Decentralised Anti-windup design approaches

with application to Quadrotor UAVs

Nkemdilim A Ofodile, Matthew C Turner, Teng D Chollom
Control Research Group, Department of Engineering, University of Leicester, UK
Motivation

Channel-by-Channel AW Features

I Anti-windup (AW) compensators improve performance in both MIMO and

SISO systems when saturation occurs.
I However in practice, MIMO AW compensators are;
. not as simple as SISO AW, the practical consequences of
L2 performance may be difficult to interpret.
. Highly complex, hence problems arise when computational resources are
limited.
. not structured.
I Key features
. Aim: Design decentralized AW compensators for a class of MIMO systems
(eg: Quadrotor UAVs fall in this category).
. The AW must ensure global stability of the entire nonlinear system.
MIMO Plant Description
I GD(s) = diag(G1(s), G2(s), . . . Gm(s)) diag(Gi(s))
I KD(s) = diag(K1(s), K2(s), . . . Km(s)) diag(Ki(s))
K
r

KD

KD

ylin
+

um

GD

Figure : Channel-by-channel AW

I Slightly modified version of figure in typical AW with (s)

having the form;


MD(s) I

X
(2)
(s) =
ND(s)
I Idea is to design m single-channel AW compensators for each Gi(s) in GD(s)
to ensure stability of the overall nonlinear system.
I Using process in [1], a linear programming solution is presented, such that
. If diagonal matrices W > 0 and V > 0 such that V = X0WX exists, AW
compensators can be designed using the LMI in [2]
I Result is structured and easy to implement with great practical appeal. Note:
[MD, ND] [diag(Mi(s)), diag(Ni(s))]

(u)
u

X1

ulin

GD

Quadrotor UAV Test Platform

Figure : MIMO system structure

I GD is a diagonal dynamic part of the plant G

I X is a non-diagonal but static, invertible matrix. Its inverse X1 can be
interpreted as a contol allocation matrix.
I When saturation inactive, system behaves as m decoupled loops
I When saturation occurs, decoupling is destroyed, system experiences windup

I System has the same structure

as earlier stated MIMO plant
I Practical quadrotor used is a
Modified 2014 3DR Quadrotor

Typical Anti-Windup (AW) Structure

ud

yd

GD

I The AW compensator, (s) has the structure and state-space realisation:



A + BF B
M(s) I
0
(s) =
F
N(s)
C + DF D
I G has right coprime factorisation G(s) = N(s)M1(s). F is chosen so that
A+BF is Hurwitz.
I A stable system with F can be obtained by satisfying the LMI described in [2]
Psuedo-Decentralized AW Features

KD

ylin

yd

vm

vd

+
+

response: From Left; 1st

Nominal response; 2nd
Saturated response no
AW; 3rd Saturation,
decentralized AW; 4th
Saturation,
channel-by-channel
AW

I Note: All plots are not exactly

alike because outdoor flights
conditions are not constant.

Desired Pitch
Actual Pitch Saturated no AW

15

15

10

10

10

10

15

15

20

20

25
1.06

1.08

1.1

1.12

1.14

1.16

1.18

Time in milliseconds

25

1.2

7.2

7.4

7.6

7.8

x 10

20

8.2

8.4
4

x 10

20
Desired Pitch
Actual Pitch Saturated with AW

Desired Pitch
Actual Pitch Saturated with AW

15

15

10

10

10

10

15

15

20

20

25

Time in milliseconds

1.08

1.1

1.12

1.14

Time in milliseconds

1.16

25

1.18
5

x 10

2.38

2.4

2.42

2.44

Time in milliseconds

2.46

2.48

2.5
5

x 10

Conclusion

(v)

vlin

20
Desired Pitch
Nominal Pitch

I Figure: Pitch angle

Figure : Full anti-windup structure

20

Pitch in degrees

um

Pitch in degrees

KD

Flight Tests Results

Pitch in degrees

Pitch in degrees

GD

Figure : Decentralized AW structure

I Here, the AW, (s)

is driven by virtual signal
v = v (v).
I By performing the stability analysis as described in [3], this LMI is obtained

ADQD + BDLD BDXUD 0 0

1
1
1

LDX
X UD X
0

He
<0
(1)

0
0
2I 0

CDQD + DDLD DDUD

0 2I
I Result is more stringent than that described in the typical AW above. Note:

He{A} = A + A0, (s)

is diagonal.

http://www2.le.ac.uk/departments/engineering/research/control

I Two approaches proposed for structured AW design.

I Pseudo-decentralised approach provides a one-step design procedure.
I Channel-by-channel approach allows independent AW design for each channel
and combines them safely for the MIMO system.
I Channel-by-channel AW is preferred due to its transparency, flexibility and
ease of implementation.
I Flight results for both designs show improved performance during saturation.
References
[1] Nkemdilim A Ofodile, Matthew C Turner, and Osichinaka C Ubadike. Channel-by-channel
anti-windup design for a class of multivariable systems. In American Control Conference
(ACC). IEEE, 2015.
[2] Matthew C Turner, Guido Herrmann, and Ian Postlethwaite. Accounting for uncertainty in
anti-windup synthesis. In American Control Conference. IEEE, 2004.
[3] Matthew C Turner and Ian Postlethwaite. A new perspective on static and low order antiwindup synthesis. International Journal of Control, 77(1):2744, 2004.

<nao9, mct6, tdc9>@le.ac.uk