Rotor Fault Diagnosis Based on ICEEMDAN Multi-Scale Entropy and NGO-HKELM
In view of the non-stationary and sensitive fault features of motor rotor fault signals,the traditional classifier parameter intelligent optimization algorithm has some problems,such as slow optimization speed,too many adjustment parameters,and easy to fall into local optimum.A rotor fault diagnosis method based on ICEEMDAN-MSE-KPCA and NGO-HKELM optimization is proposed.Firstly,the improved complete empirical mode decomposition with adaptive noise(ICEEMDAN)method is used to decompose and recon-struct the rotor vibration signals;Multiscale sample entropy(MSE)of reconstructed signals was calculated to form feature vectors.Kernel principal component analysis(KPCA)was used to reduce the dimensionali-ty of high-dimensional feature vectors;Finally,the dimensionally reduced feature vector was input into the hybrid extreme learning machine(HKELM)optimized by the northern goshawk optimization(NGO)al-gorithm for rotor fault classification.The results show that the rotor fault diagnosis model optimized by ICEEMDAN-MSE-KPCA and NGO-HKELM has an average recognition accuracy of 97.727 3%and an average search time of 1.068 1 s,with fast convergence,high accuracy and good classification effect.
NETL
NSTL
万方数据
维普
