Scribd
Upload
80 views

Y y U E: The MIT Rule For Adaptive Control: Adaptive Flight Control System - X15

The document describes the derivation and implementation of the MIT rule for adaptive control. It begins by describing the goal of finding an adaptive control gain Kc that minimizes error between the process output y and model output ym over time. This leads to deriving an expression for the gradient of the cost function and setting the control law as Kc = -μym e, where e is the error and μ is a positive gain. It then provides examples of implementing the MIT rule in Simulink models, noting the rule guarantees stability if the plant and model outputs are similar. However, modeling errors can cause instability. Averaging theory is discussed as a way to analyze stability by treating Kc as slowly varying. The

Uploaded by

maheshboobalan
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
80 views

Y y U E: The MIT Rule For Adaptive Control: Adaptive Flight Control System - X15

The document describes the derivation and implementation of the MIT rule for adaptive control. It begins by describing the goal of finding an adaptive control gain Kc that minimizes error between the process output y and model output ym over time. This leads to deriving an expression for the gradient of the cost function and setting the control law as Kc = -μym e, where e is the error and μ is a positive gain. It then provides examples of implementing the MIT rule in Simulink models, noting the rule guarantees stability if the plant and model outputs are similar. However, modeling errors can cause instability. Averaging theory is discussed as a way to analyze stability by treating Kc as slowly varying. The

Uploaded by

maheshboobalan
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

The MIT Rule for Adaptive Control: Adaptive Flight Control System X15

Derivation of MIT rule

u
Kc

Process

y e + -

Kp

Gp

y*

Adaptation Mechanism

Gm
Assume: Stable process consisting of a known linear system Gp (stable) and an unknown gain Kp. Problem: Find an adaptation mechanism which adjusts the controller gain Kc so that the process output y follows the model output ym.

Ideally we would like to find the gain Kc which minimizes the cost to go

1 J= 2

But this is very difficult to solve since the cost to go depends on the control in a nonlinear way. Solving the problem results in a strategy called dual control. To get a more tractable problem we instead try to minimize the instantaneous cost 1

e2 dt

J=

The gradient of J is given by

e ( t) 2

J = Kp Gp (s)ue Kc
The sign of Kp is positive. By assuming that Gp= Gm we achieve gradient descent by setting Where > 0 is a positve gain which determines the step size. The algorithms is quite easy to implement and stability is guarnateed.
3

Kc = ym e

Simulink model taken from Bhat at Drexel University

gamma (g) = 1 Actual Kp = 2 Initial guessed Kp = 1

Control Law (MIT rule):

= y e m

Error between model and plant output Error between actual and true gain

MIT rule and some background material is covered in CH 1 of the book Adaptive Systems, by Mareels and Polderman, Birkhauser, 1996 More complete analysis by BDO Andersons paper. Both posted on Blackboard
5

Explaining Instability

Underlying differential equation for kc is Mathieu equation. Solution regions of this equation are depicted.
One instability mechanism is interaction of excited plant dynamics with adaptive dynamics, made worse at high gain g
High (adaptive) gain instability for some Zp(s) : consider a  constant input R to display phenomenon. The MIT rule

= " g[ y " y ] y k c p m m
 leads to a characteristic equation; high g may give RHP zero  s + gkmkpR2Zp(s) = 0 From zero:
From Plant derivative of
!
kc
Dynamics

BDO Anderson, Lennart Ljung Symposium Oct 2004 6

Example of performance

g

Unshaded region is stable Sine wave input at frequency Plant is (s+1)-1

Lennart Ljung Symposium Oct 2004

Explaining Instability II

r(t)

kc(t)
kk mZm(s) m(t)

kpZp(s)

yp(t)
e(t)

ym(t)


Zm(s) is known, km is known, k p is positive and unknown, but kc(t) is known and adjustable
A second instability mechanism comes from modelling errors, here errors between Zp(s) and Zm(s)
Following two gures show case where plant and model are the same and where they are different.

8 Lennart Ljung Symposium Oct 2004

Performance: another example



g Unshaded region is stable Sine wave input at frequency Plant is e-s(s+1)-1 while model is still (s+1)-1

Lennart Ljung Symposium Oct 2004

Rescuing the MIT Rule: Averaging



= " g[ y " y ] y k c p m m
Zp(s)kpkc(t)r(t)


Averaging theory is the general analysis tool usable given separation of time scales of the plant dynamics and the learning/adaptation rate

Averaging theory treats kc slowly-varying : kc* is approximately kc for


 small g where
.
kc*=-g{Zp(s)kp r(t)}{Zm(s)kmr(t)} kc* + terms indep of kc*
Stability is ensured if the average value of
{Zp(s)kp r(t)}{Zm(s)kmr(t)}
is positive--and if Zp is like Zm at frequencies where r(t) is concentrated, then stability is achieved.

Lennart Ljung Symposium Oct 
10 2004

11

uc = y m y r e 0
integral control

Forgot Integral Wind-up!


Criticism # 1
Must include modifications used in normal PID control Anti reset wind-up Filters Bump-less transfer Variable scaling Nonlinear compensation (nonlinear adaptive control) KJ Astrom IFAC World Congress 1984

14

How to stop here - Rather than here

Poor control

Poor control

Good control

Good control

Poor control

Good control

15

Two Themes 1. Adaptive control algorithms 2. Nonlinear analysis and stability theory

B Erik Ydstie, CMU

16

B Erik Ydstie, CMU

17

B Erik Ydstie, CMU

18

You might also like