Research On Rolling Bearing Fault Diagnosis And Prediction Mathod Based On Machine Learning

Rolling bearings,as a prevalent part of complex mechanical drive systems,are extremely susceptible to failure problems in the long-term high-noise environment and high-load operation process.In the early stage of failure is not easy to be detected in time,with the accumulation of failure,will seriously affect the operation of the mechanical system.Therefore,it is necessary to monitor the operation of rolling bearings and accurately determine the types of faults to ensure the safe and stable operation of mechanical systems.This thesis takes rolling bearing as the research object,for the current rolling bearing fault diagnosis recognition rate is not high and the remaining life is difficult to predict the problem,the rolling bearing fault feature extraction and fault type identification and life prediction to carry out research,the specific main research content is as follows:(1)In view of the problem that the bearing signal is susceptible to noise interference leading to inadequate feature extraction,a rolling bearing signal noise reduction method based on variational modal decomposition(VMD)is proposed to reduce the noise of the bearing signal.An improved differential evolution operator(DE)optimized krill swarm algorithm(KH)is introduced to adaptively select the number of decomposition layers and penalty factors of the VMD to obtain the optimal combination of parameters.The effectiveness of the algorithm is confirmed by numerical simulation and simulation experiments and the noise reduction effects of EMD and EEMD are compared.The experimental results show that the algorithm can effectively improve the signal-to-noise ratio of the signal,overcome the modal aliasing problem,and provide a signal basis for fault diagnosis and life prediction of rolling bearings.(2)A support vector machine(SVM)fault diagnosis method based on sine cosine algorithm(SCA)optimization is proposed.To address the problem of difficult extraction of rolling bearing fault features,the generalized composite multiscale permutation entropy(GCMPE)is extended as the fault feature vector based on the multiscale permutation entropy(MPE).The selection of kernel function and penalty factor in SVM affects the final fault classification and recognition accuracy,and the SCA algorithm is used for adaptive optimization of key parameters of SVM to construct a GCMPE-SCASVM fault diagnosis model,use GCMPE-SCA-SVM classification model for multiple fault diagnosis and compare the classification effect of MPE-SCA-SVM,GCMPE-SVM,GCMPE-PSO-SVM and GCMPE-ELM models.The experimental results show that the fault diagnosis model has more accurate recognition accuracy and more stable recognition ability.(3)To address the problems of large number of complex degradation features extraction and low prediction accuracy in rolling bearing remaining life prediction,a convolutional neural network(CNN)based rolling bearing remaining life prediction method is proposed.The bearing signal is transformed from one-dimensional to twodimensional spectrogram using short-time Fourier transform,and the spectrogram is downscaled using bilinear interpolation,and then input into the CNN model for training and classification after denoising.The classification model is validated using the IEEE PHM 2012 RUL dataset and compared with the BP neural network prediction effect.The experiments show that the method has higher prediction accuracy and has stronger prediction ability after the noise reduction is completed for the bearing signal.