The Research Of Fault Diagnosis For Electric Actuator Base On PCA And LS-SVM

Actuators play an important role in the system of industrial automatic control, regardless of automatic control system complex, finally comes down to control of the actuator. The failures of actuator often have a direct impact on performance, therefore, detecting the presence of faults on time during operation is the basic requirements to guarantee the stability and economic of the control process.The principle and characteristics of electric actuator is studied by this thesis, on the basis of further research the common failure mechanism and fault performance of the electric actuator, finally the design the controller of electric actuators as the core of PIC18F485microcontroller, the date of fault characteristic values is collected according to the mechanism of fault select in failure of collection point.For collected fault data from electric actuator, the thesis adopts PCA to process multivariable data set, as PCA can reduce dimensions when processing linearly dependent data. Traditional PCA take standardization as the method for data preprocessing, which will cause the loss of some information. According to this point, the thesis replaces the standardization with equalization as the preprocessing method for PCA.Based on data after dimension reduced by principal component analysis, researching how to establish the model of fault diagnosis of electric actuator. Respectively compare and analyze the selected kernel function modeling for fault diagnosis, including the polynomial kernel function, radial basis kernel function, Sigmoid kernel function. Establish four multi-classification algorithm model with one to many algorithm, one to one algorithm, DAG-SVM algorithm and multi-classification algorithm based on binary tree. In the problem of LS-SVM parameters selection, the right model parameters have been chosen through the two-step grid search method combining with the method of cross-validation. Through the comparison of diagnostic results, it shows that the classification model based on multi-classification algorithm of the binary tree, in which the radial basis function is served as the kernel function, has a high rate of correct judgments. Diagnosis effect shows the classification model have the higher diagnosis rate which with RBF kernel function based on multi-classification algorithm of binary tree. Finally, the comparison of recognition accuracy between the fault diagnosis model of PCA and LS-SVM and the fault diagnosis model of BP neural network, reflects the advantage of PCA and LS-SVM fault diagnosis model on learning problems in small sample.