You are on page 1of 32

Introduction

GRED-BMRAC
Simulation Results
Conclusions

Multivariable BMRAC Extension to Arbitrary


Relative Degree Using Global Robust Exact
Differentiators

Andrei Battistel, Eduardo Nunes, Liu Hsu

COPPE
Federal University of Rio de Janeiro, Brazil

Travel Support: CAPES - AEX 2832-14-3

International Workshop on Variable Structure Systems, 2014

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Outline

1 Introduction

2 GRED-BMRAC

3 Simulation Results

4 Conclusions

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Introduction

One of the principal approaches of adaptive control is MRAC.

Early difficulty of the MIMO MRAC approach


symmetry assumption of the high frequency gain (Kp matrix)
very restrictive and fragile

To overcome this difficulty, several approaches were proposed.

Conventional MIMO MRAC: (G. Tao, P. Ioannou and K. Sun)


The knowledge of a constant matrix Sp such that
Sp Kp is symmetric and positive definite (SPD)
also very restrictive and fragile

Matrix factorization methods


Does not require the symmetry assumption
Disadvantage: leads to controller overparametrization

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Introduction

Recently a new possibility based on a generalized concept of


passivity WSPR

Assumption: Kp has a positive real diagonal Jordan form (PDJ)


less restrictive and robust condtion
Simpler control laws: does not need overparametrization
If Kp is not PDJ, a multiplier L can be used, so that LKp becomes PDJ
Disadvantage: restricted to uniform relative degree one

The B-MRAC attenuates the problems of lack of robustness and


transient performance of conventional MIMO MRAC techniques

This paper: extension of the B-MRAC for MIMO systems with


nonuniform arbitrary relative degree

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Motivation

Suppose that the system of interest has the following HFG:


 
cos() sin()
Kp =
sin() cos()

and suppose that the nominal value of is nom = 45 . Then


 
0.7071 0.7071
Kpnom =
0.7071 0.7071
which is not SPD nor PDJ

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Motivation

A possible symmetrizing matrix Sp is:


 
2 1
Sp =
2 1

such that the product is SPD.


 
2.1213 0.7071
Sp Kpnom =
0.7071 2.1213
However, if the actual value of is slightly different, say = 45.01 ,
the product Sp Kp is not symmetric anymore.
 
2.1212 0.7075
Sp Kp =
0.7067 2.1214

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Now, considering the PDJ condition, one possible choice for the
multiplier L is

 
35.3553 0
L=
0.7071 0.7071

For this choice the matrix LKp is PDJ on the interval 71 < < 73

This example illustrates that the PDJ condition is indeed robust and
less restrictive

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Binary Model Reference Adaptive Control (BMRAC)

Conventional MRAC lacks robustness and can present poor


adaptation transient.

Binary MRAC:
consists basically of the conventional MRAC modified by parameter
projection combined with high adaptation gain

tends to behave as a sliding mode controller as the adaptation gain


is increased while avoiding chattering

restricted to plants of uniform relative degree one

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Contribution of the paper

We propose a design that extends the BMRAC for plants with


nonuniform arbitrary relative degree

A combination of Lead Filters and Robust Exact Differentiators


(Levant) are used to circumvent the relative degree obstacle

The new design requires only a PDJ condition related to the


plant HFG matrix (symmetry assumption is not required)

Global stability and exact tracking are achieved for MIMO plants

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Extended BMRAC

Extending the BMRAC for plants with nonuniform arbitrary relative


degree

Key simple idea: use derivatives of the plant output to obtain uniform
relative degree one

1

0
(s + 2)(s + 3)
y = u
1
0
s+3

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Extended BMRAC

1

0

s+1 0 (s + 2)(s + 3)
y = u
0 1
1
0
s+3

s+1

0
(s + 2)(s + 3)
y = L(s)y = y + y = u
1
0
s+3

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

How to Estimate?

Hybrid estimation scheme named Global Robust Exact


Differentiator (GRED)

Combines a Lead filter with a Robust Exact Differentiator

Lead Filters can provide global stability properties, however with


residual tracking erros

REDs can provide exact tracking, however with local stability


properties

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

GRED-BMRAC

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

GRED-BMRAC

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

GRED-BMRAC

We choose a reference model with the same order as the plant

ym = M(s) r ; r , ym IRM


M(s) = diag (s + a)1 , ..., (s + a)1 L1 (s), a > 0

L(s)G(s) and L(s)M(s) have uniform relative degree one


Adaptation with a modified error

= g m
e

y is estimated by the GRED and m is known

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

GRED-BMRAC

BMRAC uses the parametrization



1
2
= vec() = . , = Im =
..
.. .

n

The control law can be rewritten as

u(t) = T (t)(t) = T (t)(t)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

GRED-BMRAC

Adaptation law with modified error:


= e
with given by a projection

0, if |||| < M or eq < 0
=
eq , if |||| M and eq 0


T e
eq = 2
||||
where M > || ||.

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

GRED-BMRAC

Figure: GRED-BMRAC blockBMRAC


Andrei Battistel, Eduardo Nunes, Liu Hsu
Multivariable diagram
Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Simulation Results
Comparison between the extended BMRAC and coventional MIMO
MRAC techniques

Uncertain plant with nonuniform vector relative degree (1 = 2, 2 = 1)

4 2
(s + 1)(s + 3) (s + 1)(s + 3)
Gs (s) = 1 2
(s 1) (s 1)

" #
1
(s+1)2
0
Reference Model: M(s) = 1
0 (s+1)

Reference Input: r (t) = [5sqw(t/3) 5sqw(t)];

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Comparison between BMRAC and conventional MRAC

Tracking errors for the conventional Direct MIMO MRAC

tracking errors
5
e
1
e2
4

2
0 200 400 600 800 1000
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Comparison between BMRAC and conventional MRAC

Tracking errors for the extended BMRAC

tracking error
5

5
0 10 20 30 40 50 60
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Comparison between BMRAC and conventional MRAC

Conventional Direct MIMO MRAC tracking performance

output signals: y1 and ym1


2

1 model
plant
2
180 185 190 195 200
time(s)
output signals: y2 and ym2
2
model
1 plant

2
180 185 190 195 200
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Comparison between BMRAC and conventional MRAC

Extended BMRAC tracking performance

output signals: y1 and ym1


2

1 model
plant
2
180 185 190 195 200
time(s)
output signals: y2 and ym2
2
model
1 plant

2
180 185 190 195 200
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Simulation Results

We consider a combination of plant/actuator G0 (s) with the following


realization

1 2 3 1 10 20 0 0
0 2 1 1 1 0.5 0 0.5
A= 0 0 1 1 B = 0

0 16 80
0 0 0 1 0 0 0 4

H =I

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Simulation Results

Plant/actuator ouptut are measured by a sensor whose dynamics is


described by

(s + 1)1 0 0 0
0 (s + 1)1 0 0
Gs (s) =
0 0 1 0
0 0 0 1
Such that the transfer function G(s) = Gs (s)G0 (s) from u to y has
vector relative degree = [2, 2, 1, 1]

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Tracking with Lead filter

tracking performance
1
y1
0.8
1 2 m1 m2

y2
,y

0.6 y
m1
y ,y ,y

0.4 y
m2

0.2

0
592 593 594 595 596 597 598 599 600
time(s)
tracking performance
2
y3
3 4 m3 m4

y4
1
,y

y
m3
y ,y ,y

y
0 m4

1
592 593 594 595 596 597 598 599 600
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Tracking with GRED-BMRAC

tracking performance
1
y1
0.8
1 2 m1 m2

y2
,y

0.6 y
m1
y ,y ,y

0.4 y
m2

0.2

0
592 593 594 595 596 597 598 599 600
time(s)
tracking performance
2
y3
3 4 m3 m4

y4
1
,y

y
m3
y ,y ,y

y
0 m4

1
592 593 594 595 596 597 598 599 600
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Tracking Errors with GRED-BMRAC

tracking error
5
6

4
5

3
4
2
3
1
2
0
1
1
0
2
1
3

2
4

3
5
0 100 200 300 400 500 600
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

GRED-BMRAC: Switching function and Control signal


switching function
2

1.5

0.5

0
0 1 2 3 4 5
time(s)
control input
10

5
592 593 594 595 596 597 598 599 600
time(s)

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Simulation Results

When only the lead filter is used, large error persists

When the hybrid differentiator is employed, zero tracking error is


achieved within numerical integration errors.

Under the same conditions, the system is unstable if only the


MIMO RED is used.

the system is unstable if the parameter projection is not used


under the same conditions.

Control signal is free of chattering

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Conclusions

We obtain an arbitrary relative degree extension to BMRAC


using the hybrid estimator GRED

This controller presents global and exact tracking with output


feedback without requiring symmetry conditions on the high
frequency gain

GRED-BMRAC allows a better transient behavior

The control signal is free of chattering

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob
Introduction
GRED-BMRAC
Simulation Results
Conclusions

Thank you for your attention

Andrei Battistel, Eduardo Nunes, Liu Hsu Multivariable BMRAC Extension to Arbitrary Relative Degree Using Global Rob

You might also like