Fault diagnosis based on EEMD-KPCA-IGSABP for motor bearing

To improve the accuracy of fault diagnosis for motor bearing with non-stationary and nonlinear characteristic, an ensemble fault diagnosis approach based on EEMD (Ensemble Empirical Mode Decomposition), KPCA (Kernel Principal Component Analysis) and IGSABP(Improved Gravitational Search Algorithm for Back Propagation neural network) is proposed. Firstly, EEMD extracts the non-stationary vibration signal data feature vector. Then, KPCA reduces the dimensionality of the vector. Finally, IGSABP classifies fault data into certain types. IGSA optimizes the weights and thresholds from input layer to output layer. To test the validity of the ensemble method, EEMD-KPCA-IGSABP is used to diagnose the motor bearing faults. The experimental results show that the accurate and adaptive of the ensemble method are improved compared with conventional single fault diagnosis methods.