L0(s) and F(s) to meet the specifications.a = G) 341 .APPENDIX D MISO DESIGN EXAMPLE D-l MISO DESIGN PROBLEM Given the plant transfer function: P(s) = s +a The specifications on \TK(jaJ)\ is given in the following table: CO 0 0 0 0.5 1 -2 1 1 2 0 5 -4 10 -10 (-40) oo 20 Lm BU(JCO) Lm BL(j(o) -20 (-80) oo 4 8 20 where ML < 3 dB. A > 3 and there is no specifications on TD. (k=l.

Ps(0..l).).. D.J.(-3. 1 where V = 40 dB and as a> increases the templates become narrower. <mdP6(3. where P. used to obtain the templates are: ..I Template . It is left to the reader to obtain the templates in order to verity the solution to this problem.(aDk. D.100).WO).. Fig.342 Appendix D D-3 SOLUTION D-3.1 TEMPLATE GENERATION The set of J plant transfer functions P. P. 2.100 The templates all have the shape shown in Fig. P4(-3.l). e £Pand l = 1.

The designs at the onset should be based upon a qualitative analysis of L(s) = G(s)P(s) = —— G(s) s +a (D.4) results in .I) and kG(s) Lo (s) = Po (s)G(s) = —— (D.( ) L(s) = ——^—^±-1- (D. A possible design of an L0(s) is shown in Fig. (D.I) and (D. (D.A Type 0 L0(s) function requires the compensator to be of the format of: G(s)= Ks( ). respectively: kK( )..( ) (D. Eq.4) and When a * O.2. D.2 DETERMINATION OF BOUNDARIES B^/co) The templates are used to determine the boundaries BR(ja>^ and..MISO Design Example 343 D-3.2.3 DETERMINATION O¥L0(s) The design of an optimum L0(s) is not unique. D-3.. D.3) which from Eqs. to determine the [/-contour as shown in Fig. along with the MLcontour and the value of V.2) Trial 1 .2) results in..

D. .344 Appendix D -200° -180° -160° -140° -120° -100° -80° -60° .2 Plots ofBdijcOi) and L0(j'co).0 -20° 4 ° Fig.

The synthesized loop transmission function is kK( ).2.. F(s) (D...7) which from Eqs.( ) . and results in 8. Trial 2 . respectively: kK( ). D. 1) and (D.(f).9) When a = 0. .504x io6 (s + l)(s + 12)0 + 120) G(s) = —————— ————-———£———— '.— (s + 2)0 + 10)0 + 26.c(co) = 0.. (D. for r(f) = u...A Type 1 La(s) requires the compensator to be of the format: (D. kK ( (D.( ) L(s) = ——^^-^^ (D..oodB which is unacceptable.2) results in.._. then the compensator GO) must be a Type 1 transfer function in order to yield a Type 1 system for a •* 0 and a Type 2 system for a = 0.6) and Lm TRl(ja)) = .. .8) and .MISO Design Example 345 TRl <*) = ———- ).5)0 + 720 ± J960) (D. as shown hi Fig. TK(s) corresponds to a Type 1 system and when a * 0 TR(s) corresponds to a Type 0 system. Assuming this is acceptable an L0(s) is synthesized.10) In order to achieve the desired specification e(ao) = r(oo) .

The simulations indicate that all the specifications on \Ts(jai)\are met except. (D. based upon Eqs. when a * 0 and K> = 0.6. The reader can simulate the system utilizing his L03 0).4 DESIGN OF F(s) Based upon the design procedure in Chap. . It is left to the reader to complete the synthesis of a L0(s).5 D-3. D-3. the following nonunique filter is obtained: 1.2. D...346 Appendix D for which three points (denoted by X) are shown in Fig...( ) 5 + 1.5 SIMULATION OF DESIGN A simulation of the system for L0(s) is made for each /*.5 kK( ). 10) and (D. where * = 1.. 3.(«&). as expected.. 12)..

that 1n<l22<1l2<l21 => 2<2 '5 Specifications: (a) There is only one command input: r2(t) (b) Lm ta < -20 dB for all co (c) For Lm t21: 347 .5<ki2<2 0. 5-7) requires.5</t22<l The diagonal dominance condition (Sec.APPENDIX E MIMO DESIGN EXAMPLE E-l DESIGN PROBLEM Given: A 2x2 plant where _!_ s (E. as co -> <x>.1) has the following independent structured plant parameter uncertainties: \<kn<2 5<k2i<10 0.

55) thru (5.3) E-2 PROBLEM Design Llo (s). and thus Lm[a22(jQ )]«-«> 0>5 (d) For LI = giqu and L2 = &#*.36) thru (5. (s) DESIGN For loop one design let/L? = 0 (i. Thus. With/. from Eqs. = 0 then the loop 1 design becomes strictly a cross-coupling rejection problem (see Fig 5.58). decouple output 1 from input 2). where for an impulse input 4 = y0: (E. respectively: (e) Lm MLl Li <3dB (E..348 Appendix E CO Lm b22(j(o) Lm a22(ja>} 0 0 0 0.1 1 2 1 0 -4 8 5 -4 20 10 -10 (-W) oo 20 -20 (-80) oo The values of a22<j(o) are considered to be zero for a> > 5.46) and Eqs.12). (5. L2f E-3 SOLUTION E-3. (5.2) <3dB (E.4) tl2~- giqu Based upon the given disturbance rejection specification .e.1 Li.5 1 -2 Table E..

7) Equation (E.5) 1+ which is rearranged to giqn \> (E.6) to yield (E.1 Since C12 = ~t 22/112 (E.2 Thus. As a consequence the cross-coupling bounds BC1 (jO. E. this constraint imbeds the cross-coupling rejections bounds specifications on ya(f).1 Equivalent MSO loop one.6) 0.9) 9.8) Applying the maximum tracking specification value for \t22 yields: (E. (E. (E.) become the com- .7) is substituted into Eq.MEMO Design Example 349 C12 O o—>—o—>—o §1 V 1 -1 Fig.

(E. 11) by qn yields: (E.) (E.5 s Eq. Therefore. (E. for 0.11) Multiplying both sides of Eq. for lower values of on. <3dB and the uncertainty in qn at o = oo. E. Thus. (E.12) 9l2 where Llo = 8i<lllo-.350 Appendix £ posite bounds B01 (y 0.4. from the specification LmMi. The nominal plant q. The template is simply a 6 dB vertical line.9) can be simplified to: (E. is chosen to be the i'* plant which is at the bottom of the 6 dB template. The worst case scenario must be chosen from the specifications in order to apply Eq.13) or Lm L]O >26 + Lm b22 = B01 (jO.10) Generally. the universal high frequency boundary can be drawn as shown in Fig.14) . (E. Also.I2)becomes \Llo\>\20b22 (E. since ]giqu\ » 1 then Eq..) For the plant uncertainty a template can be generated for qn which is good (for this particular design problem) for all a>. 12). (E.

see Fig. the tracking frequency range is effectively 0 < co < 5.2 (Ol 1 Lm B0(jw^ 26 0. An initial simple Llo (s) design containing only one pole at the origin. Based upon Eq. and the nominal qllo are Type 1 functions then Llo (s) should be chosen as a Type 1 or higher function. E. one complex-pole pair. (E. > 5.58). (5. Since the/ty elements of P. It is therefore necessary to shape the loop transmission Llo (j&) only for 0 <a><5. E. (5.36) thru (5.3 and Eqs. and in turn all the elements of Q. Note that the approximation made in Eq.y(s).12) is justified because all the values in Table E. (E.16) E-3. and one zero resulted in (0^ = 65 rps.39) and Eq. A Type 1 function is chosen for Z. 1 the bounds are: TABLE E. 14) reveals that for this example the composite bounds B010'O) are straight lines whose magnitudes are a function of frequency.15) where a>n = 140 rps and is shown in Fig. The value Lm a^ai) yields <%/a>) = Lm bn(ja>) — Lm a^jai) —> oo dB for o) > 5 and the tracking problem will always be satisfied in this frequency range.2.2 are at least a magnitude above one. In order to reduce the value of the phase margin frequency the following Lla (s) is synthesized: 54038Q + 25) (E. are .MIMO Design Example 351 Thus Eq. ~Si9na then Llo(s) _ q (s) 540838(^ + 25) (s+10)(a+84±jll2) (E. (s) The equations for loop 2..5 27 1 27 2 26 5 22 10 N/A 20 N/A Since the value of b22(ja>) is very small for a> > 5. 14) and the given specifications on tn in Table E. (E. Only the cross-coupling rejection problem remains for a.2 DESIGN (IMPROVED METHOD) FOR L2. Since .

352 Appendix £ (E. see Eq.4). (ja>) design.2 The L!r. .17) where c22 = -tJ2/q2J and (E. (7. E.18) Instead of making the necessary substitutions and performing the associated algebra the following generalized loop 2 equations. Fig. are utilized in the design.

i 12 ^2e ^2^22t (V t1\ (JL. The knowledge of the presence of RHP poles is necessary in order to interpret correctly the data for \922e(J®')\ and ^22e 0/0 / ) over the frequency range of interest (see Sec. Before generating the templates for q22e it is necessary to ascertain whether all the poles and zeros of q22e ue m me LHP s-plane over the region of uncertainty. This possibility must be verified before proceeding with the design.21) and "—— = 0 (E.This permits the loop transmission plot to "squeeze" through and achieve the desired value of 7 = 45° (for this example).MIMO Design Example 353 (E.11). Plant uncertainty case 10 of Table E. 1). The proper design of L.p. Thus loop 2 is only a tracking problem (see Fig. = 0. plant is assumed. That is. it is possible for 7 + Lj(s) .p.. and for & = 20 rad/sec it is 6 dB in height. This analysis of the plant template data reveals: .3 represents an unstable plant. At low frequencies the templates are rectangles 6 dB in height and essentially 360° wide. 345° wide at the top and 352° wide at the bottom. the template for co = 0.20) to have RHP zeros which become RHP poles of q22e (•?) and therefore of L2e (s). 18.(s) guarantees that the zeros of / + L.(s) are in the LHP. then q22e (s) will also be m.yn(s) of Eq.22 of Ref. In order to assure a negative phase angle at all frequencies for L2e (jo>). The templates for co = 130 rad/sec and co = 150 rad/sec are approximately 180° and 138° wide respectively. Since an m.22) since/. (E.5 is essentially 6 dB in height and 360° wide. it is necessary to generate the templates at high frequencies such that they shrink sufficiently in width in order to allow a gap on the right side of the ^/-contour for L2e (j®).20) 1 + Ll . 5. However.

5 0.5 0.5 2 2 kn 5 5 10 10 5 5 10 10 5 5 10 10 5 5 10 10 k* 0.5 0.75 0.25 1 1.5 1 0.5 0.5 1 0.354 Appendix E 0 §2 O—^ 7? -o -o -1 Fig.5 0..5 1 0.5 1 0. 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 k12 0.5 2 2 2 2 0.unstable plant #10 2 2 <.5 2 1.5 . E.5 1 0.75 0.3 Loop 2 MSO equivalent.5 1 <.nominal plant #15 Random points 17 18 19 20 1.5 0.5 0.75 1.5 1 0.5 1 Comments 0.25 7 9 6 8 1 0. Table E.3 Combinations using plant maximums and minimums Case 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 k.75 1.

E.4. that in going from co = 0.[l + y (0/a)] and Z.4 at "high" frequencies. These unstable plants occur for this example. E.5 to a> = 300 rad/sec the templates start to "droop" down to the right. at "low" frequencies to the shape shown by the shaded area in Fig.yl2) = -1 for which a root-locus analysis can be made as the value of Y12 is varied The "very wide" characteristics of some of these templates can best be appreciated by the reader by plotting a positive and negative real pole (a and -a) in the s-plane.4.5) that play the key role in achieving the desired value of yis given in Table E. As shown in Fig. This droop becomes more exaggerated if more template points are plotted from other unstable plants within Q. This can be seen be partitioning the characteristic equation of Eq. The templates are used to obtain the BRe (jw ) = B0e (j&i ) bounds of Fig. E.20) to yield L. Although making this assumption yields a overdesign. In order to simplify the process of generating the bounds rectangular (solid curve) templates are used as shown in Fig. A stable plant is chosen to be the nominal plant. The reader is then urged to analyze the angular contribution of Z. 2. Note that for < 750 rad/sec the bounds are considered to be essentially straight lines.MEMO Design Example 355 1.4. (E. That the width (change in angle) of the templates is the crucial factor in the shaping of L2e of the template. E. The data for the templates (see Fig. Note that at a> = 500 the template is 20 dB in height which the value of F(UHFB) used to obtain the f/-contour.a)] as the frequency is varied from zero to infinity. E. as shown in Fig.6 for loop 2 design. see .23) 1 kl2 is satisfied. That the template shapes go from being "very wide" (essentially rectangular). when the condition (E. That is.[l + j(0 1. these bounds are represented by the straight line tangent to the top of the Lm ML contour since all the templates for co < 150 rad/sec are greater than 13f in width. E.4.l(l .

24) obtain . are k. and is located at the lower right-hand corner of the templates.5 15. E. = 2.05. for the nominal plant.8 15 16. 15).6 18 138 81 61 37 10 20 -0 Table E. where yl2 = 0. 19) through (E. and k22 = 0. 2 =0. (E. = 10.24) thus from Eqs.22).4 AdB A0° 150 170 180 200 300 500 14. Table E. k. the plant gain values kv.r . Thus.2 = 2.3. (E.4 Template shapes.356 Appendix £ Fig.95 (E. k2.. For these nominal values. let a = l. and (E.5.

E. .955* _ 0.0882)(* + 39. 2 +(1960000 +540838&.95 k.5263158(/ +178 / + 21280 / + 1277676* + 27041900 *3 +178*4 +21280*3 +667303*2 + 14232578.9674)(s' + 71.26) Fig.5263158(5 + 49.9859) s(s + 17.7651 ± J96.5 Tracking templates for loop 2 (using a rectangular approximation). 0./)* +13520. (E.21) which results in a nonunique g2 being designed rather than L^ where (E.2169) (E.25) In order to make g2(s) "fairly simple" the nominal plant is used in Eq.2349 + j6.6085 ±j!01.MJMO Design Example 357 + 178 / + 21280s2 + (196000 + 540838jt 13520950 jt.3915±j24.

the phase angle is due essentially to the nominal plant. E. for this example.e. the approach in this example is to design g2(s) so that L2e (s) crosses the first few bounds at a phase angle of approximately -90°.1 dB). the location on the NC where these templates are placed result in small positive values for Lm t2(< 0.6. A second design for g2(s) can be made by choosing smaller values for the real pole and zero of Eq.8 rad/sec) of the nominal plant. In this frequency range. the templates are between 345° and 360° wide.28) whose magnitude lies on B0. i. (E. for each value of <y. E-3.38) which causes the "rapid" change in the phase angle. At2 = Lm t2ma . For example.27) where ah = 7000 and £ = 0.M) crossing for w < 150 rad/sec. (. a more complicated g2(s) may be synthesized that approaches the optimal loop transmission function Li (JO. The resultant g2(s) is: g2 (s) = ——————————-—————-——— 3.. in order to obtain a simpler g2(s). g2(s) does not contribute very much to the phase angle at the B0. However.e. and whose phase angle /LO(ja>i) lies on the right side of the f/-contour.27) in order to have L2e (s) track down the right side of the fAcontour more closely.3 PREFILTER/22(s) DESIGN The prefilter design procedure of Chap. (J<°} being on the corresponding B0e (ja>i) bound.Lm t2mln will be a very small number in the range 0 < co < 20 rad/sec. i. As a consequence. 3 requires the use of the templates to obtain the values of Lm t2mm and Lm t2min where t2 = LJ(1 + L2) over the frequency range 0 < co < 20 rad/sec. As it turns out.358 Appendix E A synthesized (nonunique) L^ (s) is shown in Fig..3155x1010(^ + 1000) (s + 500)0? + 3500 + J6062.5..1778) (E. The hump in L2f in the vicinity of a> = 150 rad/sec is caused by the complex zeros (a>n = 104. Thus. for this example. the bottom of the . In general. small variations exist for Lm t2. (j® / )• . These zeros have a low damping ratio ( £"» 0. Usually a complicated g2(s) results when adjusting /^ (ja>) so that it crosses the B0e (J150) bound near the right side of the NC with L2. V = 0(jO) = | 0(jO) | Z0C/0) (E.

for co = 20. 3. Lm TRu . (E.000 1. for three cases from Table E. These cases are chosen on the following basis: Table E. would be placed on the Lm LIO curve at the Lm L^ point to yield Lm t2mm w 0.19709 Final value 1. for a> < 20 Ati is a very small number and \t2\ ~ 1 for 9 = {P}.The loop transmission is obtained based upon the nominal plant. where a22 = 0.000 Figure E.000 1.I dB.916 ts.9 (Unstable plant) 15 (Nominal) 16 (High gain) 3.19726 2.000 t P . since this affects the value of \t.29) since it may be necessary to lower the value of \f22\ at large values of co ( > 10).3. for this example. a figure corresponding to Fig.Lm jmax = Lm b22 ^d Lm TRl .A "worst" case situation representative of the unstable plant region of 9.8 E.2\. Table E." .1 dB and a value for Lm t2min something less than O. i.000 1.Lm Tmin = Lm U22 The filter chosen is: 1 (E.913 3. etc.s 3.29) Note: care must be taken in determining the location of the pole of Eq.I.5 presents the time response characteristics of y22(t\ for a unit step forcing function.5 Case 10 M 1.000 1.19722 (1) Nominal plant . E-4 SIMULATION The computer data for ya for a number of plants from 3> resulted in \t2\ « 1 for 0 < co < 20.e.A high gain plant is chosen because it is representative of the top of the templates on the "nominal plant side. Therefore..7 E.9)].MIMO Design Example 359 template.s 2.913 2.23. (2) Unstable plant . [see Eqs. (E. are obtained by plotting the data from Table E.7) through (E. (3) High gain plant . Thus.

.6 BR. E. (jOt) and L2e (jO) plots.360 Appendix E Fig.

are met for loop 2.25 6.3 are extremely similar to one of the three cases of Table E.2 = k. y^t) exhibits respectable rise and settling times. The frequency domain specifications.4) and (E.7 Time responsesy22(t) for r2(t) = u.fjlO) = 16 dB.00 0. In designing loop 1 the cross-coupling rejections bounds are dropped to an arbitrary low value for the shaping of L^ • The synthesized Llo for a> = 20 lies on Bo. From Eqs.00 3. E. Although time response specifications are not prescribed. .5 they are not simulated.5 dB < B. Thus.(t): nominal plant case.00 076 1.(jlO) » 14.31) TIME RESPONSE .50 TIME (SECONDS) 5.50 2. a check on the cross-coupling rejection t.7) obtain: St22 111 (E.76 4. (E. (j20) = 6 dB but for co = 10 Lm L.j(ja>) performance for loop 1 is made.00 Fig. and f 22- (E.26 3. Because of the gross overdesign any other plant from the plant parameter space 9 should meet the specifications.T22 .30) where q.NOMINAL PLANT 0. as demonstrated by the simulation.UNIT STEP .. c22 = 0.MEMO Design Example 361 Since all the other cases of Table E.Js.

50 5.J5 3 0 3.7S - I ul c tO. 5 4.00 TIME (SECONDS) 37 .00 0.j(f): unstable plant case.75 ISO 2.75 - <C £0501L 0. E.50 2.UNSTABLE PLANT 1.7B .OO- w Z 0.50 6. TIME RESPONSE-T22-UNIT STEP-CASE 16 t.9 Time response ydf) for rj(t) = «_X/): high gain case. E.362 Appendix E TIME RESPONSE .T22 .50- 0.25- 0.UNIT STEP .25 6.26 6.8 Time response ydf) for r^i) = u.00 Fig.25- 0.00- $ 0.00 O.25 3. .00 Fig.75 1. 0 TIME tSECONDS) 4.OQ 0.00- 0.

As seen from this figure the -20 dB specification is met. E. and t. (E. DISTURBANCE REJECTION -t -1M.32) using the parameters for the unstable plant is shown in Fig.0579 at tf a ft 026 s.10 Frequency response t12(jco): unstable case. (E.W Itf to* FREQUENCY (RAD/SEC) Fig.IO. (E. Only the plant parameters associated with the unstable q22e plant (case 10) are used in Eq.31) for a simulation.32) 1 + L. (E. (E. E.2(<x>) = 0. The frequency response plot for Eq.1) at t. = 2.30) yields s k. The simulation yields tn(tf) = -0.2(ts)\ = ft 002 (2% of the specified maximum magnitude of 0.2.MEMO Design Example 363 Substituting Eq. .61 s. No other plants are simulated to determine if the cross-coupling rejection specification is met. \t.31) into Eq.

In order for L2f to have a lower value for its CD? it is necessary to design L. for 0. « q22 [see Eq. i. . etc. Thus. . This freedom can be used as a trade-off feature to be used according to the relative sensor noises.2«I and therefore q-zi.8 there will exist an unstable q22e. costlier. This makes L.8. An unstable plant with a pole/?.20)] which simplifies the design..5 < Y12 < 0.e.y12 does not have RHP zeros. the benefit going to L2e (or g2). (E. Note that for the example of this appendix yaMa = 0.364 Appendix E E-5 SUMMARY In some problems y. L2 has to have a large crossover frequency. > 0 requires a loop transmission with a crossover frequency77 w$ = > 2p. so that 1 + L. q22e will have RHP poles. Therefore.

Houpis. Horowitz. M. I. of the American Control Conference." Proceedings of the American Institute of Aeronautics and Astronautics (AIAA) Conference. ibid. M.1979. J. and C. vol. 2. "Linear Control System Analysis and Design. W. CA. Pachter. 7. "Design of a Subsonic Flight Control System for the Vista F-16 Using Quantitative Feedback Theory. Control. Scottsdale. M Pachter. F. vol. 3. "Proceedings of the American Control Conference. and C. pp. 350-354. "Quantitative Feedback Theory (QFT) for the Engineer: A Paradigm for the Design of Control Systems for Uncertain Systems. Control.. Wright Laboratories. M. pp. Editors: "Quantitative Feedback Theory Symposium Proceedings . 81-106. I. 1995." hit. O. 1976. and C. May. 9. Reynolds. Thompson. J. 64. 10." McGraw-Hill. and C. "Damaged Aircraft Control System Design Using QFT." IEEE Proc. "Formation Flight Control Automation. D. "Optimal Loop Synthesis in Quantitative Feed back Theory. pp 287-309.REFERENCES 1. Trosen. I. 1995 (Available from National Technical Information Service. C. Horowitz. of Control. and C.. 30. D. 33." Int. H. R. Houpis. J. Pachter. 1994. NY. ibid. vol. pp. pp. M. vol. 22.1990. Springfield. pp. 365 . H. Houpis. 123-130. H.. D'Azzo.. Loecher." WL-TR-95-3061. AF Wright Laboratory. pp. Houpis. 16. Houpis. Keating. Control. pp. 1992. 621-628. OH. Houpis.1994." hit. VA 22151. Wright-Patterson AFB. Sidi.. 6. and O. C. 97-113. 4th Ed. 5.) Horowitz I.1981. "Quantitative Synthesis of Uncertain Multiple-Input Multiple-Output Feedback Systems. H. vol. 1994. Nwokah. "Synthesis of Feedback Systems with Large Plant Ignorance for Prescribed Time Domain Tolerances. 12.1973. 1379-1404. H. "Proceedings of the National Aerospace and Electronics Conference (NAECON). J." WL-TR-92-3063. document number AD-A297571. pp.. 1. 4. OH." Int.J. 626-631. R. OH. and M. H. M. Wright-Patterson AFB. 11. San Diego.S. AZ. Chandler. "Design 3x3 Multivariable Feedback System with Large Plant Uncertainty. M. "Synthesis of Feedback Systems with Non-Linear Time Uncertain Plants to Satisfy Quantitative Performance Specifications . "Optimum Loop Transfer Function in Single-Loop Minimum Phase Feedback Systems. and P. 5285 Port Royal Road. 1973. J.. 8." Proceedings. D. vol. 677-699.

1003-1013. 81-106. Sidi. Horowitz. 44. Houpis. "The Weizmann Institute of Science. "A Quantitative Inherent Reconfiguration Theory for a Class of Systems . "Full Envelope Flight Control System Design Using QFT. R. 454-463. 15. 14. 1982. I. pp. pp.1986. Israel. 10. "QFT Digital Flight Control Design as Applied to the AFIT/F-16. "Quantitative Feedback Design for Sampled-Data System. Hardware. 27. Israel. ibid. OH 45433.1979. 1963. 36. "hit. J. 1991. 361-386. ibid. K. Pachter and C. J. M." Int. 27. O. "A Synthesis Theory for Linear Time-Variable Time-Varying Feedback Systems with Plant Uncertainty. pp. M. Software. 23. 2nd Ed.. vol. pp. "Development of a MMO QFT CAD Package (Version 2). 1986. pp. 21. and G. Palo Alto. pp. pp. M. Houpis. "Synthesis of Cascaded Multiple-Loop Feedback Systems with Large Plan Parameter Ignorance. 32. E. vol. J. 20. J." IEEE Trans.1984. M." Int. 28. "Design of Feedback Systems with Nonminimum-Phase Unstable Plants. Ibid. Baltimore." Automatica. and T. pp. and C. J. Sys. of Control. 33. 44." McGraw-Hill. K. R. M. vol. John H." 2nd Ed. Sciences. Wright-Patterson AFB. 1982. Wright-Patterson AFB.. 401^121. Thesis. 31. of Control.1986. Sept. Rehovot.1973. Houpis. WileyInterscience. of Control. Presentation at the 1993 American Control Conference. OH. Yaniv. J. 533-541. vol. vol. and O." hit. I." The Weizmann Institute of Science. 22." Int. Int.366 References 13. 350-354. 44. Horowitz. of Sys. of Control. Graduate School of Engineering. ibid. M. R. "A Quantitative Design Method for MIMO Linear Feedback Systems Having Uncertain Plants. C." Graduate School of Engineering. "Optimum Synthesis of Nonminimum-Phase Feedback Systems with Parameter Uncertainty. J. I." hit. M. Horowitz. I. Horowitz. and C.. H. D. 1981. vol. June 1994. 43. Horowitz. "Limitations of Nonminimum Phase Feedback Systems. M. 665-675." Academic Press. I. and I. Houpis. "Multivariable Flight Control Design with Uncertain Parameters.1985. pp.NY. AC-20. Air Force Wright Aeronautical Laboratories. J. 677-688. and Y. vol. and M. Blakelock.1982. Sci. 17. "Automatic Control of Aircraft and Missies. H. CA. Air Force Institute of Technology. L. ibid." hit. I. vol. Yaniv. M. ibid. 589-600. vol. USA. S Oldak. I. and Y. "Multivariable Flight Control Design with Uncertain Parameters (YF16CCV). H." AFWAL-TR-83-3036. "Important Property of Nonminimum Phase MMO Feedback Systems. Horowitz. pp. 29. "High Angle of Attack Velocity Rolls. 16. Kopelman. MD. 26. M. pp. 24. 51-57.. Air Force Institute of Technology. Sating. Horowitz. 1986. Rehovot. 40. 1994. Horowitz.1978. M.1986. "The Singular -G Method in Unstable Nonminimum-phase Feedback Systems. Liao. of Control. O.. unpublished notes. Control. Reynolds. Schneider. AFIT/GE/ENG/66D-4." Proceedings of the American Control Conference. "Improved Design Technique for Uncertain Multiple Input-Output Feedback Systems. pp. of Control. 9. pp. "Advanced Control Theory and Applications. J. Pachter.. Liao. "Digital Control Systems: Theory. 1992. Final Report. Oct.1975. I. 30.. Horowitz. Lament." Int. 18.. vol.S. et al. I. Horowitz." 13 IFAC Symposium on Automatic Control in Aerospace." M. 1377-1390. 19. M. Boyum. K. H. "Synthesis of Feedback Systems. 16. 25. . 1025-1040.

University of Strathclyde. Merino. Nov. Graduate School of Engineering. vol. pp. May 1995. 1945 41. Houpis. Bossert. pp. F.. Houpis.. AFIT/GE/ENG/92J-04. vol. Wright-Patterson AFB. W. D. University of Strathclyde. Scotland. 44.. J. and M. 39. Pachter. of Robust and Nonlinear Control. "A QFT Subsonic Envelope Flight Control System Design. R. J. 35. 315-320. Air Force Institute of Technology. Scotland. 46.. "Development Implementation and Flight of a MIMO Digital Flight Control System for an Unmanned Research Vehicle Using Quantitative Feedback Theory. 6. July 1996. Houpis.An Outline for Engineers . D.. Graduate School of Engineering. H. Wright-Patterson AFB. Osmon. H.. M. Jan. K. OH. pp. Sating. M. "Development of an Analog MIMO Quantitative Feedback Theory (QFT) CAD Package." MS Thesis. Air Force Wright Laboratories. Franchek. "Network Analysis and Feedback Amplifier Design.1997. J. 1992. C. Hamilton. Engine Speed Control. "Robust Controller Design and Experimental Verification of I. VA 22314. Pachter. Glasgow. Aug." Van Nostrand. 38. AFIT/GE/ENG/89D-2. MD. C.References 367 34. pp." vol. Sheldon. 515-531. "Active Flexible Wing Control Using QFT. Alexandria. OH. June 1994. 1987." AFWAL-TR-86-3107. 591-608. of Robust and Nonlinear Control. CA. N. 45. No." Symposium on Quantitative Feedback Theory and Other Frequency Domain Methods and Applications Proceedings. Phillips. H. Bailey." National Aerospace Electronics Conference (NAECON). 1997. J." Proceedings of the ASME Dynamic Systems and Control. E. Trosen. and K. H. OH. Horowitz. 37. "Development of an Prototype Refueling Automatic Flight Control System Using Quantitative Feedback Theory." MS Thesis. M. San Francisco. (Available from Defense Technical Information Center. 609-628." Proceedings of IEE.C. S. Aug. 40. and C." hit. and G. Symposium On Quantitative Feedback Theory and Other Frequency Domain Methods and Applications Proceedings. "Quantitative Feedback Theory (QFT): Technique for Designing Multivariable Control Systems. Helton and O.. 1989. and C. Baltimore. C. I. June 1997. H. . Graduate School of Engineering. 1997. Dec. "Design of Pseudo-Continuous-Time Quantitative Feedback Theory Robot Controllers. Jan. Rasmussen. and C.) 42. 36. Cameron Station. Wright-Patterson AFB. 48. "Application of QFT to Control System Design . AFIT/GE/ENG/93-J-03." MS Thesis." Int." IFAC 13* Triennial World Congress. "Modelling and Control of an Electrohydrostatic Actuator. "Alternative Process in Frequency Domain Design of Single Loop Feedback Systems with Plant Uncertainty. document number ADA176883. H. Wright Patterson AFB. Bode. vol. 1997. M. 7. "Quantitative Feedback Theory.. Nov. vol. Houpis. Kang. H. Pachter. 47. 7. J. IL. N. Air Force Institute of Technology. and C. Winter Annual Meeting of ASME. 1982. NY. R. Chicago. OH. 43. Osmon. M. H. C. W. 1993. Houpis." Proceedings of the American Control Conference. 129D. "Piloted Simulation of An F-16 Flight Control System Designed Using Quantitative Feedback Theory. Dayton. Air Force Institute of Technology. S. S. W. OH. 1994. Houpis. 7. Glasgow.. Pachter.

I." hi Proceedings of Quantitative Feedback Theoiy Symposium. R HA." JIEE (London). Lancaster. Jr. p. I. Sept-Oct. San Diego. pp. 1030-1035. Franklin." Proceedings of the ASME Winter Meeting. 1981. J. 65. H. "Survey of Quantitative Feedback Theory (QFT). D. OH. 12. No. 503-541. . X. Dec. and M. "Digital Control of Dynamic Systems. Reading. "Stability of Quantitative Feedback Designs and the Existence of Robust QFT Controllers. M. Zames. 1988. vol. L. Chandler." IEEE Proc. et al." Int. AFIT Lecture notes 52. Lacey. "Optimazation of the Loop Transfer Function. 50. of Control. vol. 1969.1983. 64. 2nd Ed. Wright-Patterson AFB. "A Comparison of Flight Control Design Methods. Powell. and I. Wright Laboratory. "A Method of Analyzing the Behavior of Linear Systems in Terms of Time Series. OH. S. Parsons and G. Aug. USAF. Wright-Patterson AFB. D. 61. Gera." AddisonWesley. R." IEEE Trans.. and C. vol." hit. 1996. 66. MA. Tustin. pp. "Theory of Matrices. Scotland. 1974. of Control. 255-291. Horowitz.. E. and J. Dec. WrightPatterson AFB. University of Strathclyde. H. Graduate School of Engineering. of Nonlinear Control. Y. E. "Synthesis of Feedback Systems with Non-linear Time Uncertain Plants to Satisfy Quantitative Perfomance Specifications.368 References 49. editors. Wright-Patterson AFB. AFLT/GE/EE (83D-11).. vol." M. Fontenrose. pp. R." Academic Press. Rosenbrock. 64. Tashka. 55. Graduate School of Engineering. J. and C. J." Int. 1991. "A Robust Digital Flight Control System for an Unmanned Research Vehicle Using Discrete Quantitative Feedback Theory. C. 19. Chai. and I. Jayasuriya. AC-28. 2.. SI106. No. "Multivariable Feedback Sensitivity and Optimal Robustress. pp. "Quantitative Synthesis of Feedback Systems with Uncertain Nonlinear Multivariable Plants. M. Dallas. Zhao. Hollot. 67.1980. "Object-Oriented Design and Programming of QFT CAD Environment. G. 539-563. M. Horowitz. 60. 1997. Sci.. 1990. A. Air Force Institute of Technology. 1109-1115. D. "A Comparison of QFT and H<o Techniques. 56.S. 1983. M.. 58. 51. 57. 1992. "Control of an Activated Sludge Wastewater Treatment Plant with Nitrification-Denitrification Configuration Using QFT Technique. J. AFIT/GE/ENG/91D. 63. A. vol... Application to the C-135 Aircraft. Garcia-Sanz. M. "Multiple Input-Multiple Output Flight Control Design with Highly Uncertain Parameters. Syst. 62. Thesis. ibid. 5. J. R.. Horowitz. V. L. Horowitz." MS Thesis. Bensousan." Guidance and Control Conference. J. Aug. Ewing. Houpis and P. 53.. "Development and Flight Testing of a QFT Pitch Rate Stability Augmentation System. and D. 389." Quantitative Feedback Theory Symposium Proceedings.." J. 1947. F. 94. of Guidance. pp." Symposium on Quantitative Feedback Theory and Other Frequency Domain Methods and Applications Proceedings. 1976." Academic Press. and Y. TX. Ostolaza. OH.1991. "Stability of Quantitative Feedback Designs and the Existence of Robust QFT Controllers... CA. G. 1992. Hall. 1993. 54. P. WL-TR-92-3063. Aug. Ibid. 31." Int. P. Breiner. J. Powell. Glasgow.H. 59. Control and Dynamics. W. 53. vol. Betzold. 1979. pp. OH. "Computer-Aided Control System Design. Air Force Institute of Engineering. Aug.

OH. 76. T. OH. "Strong Robustness in Uncertain Multivariable Systems. Aug. 1997. . Boje E and Nwokah ODI. Clough. I. Chait. MD. 50:1057-1069. 72." AIAA J." Rep. I.. 15(1):207- 214. of Control. Vol. June 1994. 77. 1985. No. Scott D. "Research in Advanced Flight Control Design. Wright-Patterson AFB. R. "Optimal Loop Transmission Functions in SISO Quantitative Feedback Theory. 1988. J.1984. ibid. pp. 78. MS Thesis. A Real-Time Hardware-in-the-Loop Simulation of an Unmanned Aerial Research Vehicle. TX. Control. Graduate School of Engineering.. OH. Nwokah.1992. Horowitz." Int." Lit. AC-20. M. J." MS Thesis. Swift. pp. 96-98. 70. OH. Nwokah O. AFIT/GE/ENG/95D-04. Air Force Institute of Technology. 71.. D. "Reconfigurable Flight Control System for a STOL Aircraft using Quantitative Feedback Theory. 73. V. B. AFFDLTR-79-3120. of Control. University of Strathclyde." IEEE Control Systems. Nwokah. AFIT/GE/ENG/85D-8. 1990 80.1975. Dec. 1991 79. Thompson. D. S. Horowitz. 81. Control and Dynamics.. "Synthesis of Controllers for Uncertain Multivariable Plants for Described Time Domain Tolerances. Symposium on Quantitative Feedback Theory and Other Frequency Domain Methods.. Robertson. D. Dec. F. OH: Wright Laboratory. et al. of Guidance. OH. Air Force Institute of Technology. Wright-Patterson AFB. O. 1995." IEEE Conference on Decision and Control. WrightPatterson AFB. D. (Aug 1992). Austin. I. 21-22 August 1997. Technical Report WL-TR-93-9005. 185-208. O. "Quantitative Feedback Design Using Forward Path Decoupling" and "Quantitative Multivariable Feedback Design for a Turbofan Engine with Forward Path Decoupling". Bernstein. Gerald A. Wright-Patterson AFB. Wright-Patterson AFB.References 369 68. Air Force Institute of Technology. M. and U." IEEE Trans. and D. 17. Air Force Wright Aeronautical Laboratories. I. pp. Grewel. Graduate School of Engineering. vol. "Algebraic and Topological Aspects of Quantitative Feedback Theory. O." MS Thesis.. I. S. Wheaton. 75. 69.. I. Dec. pp. 4. J. E. "A Student's Guide to Classical Control. Autom. Automatic Flight Control System for an Unmanned Research Vehicle Using Discrete Quantitative Feedback Theory.. "Superiority of Transfer Function over State-Variable Methods in Linear Time Invariant Feedback System Design. D. Air Force Institute of Technology. Nwokah. 84-97. Model Identification and Control System Design for the Lambda Unmanned Research Vehicle." Proceedings of the American Control Conference. 1189-1206. 40. Shaked. Sept. S. Cacciatore.1989. Wright-Patterson AFB. Baltimore. vol. 1979. 74. Jayasuriya and Y. Nordgren and G. Dec. "A Quantitative Feedback Theory FCG Design for the Subsonic Envelope of the VISTA F-16 Including Configuration Variation and Aerodynamic Control Effector Failures. "Parametric Robust Control by Quantitative Feedback Theory. David G. MS Thesis.

(b) For the upper bound Mp = 1.3 4 4 1.9 3 4 1.5 0. 5 < a> < 70.62s.9 is: where P« s(s + a) where the parameter variations are given by the following table.1 1 2 4 1.2 0.65s. (d) For the upper bound Mp = 1. 370 .5 1. 2 and for both the upper and lower bounds 4 = 1.PROBLEMS CHAPTER 2 2.5 1.2 and for both the upper and lower bounds 4 = 1.6 0.8 0. 12 and for both the upper and lower bounds ts = 1. 6s. 2.7 6 0.6 1.2 and for both the upper and lower bounds ts = 1.1. Obtain the tracking models TR[j(s) and TR^S) to satisfy the following specifications: (a) For the upper bound Mp= 1. CHAPTERS 3.4 5 0.1.8s.5 7 0. The plant transfer function of Fig. (c) For the upper bound Mp= 1.7 Plot Lm P.5 0. Plant k b a 1 1 1.6 0.(ja>) and ZPX/w) vs w on semilog graph paper for all 7 plants for the frequency range of 0.

3. 30.20)]. 2. Figure 3. (b) Hint: Recommend that the following transfer function be synthesized: Kxs gcons where Ks a)2+b2 and where a value of ^ be select in the range of 0. (c) Determine the values of 8r for the frequencies of 30.7 (see Ref.7 s. 4.1 is modified by inserting a unity-feedback loop around the plant 9 as shown in the figure below.Problems 371 (e) For the upper bound Mp = 1. 16.3. 50.5 < £< 0.3 and for both the upper and lower bounds 4 = 1.0. Note: In synthesizing these models ensure that the magnitude of Sffjai) is always increasing as (o increases within the desired BW. For this plant: (a) Obtain templates SPQco.4. 8.1.0. *> Y(s) . 1 for tp = 60 ms. 1). that satisfies the following specifications: (a) \yD(tP) I = 0. 3.0.0. and 70. Determine the disturbance bound model TD(s\ for Case 2 disturbance. (b) Determine the value of V [see Eq.1 is the same plant given in Prob. 3. 1. 2. 50. Compare these values with the value of I7determined in (b).) for the following frequencies: ft 5. and 70. (3. 3. Consider that the plant of Fig.2.

1 (a). hi all these design problems show all open-loop Nichols Chart plots and closed-loop Nichols chart and Bode plots as shown in Design Example 1. 3.5 but where the plant Pe and the feedback structure are given in Prob. B). 3.7. Compare the order of the compensator and prefilter of this problem with those of Prob. (c) Prob. 3. 3. 3. to satisfy the tracking specifications of Prob. (b) Prob.1(e).1. 3-18 for the tracking specifications of (a) Prob.1 (a). 2. 3. and (e) Prob.3.1(d).4. 3. For the analog control system of Fig. and (e) Prob.5. 3. 3. Repeat Design Example 1 (Case 1 disturbance) of Sec. 3. 3-6 through 3-20 except for Fig.3(a).372 Problems The effective plant that is to be utilized for a QFT design is given by: I(s) 1 + P.6. 1 the uncertain plant is . Repeat Prob. 3. Also. 3. 3. 3. 3-18 for the tracking specifications of (a) Prob.1(d). (c) Compare the size of the templates of this problem with those of Prob.1(b). 3-17 B0 = BD.10. as done in Sees. In Design Example 1 (Case 1 disturbance) of Sec.1(b).5. (a) Obtain the transfer functions Pe for each of the 7 plants of Prob. show the time response plots for all J plants. Repeat Design Example 2 (Case 2 disturbance) of Sec. 3.8. 3-11) specification of Prob. By putting a unity-feedback loop around the plant ff what effect did it have on the size of the templates for the plant fPel The reader is highly urged to do a few of the following QFT design problems with a minimum use of a QFT CAD package. 3. How close does the resulting QFT designed system satisfy the disturbance response specification for both Design Examples 1 and 2? 3.2(b).1 (a) and the Case 2 disturbance (see Sec.1 synthesize G(s) and F(s).1. 3.1(e). 3. 3. (c) Prob. By doing Steps 4 through 1 1 of Sec. 3-4 by use of TOTAL-PC (App. (b) Prob. where the plant is given in Prob. 3. (b) Obtain the templates 3Pe(/«$) for the same values of frequencies given in Prob. 3. 3. 3. For Fig. 2.26. will enhance the readers' understanding of the basic QFT concepts of Chap. 3. Repeat this example for which the disturbance rejection specification is ignored which results in B0 = BR.9.

3. 3.14.17. Given the plant of Prob. 3. For a Case 2 disturbance rejection specifications given by Prob.10(2)-(a) and -(b) are only to be satisfied. 3.2 for TRv and ts = 1.12._ F Prefilter G C ompensator P ^ Plant . 3. 3. 3.11. 3. 3. (2) The design specification for Case 2 disturbance rejection are (a) Op = |xo(/p)| = 0. Repeat Prob.1 (a)].3. 3.16.1..12 where the Case 2 disturbance specifications of Prob. 3.10 where 2 < a < 10 and 1 < k < 5. 3.(0 and compare their respective FOM with the desired FOM. which sets the value of ML is 4tf'. (1) The tracking specifications are given by Prob. Repeat Prob.15.1 (b)>b(oo) = 0 (c) Repeat the design where yD(x>) * 0 (3) The phase margin angle (y) of the design. 3. l(d). 3. Repeat Prob. 3.10 for the uncertain plant of Prob. 3. Repeat Prob.that will satisfy the desired performance specifications.10 for a Case 1 disturbance. (b) Obtain yR(f) and>-D. 3. (a) Synthesize the compensator G(s) and the prefilter F(s) for the uncertain plant of Fig. Mf = 1. Repeat Prob.65 s (for both the upper and lower tracking bounds) [see Prob. E EP . 3.4. 3.13.10 for I < a < 4 and 1 <k<4 where the Case 2 disturbance specifications of Prob.10 for the uncertain plant P.10(2)-(a) and -(b) synthesize G(s) and F(s).Problems 373 P. 3.(s)=- ka s(s + a) where 7 < a < 15 and 2 < k < 7 represent the bounds of the uncertainty in the plant.10(2)-(a) and -(c) are only to be satisfied. A design has been made for the following figure: . 3. of Prob.

3. Find Sp (s) . B below P(s) 1). A P(s) = Us and Gt(s) = ll(s + 2).) and for P(jO.)H(f .17 the plant transfer function is P(s) = K/[s(s + a) where 1 < K < 5 and 1 < a < 10.6. H!(S). G(s).2. the system sensitivity to variations in P. and G(s) are known. F(s).18. Find Ft(s) and G^s) as functions of the known transfer functions F(s).374 Problems Where the transfer function P(s). and S(s) so that the two figures are equivalent.] (a) Obtain templates for Pd'co. Note: the loop transmissions for both control systems must be identical. [Note: see Fig. which one yields smaller values of I SP (jo) I over a range of finite values of col Which configuration is the best for minimizing the parameter variation effects on 7(s)? B 3. For Prob. for each configuration. . 3. It turns out to be more convenient to implement the system by the one shown in the following figure: ///(*) is a tachometer and S(s) is a sensor. In the following Fig. In Fig. If these are different. and these transfer functions are known.19. (a) (b) (c) Find G^s) so that Y(s)/R(s) is the same for both configurations.

3.l<a< 10 and 1< k < 12.^ =0.6.6. (b) Obtain yR(f) and^(/) and compare their respective FOM with the desired FOM. 3. 3. See the discussion on sensitivity in Sec.1 and yD(oo) = 0 (a) Synthesize the compensator G(s) and the prefilter F(s) for the uncertain plant of Fig. for values of co.21.20.2. l(c) and for the Case 1 disturbance specification (a) Given in Prob. 3.455 CHAPTER 4 Reference should be made to Fig.10(2) . Obtain template data for points^.1(a) (2) See Prob.1. 3.01s.5. and 2. 3.10 the bounds of the structured incertainty in the plant are given by 3<a<9 and 2<k<8 The tracking specifications are given in Prob. Synthesize the controller G and the prefilter F for the uncertain plant of Prob.) are different.1.20 for the tracking specifications given by Prob. 3. = 0.873 5 +11. 3.592+ J15. Repeat Prob.1(e) and the Case 1 disturbance specifications are OP = MtpJln.10 where T=Q.Problems 375 that is defined as the effective plant and where Hj(s) = s. 6-2. 4. 3.20 (b) Given by the transfer function TD(s)=3. B. C. 3. For the analog plant of Prob.8 for the QFT MISO sampled-data control system design problems of this chapter.1. 4. 3. (b) Explain why the sizes of the templates for 3P(/a)^ and 3Pe(jco. and/) of Fig. The problem design specifications are: (1) See Prob. 3.

CHAPTERS 5. 4.1(a)]. .1.4. 4. Mp = 1.5s T= s+2 0 0 ——— 2 s + 1 (s + l)(s + 2) s+2 The one degree of freedom is shown hi the figure where / is the identity matrix and / = F.14.1 for 2 < a < 10 and 1 < k < 5. (See Prob.12.1 for the uncertain plant Pe of Prob. IVflCp)! ~ 0. Compare the results with Prob. only yD(oo) = 0 is to be satisfied for a Case 2 disturbance. The plant matrix and the desired control ratio matrix for a 2x2 system are: -1 \—— T 1 0.1 andyD(<x>) * 0 are specified for this problem.65 s (for both the upper and lower tracking bounds)[see Prob. 3. 4.2. Repeat Prob. 2. Repeat Prob.376 Problems 4.1 for / < a < 4 and 1 < k < 4. Repeat Prob.4. 3. 4. 4.1 andyD(co) * 0. Compare the results with Prob.6.2 for TR(f and ts= 1. 3.2 only yD(<x>) ^ 0 is to be satisfied for the Case 2 disturbance response.20 as a discrete control system for T= 0. 3. Repeat Prob. 4. Repeat Prob. Repeat Prob.3) 4. 3. 3.5 (s + l)(s + 2) 0.1 for the uncertain plant of Prob.3. 4.01 s and for a Case 2 disturbance where \yD(tp)\ = 0. 4.2 2 ——— p= s+1 0. A y= 40°.

are the zeros of A = det[7 + PG]. Determine G(s) and compare it with G of part (b). Assume that G is a diagonal matrix. This reveals that for most practical MIMO plants of the above form if P~' is derived from P then . 5.Pnpn.Problems 377 (a) Determine P1 (b) Determine G that results in achieving the desired T. Prove that the zeros of interest are those of Hint: show that (Npu) = (b!lqn). In 2x2 MIMO system.1 represent a 2x2 MIMO plant where the plant is described by: (a) Determine P.0. 5.5+gy)5 g12 s+2 where EI > 0. (b) Determine P ' directly from P. Let Fig. especially PG (d) For this part assume tn(s) = t22(s) = (s .3.Qs + E£) and no remainder. 5.2. s2 > 0 •(0-5+g2) g22 5-0. Suppose P has the form Pu P= P21 P12 P22 P22 ~ PI2 Pll thus PI = ~P21 where A = pup22 . (c) Suppose the actual compensator transfer function is: (0. Do these control ratios yield stable responses? Hint: Analyze T=[I + PG\'1PG.2 Determine tn(s) and t2i(s).2)/[(s + 2) (is + 1)] and t12(s) = = 0 where T> 0. the system poles due to feedback.By dividing d into A you should obtain the quotient E.



the common poles of the elements of P should not appear as common zeros of the elements of ,F;. (c) Utilizing Eq. (5.6) prove that the remainder is zero. Hint: P = D'!N and U =

5.4. A MISO control system involving the uncertain plant £P for which the compensator G7 is to be designed to satisfy the desired performance specifications is shown in Fig. A. It is suggested that before the compensator is designed, that the system structure be modified by inserting a feedback unit around the plant as shown in Fig. B. It has been stated that by so doing will enhance the degree of system robustness to plant parameter variations. Given:







1 5+2



H(s) =

S S +1

Find G2(s) so T(s) = Y(s)/R(s) is the same for both system structures.
5.5. Given





P(s) =

5 + 1



s +2 The following weighting function W is selected to achieve an effective

decoupled plant Pe = PW.
W =

"9 10"

With no parameter uncertainty in P and/or W the det [/ + PfFjhas excellent stability margins. Show that with uncertainty in the parameters in W may lead to an unstable system. Hint: let
W = 9 +a 8+c
10 + b 9+d

and suppose a = d= -0.1 andb = c= 0.1.


NOTE: For all MIMO problems for Chapters 6 through 8 only diagonal G(s) and F(s) matrices are to be used. 6.1. For Prob. 5.4 find the system sensitivity function SJ for each structure. If these sensitivity functions are different, which structure is least sensitive to plant parameter variations and why?

6.2. In a 2x2 MIMO system rt * 0 and r2 = 0, thus only tu ant t2i are involved in the design. The specifications on tn are:

1.02 0.95

0.2 1.07 0.92

109 0.80


2 0.65 0.20

5 0.25 0

10 0.05 0


0.97 0.55

The specification on t2i is that b21 = 0.01 for all values of a> and the specification on each Lt satisfies






The plant is described as follows:

0.2<kn<l, 0.5<k12<2, 2<k21<5, 0.1<£ 22 <0.

Notes: (a) The q's are assumed as real numbers (gain uncertainty only) thus ymail = [(7)(ft5)]/[(ft5)(2)] = ft5. (b) It is intended that this problem be done without the use of a CAD package. See Sec. 6-5.3

Choose the following nominal values:

(a) (b)

Find the bounds on Llo at co = ft 1, 0.2, ft 5, 1 and 2. Use
b*L 1 ill

to find the bounds on L2o for the same values of w as in part (a). Hint: Use the "inverse NC" for this purpose, i.e, let (2 = l/L^ to obtain the form 6.3. Given a 2x2 plant whose Q matrix is

<&,,<2, 0.5<k,2<2,
qnq22<q12q21 as

<lO, 0.5<k22<l
, r;(0=0

The specifications based upon input r2(t) = u.,(t) are:



t,3\ < 0.1 for all®

1.4432(s + 7.78) s + 1.95 ± y'2.725 ______21.50525________
«22 =

(s + 2.036)(s + 1.8 ± y 2.70601)

Additional specifications on L, and L2, respectively, are
Lm Li
<3dB Lm



(a) Design Llo(s), L2o(s\ andy^O) using Method 1 and letting f12 = 0. (b) Determine gt(s) and g2(s).
6.4.Repeat Prob. 6.3 by redesigning L2(s) where 0.5 <k22< 1.2 aadLi(s)a£Prob. 6.3 is not affected by the change in k22.
6.5. Given the 2x2 plant

with the uncertainties:
l<kn<2 0.5<k,2<l 0.25<k21<0.5 2<k22<4

and with one command input r2(f) = u.i(t). Design LIO(S), L2a(s), andf22(s), by use of Method 1, to satisfy the following specifications:
1.4432(5 + 7.78) s + 1.95 ± y'2.725 ______21.50525______
(s + 2.036)(5 + 1.8 ± y'2.70601)

< Lm

< 3 dB

LmML2 < Lm




and where \t12\ < -20 dB for all <o. Use the MIMO QFT CAD package of App. A to do the design and obtain the Nichol plots for both loops, the openloop and close-loop Bode plots for this 2x2 system, and simulate your design in order to obtain the time domain performance results. Note that t2i need not be specified since r,(f) = 0. Note the phase margin frequency cot for each loop.
6.6. Repeat Prob. 6.5 where



5 + 0.01

5 + 0.01


Repeat Prob. 6.5 where 0.5 < kn < 1, 0.5 < kl2 < 1, 0.1 < k21 < 0.2, and 1 < k22 < 2


Given the 2x2 MIMO control system of Fig. are:

5.5 whose i = 1,2, 3,J= 4 plants
5 + 1.4 P2(5) =

r i
5 + 1 0.11
5 + 2

0.095 1
5 + 2

5 + 1.9

5 +1.5

5 + 2.1

5 + 1.8

5 + 1.8

5 + 1.8



P3(s) =

5 + 2.2


1.9 5 + 2.2

0.13 5 + 2.3


5 + 2.2

5 + 2.6

11.70 *11 = 5 + 1.95± y2.81
3.0825 + 21.574
5 2 + 4.885 + 21.574

(5 + 3.398)(5 + 2.7±/2.1) ________691.5_______

53 +S7.7552 +387.55 + 691.5



by =0.1 where /' *j, a>^< 10 radlsec, and co^ < Sradlsec.

(a) Verify diagonal dominance for this plant, (b) Using METHOD 1 synthesize the compensators and prefilters, assuming diagonal G(s) and F(s) matrices, where the command inputs are: r;(0 = r2{t) = «./(/)• Note: the stabihty specicification for each loop is to be based upon the peak overshoot value of bu and b22, respectively, (c) Obtain the y(i) responses. Note the BW frequency co^ for each loop. 6.9. In order illustrate the conservatism of the QFT design technique repeat Prob. 6. 8 by ignoring the specification be; that is, the optimal bounds will be based only upon the stability and tracking specifications. Compare the results with those of Prob. 6.8.


7.1. Repeat Prob. 6.3 by Method 2 and compare the compensators and the prefilters for each loop. 7.2. Repeat Prob. 6.4 by Method 2 and compare the compensators and the prefilters. 7.3. Repeat Prob. 6.5 by Method 2 and compare the results.
7.4. Repeat Prob. 6.6 using Method 2 and compare the compensators and the prefilters.
7.5. Repeat Prob. 7.2 for Prob. 6.8 and using Llo of Prob. 6.8. Compare the results with those of Prob. 7.2.

7.6. Repeat Prob. 6.3 to redesign L2o by use of Method 2 and using Llo of Prob. 6.3 where 0.5 < k22 £ 1.2. Compare the results with those of Prob. 6.3.
7.7. Repeat Prob. 6.8 using Method 2 and compare the results with those of Prob. 6.8. 7.8. In order illustrate the conservatism of the QFT design technique repeat Prob. 7.7 by ignoring the specification b,j, that is, the optimal bounds will be based



only upon the stability and tracking specifications. Compare the results with those of Prob. 7.7.
7.9. In order illustrate the conservatism of the QFT design technique repeat Prob. 7.8 by ignoring the specification b,j, that is, the optimal bounds will be based only upon the stability and tracking specifications. Compare the results with those of Prob. 7.8.



External disturbance rejection problems where[r(0 = 0].


The P matrices of Fig. 8.1 are those given in Prob. 6.8. The PD matrices are: 1
s +2
5 + 1.6

0.095 "

5 + 4
PD2 =


5+ 3

5 + 1.1

5 + 5

3 2 0.081
o 1 ~*

5 + 1.7

PD, =

5 + 3

5 + 1.8

PD< =
0.169 2.5
5+ 6

5 + 1.2

5 + 1.3

5 + 4

The specifications are as follows: (a) use the ML specifications of Prob. 6.9, (b) b,} < 0.1 where i* j, (Bde)y = 0.1>\tdll \ [see Eq. (8.33)], coh < 30 rad/sec, and «^ < 40 rad/sec. Use Method 1 to synthesize G(s). For dj(f) = u.,(t) simulate your system to obtain theXO responses.
8.2. Repeat Prob.8.1 using Method 2 and compare the results with those of Prob. 8.1.



Tracker/External disturbance rejection problems: unit step tracking and external disturbance forcing functions are to be utilized.

8.3. Use the tracking specificationss of Prob. 6.8 and the external, disturburbance specifications of Prob. 8.1. Use Method 1.

8.4. Repeat Prob. 8.3 using Method 2 and compare the phase margin frequencies with those of Prob. 8.3.

1211 -96.4347 -100.3373 -42.103 ®(RAD/SEC) 70.0876 21.28708 -117.5 and 70 is shown.8602 -24.9205 -42.9018 -24.371 7.4459 22.8595 -41.827 RL 25.1.79) 386 .646 -107.70973 7.5728 -90. The following TOTALPC printout for the template data only for a> = 0. (d) 0. w(RAD/SEC) 0.4092 4 5 6 7 -24.977* + 12.07) u s+1.0000 PLANT 1 2 3 DECIBELS -36.9210 -90.8598 DEGREES -90.9001 -90.50000 PLANT 1 2 3 4 5 6 7 DECIBELS 6.952 ±j2.Answers to Selected Problems CHAPTER2 2.1. TOTAL-PC was used to obtain data for this problem.0818 -90.7365 -90.2455 -90.277 (s + 2.145)(s + 2±j2.884 -111.69503 20.6546 CHAPTERS 3.49736 5.968 -105.8134 DEGREES -92.

See Sec. (a) 0 < v. (b) From Eq. The solution is for the disturbance rejection specifications of (3): (a) and (b).574 TR L V R (w)«TR L (s) = —————— 7^^————— — R W s 3 + 57.200)(w + 200)(w +120000) w 2 (w +0. (3. Step 1. only the stability bounds can be satisfied.082s + 21.99999167) Step 6.0412 .0833jdO~7(w . The TOTAL-PC CAD package (App.057 dB « V CHAPTER 4 4.8dB Steps 7 through 9.882s + 21. TD(w) = 0.0618 dB = V (c) For (0=16 rad/sec: 8P = 18.(-6.) is utilized to obtain the data for drawing the templates. 4-7.2.LmK^ = 12.1. .5 Step 2. „ T.574 + 4.Answers to Selected Problems 387 3.3.0206) = 18.75s2 + 387. Step 4. C.20) ALmP = LmK^ . ML*3. (W) » TR > 2 S 3. . Step 5. = . < 20/1 = 20 .1(-20dB) Step 3 Region I Region ffl: 0tov a ^«200 OtovK= 746 For this problem.5s + 691.

11 0.248 1. TR FOM (Final value =1.515 1.228 z "W z w 1.38 1.1209 0.5250 0.388 Step 11.2597 2.46 1.024 1.158 tp 0.267 z w 1.3998 0.7941 0.607 1.000) Plant 1 2 3 4 5 6 7 Domain w z Vf ML 1.2795 0.066 1.266 1.25 0.158 1.2642 0.04 —— 0.1927 0.52 0.0558 0.0528 0.79 0.02 0.61 ts 2.6086 0.000 1.248 1.28 0.150 1.028 1.39 0. + 1000) w + 150 Step 13.03+ 1.152 1.53 0.001 1.229 1.38 0.000 1.05 0.028 1.0279 0.028 1.8795 0.1786 0.244 1.19 0.26 0. Answers to Selected Problems 36009(>v + 0.14+ 0.6+ TD FOM (All overdamped responses) .1376 0.99999167)(w + 12)(w + 436) (w + 2)(w + 200)(w + 3550)(w + 4000) (w + 500)(vf + 35000)(vf + 60000) (w + 4500)(w + 120000) Step 12.96 0.88 0.069 z w z w z 8 9 10 w z w z w z 1.

1535 0.1539 0.2 .1492 0.1551 0.1507 0.2 s .1535 0.2 .1536 0.1468 0.99999) CHAPTERS 5.1553 0. (a) s(s +1) -0.1.1528 0.(s + 1)(5 2) 5-0.1536 0.0.Answers to Selected Problems 389 Plant 1 2 3 4 5 Domain w z w z w z w z w Final value 0. 1531 In order to satisfy the disturbance response specification (3): (a) and (c) the starting loop transmission function is obtained by dividing P'eo by w which will ensure that Y(w)/D(w) has a zero at the origin in order to satisfy X°°)= 0.1521 0.1674 0.1528 0.2QQ)(w +120000) w 2 (w +0.2 2(5 + 1)(5 + 2) 5-0.1614 0.1575 z 6 7 w z w z 8 9 10 w z w z w z 0.4(5 + 1) 5-0.1535 0. Thus _K(w.1545 0.1535 0.

h.5 (s + 2)gl2 + 0. .m.5g 2 ) (s + l)(s + 2) Analyzing the elements of the first column it is noted that they can be n. zeros of det P to each element of T. is required.5.5sg12 PG (s + l)(s + 2)(s .2).p.2) 0.2 -1 (* s -0. Thus.l) -°.4 G(s) = s+2 2 which does not have any RHP poles.0.5 + gj)(.5(s + 2)g.2g22 (s + l)(s + 2) 0.m. system. . The problem can be foreseen by examining del P(s) which has a RHP zero thus resulting in an n.y . This imposes restrictions on the elemenets of T because (AdjP)L (j = ————— det/> . (d) Since G = PlL = R'T\I-T\l. one can assign the r.2 . substituting the specified T and P1 of part (a) yields 2) s+2 Ts1 +2rs + 0.p. The pole at the origin in the G of part (b) has been replaced by the LHP poles zs2 + 2rs + 0.y(gj -0.0.2 + 0.2(0.0.p.? .390 Answers to Selected Problems (b) Since the desired T(s) is specified then from T = [I + PG\1PG= [I + L]~1L solving for L yields 1 0 Thus: 2Qy + l) 5 0 1 PG 1 2 -0. which in turn impose restrictions on the elements of T(s) in order to avoid in designing an unstable system.4 s(s-O.(* + ) 2(5 + 2) (c) With the specified G the following is obtained: (0.

max u/. CHAPTER 6 6.02qlla 0. (b) The sensitivity functions for structures are. for Fig. Setting these two equal to one another and substituting in the known transfer functions yields G2(s) = ll(s + I). (a) For Fig.P/(1 + G. respectively.44): at low frequency where 0. s(g + 2) / s 5 + I)2 The ratio of these two functions is: < 1 for all finite a> s +2 Thus.4.2. si where the maximum value of \b21lq12\ is used Thus.004 . B. which is a 2 degree of freedom structure. 0. A: TA = G.P).Answers to Selected Problems 391 5.5. will yield a better design and can be made even better in e. w ye a et terms of the system sensitivity to P. since | Sp \B < \ Sp \A then the structure of Fig.02 »1 vo. (a) From Eq. (5. B: TB = G2P/[1 + P(H + G2)}.

with its ML = 3 dB contour. It is left to the reader to determine the bounds with bn = 0.g. 6.4084xl05(s + 25) (s + 10)(s2 + 168s+ 19600) .65 -> -36 dB at a> = 2. a "trade-off' is in order to improve (lower) the bounds on L2o at the expense of raising the bounds on Llo.3.004 For the frequency range of 0.1 loop being essentially the same as the tracking bound due to rlt. Thus. the templates obtain by hand. b2i/bn = 0. This gives the bound on L2o (j2) at approximately +10 dB at -100° which is much tougher than that on Llg(j2). Note that 0." For e. For values of frequency greater than 2 rad/sec the bounds are obtained based on the following: b2! 9n bn 912 '11 922 ''21 921 (b) For the bounds on L2 use b2i bn 921 922 for the "low frequency range. depending on the accuracy of the cutout templates and the graphical design.1 to 2 rad/sec the bounds on L\o are obtained on the basis of r n . the following are obtained: 5..4084xl05(s s(s + 10)(s2+168s+ 19600) (a) gi(s) = 5.18 and compare them with those determined for the original value of bn.01/0. By use of a NC.392 Answers to Selected Problems Therefore the qn template is 14 dB in height and the Tc bound can be ignored resulting in the optimal bound for the 1.

Answers to Selected Problems 393 CHAPTER? 7.2 (a) See Prob.97)(s + 71.61 ± jl01. (b) A student solution without the use of a CAD package: 7.044xl09 (s + 49. 6. .2)(s + 300)(s + 3000 ± j'5196) 1.3(a) for L. smdg^s).99)(s +1250) s(s + 17.088)Q + 39.23 ± y6.338x1010(5 + 1250) (s + 300)(*2 + 6000* + 36000000) /22 = ' 1 s +1 (c) All performance specifications are satisfied With the use of a CAD package a simpler compensator may be achievable.77 + j96.39 + j24.

302 Aircraft. 253-255 methodology. 315 A-8 MISO loops of the tracking control problem. nonminimum phase. 250 design objectives. 101-105 design procedure (MISO). 248 control system. 294 templates. 311 A-6 Formation of plant models for tracking control problems. 296 unmanned research vehicle (URV). 314315 problem statement. 307-327 A-l Introduction. 308 A-3 Continuous-time vs. 243 QFT design. 324-325 A-15 Prefilter design. 250-251 external disturbance problem. 317 A-10 0 matrix validation checks. 328 B-2 Overview of TOTAL-PC. 329 options. 10 VISTA F-16. 318-319 A-12 Specifications. 325-326 B TOTAL-PC CAD package. 243 C-135B modeling.Index Acceptable sets: command. 10-12 yaw rate. 149-264 disturbance input-output relationships. 308 A-4 Overview of the multivariable external disturbance rejection problem. 285-305 Aircraft digital controller. 328-334 B-l Introduction. 244-245 overview. 347-364 Asymptotic results. 243-244 background. 248 scope. 13-14 Analog control system. 325 A-16 Design validation. 311-313 A-5 Open-loop structure. template. 285305 variable parameter. 102 D MISO design example. 296. 225-264 air-to-air refueling PCS design concept. 252-258 channel 3 (SISO) design. 12 lateral/directional. 276 Augmented tracking models. 101-105 Answers to selected problems. 12 All-pass-filter (a. 246-247 disturbance rejection specification. MIMO.p. 5 control ratio. 260-262 Aircraft digital control. 259-260 nonlinear. 244 simulations. 103-105 Analog system. 259-262 linear. 40 394 . discretetime design. 319-320 A-13 Bounds on the NC. 335-340 Airplane control surfaces. 33-91 Analog plant. 243-264 engineering visualization. 258 control problem approach. 5 tracking input-output relationships. 155. 252-253 "loop 2" design. 244 disturbance modeling. 245 MISO.164 Amplifier: operational (OP-AMP). 341-346 E MIMO design example. 294 Aircraft flight control system: A-7 Inverse of P.. 115-116 lag characteristic. 315-317 A-9 MISO loops of external disturbance rejection problem. 250-262 refueling boom. 245-246 closed-loop Lm plots. 242 loop shaping. nonminimum-phase QFT design. 245 assumptions.f). 102-103 angle characteristic L4'(w)]. 386-393 Appendix: A MIMO QFT CAD package. 44 A-11 Improved method. 328329 B-3 QFT CAD package. 307-308 A-2 Introduction: overview of multivariable control. 6 Aerial refueling. 331-334 C TOTAL-PC discrete QFT design process. 3l 8 aerial refueling. 289 flexible wings. 255-258 "loop 1" design. 320-324 A-14 Compensator design.

129. 92 w-domain. 239 interaction.24. case 2. 203-207 example 6.295 optimal.132 from root locus. 47 resonant frequency. 24. 209ess for a robust control system) 212 Benefits of QFT. procedure. 129. 61 -62 Bending mode(s).304 of disturbance-rejection models. 155 395 Bandwidth (BW). 52-61 disturbance rejection (MISO). sensitivity. 222 a. 92-100 procedure for determining. control ratio from. 271-277 Bound(s): composite. 47-48 M contours. 23-24. 22 aliased noise. 57-61 external disturbance. 83 C-135B modeling.258 gain adjustment.23-24. 52-57 Closed-loop system specifications. 18 sampled-data system.266. 97-100 Tustin. 93 s. 123 tolerance. 28 Closed-loop frequency response: examples of. 24. 207-209 performance specifications (see Performance specifications and Bounds) transfer function. 31-32 Bilinear transformation. 285 Bridging the gap. 93 w'-domain. 79-81 (see also Design and implementation proccross-coupling. 170-173 stability. 29 from log magnitude-angle diagram. 29 Closed-loop control system: characteristic equation. 50-52 model (cot).132 Closed-loop characteristics.to w-plane. 292 wing leveler.5. 18 universal high-frequency (UHFB). 39 upper Bo-. 45-47 Bridge the Gap.4.147. Boundaries (see Bounds) Boundary (frequency response): 185.99 Binet-Cauchy formula. 113-115 system (tot ). 241 interaction term.239 determination. 37-41 Closed-loop systems: . 282 Body of engineering QFT knowledge. 245-246 CAD packages (see Computer-aided-design (CAD) packages) Characteristic equations: from control ratio. 68 theorem.209-212. 301 construction of. 100. 169-173 example 5. 244. 129. 245 waypoint directed. 39 Basically non-interacting (BNIC) loops.15. 188 procedure for determining.3 8. 24 stability.227 composite.to z-plane.s. 45-47 stability (w-domain). 296 arbitrarily large.s. 239 lower BL. 39 w-domain. 304 property. 22 tracking BR. Ba. 96.286. Bh. 129 model response. 7-9 Autopilot(s).6. 47-48 maximum value of. B. procedure. 230-233 Block diagrams.295 performance specification. 24. 50-62 characteristic of. 92-94 warping. 40. 194 Q(z) 129 roots of. 21. 174. 209-212 disturbance BD. 47 example 6..297.Index Automotive fuel injected engine. iii. 61 -62 cross-coupling.72. determination (3x3). 40 Bode: Diagram. 222. 181.3. 2-5. 47-48 from log magnitude plot. 106 Bode Plots. 41 drawing. 96. 94-96 .225-227 frequency. idle speed control. 52-61 case 1. 203-212 example 6. 36 Closed-loop formulation. 61-62 performance.47-52 loop transmission. stability.287 examples of. 47-48.

36-42 tracking. 274 Tustin model of.118 tracking (see Tracking system) welding. 37-41 Command input. 147.s. 328-340 Condition 1.3% performance. 52-61. 21-23 pseudo-continuous-time (PCT).118 Type/w.29. 45-46 matrix. 286 implementation issues. 294 implementation.280-282 robot. 300 performance specifications. 269. 307-327 MISO QFT CAD. 129-130 Constraints for a. analysis of. 157 MIMO. 270 discrete-data (see Discrete MISO) disturbance-rejection. 178-180 Condition 3. 64. 38 disturbance. 6-7 Controllability. 3x3 plant matrix.25 tracking/regulator (MIMO).plane to z. 41-42. 283.45-46. 2x2 plant matrix. 92 flight. 61-62 Computer simulation (see Simulations) Computer solution (see Digital computer solution) Computer-aided-design (CAD) packages: MIMO QFT CAD.274 nondiagonal G. 155. 36-42 poles of. 92-148 single-input single-output (see Single-input single-output closed-loop control system) Type 1. 262 VISTA F-16 flight. 176-178 Condition 2. 291 robust. 180-181 Configurations. 1-5 Control systems: aerial refueling.276. 222-227 Control authority allocation. 10-12 basic sampled-data. acceptable set. 36-42 acceptable sets of. 294 minimum gain. 41-42. 165 Controller: accuracy. 21-23 redesign. 5. 129. 5. 36-41 sampled-data (S-D). 2. 37-41 time response of. diagonal compensator. 37 Closed-loop tracking performance. 37-40 synthesize.76-77 disturbance-rejection (model).plane. 64. 165 Controllable system.269. 129-132 s. 155-158 MISO. 136145. 154 modeling of. 107 matrix. 9-10 minimum order. 92-148 DIR.64.286 Control problem: approach. 4.19. 12-13 robust control. 242-264 aircraft flight. 164 augmented. 227-228 Complete controllability. 307-327 TOTAL-PC CAD (MISO). 7 Control ratio. 276 digital (see Digital controllers). between time and z -domains. 40-41 desired. 33-148 (see also Quantitative feedback theory) Compensator. 2. 225264 linear time-invariant (see Linear timeinvariant control systems) multiple-input multiple-output (see Multiple-input multiple-output control systems) multiple-input single-output (see Multiple- input single-output control systems) muitivariable (see Multivariable control systems) nonlinear. 92.286 Control design process. 36. 250 design process. 41-42. 106-107 computer-aided-design (CAD) packages (see CAD packages) continuous (see Analog control system) design approach (see Design procedure) digital computers in. 5 Compensation: DIG. 142 (see also Closed-loop frequency response) Control system: actuator plant. 248 engineering. 122 simple second-order. 67 Composite boundary. minimum. sensitivity. 274 Index aerial refueling. 10-12 minimum order. root locus. 221-221. 97-100 .5 performance specifications. 274 minimum gain. 2x2 plant matrix.132 psuedo.274 Correlation. 9-10 Control theory background. 11. 34 gain. 165 Complex-conjugate poles. Constraints on the plant matrix.

296-297 fourth design cycle.m. 290 linear. selection of. 289 functional requirements. 175 condition. 297 example. 297 unmodelled behavior.296 system test. 294-295 third design cycle. 291-298 Design techniques.297 #4. 292-294 second design cycle. 290. 72-75 MISO analog. xi-xvi Definitions: controllability. 165 phase margin. 173.301-302 porpoising. 304 #1. 285-305 aircraft model. 75-85 Current. 67.165. 290 engineering visualization. 103-105 outline.297 selection of design envelope. 2 control system.294 #2. 296 Design envelope.293 nonlinear. 85-88 Design method 1.301 introduction. 150. 70-72. 290 requirements. 300-301 Design example 2 (MISO).p. 19-22 Decibel (unit). 296 Design equations for: the 2x2 system. QFT (see QFT) Desired performance. QFT. 160-181 m> 3. 288 control system design process. 293 operator-in-the-loop.221-222 Design example 1 (MISO). 290. 280 example. 302-305 control authority allocation.293-294 hardware/software consideration.296. 292. 209-212 effects).292 complex-pole pairs. 296 PCX design. 290 specifications.245 Design and implementation process for a robust control system. 178-181. 288-289 redesign. 286-288.294. 176-180. per octave.295 #3. 297-298. 286 PCS design. 219-221 Design of the prefilter. angle. 165 observability. 172 boundary determination (3x3). 289. 20 D operator notation. 152 Damping ratio (£). field. 292. 273-274 Design trade-off (see also Trade-offs). 289 nonlinear simulation. 291-298 first design cycle. 290 template contours. 37-41 Diagonal dominance. 298-300 template contours. 298 hardware-in-the-loop simulation/implementation.293. 145 Design process. 133-136 n. 183 Design procedure for a MISO system. 73 Degrees of freedom (DOF). 298-300 simulation. 290. 297-298 dynamics model.294 inner loop PCS. 297. 35-36 PCT control system. 292-293. 292 QFT control system design. 137-139. 72-75 MISO discrete. 184-192 Design method 2. 24.72-75 Design procedure: basic. 300 controller implementation. longitudinal.301 bending modes. 74 Design. 293. 293 DC (direct-current) servomotor. QFT. 296 intrinsic turn coordination.214-216 the 3x3 system. 288 engineering interactive simulation. 34. 36. 289-290 interactive.186 CTRL-C program. 280-281 397 implementation. 294 design process example. 183-212 example.213-234 example. 292. 290 performance specifications. 292 DATCOM.227 bound. 67 relation to resonant frequency. 64 frequency. 289 operator-in-the-loop simulation. 21 DC (direct-current) shunt motor. 163.296 flight test. 285-286 linear simulation. plant. 65 Definitions and symbols. 296. 296. 297. 286 control system implementation issues. hardware-in-the-loop simulation.304-305 Design performance trade-offs (see also Tradeoffs). determination of.Index Cross-coupling. 173-175.193-194 . 21 transfer function.

158-165 justification of. 10 optimization and simulation run time.9 nominal plant determination.6 minimum order compensator (controller).R. 1 weighting matrix. 1-5 Engineering interactive simulation.57 MIMO system with external. 34 Distributed uncertain systems. 34. 53. external.8 basic mxm plant P preconditioning.57 definition. 272 E. 225264 EASYSx CAD. 202-203 Equivalent method. 136 QFT w-domain design.200-202 theorem. 290 Engineering method.R.R. 152 linear constant-coefficient. qu 's. 41-42. 198 Disturbance response characteristic. 175.2 n. 33.398 Diagonal matrix.R.R.5 QFT method 2. 235. 198 trade-off. 276 E.R. 152-153 Digital control systems (see Discrete MISO) Digital controller: algorithm. 225-264 design example. 36. 278-280 matrix inverse. 278-280 DFCS (digital flight control systems). 150 system linear. 41-42 generic.R.R. 242-264 model(s).76-77 Disturbance input(s). 6 DOF (degrees of freedom). 147. 275276 E. 58 Disturbance-rejection: a. set of plants. 269-270 Digitization (DIG) digital control analysis technique. 152 Engineering approach. 118-136 Direct (DIR) digital control analysis technique.m. 52-61.178-181 general. 286 Digital-computer solution: factoring of polynomials. 41-42.271-277 E. 15. 272-273 E. 276 hardware. 5 matrices. 158 Example of: . 115-117 prefilter design. 274 E.266. 150.R. 246-247 Disturbance bounds (see Bounds): case 1 (MISO) design procedure. 274 software.3. 13 non-ideal step function for implementation. 12 controller implementation.s. 222 universal design features. 126-127 simulation. sensitivity. 52-57 case 2 (MISO) design procedure. 45-46 control systems. 106-107 nonmimimum phase.7 minimum compensator gain.290 minimum order.R. 156.45-46. 133-136 Discrete MISO: basic design procedure. 271 E. 197 example6. 20 Engineering visualization.2. 160168 Electric circuits. 92-148 (see also Quantitative feedback theory) Disturbance: input-output relationships. Effective MISO: equivalents loops of MIMO system.228-229 electric circuits. vi. 266 Engineering rules. 92-148 Discrete design procedure.274 E. 276 E.R. 152 Index bounds (see Bounds) control ratio(s). 275 E. 276 E. 225-264 Disturbances. 277-278 Engineering control problem.1. 92-148 PCT design technique.247 modeling. 239 Disturbance control ratio.3 templates. 1.245 Dominance. 52.R. 274 E.157 Differential equations. 76-77 aerial refueling. 57-61 external disturbance effects. 133-136 error in the design. 7 Laplace transform of.196-197 example 6. 127-132 w-domain DIG design. 52. 195-212 definition.195-196 example 6. 244. 273-274 E.4 design techniques.p. 289 Equilibrium and trade-offs. 251 Disturbance-rejection problem. 107 Discrete control system.R. sets of. 11 asymptotic results. diagonal. MISO. 286 implementation. 118-136 Discrete-data feedback control system (see Discrete MISO) Discrete quantitative feedback technique. 276 Environmental conditions. 123-126 model with plant uncertainty.

68 Gyro. 290.47-48 maximum (see Maximum frequency response) resonant frequency. 37-38 log magnitude diagram. 228 Gain scheduling. 119 Frequency response: Bode plot. 22-23. 12. 158 MIMO plant. 34 a. 70-72 hardware noise. 298 design of flight control laws for aircraft with flexible wings. W. 102 (see also All-pass-filter) Fixed point mapping. 22-23. 151. 109. 12 lateral/directional.304 Flow graphs. 283-305 VISTA F-16 flight control system. 37-41 closed-loop. 222 Feedback structure. 33. 34 Feedback theory.f. 106 Idle speed control for automotive fuel injected engine. 41 summary. W.41 analysis. 22 bounds (see Bounds) corner. 7-9 . 73.Index 2x2 plant. 186-189 templates. 286 Gain adjustment. 16-19 Figures of merit. 293 unmanned research vehicle digital flight control system.165-168 3x3 plant.) range. 258 closed-loop tracking performance based on. 102-103 design of prefilter. 37-41. 186-189 high.162. iii. 37-41 disturbance rejection.37-38 control system. 245: Functional requirements. 150. 47-48 Gain matrix. 34. 162 Fixed point theorem. 152-154 Experimental determination. 290 systems (FSC) design.293 Hardware/software consideration. 37 peak overshoot. 211-212 Hold device: zero-order hold (see also Zero-order hold). 294 manual. 39. 22 Filter(s) (see also Prefilters): all-pass.267 Graphs. roll rate. 294-298. 113 too small. 225264 (see Aerial refueling) External disturbances. 68-70 Bode diagram. 162 Fixed point theory. 39. 133-148. 10-12 Flight tests. 186-189 Frequency: 0-dB crossing (<%). signal flow (see Signal flow graphs) Guidelines for loop shaping. Schauder. 33 Frequency sensitive matrix. 65 phase-margin frequency (aty).s. 22-23. 228 Fuel transfer model. signal (see Signal flow graphs) Frequency range: analysis. 106 Horowitz. 22 settling time. 22 rise time. 3741 figures of merit. 296 transfer function. 83 CAD program. two degrees-of-freedom. 32 insight.269 Hardware-in-the-loop simulation / implementation.186-189 low. quantitative (see Quantitative feedback theory) Feedback. 175 Ideal impulse transform. why. 43 bandwidth. sensitivity. 47 302-303 External disturbance problem. 47 sampling. 300-301 High-frequency (h. 43 trade-off. 106 Ideal sampler. 92 minimum.41. 22 peak time. Flight control: aerial refueling control system. 41 effect of additional poles and zeros on. 225-264 (see also Loop shaping) Frequency-response method (see QFT). bending modes. 399 resonant (see Resonant frequency).

294 Limiter. characteristics.252-258 Bode diagram. 298 M(o>) contours.p. 150. 149 minimum-phase (m. basically non-interacting. 183 invertibility. 160168 £77 (see Linear-time-invariant control systems) Implementation issues. 157 compensation. closed-loop. 280-281 Mapping: between . 293 n.186 Linear time-invariant (LTI). 150-155. electric circuits. 169. 294 Linear constant-coefficient differential equations.s-and w-domains.281 Matrices: disturbance. MISO. integrator wind-up. 146 shaping of Lmo(w).183.). loop method. 291-306 Lateral/directional flight control system. controller. 152 Input: command.278-280 Invertibility.p. 47-48 log magnitude-angle diagram. 292. control system. 154 transfer functions. 300 Implementation process (see Design and implementation process for a robust system) Impulse function input.290 Implementation. £0(w). 2. 181 Loops. disturbance (see Disturbance input) prefilter. 162. 147 Index drawing of. 150 Integrator wind-up limiter. effective. 155-158 MISO equivalents. 47-48 Log magnitude plots.244 feature. 94 between w. 62 shaping of L(w). 162 MATHEMATICA CAD package.m. 4.238 Initial conditions. 152 Loop method.and z-domains. 259-260 Linear superposition. 161.p. 94 fixed point. 258 Log magnitude and phase diagram. 48 tangents. controllability. 22 plants. 162. 22 non-minimum-phase (n. 273 Log magnitude (Lm). 150 MACSYMA program. 255 Lambda unmanned research vehicle (see URV).) plant. 68-70 optimum. 73 (see also Bode plots) Log magnitude-angle diagram.280 MATLAB. 22 MIMO plant. 118-126.6.296 Linearized plants. gain adjustment by use of. 241 Intrinsic turn coordination. 244. 135 shaping of L0(ja>).247 Matrix (matrices). 285 Linear simulations.63-68 Loops.400 Implementation and testing. 188 function. 83-84 CAD program. control ratio. 16-32 Inverse matrix.238 vector.m. 115-117 nominal. 278-280 . 35 gain. 152 Lag-lead compensator. 83 phase margin: angle. 286. 62-70.294 Interaction bound. 157 inverse. 158-159 QFT. 1 Linear control systems (see Linear-timeinvariant) Linear multivariable control systems (see MIMO and MISO) Linear plant models. 152 Loop shaping. 47-48 gain setting. 165 diagonal. 276 digital. 34 unit impulse. theorem. 280 Kirchhoffs law. 83 guidelines for. 161. 62-70 Loop transmission: BW. 64 frequency. of MIMO system.22 control systems. 47 nf equivalent single-loop MISO control system (see MIMO) nf prefilter/cross-coupling problems. 6. 296 Introduction to MIMO compensation. 307-327 data for. 251-252 Loop equations. 247 plant.

control ratios: (see Tracking models) (see Disturbance rejection models) Models. 41-42. 150-155 Matrix polynomial. 150. 184-192. 151. 247 Matrix equations. 47 Maximum overshoot (see Peak overshoot Mp). 328-340 Mm. 225-226. 307327 MIMO system design method 1 (see Method 1.Index observability. 158..274 Minimum-phase (m. 225-264 . 253-255 design example.183-212. 150-151 uncertain plant. 149 effective. 36-42. 217.): effective det P.s. 228-233 unstable plants. bending. 165 plant. plant transfer function. constraints on. 165 desired. 34. 307-327 TOTAL-PC CAD. 219-221. QFT) MIMO system with external disturbance inputs. 157 rank of. 213. 47-48 Mode. 173-175. 155-158 MIMO design example. 178-181. 33. 37-42 la.280-281.p. 176-181. 226-227 design equations for the 2x2 system.247 weighting. 285 Modified single-loop equivalents. 22. QFT. 22 Method 1. 223 constraint 2.73 QFT system performance. 37-41 disturbance matrix. 37-41. 85. 183-212.75-76 Models. 19-22 Multiloop electric circuits. 149-306 2x2 system. 219-221 analog control system.252-253 design example. 228-233 plant. 160-168 MIMO QFT CAD. 213-234. 42-44. 239-242 example.213-234 applications of. 2. 239-242. 214-216.218-221 effective MISO loops of. 165-168.1. 64. 223 401 system performance specifications. 278 MATRIXx CAD. plant. m''. 1. QFT) MIMO system design method 2 (see Method 2. ISO181. 160-168 external disturbance rejection problem. 202-207 3x3 system. 154 system. 269. 173. 209-212 design method 2. 246-247 external disturbance control ratio.214-216.225-264 Minimum order compensator (controller). 288 wind gust. linear. 76-77 tracking control ratio.213 example. 152 Multiple-input multiple-output (MIMO) systems. 149-264 compensation. 156 diagonal. 173-175. 154 transfer-fiinction(s). 213-234 Motor: dc (direct-current) servo. MIMO. 165 singular.152-154 MIMO QFT CAD package. 239-242 constraint 1.297 Model: dynamics.209-212. 165168. 33-91 design example. 146 MISO (see Multiple-input single-output control systems): analog control system. 150-155 example 5. 202-209. 224-225 design guidelines. 218-221 example. 304. 174.245-246 cross-coupling response. adjustment.. 150-151 polynomial. 219-221 MIMO (see Multiple-input multiple-output control systems) MIMO compensation.209-212. 92-148 equivalent: control system. 347364 MIMO plant. 173.239-242 design equations for the 3x3 system. 218-219 example (3x3). 222 conditions for "a. 155-158 conditions for existence of a solution. 245 Modeling: C-135. 21 shunt. QFT.297 Maximum frequency response (Mm). 228-233. 278 prefilter. 244. 173. 219-221 inherent constraints." 222-226 design method 1. 341-346 discrete control system. 184-192 Method 2. 216-217 equations (2x2). sensitivity. 165 tolerance. 150-155 Matrix notation. 21 position control.

45 Nondiagonal G. justification of QFT. 103-105 basic design procedure for.'s. 227-228 Nonlinear control system. 22. 170-173 (see also Bounds) reduced overdesign. 289. 225-264 synthesis of. 293 Nominal loop transmission function. 236-262 synthesis problem.213 performance bounds. 133-136 error in the design. 113-115 design procedure. 2.62. 277-278 Nonlinearity: linearization. 68-70 synthesizing. 221-222 m2 equivalent single-loop MISO system.217 regulator control system. electric.22. 4.37.p.6 Nonlinearities. 228-233 m>2 system. 33-306 analog plant. 23 Nonlinear simulation(s).44 determination. 296 feedback. MISO) model. 33-306 Multivariable control systems: quantitative feedback theory (QFT). 24. is too small. 5-6 saturation. 122 examples of. 47-52 (see also Log magnitude-angle diagram) Nodal method: electrical circuits. 4-6. structured. effective plant del P. 33 Network. 36 design example 1. all-pass-filter. 152 Noise. design of. 11.185 discrete plant. 165 m. 33-224 Observability. 62-68 type 1." 265284 Nyquist's stability criterion. 92-148 bounds. 106 plant templates. 34.p. 107-136 templates. 146 Now the "practicing engineer takes over. 24. 149-306 with external disturbances. 5 Nonlinear plant.m. uncertain plants.37. using direct polar plot. 150 overdesign. 65 a^ (see Resonant frequency). 19 bound. 158-159 matrix notation. 2.293 Nonlinear systems.): (see also Templates) prefilter. 33 tracking/external disturbance. 63 optimun. 169-173 example 5. 91. 123-126 QFT w-domain design. 75-85 design example 2. 33-91. 272 plant(s). simulation (see Simulation) synthesizing Lmo(w) (see Synthesizing). 62 shaping of.262 uncertain plants. 44.42-44 selection of. 1-7. 158-165 frequency response method. 174. 113 (o. 92 La(w). 301 aliased.165 systems.402 interaction between the loops.24 characteristics.. 72-75 basic QFT design procedure for. 165 Octave(s). 275-276 loop (see Loops. . 62-70 guidelines for.p.. 33-306 trackers. 115-117 9. 20. Index variable parameters. 33-306 33. 225-264 Multiple-input single-output (MISO) systems. 174.11. 235. 173.74. 19. 260-262. 119-122 equivalent method. 28-30 parametric variation.2. n.74 Nonminimum-phase (n.101-105. 146.91. 2. 158-168 introduction to. 102-103 analog control system.64 Nominal plant(s). 258.28-31 insight to the use of the. 85-88 discrete (sampled-data). 149. 194 regulators. 70-72 (see also Prefilter) psuedo-continuous-time system (see PCT system). unstable. 149-306 bounds (see Bounds) analog plant. 1 the engineering approach. 6 structured parametric uncertainty.m. 33 introduction.275-276 template. 29. 301 sensor. 29-31 rotated. 101-105 QFT design procedure. 152 Nichols chart. 55-57 tracking bounds on.

34-35 region of structured. 137-139 of MISO discrete-time systems. basic explanation. 275 set. 19-22 transfer function.Index co. 292-305 simple PCT example. 136-145.2. 5. 145 DIG design method. 180-181. high frequency range. 222 MIMO.34 templates. 151. 22. 173. 137 Parameter sensitivity (structured).p. 42-44 nominal. example 6.22 Overview. 28-29 Open-loop control systems. 21 Operational amplifiers (OP-AMP). 42-44 w-domain.213 reduced. 150 Overdamped response(s). 209-212 input vector. 202-207 3. 37 Overdesign. 34-35 Plant uncertainty set. 150 Plant(s): 2x2. 150. 150 conditions for existence of a solution. 185-186. 158. 163 example 5.21 preconditioning. 217 example. constraints on. 22 effect of zeros on. (see Templates) transfer function. 21-23 what is. 42-44 construction of.189-192 tracking. 150 parametric uncertainty.4. 27 Open-loop systems: dc shunt motor. 21 sensitivity analysis. 113 Plant uncertainty. 136 introduction to.176-180. 22 peak. 37-39 Peak time (tp). 275-276 open-loop. (see also Structured parametric uncertainty). 88-91 Open-loop plant.250-251 minimum-phase QFT system (see Bounds). 152 Outline of text. 34. 280 MATHEMATICA.269 Phaser representation.45 determination.218-221 example 6. D. 228-233 matrices. 22 Performance specifications. 23-24 403 bounds. 228-233 . 92 Peak overshoot. 29.170-173 external disturbance. effective. 192194 Peron root interaction measure.280-282 applicability of design technique to other plants.34. 150 minimum-phase. 36-42. 17. 139-141 PCT system design. 296 bandwidth (see Bandwidth) m.292-305 sampled-data control system example. 17-19. 21. 17. structured. 29-31 PCX (pseudo-continuous-time) control systems. 151. effective. 34. tolerances. 19-21. 150 inversion. 278-280 invertibility.48-50 Plants: matrix. 59 Plant templates. 19-21 simple mathematical description. 64 Phase-margin frequency (»„>). 19-22 region of plant. nonminimum (see Nonminimum phase) Phase-margin angle (f). 52-61. 21-23 cross-coupling effect. HI144.145 QFT applied to. 14-15 Output vector. 178-180 conditions. 180-181 matrix. 229-233 MIMO. 169-173 control system. 27 output vector. 33. 13-14 Operator notation. 92. 145-146 design summary. 176-178 condition 2.165-168. 5. 280 Pade' approximation.d. 186 simple example. 36-42 Performance analysis. 34 uncertainty. achievability of a. 19 Parameteric variation NC characteristics. 228 Phase. (see Undamped natural frequency) Open-loop characteristics. 40 second-order system. 286-288.136-145.6. 149-155 models. 165-168. condition 1. 19-23 Parametric uncertainty.247 matrix. 73. QFT. unstable.174. 92.217 Overshoot: maximum. 186189 examples.

304-305 design of the [F(w)]. 21-23 (see also Performance specifications) conditions for: for "a. 23-24 design procedure for a: MISO system. 7 sewage system. 22. 100 Polynomial matrix. 24 overview. 92-148 n. 222-226 inherent constraints. vii. 175. 165. 22 MIMO. RHP. 107-132 example 1. 126-127. xi-xvi. 85-88 design objective. 33-148 analog. 222 design example. 7 Pseudo-continuous-time control systems (see also PCT) approach. 113-115 closed-loop specifications. 149-305.p. 149-306 Polar plot analysis. 72-75 MISO system. 19 design overview. 185 set of possible. MIMO.267-270 w-domain DIG design. 16-32 L. 122 Quantitative feedback theory (see also QFT). 5 uncertainty. 31-32 bounds (see Bounds) closed-loop system specifications. 225. 25-27 design objective. 146 nonminimum-phase. 291-305 operational amplifiers (OP-AMP). 133-136 n.m.146 with variable parameters. 6 introduction. 242-263 flight control laws for aircraft with flexible wings. 236 robust design technique. 23-24 . 156 transfer function. 1-15 unstable. 225-264 MISO.m. 136 frequency bounds on. frequency domain technique. 10-12 idle speed control for automotive fuel injected engine.p. 110-112 plant templates. 5 unstable. Prefilter design (see Prefilter) Prefilter matrix. 113 QFT applications. 103-105. 146. 101-105 QFT design procedure.209-212 design method 2. now the. 7-9 lambda unmanned research vehicle digital flight control system. sensitivity" in single-loop design. 107-136 bounds. 6 regulator control problem. discrete. 227-228 problem classes. analog. 19-27 design. 88-91. 156 Prefilter: design of the [F(s)]. 12 flight control system. with external (input) disturbance. 129 correlation of. 6 Process control. 7 transparency of. 129-131 stability. introduction to. 14 welding control systems. 33-91 discrete. 9-10 QFT: basics. 129-131 good Tustin approximation. 75-85 example 2. robust. MISO. 34 nr. 12-13 robust process control.s. 5-6. 96 Problem classes.I. plant. 265-284. 25 sampled-data systems. 34. 14 wastewater treatment control system. 71-72 input.T. 214-216 design example. 129 Pole-zero locations: closed-loop. 5. analog plant. 37.150 matrix. 91. 92 Pseudo control ratio. 103-105 nondiagonal G.233 Pole-zero diagrams. 278 Practicing engineer takes over. 70-72. 16-306 applications of (see QFT applications) benefits of. 122 Poles. 226-227 existence of a solution. 34 Prewarping. 156 diagonal. 129 effect on performance. 13-14 robot controllers. 34-35 standard symbols & terminology.404 Index discrete. VISTA F-16. design method 1. 7-14 aerial refueling control system (see also Aerial refueling). 92-148 (see also Discrete MISO) single-loop MISO systems. with time response. MISO analog plant.

223 constraint 2. 271-277 QFT method (single-loop MISO system). 245 reshaping.. 187 parameter (see Parameter sensitivity) system. 37 underdamped. 19. 47 Response characteristics. 33-91.. 164 disturbances.s.74. 139-141 MISO model. 37-38 frequency of oscillation.174. 92 too small. 37 Sensitivity analysis. 129-132 Resistor-inductor circuits. insight to the.296 Sensor(s): angular rate. 19-22 simple example. 225-226 constraint 1.). 165 References.. MIMO.).Index design overview. 262 Reshaping of root locus. 276 s plane to 2-plane transformation methods. QFT. 17-19 Sensitivity fimction(s). 183. 236-262 Regulator/tracking problem. conservatism of the. 92-148 basic or uncompensated. 39-40 second-order. analysis. 25 Root locus. 5. 216 conditions for. 19. 186 Sensor noise. 85 effect of real pole on.186. 248 Regulator control problem. 23-24 single-loop system. aerial. 5.113 time. 37 Rise time (tr).). system. 19-21 simple mathematical description. 22 Robot: arm control. 2123 QFT knowledge. 37-38 peak overshoot. simulation. QFT technique. discrete MISO) (see also Discrete MISO model) example (PCT). 25 Robust design technique. 106-107 (see also Quantitative Feedback Theory.232 design techniques. 37-41 response characteristics. 25 Robustness. 37 frequency response. 129-132 Roots of characteristic equation. dc. 27 QFT technique. body of engineering. 94-96 Sampled-data (SD) feedback control system. 129-132 z-domain. 293. 365-369 Refueling boom. 276 Rank of matrix.72-75 discrete. 162 Scientific method. 129-132 time response from analysis of. 216. 37 performance characteristic. 4. 92-148 Ramped-up step function. 222-226 high frequency range. 139-141 Sampling: frequency (co. MISO. 129 Run time. sensitivity. 37-41 correlation of sinusoidal and time responses. 92 minimum (fii)min. 265 S-D (see Sampled-data control system) Second-order systems. 34 structured parametric uncertainty (see also Structured parametric uncertainty): basic explanation. 40 effect of real zero on. 106-107 simple PCT example. time. 293 angle. 12-13 Robust control system design. 226-227 a.222-226 application of. 293 Schauder fixed point theorem. 34 . 188 low frequency range. 152 Resonant frequency (co. 21 Set of: acceptable of regions % 155. 37-39 Response: overdamped. 18-19. 119 Saturation.s.74. 296 yaw rate.25 design and implementation process for a (see also Design and implementation process for a robust control system) Robust control systems. 11. 37-41 determination of Mm and aj. 294 sideslip. 186-189 applications of. 37-39 transients. 224-225 inherent constraints. 296. 255 arbitrarily small (a. 62. 109. 294 roll angle. 27-31 QFT. 129-132.302 Servo motor. 37-41 simple. 405 analog.

dc. 183-212 Single-loop structures. 94 Stable uncertain plant. 93. 158 3x3 system.).293 operator-in-the-loop. 154 Software considerations. 2. 37 Simulator. 14 SFG (see Signal flow graphs) Shunt motor. 37-41 representation. 258. 135 regions.150 lower bounds. 37. uncertain. 149-165 State space representation: LTIMIMO system. 1. specifications) Speed control.295. 5 MIMO (see Multiple-input multiple-output control systems) minimum-phase. 72 Step function: MISO (see Multiple-input single-output control systems) . 115 Standard symbols and terminology. 19-21 simple mathematical description. multiple-input multiple-output. 225-264 System: analog.406 plants. m. 107 Step input. 6 distributed. 194-195 bounds. linear. 21-22 Superposition theorem.269 2x2 system. 17. xi-xvi Synthesizing: Lm(w).298. 34 transfer function matrices. simplification of the.293 linear. 62-70 tracking models. 160 nonlinear systems. 150-155 Steady-state error. 300-301 Specifications. Structured parametric uncertainty. QFT. 150 response model. dc shunt motor.304 discrete MISO system. 12. design method 1.136 engineering interactive.276 Index non-ideal. 39 controllable. 225-264 linear. 37-41 System performance specifications (see Performance specifications) System with external disturbance inputs. 94-96 prewarping. 19-22. 41-42. 165 performance specification. 290 hardware-in-the-loop. performance (see Performance. 154 upper bounds. Lamars. 149-264 bandwidth.293 non-ideal step function for. 21 simple example. 45-47 z-domain. 33-91.183 Symbols.289. 19 constrained. 158 Simulation of design (see Simulations)) Simulations). 150 Settling time («. 194 U-contour. 129-132 Nyquist's criterion (see Nyquist's stability criterion). 149306 basic explanation. 1 linear differential equations.127-132. 45-47 location of poles and (in z-plane). 259-260. 22 Sewage system. 45^7 universal high-frequency boundary (UHFB). 276 unit. 23-24 Systems: closed-loop (see Closed-loop control system) control (see Control systems) distributed. polar plots. 276 nonlinear. 24 design procedures (see Quantitative feedback theory). 146 Lmo(w). xi-xvi State equations. QFT. 165 representation. 165 cross-coupling effect.289. 92-148 external disturbance. 169. discrete-data feedback control. 96 Stability: analysis. 290 run time. 118-126. 19-21 s-plane and z-plane relationship.290 Single-input single-output (SISO) closed-loop system: disturbance-rejection design principles for. 23-24 matrix. 150 observable. 290. 19-22 Signal flow graph(s) (SFG).72 Straight climb maneuver. 2. 118 L0(ja>) (shaping of)..p.297. 2 region of. 189192 Singular matrix. 238 Single-loop (MISO) equivalents. 24 bound (U-contour). 22. 36-42 tracking. 154. 37-41 uncontrollable.260-262.

97-100 Type 1 system. 37 and correlation of pole-zero diagram with.2. 22-23. 196-197 example 6. 88-91 w-domain characteristics. 36-42 Tracking/regulator control system. 5 Tracking control system.217 Transparency of QFT. vii. 298-300 special case. 74. selection of design envelope. 20 matrix. 94 Tustin model of control ratio.37 sensitivity analysis. 37 rise (tr). 195-212 example 6. outline of. Tracking bounds (see Bounds). good. 37-41 augmented. 48-50 expansion process. vii. 5 Tracking models.3.200-202 universal design features. high frequency range.63 Transmission zeros.42-44. 97-100 w-plane to w -plane. 296 frequency range for.296 characteristics of. 36-42 second-order system. 92 approximation. 185-186. 37 Tolerance(s). 88-91. 24.1. 37 peak time (tp). 130-132 and correlation of sinusoidal responses. 202-203 Transparency of QFT. 123-125 example. 98 Transient response: correlation to frequency response. 268. 93-94 (see Tustin transformation) Transform: Z. 24-26. 44 contours. 123 width. 35 nominal loop. 328-334 Tracking control ratio(s). bounds (see Bounds) matrix.37 plant. 183 equilibrium and.272-273 360° wide. 92. 42-44 plant uncertainty.267-270 Tustin.transform. set of. 253 aircraft. 40 simple second-order. 43. 14-15 Time response: best. 18-19 407 Temperature. 37 Time: rise.299 nominal plant. 142 s. 175. 37-41 of closed-loop system.Index nonminimum-phase. 22. 64. 165. 22 second-order disturbance-rejection model. set of. 22-23. 76-77 settling time (see Settling time) specifications. 22. 37-39 synthesize. 21 environmental conditions. 100 w. multivariable control system. 93 options. 19.199-202 Trade-offs (see also Equilibrium and trade-offs). 150-151 minimum phase. dc servomotor.38. 37-39 Transmission function: loop.63. simulation. 195-196 example 6. 276 settling (see Settling time) Time-response specifications. 34 Transfer-function matrices. 262 Trade-off. design performance.267-270 Transfer fiinction(s). 22 run. 142 Tustin transformation. 3741 for unit-step input. 217. 118 Tracking input-output relationships. 150 Transformation: bilinear. 34 prefilter. based on frequency response. 262 Tracking/regulator MIMO system. 92-100 s.plane to z-plane. 44-45 plant.47-52 procedure for determining. QFT.118 . 192-194.and w -domain. 113 very high frequency. 68 design. closed-loop. xi-xvi Text. 19. 189-192 TOTAL-PC CAD package. 21. 34 nonminimum phase. 298300 discrete systems. 175.267-269. 50-52 w-plane to z-plane.to z-plane. 154 performance. 37-41 Tracking performance.223 Terminology. 44.29 Template(s). 331-334 Trackers. 43-44 selection of design envelope.212 unstable plants.

45 Uncertain plant. 39 URV bending mode. 34 Vector(s): input.p. 291-305 bending modes. 297-298 stability. 115 Uncertain plants. 41 effect ony(t). lambda (see URV) Unmodelled behavior. 137 transfer function. 150.247 Uncompensated sampled-data control system. 94-96 hardware/software considerations.136 plant templates. 300-301 implementation issues. plants.p. 129-131 z-plane. 150 state. 9-10 Why feedback. 93 Weighting matrix W. 37 Unit circle. 260 generic.408 Type m system. 146 Unstable plants. 98 matched. 16-19 Wind gust model. 151 VISTA F-16. 93 w(s)-plane to w '(i)-plane transformation. 88-91 Unstable plants.238 Unit-step input.). 94.247 stable. 302 Variable parameters. 92 characteristics. 292 first. 34 m. 37 Universal high-frequency boundary (UHFB). 251 z-domain. 100 unit circle. 298-300 unmodelled behavior. 98-99 Wastewater treatment control system. 150 output. 228 gain matrix.m. 93-94 . exact.233 transmission.301-303 design cycle. 297.247. 122 U-contour. 129-131 Unit impulse input. time response for. 154. MIMO. 165. and n. 162. 106. 294-295 third. 17. 99 Index Warping.228-233. responses. 111.226. 165 CMa). (see a. Unmanned research vehicle. 216. 93 property.so Damping ratio) second. 245 Dryden. 92. 94-96 S-transform. 291 -305 z-plane and s-plane relationship. 88-91 URV. 99 Zero-order-hold (ZOH). template generation for. 300 selection of design envelope. digital flight control system. 45. 303 Underdamped. 228 Welding control system. 150. 96 z-plane to s-plane correlation. 123 property. 292-294 fourth. plant with. 165 Undamped natural frequency (tu. 106 Zeros: dominant. z-domain. 45-47 UHFB (universal high-frequency boundary).271 frequency sensitivity. 10-12 w(s)-plane. 92 root-locus. 22-24 matrix. 39^0 RHP. 94-96 prewarping. 118.217. 297 Unstable P(s). 14 w -domain. 296-297 design process example. 34. 136-144 Uncontrollable system(s). 37 systems.

Sign up to vote on this title
UsefulNot useful

Master Your Semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master Your Semester with a Special Offer from Scribd & The New York Times

Cancel anytime.