Robust Control

Spring, 2019
Instructor: Prof. Masayuki Fujita (S5-303B)

5th class
Tue., 14th May, 2019, 10:45～12:15,
S423 Lecture Room
5. Control
5.1 General Control Problem Formulation
[SP05, Sec. 3.8]

5.2 Control Problem and DGKF Solutions

[SP05, Sec. 9.3]

5.3 Structure of Controllers

Reference:
[SP05] S. Skogestad and I. Postlethwaite,
Multivariable Feedback Control; Analysis and Design,
Second Edition, Wiley, 2005.
General Control Problem Formulation [SP05, p. 104]
: control inputs
: measured (or sensor) outputs
: exogenous inputs
(disturbance and commands, etc.)
: regulated outputs

Generalized Plant

Closed-loop Transfer Function (LFT) lft(G,K)

3
General Control Problem Formulation
[SP05, Ex. 3.18] (p. 105) Building Interconnection

Exogenous Inputs

Regulated Output
Measured Output
Control Input

Generalized Plant
Remark

4
General Control Problem Formulation
[SP05, Ex. 3.19] (p. 107) Including Weights
Exogenous Input
Regulated Outputs
Measured Output −
Control Input
Generalized Plant

Remark
−

5
LQG Type Control Problem Formulation [SP05, pp. 344, 356]
Regulated Outputs Exogenous Inputs

Generalized Plant

LQG (Linear Quadratic Gaussian)

Norm

6
［Ex.］ Spinning Satellite: Building Interconnection sysic

Nominal Model

Multiplicative (Output) Uncertainty

Uncertainty Weight

Performance Weight

7
［Ex.］ Spinning Satellite: Building Interconnection sysic

Interconnection (Mixed) MATLAB Command

%Generalized Plant%
systemnames = ‘Pnom WP WM';
inputvar = '[w(2);u(2)]';
outputvar = '[WP;WM;-w-Pnom]';
input_to_Pnom= '[u]';
− input_to_WP = '[w+Pnom]';
input_to_WM = ‘[Pnom]';
G = sysic;
8
Examples of Control Problem [SP05, pp. 104-114]
1

• Sensitivity Minimization Problem

• Robust Stabilization Problem
• Mixed Sensitivity Problem
• LQG Type Control Problem
• Feedforward Problem Interconnection
• Estimation Problem Nominal Plant Model
Performance Weight
Uncertainty Weight

“If the Robust Control Toolbox of MATLAB complains,

then it probably means that your control problem is not
well formulated and you should think again”
9
 Control Problem [SP05, p. 357]

Optimal Control Problem

Find all stabilizing controllers which minimize

Sub-optimal Control Problem

Given , find all stabilizing controllers such that
-iteration 10
The “1984” Approach (1984 ONR/Honeywell Workshop) 2

Generalized Plant

B. A. Francis and J. C. Doyle,

All Stabilizing Controllers SIAM, 25-4, 1987

: Stable Transfer Function Matrix

Closed-loop Transfer Function (LFT)

Model Matching Problem

−
Affine in 11
State Space Approach [SP05, p. 357]

Generalized Plant

Closed-loop Transfer Function (LFT)

Control Problem
Given , find all stabilizing controllers such that
12
A Simplified Control Problem [SP05, p. 353]
Generalized Plant

Assumptions 3 4

(A1) is stabilizable and is detectable

(A2) is controllable and is observable
Full rank on the imaginary axis

(A3) and

13
DGKF Solution [SP05, p. 357]
There exists a stabilizing controller such that
if and only if the following three conditions hold:

(i) There exists a solution to

such that

(ii) There exists a solution to

such that

(iii)

14
Central Controller [SP05, p. 358]

All Controllers [SP05, p. 358]

： Stable Proper Transfer Function Matrix such that

Parameterization of Control 15
 DGKF 5
Doyle, Glover, Khargonekar, Francis, IEEE TAC, 34 - 8, 1989
(1988 ACC)
State-Space Solution to
Standard and
Control Problems

J. C. Doyle K. Glover

P. P. Khargonekar B. A. Francis
16
Sketch of Proof (sufficiency) [Zhou98]
By using the following lemma, we propose one of for
feedback systems which consists of and .

Bounded Real Lemma

For , , the following
two conditions are equivalent.

(i)

(ii) There exists a such that

and has no eigenvalues

on the imaginary axis.

[Zhou98] K. Zhou with J.C. Doyle, Essentials of Robust Control, Prentice Hall, 1998. 17
Case 1: state
Consider the feedback loop with the state xc = [x' xˆ '] ' .

Generalized Plant

Central Controller
Closed-loop Transfer Function (LFT) from to

18
 Define by

Then, we have

and
positive semi-definite

positive definite
does not have eigenvalues on the imaginary axis.
Hence, from Bounded Real Lemma, there exists a stabilizing
controller such that .

M. Sampei, T. Mita and M. Nakamichi, “An Algebraic Approach to Output

Feedback Control Problems,” Systems and Control Letters, Vol. 14, pp. 13-24, 1990. 19
 Case 2: state

Consider the feedback loop with the state

Closed-loop Transfer Function (LFT) from to

Set as follows

The remaining part is the same as the previous one.

K. Uchida and M. Fujita, “On the central controller: Characterizations via

differential games and LEQG control problems,” Systems and Control Letter,
vol. 13, no. 1, pp. 9-13, 1989. 20
 Case 3: state

Consider the feedback loop with the state

Closed-loop Transfer Function (LFT) from to

c

Set as follows

The remaining part is the same as the previous one.

K. Uchida and M. Fujita, “Finite Horizon Control Problems with Terminal
Penalties,” IEEE Trans. on Automatic Control, vol. 37, no. 11, pp. 1762-1767, 1992. 21
Structure of Central Controller [SP05, p. 358]
, ： State Feedback

： Worst Disturbance Estimation

： Worst State Estimation

Minimum Entropy Controller

Entropy

22
From LQG Control to Control
LQG Controller Central Controller
( Controller)
5 6
State Feedback State Feedback

Worst Disturbance Estimation

State Estimation Worst State Estimation

Riccati Equations: Riccati Equations:

23
LQ Theory Game Theory
Linear System Linear System

Cost Function Cost Function

minimizes cost function minimizes cost function

maximizes cost function

For Certain For Certain

Output feedback
： Uchida and Fujita, 1989
LEQG Control
“E”: Exponential 24
From LQG Control to Control

Linear System
Theory
R.E.Kalman
L.S.Pontryagin
Optimal Stability
Control LQG Theory
Theory
A.M.Lyapunov
“Gap between Theory and Practice”
R.Bellman
Feedback
Theory

H.W.Bode H.Nyquist 25
Completion of Modern Control Theory
A stabilizing controller
State feedback/Observer Transfer Function
An optimal controller Pole/Zero
LQG(=LQR+Kalman Filter) Structure
All stabilizing controllers Controllability,
(Youla) Parametrization Observability
All optimal controllers
controller State Space Form
( : =LQG) (Data Structure)
State

-
26
,
Robust and Optimal Control
1940’s –1950s→ 1960’s –1970s→

Tough and Strong Smart and Intelligent

1980’s –1990s→

Tough and Smart 27

[ZDG96] K. Zhou, J. Doyle and K. Glover, Robust and Optimal Control, Prentice Hall, 1996
General Solutions
6 7
“State-space Formulate for All Stabilizing
Controllers that Satisfy an Norm Bound and
Relations to Risk Sensitivity”
K. Glover and J.C. Doyle, Systems and Control Letters, 11, 1988.

hinfsyn h2syn
5
[k, cl, gam, info] = hinfsyn (p, nmeas, ncon, key1, value1, key2, value2, …)
input argument output argument
p generalized plant k LTI controller
nmeas number of measurement outputs cl closed loop system which
ncon number of control inputs consists of and
gam norm of closed loop system
info information of output results
Key setting
Gmax upper limit of Gam Method Ric ：Ricatti solution
Gmin lower limit of Gam Lmi ：LMI solution
8
Tolgam relative error of Gam Maxe ：max entropy solution
So frequency at which entropy Display Off ：not show setting process
is assessed On ：show setting process 28
Robust Control Toolbox

Robust Control Toolbox - Analysis and Synthesis Toolbox LMI Control Toolbox
(1988) (1993) (1995)

Robust Control Toolbox ver. 3 (2005~) R2019a

(Eds.) Gary Balas Andy Packard Michael Safonov

30
After DGKF

[DP05] G.E. Dullerud and F. Paganini,

A Course in Robust Control Theory:
A Convex Approach,
Text in Applied Mathematics, Springer, 2005.

• Linear Matrix Inequality (LMI)

• Linear Parameter Varying (LPV) Systems

• Integral Quadratic Constraints (IQC)

• Sum of Squares (SOS)

30
System Level Synthesis
1940’s –1950’s→ 1960’s –1970’s→

Tough and Smart and

Strong Intelligent

2000’s – 2010’s →
1980’s –1990’s→

Tough, Smart
Tough and
and Elegant
Smart
System Level
Synthesis (SLS)

K. Zhou, J. Doyle and K. Glover, Robust and Optimal Control, Prentice Hall, 1996 31
J. Anderson, J. Doyle, S. Low, and N. Matni, “System Level Synthesis,” Annual Reviews in Control, to appear, 2019
A.I. = Actionable Intelligence

P.P. Khargonekar, and M.A. Dahleh,

Model-based
Model free LQR
LQR Advancing systems and control
research in the era of ML and AI,
Annual Reviews in Control, Vol. 45,
minimize pp. 1-4, 2019

s.t.

Reinforcement Learning
32
5. Control

5.1 General Control Problem Formulation

[SP05, Sec. 3.8]

5.2 Control Problem and DGKF Solutions

[SP05, Sec. 9.3]

5.3 Structure of Controllers

Reference:
[SP05] S. Skogestad and I. Postlethwaite,
Multivariable Feedback Control; Analysis and Design,
Second Edition, Wiley, 2005.
6. Design Example

6.1 Spinning Satellite: Control [SP05, Sec. 3.7]

6.2 2nd Report

Reference:
[SP05] S. Skogestad and I. Postlethwaite,
Multivariable Feedback Control; Analysis and Design,
Second Edition, Wiley, 2005.
An Example of Generalized Plant
1

F-18 High Alpha Research Vehicle

Flight Control Design Using Robust Dynamic Inversion and

Time-scale Separation
J. Reiner, G. J. Balas and W. L. Garrard, Automatica, 32 - 11, 1996 35
Model Matching Problem in SISO Systems
2
: Stable
B.A. Francis,
All Stabilizing Controllers Springer-Verlag,
1987
, : Proper Stable

Decision of Parameter
Nominal Performance

Robust Stabilization

Model Matching Problem

−
36
Assumptions of Control Problem for simplicity
3
[SP05, p. 354]
The following assumptions are typically made in and problems:
(A1) is stabilizable and is detectable
[ Requirement for the existence of stabilizing controllers ]
(A2) and have full rank
[ Sufficient to ensure the controllers are proper and hence realizable]

(A3) has full column rank for all

(A4) has full row rank for all

[ To ensure that the optimal controller does not try to cancel poles or zeros
on the imaginary axis which would result in closed-loop instability]
(A5) and
[ Conventionality in control. makes strictly proper.
makes strictly proper and simplifies the formulas in the algorithms.]
[ Neither of them are required in control. For significant simplicity.]
37
Assumptions of Control Problem for simplicity
4
It is also sometimes assumed that and are given by
(A6) and

[ This can achieved, without loss of generality,

by a scaling of and and a unitary
transformation of and . ]

In addition, for simplicity of exposition, the following assumptions

are sometimes made
(A7) and
(A8) is controllable and is observable
[ If (A7) holds, then (A3) and (A4) may be replaced by (A8) ]
[ From (A2), (A6) and (A7), the following equations hold ]

and
38
 Control Problem [SP05, pp. 344-351, 355-357]
5
Optimal Control Problem
Find a stabilizing controller which minimize

Regulated Outputs

Exogenous Inputs

h2syn [k, cl, gam, info] = h2syn (p, nmeas, ncon)

39
 Control Solutions
6
synthesis
h2syn hinfsyn h2hinfsyn
controller synthesis controller synthesis Mixed
controller synthesis

sdhinfsyn ltrsyn
Sample-data LQG LTR controller synthesis
controller synthesis （LTR: Loop-transfer Recovery）

Delay
hold Sampler

40
 Control Solutions
7
Loop-shaping synthesis
loopsyn mixsyn
loop shaping mixed sensitivity
controller synthesis controller synthesis

ncfsyn
irreducible decomposition
controller synthesis
(using Glover-McFarlane Method)
−

synthesis
dksyn
Robust controller design
using -synthesis
41
LMI Solutions [SP05, Sec. 12]
8
Linear Matrix Inequality (LMI)

: Constant symmetric real matrices

Riccati Inequality

LMI

Control Problem

: State Feedback
There exist such that

42