Sunteți pe pagina 1din 19

Anti-windup: basic concepts for

advanced design
One (only one!) of the many possible
ways to deal with actuator saturation
Luca Zaccarian

University of Rome, Tor Vergata (Italy)

Outline

What is anti-windup synthesis?


An example
Simple ideas vs. advanced synthesis
Technical issues
Synthesis approaches

What is anti-windup synthesis?

Given; and
interconnection
without saturation
performs well.




Synthesize this; to
regain performance
lost due to saturation.

Not about synthesis for saturation


starting from a clean slate

Bumpless Transfer and AW


Manual
input




Input substitution instead of input saturation


In both cases controller output differs from plant input

An example
p0 +
p0
pA

pB

F
FL
mg

Small signal time domain specs are met


with double lead plus integral action:

Example (contd)
p0 +
p0
pA

pB

F
FL
mg

Because of actuator limitations, controller


is ill-suited for large step changes:

Example (contd)
p0 +
p0
pA

pB

F
FL
mg

If actuators were made five times more powerful,


controller would be well-suited for large step changes:

Example (contd)
p
p0 +
p0
pA

FL
mg

If actuators are left at original strength and controller is


augmented with anti-windup, small signal specs are
maintained and controller is suitable for large step changes:

Simple ideas vs advanced synthesis


What you find in an undergrad textbook:
AW

sat (u ) u
u

k AW

kI

kp

Simple vs. advanced contd


Effect of extra feedthrough (MIMO academic example)

sat (u ) u

k FT

k AW
kI

AW

kp

unconstrained
undergrad text solution
extra feedthrough

Simple vs. advanced contd


Effect of dynamics in AW compensator (Linearized F-8)

Flight path angle

Unconstrained

Static anti-windup

Dynamic anti-windup

Technical Issues
Sensing and actuating
Well-posedness

Quantifying good
For exponentially
unstable plants, a
nonlocal problem
with a local solution
(A bit messy)




Sensing




Requires minimal
extra sensing (can
be implemented in
software).
Ensures that extra
compensation
doesnt activate
until saturation is
reached.

For exponentially unstable or nonlinear,


additional plant measurements are typically used

Actuating on the controller


External


 

Full authority



Controller states
are not accessible

 

Controller states
are accessible

Quantifying good anti-windup I


Input/output gain (I/O)
Make L2 gain from w to z small.

v(t )

dt




v2=

open-loop gain
2

w2

unconstrained closed-loop gain

Quantifying good anti-windup II


Unconstrained response recovery (URR)

Make small the L2 gain from [ul sat (ul )] to [z zl ]




Energy of what is not


feasible in ideal response




to

energy in difference of
ideal and actual responses.

Unconstrained response recovery (contd)


A nonlinear robot with Nonlinear PID






A nonlinear robot example (contd)









Lending a helping hand


only when needed

Synthesis approaches (2000-2006)


Optimized quadratic performance via LMI (linear
matrix inequality) based synthesis (mostly in
continuous-time; discrete-time also possible)
Nonlinear performance via hybrid switching and
discrete-time MPC based synthesis (sampled-data
solution in the continuous-time case)
For nonlinear systems and exp. unstable linear ones,
nonlinear anti-windup solutions from the field of
nonlinear control
Constructions available for magnitude and rate
saturation
All results apply to MIMO problem settings
Joint work with Andy Teel (UCSB) and other

S-ar putea să vă placă și