Rolling Bearing Fault Diagnosis Based On S-transform And Deep Learning

There are many key components of mechanical transmission device,but rolling bearing,as one of the core parts,its operating environment is particularly bad,so it has an important impact on the service life of mechanical equipment,and it is also the key factor for the normal and reliable operation of the whole equipment.Rolling bearing is also one of the most vulnerable parts of mechanical transmission components.This is because the bearing bears the load during mechanical transmission and avoids the wear between components.The good running state of bearing performance directly affects the running performance of the whole mechanical equipment.Bearing is the joint of rotating machinery,which is used in many fields,such as automobile,electrical,medical equipment,aviation and so on.When the rolling bearing breaks down,it will affect the production efficiency of the whole mechanical equipment,and there will be hidden dangers threatening the safety of personnel,and the failure frequency of the rolling bearing is very high.Therefore,the diagnosis of the actual running state of the rolling bearing is an important research topic.In this paper,the bearing data set provided by Case Western Reserve University is selected to carry out the simulation analysis and example verification of the proposed model.The bearing data set is directly used for feature extraction,because the signal features reflected in the time-domain signal are not sufficient,which will lead to serious loss of the final extracted features.Therefore,in this paper,the original time-domain signal data is used to get the time-frequency characteristics of the original signal through S-transform.After processing the two-dimensional time-frequency characteristics,the convolution neural network model(CNN)is trained by using the time-frequency characteristics.Through the comparison of performance evaluation indexes,the effectiveness of the method of bearing fault diagnosis based on the combination of S-transform and CNN model is proved.Then,under the same experimental conditions,the diagnosis model proposed in this paper is compared with other traditional models.The simulation results show that the average accuracy of the proposed model is 94.62%,and the extracted features have high discrimination.Compared with long-term memory network(LSTM),convolutional neural network(CNN)and support vector machine(SVM),this model has higher diagnostic accuracy and better stability.Because the model proposed in this paper has to set a large number of parameters artificially before training,and the model is not universal when encountering new fault diagnosis problems,and the model is also lack of self adaptability,this paper introduces the sub group algorithm(PSO)to optimize the selection of key parameters to change the shortcomings of the lack of self adaptability of CNN network.Finally,the adaptive model is verified by the performance index,and the simulation results show that each index of the adaptive model is very high.