Research Of Data Driven Fault Diagnosis Approach For Actuator Of Civil Aircraft

Modern industrial system has become more and more complicated to meet the increasing demand on production quality and system performance,causing the problems of safety and reliability,fault diagnosis technique is then proposed.System nonlinearity and disturbance diversity make it difficult to model according to system mechanism,and traditional model-based fault diagnosis algorithm is therefore inappropriate.On contrast,data driven fault diagnosis method is capable of utilizing historical system data for establishing fault diagnosis model,and is more suitable for modern industrial system without the effort for analytical modeling.As the executing agency of the flight control system,the actuator is a mechanical-electro-hydraulic highly-coupled,multiinput-multioutput and severe nonlinear closed-loop control system.Because of the complex structure,it has a high failure rate.Thus,it is of great theoretical and realistic significance to establish a data driven monitoring model for the actuator.Meanwhile,as the operation process of the actuator is a typical batch process,it is also conducive for improving monitoring accuracy and acquiring better understanding to separate the process into sub phases accurately.The basic principle of multiway principal component analysis(MPCA)is introduced,and its performance is verified and compared under different parameters.Aiming at the shortcomings of MPCA in practical application,a fault diagnosis method for actuator based on multiway information incremental matrix(MBEAM)is put forward,which can directly take advantage of the correlation changes in historical batch data without conducting eigenvalue decomposition of the covariance matrix.The algorithm can provide more direct interpretation of faults,reducing the amount of computation.Experimental results have illustrated the superior performance of the method in online monitoring.In addition,the performance of the method is further improved by combining the adaptive kernel principal component analysis(KPCA)method and the MBEAM method.MBEAM is adopted for separating the operation process into sub stages in the first place.Then the kernel parameter of the KPCA algorithm is optimized according to the maximumvariance principle(MVP)and the nearest neighbor distance reservation principle(NNDPP).Because the information of the data is maximized through the projection represented by the kernel function,more sufficient sample feature can be extracted for fault diagnosis.Experiment results have illustrated the superior performance of the algorithm in feature extraction and fault diagnosis.Finally,the performance of the methods mentioned are validated on a real actuator.The basic principle,hardware structure and available experimental data of the actuator are introduced,and the data are further selected considering the characteristics of the actuator and the principle of the algorithm.Furthermore,experiments are conducted under two different operation patterns to validate the performances of the methods.At last,through comparing and analyzing experimental results,the principle of selecting monitoring method for the actuator is proposed that appropriate method should be chosen by balancing model accuracy and computation complexity,so that the requirements of actual operation conditions can be satisfied.