Զѧ
ACTA AUTOMATICA SINICA
1997ꡡ234ڡVol.23No.41997



ѧϰƵȶ
Ӧ
ƽ»á¾
ժҪһѧϰƵȶоݣӦһȶоݽѧϰתΪѰɢˣӶش⣬һʲôͨѧϰԿƣѧϰʲô˲ڵ벻Ӱѧϰ.һоһֻ˲ѧϰɣ֤ٹȫ̵.
ؼʡѧϰƣȶԣе.
STABILITY ANALYSIS FOR ITERATIVE LEARNING CONTROL
AND ITS APPLICATION TO MANIPULATORS
JIANG PING CHEN HUITANG WANG YUEJUAN
(Dept. of Electrical Engineering,Tongji University, Shanghai 200092)
AbstractThis paper presents a stability criterion for design iterative learning controller.From the criterion,one can change the design problem into finding a positive definite discrete matrix kernel.It has been shown that(1) what kind of variable can be controlled by the iterative learning,(2) and what kind of filter can be included into the controller.As an example,we have designed a parameter learning controller for manipulator trajectory tracking which leads to a perfect learning.
Key words Iterative learning control,stability,manipulator.
1
ǿظ¼лþ飬ݴ˽ȷ.һ˼Ӧڿоͳѧϰƣ1ݣͼ1ʾһеֽظʱһϵͼͨķƲʱtеģѧϰrУʷԸƹ߸.ѧϰƵĹ̿ԿȰ(˶)ְɢ(ѧϰ).ͨķǻɢLyapunovȶоݽеģ23ݣȽģ㴦ԻԱ֤ʷ˶ĲԣȻһӦڻϵͳ޷ȷѧϰƵԣһѧϰƷһɢ4ݣѧϰƵһϵͳзýһδϵͳ״̬ڹյ(t=nT)Ĳԣһ㷴ͬ˵ʱʱԣûй߸ȫ̵Է.


ͼ1ѧϰͼʾ
Ϊ˱һֽѧϰƵȶԷ˼ɢ̵ѧϰʣ֤߸ȫ̵ԣõĿɲһǼ򵥵ϣݾ󣬳ȡظѧϰѧϰʽҲֻǼ򵥵һΨһʽ֮ɢ˵Ĺ㷺ʽ.ݴ˱һ˲ѧϰɣйƵӦƷۣ5ݣҪԷһ״̬⣬һʵϵͳϵͳƣѧϰڿ㷨еλ.Ϊ˱֤Էȫѧϰܹ֤ͬ˹߸ȫ̵Ľ.
2ѧϰȶо
ͼ1ʾѧϰпΪkظϵͳɣ
i=fi(t)+ui(t) i=1k, 0ttf
(1)
Сfi(t)нظfi(t)=f(t)iѧϰΪui.
ѧϰƾ(1)ʽʼϵųЧƲ֣iѧϰxiϵͳʾ.ѧϰĿΪѧϰuiʹϵͳ״̬ʱ켣xd(t),0ttf.һĿϵͳʼxi(0)=xd(0).
оݡϵͳЧsi(t),0ttfsi(0)=0,ʱtʹ

֤.iϵͳVi(t)=(t)si(t),
򡡡
ɳʼsi(0)=0
ȫkϵͳ

߶kȡޣ


򡡡
tfн磬(t)н磬
Vi(t)=0,si(t)=0, 0ttf.
֤.
һ۶ϵͳ(1)ѧϰƣԷһתΪȡоݲʽɢ.
Чsi(t)=xd(t)-xi(t),


fi(t)ظá


(2)

d(t)f(t)нʱɵ
򡡡[xd(t)-xk(t)]=0, 0ttf.
ʵһʽһΪ

(3)
F(i-j)Ϊɢˣ6.
Ȼ(2)ʽΪʽؽ⣬ΪZ任Z(F(i))=diag(z/z-1,,z/z-1)Ϊʵɢݺz=1һ.һѧϰ㷨ɽһʾΪ
ui(t)=ui-1t)+W(z)si(t).
(4)
si(t)W(z)˲ѧϰ뱣֤G(z)=W(z)z/(z-1)Ϊʵɢݺͼ2.˲һѧϰƵ³ԣͬʱӰϵͳ.ȡW(z)=(1-a)z/(z-a),(0a1)Ϊһõĵͨ˲̣ʱѧϰ㷨Ϊ

(5)


ͼ2ѧϰɢݺʾ
潫֤G(z)ʵԣ

(1)G(z)z=1һ򵥼㣬0G(z)ڵԪ
(2)z=(1+j)/(1-j),G(z)ʵΪ

Ϊʹ,ReG(j)]0Ҫa1/3ԲӰѧϰԵ˲ϵӦ0a1/3.
3˲ѧϰ
еֽظҵʱƷӦڻеֲѧϰ.еֶѧ
M()+C(,)+G()=.
(6)
Ӧѧϰƿئii=1k,
M(i)i+C(i,i)i+G(i)=i,i=1k
ʹеָ켣k(t)=d(t),k(t)=d(t).
ei(t)=d(t)-i(t),ϵͳЧ
si(t)=i(t)+Tei(t),T0  i(t)=A(i)si(t),
A(i)ΪM(i)ƽ.


(7)
Сri(t)=d(t)+Tei(t).
ɷ(7)֪ѧϰfiеʵ״̬йأǲظģǰӦڿѧϰҳظѧϰ.ѧϰfiֽΪ
fi=Yi(i,i,d,d,d)a(t).
СYiϵͳ״̬ĻеȾa(t)ɻеֶѧɵĲʱĶӦ޸ر仯ҲʱĶӦڻеֹиذһʱɷ仯.ѧϰ

(8)
F(i-j)Ϊɢ.

(9)
оԦk(t)=0, 0ttf.
A(i)0ʦi(t)=d(t), i(t)=d(t),0ttf.
ӦĲѧϰɿɱʾΪʽΪ

(10)
豣֤G(z)=W(z)z/(z-1)Ϊʵɢݺ.
ǵõȫѧϰĻ˲ѧϰɣģ͵ĻӦƣ7ݾзǳƵʽͶżϵ.ӦʱtеģѧϰrеģӦƱԶѧĲΪýȶۣѧϰƿӦڶѧ仯ϣֻҪֱ仯ڷѧϰп֣ӦƵֻ֤켣ԶĽԣѧϰƽ֤ȫ켣Ľԣڻеָٹضʱ̼أӦƽ½ɹ̰нϴƫ֣ѧϰƾɴѧϰ󲻻ᷢһ.
4
ʵƵƽؽֱβи˽еģиԵظҵص㣬رʺѧϰƵӦ.ͬʱڱеµѹÿʽиȲͬʱ̨治ˮƽʱиᷢϴ仯Ӱиȣڲǰ²и뷽ǲѧϰƷڲкͷƺ8ݣи̽иӰ죬ѧģΪ

Сf=[fx fy]T=JTt, t=[tx ty]TΪиߵλJΪеJacobian.
ѧϰΪеֶѧDij, i,j=1,2,иСc.
ڱ۳Ϊ1ףѧΪ
D11=3.55(kg.m2), D12=1.2(kg.m2), D22=1.8(kg.m2), c=6(N).
е1m/sٶи̨(0,1m)(1m,1m)ֱߣٶγΪ0.5m.(10)ʽƲѡȡΪ
W=diag(600,600,600600), T=diag(50,50).
Ӧ޽緽ȡ

ʼֵΪ0κ֪ʶһοƽͼ3ʾΪe1M=0.082,e2M=0.16.ڶѧϰƣΪe1M=0.047,e2M=0.070.и10κѽe1M=0.014,e2M=0.025㣬ͼ4ʾﵽĿƽ.


ͼ3һο


ͼ4ʮο
5
ĸ̵ɢص㣬һѧϰƷھظѧϰ.ѧϰپۼӺʽԵõһ⡪ɢ.һȶоݣ˻еֲѧϰ֤еֶԹȫ̵Ľ.֤Ч.
1) ĵõȻѧ.
߼飺ƽ19631992ҵͨѧϢƹϵѧʿѧλΪͬôѧϵоԱҪо˿ܻ.
»á19331953ҵϺͨѧϵΪͬôѧϵڡʿʦҪо˿ܻ.
¾ꡡ19351955ҵϺͨѧϵΪͬôѧϵڡʿʦҪо˿ܻ.
ߵλͬôѧϵϺ200092
ο
1ݡArimoto S,Kawamura S,Miyazaki F.Bettering operation of robots by learning.Journal of Robotics System,1984,1(2):123140.
[2] Kawamura S,Miyazaki F,Arimoto S.Realization of robot motion based on a learning method.IEEE Trans on SMC,1988,18(1):126134.
[3] Kuc T,Nam K,Lee J S.An iterative learning control of manipulators.IEEE Trans on Robotics and Automation,1991,7(6):835841.
[4] Kuc T,Lee J S.An adaptive learning control of uncertain robotic systems.Proc of the 30th Conf on Decision and Control,1991,12061211.
[5] Jeon D,Tomizuka M.Learning hybrid force and position control of robot manipulators.IEEE Trans on Robotics and Automation,1993,9(4):423431.
[6] Landau Y D.Adaptive controlthe model reference approach.New York:Marcel Dekker,1979.399402.
[7] Slotine J J E, Li W.Adaptive manipulator controla case study.IEEE Int Conf on Robotics and Automation,1987,13921400.
[8] ƽ»ã¾.и˵ͷƷо.ˣ199618(3)167172.
ոڡ1995-01-04
