ѧ 
JOURNAL OF ASTRONAUTICS
1998 19 1 No1 Vol.9 1998Ƽڿ

߽ģķԶ̬
ϵͳдϼⷽ
ťʤ
(ҵѧԶϵ**150001)
ժҪһֻ߽ģĶ̬ϵͳдϼⷽ,߽̬ϵͳĳǰһԤģ,ȻԴԤʹʵ֮һԤֵȽԼ⴫ϡĵŵǿԼ,ͬʱڲѧϰʽ,ǳںϵͳϼҪ,ϼֵѡȡҲȽϼ򵥡Ϊ֤ķ,һϵͳͬʱƯƹϵϼ̡,ʮЧ
ʡϵͳϡϼ⡡硡ѧϰ
MULTIPLE SENSOR FAILURE DETECTION 
IN NONLINEAR SYSTEM BASED 
ON SYSTEM IDENTIFICATION METHOD 
USING ON-LINE LEARNING NEURAL NETWORK
Niu YongshengZhao Xinmin
(Dept.of Automated Test,Measurement and Control.,
Harbin Instiute of Technology*Harbin*150001)
AbstractThis paper presents an innovative approach for multiple sensor failure detection in a nonlinear system based on system identification using an on-line learning neural network.The approach builds an one-step ahead prediction model for the nonlinear system using an on-line learning neural network firstly,and then the discprepancy between the on-line estimations of the sensors using the built neural system model and the actual values of them is compared with a predetermined threshold to detect sensor failures.One advantage of the approach is the ability to detect multiple sensor failures.Another advantage is that it may be suitably used in autonomous systems with the neural network learning and working on-line.The method is proved to be very effective by a simulation result of detecting two sensor failures for a space robot system.
Key wordsNonlinear systemSensor failureFault detectionNeural networkOn-line learning
1
ſѧĲϽ,ܵĺͿռվ߱ԴϵݴΪϵͳĹؼ,,ӲϺ,ͻӰϵͳСʵ,ǿϵͳбȽ𻵵1,˿ϵͳ⴫Ƿ,˹,򱨾,ñݴ,عȷƹ̵˳СNASA涨ռ˿ϵͳӦһϵͳ1ݡĿǰϷж,Ƚ϶ǲ÷෽1,˹緽2-6ݵȡ÷෽Ҫϵͳľȷѧģ,ӦھģͲȷԵϵͳͷϵͳеӦ⴫Ϸȱ:1.Լϡ2.ϵͳѧģϵͳ״̬ʸ3.ֻɾֹӳ,ͬһʱ̡4.ûпϵͳΪ˱罨ģĴϼⷽ˷ŵǿԼ,ͬʱڲѧϰʽ,ǳںϵͳϼҪ,ϼֵѡȡҲȽϼ򵥡ͨһʵʿϵͳ1ƯƹϼķķЧġ
2߽ģĴϼⷽ
һԶ̬ϵͳ:
y(k+1)=F(y(k),,y(k-m+1),u(k+1),,u(k-n+1))(1)
(u(k+1)һpάʸ,y(k+1)һqάʸ,n,mӳ,ϵͳn=m=1,F()һķԺ)һǰʹӳٵ롢ʾ8:
(2)
ѵǰ׼е,ȻȡѵϵͳĶ̬ѧģ,ҲԲȡߵƷʽ,ԭͼ1ʾ,Ϊ֪ģͣ3,8,ͼ2ʾ,i,j,kֱʾ롢㡢,Wij,Wjkֱʾ֮Ȩֵ֮Ȩֵ


ͼ1̬ϵͳģ߱ʶ緽ʹϼԭ


ͼ2Ԥṹ
߽Ԥģͺʹϼľ岽:
(1)Ѿ۲⵽k+1ʱ̵Ĵy(k-m+1),,y(k),y(k+1)ϵͳu(k-n+1,,u(k),u(k+1),k+1ʱǰkʱǰu(k-n+1),,u(k),u(k+1),y(k-m+1),y(k)Ϊ,k+1ʱ̵y(k+1)Ϊ,Ӧ򴫲㷨(BP㷨)ѧϰ,ֱһһѵȨֵ
(2)ǰһ,Ȩֵ,k+2ʱǰk+1ʱǰu(k-n+2),u(k+1),u(k+2),y(k-m+2),,y(k+1)Ϊ,Ԥk+2ʱ̵ϵͳֵ
(3)޴ʱy (k+2),ʮֽӽ,֮Ϊʹ,ʱ,ӦĲв,k+2Ĺ ʵy(k+2)֮(в)ľֵy i (k+2)Ϊһȷֵ,жϵiϡ
(4)ûй,,k+2ʱǰk+1ʱǰu(k-n+2),,u(k+2),u(k+2),y(k-m+2),,y(k+1)Ϊ,k+2y(k+2)ΪٴѧϰֱȨֵһȲص3ֱ⵽ϡ
ÿֻѧϰһ,߷ѧϰе,ٶȺܿ,Ա֤ʵʱϵҪ
ڽģǰṹݾԴѡ񡣱ĸѡṹֵѡڹһѵ1,ȡùС,Ƚϴʱ󱨾,ֵȡù,ԼСĴ,©ĸԤ׼ƫĴСȷֵ,Ԥ3׼ƫ֮ͦ+3Ϊֵڵƹʹʵڸ㲢һ,ԱôбʦѺͲt֮ѡtΪԤ,ʵ,ԤҲһֵѡֵѧϰ󱨾,ȻҲп΢ѡƫ,ӶԼСֵĴ΢Сϡв仯,Զ϶ϡ
3ʵ
Թҵ˴ϵʵʱϷ˵ķ,Ҽٶϵͳ,ϵͳ·ġõĻѧģһǿϡϵͳģ,ɢΪ1:
(3)
ʽ
D(k)=a1+a2cos(x4(k))
E(k)=a3+a2cos(x4(k))/2
A(k)=a2sin(x4(k))x1(k)x2(k)+x22(k)/2-a4cos(x3(k))
-a5cos(x3(k)+x4(k))+u1(k)
B(k)=-(a2sin(x4(k)))x21(k)/2-a5cos(x3(k)+x4(k))+u2(k)
C(k)=(a1+a2cosx4(k))a3-(a3+a2cosx4(k)/2)2(4)

u1(k)=125-5(1-exp(-0.2k))
u1(k)=125-10(1-exp(-0.2k))(5)
u1,u2Ϊ,eΪֵ,Ϊdiag0.00001,0.00001,0.00001,0.00001ݵİ
y1,y2,y3,y4ֱǲؽڵĽٶȺͽλƴTΪ,ֵΪ0.001,100㡣a1=3.82,a2=2.12,a3=0.71,a4=81.82,a5=24.6ǰĹ2,ѡ񵥸ṹ,ڵΪ8,ֱΪy1(k),y2(k),y3(k),y4(k),u1(k),u2(k),u1(k+1),u2(k+1),ڵΪ4,ֱy1(k+1),y2(k+1),y3(k+1),y4(k+1),ѡṹΪ854߽˶̬ģͲ⴫ϡļٶٶȴy2ͽǶȴy4ڵ50ͬʱʽϴƯƹͼ3,5ʾͬʱϵڿϵͳǽص,ñķҲԶԸϡͼ4,6ʾѧϰԤģԤʵĲвֵĹϵ,ֱֵΪ0.024,0.0182,ͼпԿķڵ51Ưƹϡ


ͼ3Ԥͽٶ
Ưƹ


ͼ4ٶȴϹ


ͼ5Ԥͽٶ
Ưƹ


ͼ6ٶȴϹ
4
о˽̬ϵͳԤģ͵ѧϰͼ⴫Ϸ,ͨһʵ,˵ķ߼⴫ϡķص:
1.ڸϵͳ,رǷϵͳеĴϼ⡣
2.ܹԤǷ,Լഫϡ
3.ķֻܼʽϴƯƹϺͻ͹ϡ
4.ڼٶϵͳоϼ,ϵͳʹͬʱ,һоϡʵ,йһڹǸѵ,ǽۡ
ο
1ܶ,.ϵͳĹϼϼ.:廪ѧ,1994
2Guo T.-H. and Nurre J.Sensor Failure Detection and Recovery by Neural Networks.IJCNN,1991:1221-1226
3Zhao Xinmin,Ye Xiaochun,Zhang Chen.Artificial Neural Network for Sensor Failure Detection in Hovercraft Engine.IEEE Annual Conference on Instrument and Measurement,Hamamatsu,Japan,1994:167-170
4Napolitano M R,et al.Neural-network-Based Scheme for Sensor Failure Detection,Identification,and Accommodation.AIAA Journal of Guidance,Control,and Dynamics,1995,18(6):1280-1286
5Napolitano M R,et al.A comparison Between Kalman Filter and Neural Network Approaches for Sensor Validation.AIAA paper 96-3894
6Bulsari A,et al.Sensor Fault Detection Using State Vector Estimator and Feedforward Neural Networks Applied to a Simulated Biochemical Process.Actapolytech Scan,Chemical Technology and Metallurgy Series,1991:1-20
7Patton R J,et al.Neural Networks Based Fault Diagnosis for Nonlinear Dynamic Systems.AIAA Paper 95-3219
8Զ.ķϵͳʶ㷨.ҵѧʿ,1995
