Review
From static output feedback to structured robust static output feedback: A survey

https://doi.org/10.1016/j.arcontrol.2016.09.014Get rights and content

Abstract

This paper reviews the vast literature on static output feedback design for linear time-invariant systems including classical results and recent developments. In particular, we focus on static output feedback synthesis with performance specifications, structured static output feedback, and robustness. The paper provides a comprehensive review on existing design approaches including iterative linear matrix inequalities heuristics, linear matrix inequalities with rank constraints, methods with decoupled Lyapunov matrices, and non-Lyapunov-based approaches. We describe the main difficulties of dealing with static output feedback design and summarize the main features, advantages, and limitations of existing design methods.

Introduction

Static output feedback design is a theoretically challenging issue in control theory and it has attracted considerable attention due to its great importance in practice. However, so far, there has been no exact solution to this prominent problem which can guarantee the design of static output feedback or determine that such a feedback does not exist. The fact is that the problem is intrinsically a Bilinear Matrix Inequality (BMI) problem which is generally NP-hard (Toker & Ozbay, 1995); furthermore, it becomes non-smooth in the case of problem formulation in the space of the controller parameters (Toscano, 2013).

To solve the static output feedback design problem, well-known bilinear matrix inequality (BMI) solvers such as the commercial software package PENBMI (Henrion, Lofberg, Kocvara, Stingl, 2005, Kocvara, & Stingl) and the free open-source MATLAB toolbox PENLAB (Fiala, Kocvara, & Stingl, 2013) can be applied. The algorithms behind these solvers combines the ideas of the (exterior) penalty and (interior) barrier methods with the augmented Lagrangian approach (Kocvara & Stingl, 2003). These solvers can locally solve all kinds of BMI problems, including static output feedback. Since our aim is to survey dedicated static output feedback design methods, we no longer discuss these general BMI approaches. Note however that BMI solvers most often fail to provide a solution for the static output feedback BMI problems, and the choice of an initial guess is very crucial for these solvers.

The only survey dedicated to static output feedback has been conducted in Syrmos, Abdallah, Dorato, and Grigoriadis (1997). Since then, the past two decades have witnessed much theoretical progress on static output feedback design which has not been covered in that survey. A large amount of research has been carried out on the development of the static output feedback controllers according to Lyapunov theory via linear matrix inequality based (LMI-based) approaches (e.g. Agulhari, Oliveira, Peres, 2012, Apkarian, Noll, Tuan, 2003, Arzelier, Gryazina, Peaucelle, Polyak, 2010, Arzelier, Peaucelle, 2002, Benton, Smith, 1998, Cao, Sun, 1998, Cao, Sun, Mao, 1998, Dabboussi, Zrida, 2012, Dong, Yang, 2013, Du, Yang, 2008, Ebihara, Hagiwara, 2003, Ebihara, Tokuyama, Hagiwara, 2004, Geromel, de Souza, Skelton, 1998b, Ghaoui, Oustry, Ait-Rami, 1997, Ghaoui, Balakrishnan, 1994, Grigoriadis, Beran, 2000, Grigoriadis, Skelton, 1996, Hassibi, How, Boyd, 1999, Iwasaki, 1999, Iwasaki, Skelton, 1995b, Karimi, Sadabadi, 2013, Kim, Moon, Kwon, 2007, Koroglu, Falcone, 2014, Lee, Lee, Kwon, 2006, Leibfritz, 2001, Mehdi, Boukas, Bachelier, 2004, Moreira, Oliveira, Peres, 2011, Noll, Torki, Apkarian, 2004, Peaucelle, Arzelier, 2001a, Sadabadi, Karimi, 2013a, Sadabadi, Karimi, 2013b, Sadabadi, Karimi, 2015, Sadeghzadeh, 2014, Tran Dinh, Gumussoy, Michiels, Diehl, 2012). Most of these methods present an iterative algorithm in which a set of LMIs is iteratively repeated until some certain termination criteria are met. In addition to the Lyapunov-based approaches, there exist non-Lyapunov-based static output feedback control strategies (see, e.g. Apkarian, 2013, Apkarian, Bompart, Noll, 2007, Apkarian, Noll, 2006, Arzelier, Deaconu, Gumussoy, Henrion, 2011, Burke, Henrion, Overton, 2006b, Chesi, 2014, Gumussoy, Henrion, Millstone, Overton, 2009, Gumussoy, Overton, 2008, Peretz, 2016).

The objective of this paper is to provide a comprehensive review on the existing static output feedback design methods. The main focus is on pure stabilizing static output feedback design with no other specification. But the paper also addresses the problem of structured feedback, simultaneous stabilization, multi-performance, and robust control design. All methods and approaches described in the survey are gathered in order to provide a comprehensive classification. All results have been reinterpreted and rewritten so as to fit a common notation/framework. The notation uniformization allows a simplified overview on the differences and resemblances of the results. It allows as well to provide direct extensions of the existing results for example using system duality. Due to the fact that fixed-order dynamic output-feedback can equivalently be transformed into static output feedback by introducing an augmented plant (Ghaoui et al., 1997), this survey paper can also be used for fixed/low-order control design problem.

The paper is organized as follows. Section 2 presents problem statement and main difficulties associated with stabilizing static output feedback design and its extensions to structured feedback, simultaneous stabilization, multi-performance, and robust control. The five sections that follow provide our classification of SOF design methods. Section 3 focuses on special cases where under specific structures of the open-loop system, the SOF problem becomes convex. Section 4 reviews the available literature on iterative LMI heuristics for the intrinsically BMI nature of SOF design. Section 5 covers the heuristics related to a reformulation of the SOF design as LMIs with rank constraints. While all the previous sections describe results build out of classical Lyapunov conditions, Section 6 is devoted to methods with decoupled Lyapunov matrices that have better characteristics with respect to robustness. Section 7 exposes alternative approaches which are non-Lyapunov-based. All the classes of results are analyzed in terms of their known or claimed numerical characteristics, well as in terms of their ability to address the structured feedback, simultaneous multi-performance, and robustness issues. The paper ends with global concluding remarks in Section 8.

The notation used in this paper is standard. In particular, matrices I and 0 are the identity matrix and the zero matrix of appropriate dimensions, respectively. The symbol ⋆ denotes symmetric blocks in block matrices. The symbols AT, {A}S, A, ‖AF, and A12 are respectively notations for the transpose of A, {A}S=A+AT, the maximal rank perpendicularity such that AA=0, Frobenius norm of A, and the unique nonnegative-definite square root of positive-definite matrix A. For symmetric matrices, P > 0 (P < 0) indicates the positive-definiteness (the negative-definiteness).

Section snippets

Main SOF stabilization problem

Consider a linear time-invariant (LTI) continuous-time system x˙(t)=Ax(t)+Bu(t)y(t)=Cx(t)and a static output feedback controller u(t)=Ky(t)where xRn is the state, uRni in the control input, and yRno is the output of the system. The state-space matrices A, B, C, and the control gain K are of appropriate dimensions. The closed-loop system is described as follows: x˙(t)=(A+BKC)x(t)and its stability is equivalent to that of the dual system x˙d(t)=(A+BKC)Txd(t).

Theorem 1

The following statements are

Convex cases

From our study of the literature we have been able to extract five types of results in which a particular structure of the data enables to convert the non-convex static output feedback design to a convex optimization problem. The first of these results is the well known state-feedback case (and its dual case related to observer design).

Proposition 1

In the two following cases a change of variables involving the Lyapunov matrix makes the problem convex:

  • 1.

    In the full-actuation case,B=I, if there existP=PT and L

Iterative LMI heuristics

As seen in the previous section, only special cases provide convex LMI conditions for the design of SOF gains. Meaning that only subsets of all possible SOF gains can be designed with these methods. Some of these subsets can be empty, even when there exists some stabilizing gain. To go further, one can address the original problem that happens to be bilinear in the variables. A natural approach, largely explored in the literature, is to proceed as in Proposition 5 by freezing some of the

Heuristics for solving LMIs with rank constraints

As seen in the last proposition, there is a high potential of algorithms that would be based on the two dual constraints (5) and (6). It is the case of all the algorithms that follow which are based on the following reformulation of the problem, where the first two conditions are noting but (6) and (5) after applying the elimination lemma (Iwasaki, Skelton, 1994, Iwasaki, Skelton, 1995a, Skelton, Iwasaki, Grigoriadis, 1998) and the last two guarantee that QP=I and hence the SOF problem has a

Methods with decoupled Lyapunov matrices

In the previous sections we have recalled many heuristic techniques based on the formulations (5) and (6) that in general do not allow during iterations to optimize simultaneously over the control gain and the Lyapunov matrices. The reason for this difficulty is in the products between these decision matrices in the matrix inequalities. In the following, we expose two alternative formulations that allow decoupling of the control gains (or some matrices which in turn allow to build the SOF

Non-Lyapunov-based approaches

In addition to Lyapunov-based approaches, there exist non-Lyapunov-based methods. These approaches focus on solving the following optimization problem: minKα(A+BKC)where α is the spectral abscissa of the closed loop system, i.e. the maximum real part of its eigenvalues. The above optimization problem is non-convex and non-smooth. In fact, the lack of convexity and smoothness of the spectral abscissa and other similar performance criteria make the above optimization problem difficult to solve (

Conclusions

This paper reviews the existing methods for the design of a static output feedback. It does not claim to provide an exhaustive list of the many contributions on the topic, but we believe it covers all main methods. The survey proposes a comprehensive classification of these methods: convex cases; iterative LMIs for the BMI problem; iterative LMIs for the rank constraint formulation; achievable improvements with decoupled Lyapunov matrices; non-Lyapunov approaches. The global conclusion is that

References (136)

  • J. Frechard et al.

    Drive requirements for elastic web roll-to-roll systems

    Mechanism and Machine Theory

    (2013)
  • S. Gumussoy et al.

    Multiobjective robust control with HIFOO 2.0

    6th IFAC symposium on robust control design. Haifa, Israel

    (2009)
  • A. Hassibi et al.

    A path-following method for solving BMI problems in control

    American control conference (ACC), San Diego, CA, USA

    (1999)
  • D. Henrion et al.

    Solving polynomial static output feedback problems with PENBMI

    44th IEEE conference on decision and control (CDC) and the European control conference (ECC), Seville, Spain

    (2005)
  • D. Henrion et al.

    Rank-one LMI approach to simultaneous stabilization of linear systems

    System and Control Letters

    (1999)
  • G. Hilhorst et al.

    Sufficient LMI conditions for reduced-order multi-objective H2/H control of LTI systems

    European Journal of Control

    (2015)
  • T. Iwasaki

    The dual iteration for fixed-order control

    IEEE Transaction on Automatic Control

    (1999)
  • T. Iwasaki et al.

    Parametrization of all stabilizing controllers via quadratic Lyapunov functions

    Journal of Optimization Theory and Applications

    (1995)
  • A. Karimi et al.

    Robust control of polytopic systems by convex optimization

    Automatica

    (2007)
  • A. Karimi et al.

    Fixed-order controller design for state space polytopic systems by convex optimization

    5th IFAC symposium on system structure and control, Grenoble, France

    (2013)
  • P.P. Khargonekar et al.

    Robust stabilization of uncertain linear systems via linear control: Quadratic stabilizability and H control theory

    IEEE Transaction on Automatic Control

    (1990)
  • R. Lind et al.

    Evaluating dk iteration for control design

    American control conference, Baltimore

    (1994)
  • T. Mitchell et al.

    Hybrid expansion-contraction: A robust scaleable method for approximating the H norm

    IMA Journal of Numerical Analysis

    (2016)
  • C. Agulhari et al.

    LMI relaxations for reduced-order robust H control of continuous-time uncertain linear systems

    IEEE Transaction on Automatic Control

    (2012)
  • D. Ankelhed et al.

    A quasi-Newton interior point method for low order H-infinity controller synthesis

    IEEE Transaction on Automatic Control

    (2011)
  • D. Ankelhed et al.

    A partially augmented Lagrangian method for low order H-infinity controller synthesis using rational constraints

    IEEE Transaction on Automatic Control

    (2012)
  • P. Apkarian

    Tuning controllers against multiple design requirements

    American control conference (ACC), Washington, DC, USA

    (2013)
  • Apkarian, P. Software (hinfstruct, looptune, systune), Available:...
  • P. Apkarian et al.

    Non-smooth structured control design with application to PID loop-shaping of a process

    International Journal of Robust and Nonlinear Control

    (2007)
  • P. Apkarian et al.

    Parametric robust structured control design

    IEEE Transaction on Automatic Control

    (2015)
  • P. Apkarian et al.

    Nonsmooth H synthesis

    IEEE Transaction on Automatic Control

    (2006)
  • P. Apkarian et al.

    Fixed-order H control design via a partially augmented Lagrangian method

    International Journal of Robust and Nonlinear Control

    (2003)
  • P. Apkarian et al.

    Concave programming in control theory

    Journal of Global Optimization

    (1999)
  • P. Apkarian et al.

    Robust control via concave minimization local and global algorithms

    IEEE Transaction on Automatic Control

    (2000)
  • D. Arzelier et al.

    H2 for HIFOO

    International conference on control and optimization with industrial applications

    (2011)
  • D. Arzelier et al.

    Mixed LMI/randomized methods for static output feedback control design

    American control conference (ACC10)

    (2010)
  • D. Arzelier et al.

    An iterative method for mixed H2/H synthesis via static output-feedback

    41st IEEE conference on decision and control (CDC), Las Vegas, Nevada, USA.

    (2002)
  • G.J. Balas et al.

    μ-Analysis and synthesis toolbox

    (1993)
  • B. Barmish

    Necessary and sufficient condition for quadratic stabilizability of an uncertain system

    Journal of Optimization Theory and Applications

    (1985)
  • B.R. Barmish

    Stabilization of uncertain systems via linear control

    IEEE Transaction on Automatic Control

    (1983)
  • R.E. Benton et al.

    Static output feedback stabilization with prescribed degree of stability

    IEEE Transaction on Automatic Control

    (1998)
  • J.V. Burke et al.

    Satibilization via nonsmooth, nonconvex optimization

    IEEE Transaction on Automatic Control

    (2006)
  • J.V. Burke et al.

    HIFOO: A MATLAB package for fixed-order controller design and H optimization

    Fifth IFAC symposium on robust control design, Toulouse

    (2006)
  • J.V. Burke et al.

    A robust gradient sampling algorithm for nonsmooth nonconvex optimization

    SIAM Journal on Optimization

    (2005)
  • Y.-Y. Cao et al.

    Static output feedback simultaneous stabilization: ILMI approach

    International Journal of Control

    (1998)
  • G. Chesi

    Robust static output feedback controllers via robust stabilization functions

    IEEE Transaction Automatic Control

    (2014)
  • F.H. Clarke

    Optimization and nonsmooth analysis

    (1983)
  • D. Coutinho et al.

    Guaranteed cost control of uncertain nonlinear systems via polynomial Lyapunov functions

    IEEE Transactions on Automatic Control

    (2002)
  • C. Crusius et al.

    Sufficient LMI conditions for output feedback control problems

    IEEE Transaction on Automatic Control

    (1999)
  • K. Dabboussi et al.

    Sufficient dilated LMI conditions for H static output feedback robust stabilization of linear continuous-time systems

    Journal of Applied Mathematics

    (2012)
  • Cited by (182)

    View all citing articles on Scopus
    View full text