Linear Matrix Inequalities in Robust Control. Venkataramanan (Ragu) Balakrishnan School of ECE, Purdue University MTNS 2002

Similar documents
Semidefinite Programming Duality and Linear Time-invariant Systems

Modern Optimal Control

FEL3210 Multivariable Feedback Control

LMI Methods in Optimal and Robust Control

Fast Algorithms for SDPs derived from the Kalman-Yakubovich-Popov Lemma

Robust Stability. Robust stability against time-invariant and time-varying uncertainties. Parameter dependent Lyapunov functions

State feedback gain scheduling for linear systems with time-varying parameters

Robust Multi-Objective Control for Linear Systems

H 2 Suboptimal Estimation and Control for Nonnegative

June Engineering Department, Stanford University. System Analysis and Synthesis. Linear Matrix Inequalities. Stephen Boyd (E.

IN many practical systems, there is such a kind of systems

Convex Optimization Approach to Dynamic Output Feedback Control for Delay Differential Systems of Neutral Type 1,2

José C. Geromel. Australian National University Canberra, December 7-8, 2017

Research Article An Equivalent LMI Representation of Bounded Real Lemma for Continuous-Time Systems

Marcus Pantoja da Silva 1 and Celso Pascoli Bottura 2. Abstract: Nonlinear systems with time-varying uncertainties

LINEAR QUADRATIC OPTIMAL CONTROL BASED ON DYNAMIC COMPENSATION. Received October 2010; revised March 2011

Robust Anti-Windup Compensation for PID Controllers

Hybrid Systems Course Lyapunov stability

EE363 homework 8 solutions

Parameterized Linear Matrix Inequality Techniques in Fuzzy Control System Design

Analysis of undamped second order systems with dynamic feedback

ECE504: Lecture 8. D. Richard Brown III. Worcester Polytechnic Institute. 28-Oct-2008

EE363 homework 7 solutions

Event-triggered control subject to actuator saturation

Controller Coefficient Truncation Using Lyapunov Performance Certificate

Optimization in. Stephen Boyd. 3rd SIAM Conf. Control & Applications. and Control Theory. System. Convex

SYNTHESIS OF ROBUST DISCRETE-TIME SYSTEMS BASED ON COMPARISON WITH STOCHASTIC MODEL 1. P. V. Pakshin, S. G. Soloviev

Static Output Feedback Stabilisation with H Performance for a Class of Plants

STABILITY AND STABILIZATION OF A CLASS OF NONLINEAR SYSTEMS WITH SATURATING ACTUATORS. Eugênio B. Castelan,1 Sophie Tarbouriech Isabelle Queinnec

6.241 Dynamic Systems and Control

1. Find the solution of the following uncontrolled linear system. 2 α 1 1

Introduction to linear matrix inequalities Wojciech Paszke

Hybrid Systems - Lecture n. 3 Lyapunov stability

A new robust delay-dependent stability criterion for a class of uncertain systems with delay

State estimation of uncertain multiple model with unknown inputs

Solution of Linear State-space Systems

Multi-Model Adaptive Regulation for a Family of Systems Containing Different Zero Structures

Constrained interpolation-based control for polytopic uncertain systems

Linear Matrix Inequality (LMI)

Stability of linear time-varying systems through quadratically parameter-dependent Lyapunov functions

Stabilisation of network controlled systems with a predictive approach

arzelier

Cherry-Picking Control Nodes and Sensors in Dynamic Networks

Multiobjective Optimization Applied to Robust H 2 /H State-feedback Control Synthesis

From Convex Optimization to Linear Matrix Inequalities

7.1 Linear Systems Stability Consider the Continuous-Time (CT) Linear Time-Invariant (LTI) system

Constructing Lyapunov-Krasovskii Functionals For Linear Time Delay Systems

Integral action in state feedback control

NONLINEAR PID CONTROL OF LINEAR PLANTS FOR IMPROVED DISTURBANCE REJECTION

A New Strategy to the Multi-Objective Control of Linear Systems

Analytical Validation Tools for Safety Critical Systems

Robust Observer for Uncertain T S model of a Synchronous Machine

Optimization based robust control

Module 02 CPS Background: Linear Systems Preliminaries

MATH4406 (Control Theory) Unit 1: Introduction Prepared by Yoni Nazarathy, July 21, 2012

Stabilization of a Pan-Tilt System Using a Polytopic Quasi-LPV Model and LQR Control

On Computing the Worst-case Performance of Lur'e Systems with Uncertain Time-invariant Delays

Lecture Note 5: Semidefinite Programming for Stability Analysis

Robust Output Feedback Controller Design via Genetic Algorithms and LMIs: The Mixed H 2 /H Problem

Graph and Controller Design for Disturbance Attenuation in Consensus Networks

Control Systems (ECE411) Lectures 7 & 8

Stability and performance analysis for input and output-constrained linear systems subject to multiplicative neglected dynamics

Two Challenging Problems in Control Theory

Memoryless Control to Drive States of Delayed Continuous-time Systems within the Nonnegative Orthant

ECE 680 Linear Matrix Inequalities

ON POLE PLACEMENT IN LMI REGION FOR DESCRIPTOR LINEAR SYSTEMS. Received January 2011; revised May 2011

Denis ARZELIER arzelier

Robust optimization methodology and applications

Stability and Robustness Analysis of Nonlinear Systems via Contraction Metrics and SOS Programming

Quadratic Stability of Dynamical Systems. Raktim Bhattacharya Aerospace Engineering, Texas A&M University

H State-Feedback Controller Design for Discrete-Time Fuzzy Systems Using Fuzzy Weighting-Dependent Lyapunov Functions

Lecture 10: Linear Matrix Inequalities Dr.-Ing. Sudchai Boonto

Modern Optimal Control

Appendix A Solving Linear Matrix Inequality (LMI) Problems

Research Article Convex Polyhedron Method to Stability of Continuous Systems with Two Additive Time-Varying Delay Components

GLOBAL ANALYSIS OF PIECEWISE LINEAR SYSTEMS USING IMPACT MAPS AND QUADRATIC SURFACE LYAPUNOV FUNCTIONS

H Synchronization of Chaotic Systems via Delayed Feedback Control

ROBUST PASSIVE OBSERVER-BASED CONTROL FOR A CLASS OF SINGULAR SYSTEMS

LMI Methods in Optimal and Robust Control

Optimization Based Output Feedback Control Design in Descriptor Systems

Agenda. 1 Cone programming. 2 Convex cones. 3 Generalized inequalities. 4 Linear programming (LP) 5 Second-order cone programming (SOCP)

Control Systems Design

Research Article Partial Pole Placement in LMI Region

Module 07 Controllability and Controller Design of Dynamical LTI Systems

Dual matrix inequalities in stability and performance analysis of linear differential/difference inclusions

Dissipativity. Outline. Motivation. Dissipative Systems. M. Sami Fadali EBME Dept., UNR

Tracking Control of a Class of Differential Inclusion Systems via Sliding Mode Technique

Robust Anti-Windup Controller Synthesis: A Mixed H 2 /H Setting

Mathematics for Control Theory

Anti-Windup Design with Guaranteed Regions of Stability for Discrete-Time Linear Systems

Semidefinite and Second Order Cone Programming Seminar Fall 2012 Project: Robust Optimization and its Application of Robust Portfolio Optimization

Approximation and Complexity Trade-off by TP model transformation in Controller Design: a Case Study of the TORA system

An LMI approach for robust stability of genetic networks

OPTIMAL CONTROL. Sadegh Bolouki. Lecture slides for ECE 515. University of Illinois, Urbana-Champaign. Fall S. Bolouki (UIUC) 1 / 28

EE C128 / ME C134 Feedback Control Systems

Synthesis of Static Output Feedback SPR Systems via LQR Weighting Matrix Design

Efficient robust optimization for robust control with constraints Paul Goulart, Eric Kerrigan and Danny Ralph

Control Systems I. Lecture 2: Modeling. Suggested Readings: Åström & Murray Ch. 2-3, Guzzella Ch Emilio Frazzoli

ROBUST CONTROLLER DESIGN: POLYNOMIALLY PARAMETER DEPENDENT LYAPUNOV FUNCTION APPROACH

IE 521 Convex Optimization

Transcription:

Linear Matrix Inequalities in Robust Control Venkataramanan (Ragu) Balakrishnan School of ECE, Purdue University MTNS 2002

Objective A brief introduction to LMI techniques for Robust Control Emphasis on robust control, not LMIs Only techniques that use LMIs Tutorial, no new results General ideas, little detail

What is robust control? Modeling often first step in engineering Models often inexact: Exact models difficult, expensive, sometimes impossible Simpler models often desirable Need to analyze and design for inexact models Good designs, designed for one model, may fail for even a nearby model Hence robust control

Control system analysis and design framework w u Inexact system model z y w is vector of exogenous inputs z is vector of outputs of interest u is vector of control inputs y is vector of observed outputs Analysis problems: Study of properties from w to z Design problems: Design control law u = K(y, t) such that map from w to z is good

Polytopic models S_2 S_1...... S_n... Actual model lies inside a polytope Vertices S 1,..., S n represent extremes of operating conditions or parameter values, or linear models from linearization of a nonlinear system Goals of robust control: Analyze worst-case behavior over the polytope Design for worst-case behavior over the polytope

Feedback uncertainty models w u p Linear model y z q models uncertainties, nonlinearities, parameters Assumed to have additional properties: bounded in some norm, passive, etc structured (diagonal, block diagonal, etc) Goals of robust control: Analyze worst-case behavior over all Design for worst-case behavior over all

Robust stability Question: Inputs identically zero. Does state go to zero for all ICs irrespective of? Sufficient condition: There exists V (ζ) = ζ T P ζ, P > 0, s.t. dv (x(t))/dt < 0 a.k.a. quadratic stability condition yields LMIs for a number of robust control models Thus: Search for quadratic Lyapunov functions for robust stability is an LMI

LMIs: A very brief introduction Linear Matrix Inequality (LMI) F (x) = F 0 + x 1 F 1 + + x m F m 0, ( ) where x R m is the variable, and F i s are given symmetric matrices A convex constraint on x R m Feasibility problem: Find x such that ( ) holds Optimization problem: Minimize c T x over ( ) Fundamental complexity is low, polynomial-time algorithms available Efficient solvers available

Example: Robust stability of polytopic systems ẋ(t) = A(t)x(t), A(t) Co {A 1,..., A L }. There exists V (ζ) = ζ T P ζ, P > 0, s.t. dv (x(t))/dt < 0 iff P = P T > 0, A(t) T P + P A(t) < 0, A(t) Co {A 1,..., A L }, or P = P T > 0, A T i P + P A i < 0, i = 1,..., L. (An LMI) Technique can be generalized to more sophisticated Lyapunov functions ( parameter-dependent ) Can account for rate of parameter variations

Example system: Example: Robust stability of feedback-uncertain system ẋ(t) = Ax(t)+Bp(t), q(t) = Cx(t), p(t) = (t)q(t), (t) 1. There exists V (ζ) = ζ T P ζ, P > 0, s.t. dv (x(t))/dt < 0 iff x(t) T (A T P + P A)x(t) + 2x(t) T P Bp(t) < 0 whenever p(t) T p(t) q(t) T q(t). Equivalent to x(t) T (A T P + P A)x(t) + 2x(t) T P Bp(t) + q(t) T q(t) p(t) T p(t) < 0

Equivalent to P > 0, λ 0, AT P + P A + C T C B T P (An LMI) P B I < 0. Stability condition has connections with small-gain theorem Technique can be generalized to more sophisticated Lyapunov functions Can handle a much larger class of Can handle structure information on

Robust performance analysis Question: What is worst-case value, over uncertainties, of a performance measure 0 φ(x(t), w(t), z(t)) dt? Approach: Suppose V (x(t)) = x(t) T P x(t), P > 0, and for all trajectories d V (x(t)) φ(x(t), w(t), z(t)) dt ( ) Then, x(0) T P x(0) 0 φ(x(t), w(t), z(t)) dt, or x(0) T P x(0) an upper bound on robust performance Condition ( ) is an LMI for a large class of functions φ Approach applicable to both polytopic systems and feedback-uncertain systems Approach can be extended to handle more sophisticated Lyapunov functions

Robust control design w u Uncertain System y z u=k(y,t) Problem: Find the optimal control strategy, u = K(y, t) s.t. closed-loop system enjoys good robustness properties Some feedback possibilities: y = x, i.e., state feedback Only a linear combination of state components available, i.e., output feedback The uncertainties are measurable in real-time, and can be used for feedback, i.e., gain-scheduled

Constant state feedback Feedback strategy is u(t) = Kx(t) Yields LMI problems in several important cases For example, polytopic systems: ẋ(t) = A(t)x(t)+Bu(t), u(t) = Kx(t), A(t) Co {A 1,..., A L } Closed-loop system stable if P = P T > 0, (A i + BK) T P + P (A i + BK) < 0, i = 1,..., L. With Q = P 1, Y = KP 1, Q = Q T > 0, QA T i + Y T B T + A i Q + BY < 0, i = 1,..., L. (An LMI)

Output-feedback design with LMIs Feedback strategy y = K(y(t), t) In general much harder than state feedback, get LMIs only in a handful of cases Underlying difficulty illustrated with constant state-feedback design for polytopic systems: ẋ(t) = A(t)x(t) + Bu(t), y(t) = Cx(t), u(t) = Ky(t), Closed-loop system stable if A(t) Co {A 1,..., A L } P = P T > 0, (A i +BKC) T P +P (A i +BKC) < 0, i = 1,..., L. (No convexification possible)

Gain-scheduled-feedback design with LMIs Feedback scheduled by the measured uncertainties Example: feedback-uncertain systems (t) q y Linear system p u v Klti f (t) Determining K lti and a quadratic Lyapunov function for the closed-loop system is an LMI problem

Concluding remarks LMIs offer numerical solution of many robust control problems Interest began in the 1990s when efficient LMI algorithms became available, and continues to grow Rich literature available, many specialized results for special robust control problems

Challenges Many practical problems yield large-scale LMI problems, current solvers often inadequate. Special-purpose algorithms and solvers desirable Study of applications of LMIs to robust control fairly mature, other applications remain to be explored

Some references SIAM monograph Linear Matrix Inequalities in System and Control Theory, by Boyd et al. Kluwer Handbook on Semidefinite Programming, Wolkowicz et al. (editors) SIAM collection Recent Advances on LMI Methods in Control, El Ghaoui et al. (editors) Most recent control conferences ACC and CDC