Research On Machine Sound Fault Diagnosis Models With Deep Neural Networks

Engine faults are existing in the form of sound signal,and characterized by diversity and diagnosis difficulty.Sound fault diagnosis methods based on neural network can overcome handcrafted shortcomings such as high labor cost and low diagnosis efficiency,which become the research hotspot in the field of machine fault diagnosis in recent years.We carry out fault diagnosis research on motorcycle engine sound data based on convolutional neural network and recurrent neural network.The main work is as follows:We research the problem of motorcycle engine fault diagnosis based on a Convolutional Neural Network(CNN)that is used to classify sound emotions.Several unified time data processing tricks and data augmentation methods are used on the motorcycle engine sound data,the FBank(Filter Bank)spectral features are extracted,and the classification model is trained on the features.The experimental results show that random down-sampling and constant continuation strategies on data set augmentation improve performance on the model.The trained model is evaluated on the test set,and the accuracy,precision and recall rate are 91.0%,92.8% and 90.7%.In order to improve the accuracy of fault diagnosis,the corresponding research on the above fault diagnosis problem are carried out based on several recurrent neural networks.On the basis of the above-mentioned data supplement and data augmentation methods,the spectral features are extracted along with the first-order difference.We construct classification models for motorcycle engine sound data based on simple recurrent network,gated recurrent unit(GRU)network and long short-term memory network(LSTM)respectively.A large number of comparative experiments show that the accuracy rate of the trained LSTM on the test set can reach 99.7% by using random down-sampling and adding white noise to expand the train set,which basically meet industrial request.