The flywheel fault detection based on Kernel principal component analysis

Abstract

Simulink model is built according to the mathematic model of flywheel system which is a closed-loop system. Considering the fault modes and the corresponding fault parameters of flywheel, the training data and testing data for two faults are collected from the Simulink model. And then the Kernel Principal Component Analysis (KPCA) is proposed to analysis the correlation of five parameters in the flywheel system. The faults of flywheel will cause some abnormal changes of testing data. This method can solve the problem of lack of fault knowledge and complex mathematic modeling. The behaviors of flywheel can be learned from the training data and the correlation is the interactions of five parameters in the flywheel system. However, the variable correlation is classified as two types, such as the nonlinear and linear relationship. The Principal Component analysis (PCA) is used to build the linear model of training data. The chosen five variables is a nonlinear relationship. In order to demonstrate the effectiveness of proposed algorithm, it is necessary to compare the detections results of KPCA with the results of PCA. Numerical simulation results show that the SPEKPCA index can detect the faults of flywheel without complex mathematical modeling, and better than the detection results of PCA model.