Research On Fault Diagnosis Of Rolling Bearing Based On Time-Frequency Method And Deep Learning

As an important part of rotating machinery,rolling bearings are prone to fatigue and wear due to their harsh working environment and complex and changeable working conditions,which lead to serious failures and frequent accidents.Therefore,it is of practical significance to detect the early failure and take corresponding measures to ensure the safe operation of the whole machine before the failure situation of the rolling bearing has not yet developed.In this paper,the traditional time-frequency method is combined with the intelligent Deep Learning algorithm,and several new fault diagnosis methods are proposed for the problem of how to efficiently and accurately perform feature extraction and fault identification of vibration signals,which are summarized as follows:A rolling bearing fault diagnosis model based on MOMEDA-ASO-VMD was proposed to solve the problems such as the coupling effect of preset key parameters in Variational Mode Decomposition affecting the decomposition effect,eliminating the interference of human factors and improving the anti-noise ability of VMD.Firstly,Multipoint Optimal Minimum Entropy Deconvolution Adjusted(MOMEDA)was used to filter and de-noise the signals.Secondly,Atom Search Optimization(ASO)is used to make VMD adaptively set key parameters of decomposition layers and penalty factors and decompose noise reduction signals.Finally,the Intrinsic Mode Function component is demodulated to judge the fault type.Simulation signals and experimental data show that this model can effectively distinguish single fault of rolling bearings in early stage.A rolling bearing fault diagnosis model based on multi-scale features and Long Short Term Memory(LSTM)network was proposed to simultaneously identify multi-position and different damage degree faults of rolling bearings and avoid the need for a large amount of prior knowledge reserve.Firstly,the vibration signal was decomposed by MOMEDA-ASO-VMD and reconstructed according to the envelope entropy.Secondly,the time domain characteristic statistics,frequency domain characteristic statistics and Composite Multiscale Permutation Entropy of reconstructed signals were extracted to construct a multi-scale fault feature data set.Finally,LSTM is used to diagnose rolling bearing faults with different damage degrees and multiple positions.Compared with the Support Vector Machine algorithm,the confusion matrix showing the diagnostic results proves that the proposed method has more real-time performance and diagnostic accuracy.The timing characteristics of the fault vibration signal of the rolling bearing are ignored by the ShufflenetV2 network.For the bearing fault problem of multiple working conditions,feature extraction cannot be performed efficiently and accurately.The ShufflenetV2 network has a deep number of layers and a large amount of parameters,which causes the network to be prone to overfitting.In response to the above problems,a rolling bearing fault diagnosis model based on the improved ShufflenetV2-LSTM network is proposed,in which the LSTM network and Dropout layer are serially embedded in the ShufflenetV2.This method preserves the ability of the ShufflenetV2 network to extract features,and the advantage of the ability of LSTM to strengthen the data sequence is also inherited,so this improves the accuracy of fault diagnosis.The generalization ability of the model can be enhanced by Dropout,which can effectively suppress the degree of overfitting.In addition,the activation function and optimizer are also optimized in the model,which makes up for the additional time cost brought by the Dropout layer,so that the robustness of the system is improved and fault diagnosis can be analyzed efficiently.The experimental results show that compared with other mainstream algorithms,the proposed algorithm has a shorter training time and a diagnosis accuracy of 97.38%,which can effectively discriminate the faults of rolling bearings under multiple working conditions.