Fault Diagnosis of Gear Rotor System Based on Collaborative Filtering Recommendation Method

Abstract

The gear rotor system is the core component of high-end energy power equipment such as compressors, wind generators and other speed-increasing box devices, and failures are inevitable. However, the gear-rotor system often suffers from pitting, cracks and even broken teeth or their combination failures, and the vibration signals exhibited have weak, very strong random interference, nonlinear and non-stationary characteristics. In the fault diagnosis of gear rotor system based on vibration signal, the fault feature extraction and fault identification are particularly difficult. Based on the strong characterization ability of multi-scale permutation entropy for time series signal complexity and multi-level wavelet energy entropy for non-stationary time-varying signal, the entropy change law of vibration signal, under different fault states of gear, are analyzed; In order to solve the problem of multi-class state data imbalance, an adaptive oversampling balance method is proposed; In order to solve the problem that the collaborative filtering recommendation algorithm cannot construct a scoring matrix as used in fault diagnosis, design an innovative design of the gear fault feature-state joint scoring matrix. Combining time-domain and time-frequency domain entropy, a new collaborative filtering recommended gear fault diagnosis model is constructed, and the model parameters are optimized by using gradient descent algorithm and alternating least square method; Finally, a collaborative filtering recommended gear fault diagnosis method based on adaptive minority oversampling technology and multi-scale and multi-domain entropy fusion is formed. The results show that the diagnostic accuracy of collaborative filtering recommendation based on multi-scale permutation entropy has been improved from about 70% to about 90%, after adopting adaptive minority oversampling to balance data types, and collaborative filtering recommendation based on multi-layer wavelet energy entropy is accurate. Moreover, the rate has risen from around 75% to over 90%, and fusion of multi-scale permutation entropy and multi-layer wavelet energy entropy as the feature quantity for gear fault diagnosis, the diagnosis and recognition rate of unbalanced data can reach more than 90%, and the balanced data can reach more than 95%.