Research On The Intelligent Diagnosis Method Of Rolling Bearing Fault Under Multiple Working Conditions Based On Machine Learning

As one of the key components of machine tool electric spindle,the condition of rolling bearing directly affects the operation status of machine tool spindle and the whole computer numerical control machine tool.An effective intelligent diagnosis method of rolling bearing failure is essential to improve the reliability of computer numerical control machine tool processing.However,in order to meet the processing requirements of different products,the operating conditions of electric spindle bearings are complex.The machine learning intelligent diagnosis model established in a certain working condition(source domain)is difficult to ensure better diagnosis when applied to new or unknown working conditions(target domain).In addition,the monitoring signal is easily disturbed by other equipment and environment,and there is noise information unrelated to the fault state,which leads to the decrease of the diagnostic accuracy of the model.In view of the above problems,this paper takes rolling bearings of electric spindles as the research object and carries out the research on the intelligent diagnosis method of rolling bearing faults under multiple working conditions based on machine learning,as follows.For the problem that rolling bearing fault vibration signals are easily disturbed by other equipment and environment,etc.,a noise reduction preprocessing method based on improved variable fractional modal decomposition-adaptive wavelet thresholding(IVMDAWT)is proposed.The method firstly improves the variational modal decomposition based on the dual determination criteria of sample entropy and correlation coefficient to avoid the problem of inaccurate IMF component screening.Secondly,an adaptive wavelet threshold function is constructed,which can adjust the noise reduction form adaptively according to the noise content of the noisy components and clean the screened noisy components with secondary noise reduction.Finally,the joint noise reduction is realized by reconstructing each component.Through the experimental verification of the simulated and measured signals,IVMD-AWT has better noise reduction effect and adaptivity,and can improve the diagnostic accuracy of the machine learning model to a certain extent.Existing intelligent diagnosis models usually design deep and complex network structures to extract richer fault features,which leads to large number of parameters and computation and low diagnosis efficiency.To address this problem,an electric spindle bearing fault diagnosis method based on convolutional neural network with improved inverse residual structure based on efficient channel attention(EIR-CNN)is proposed.This method is based on the lightweight network Mobile Net V3,and further light-weight optimization of its inverted residual structure,replacing the squeeze excitation(SE)module in the inverted residual structure with the efficient channel attention(ECA)module to construct the improved inverse residual structure based on efficient channel attention(EIR)which leads to the proposed EIR-CNN.The EIR-CNN has been verified by the experiments of rolling bearing and spindle bearing fault diagnosis under single working condition,with smaller number of parameters and computation,shorter training time and higher stability of the model while maintaining high diagnostic accuracy.To address the problem of large differences in data distribution under multiple operating conditions and the low diagnostic accuracy when training machine learning models under a certain operating condition and applying them to new or unknown operating conditions,we propose an efficient channel attentive improved inverse residual structure of convolutional neural networks joint domain-adversarial training of neural networks(EIR-DANN)is proposed for the diagnosis of electric spindle bearing faults.Based on the research of EIRCNN,this method constructs the feature extractor of domain adversarial training of neural networks(DANN)based on EIR-CNN,and makes the feature extractor continuously extract common features under different working conditions based on the idea of domainadversarial,and then completes the unsupervised migration fault diagnosis under multiple working conditions.It is verified that EIR-DANN can effectively solve the problem of crossconditions fault diagnosis under multiple working conditions with high diagnostic accuracy and generalization performance by the experimental diagnosis of rolling bearings and spindle bearings under multiple working conditions.