Study On The In-depth Identification Of Acoustic Emission Signals From Bearing Failure Modes

As an indispensable component of the modern industrial system,rolling bearings often fail or damage due to human or objective reasons,hindering the normal operation of social production and life,so ensuring the normal operation of rolling bearings is an important part of industrial safety.In this paper,the bearing fault is diagnosed by acoustic emission signal,and through the analysis of the principle of acoustic emission signal generation and the meaning of each feature parameter,the acoustic emission signal of the fault rolling bearing is extracted,and combined with deep learning,the deep mapping relationship of the essence of the original acoustic emission signal in each fault state is excavated,which solves the traditional fault diagnosis method relying on manual experience,pre-processing complex problems,and completes the feature extraction and intelligent diagnosis of the acoustic emission signal of rolling bearing fault.Taking rolling bearings as the research object,this paper collects five kinds of fault acoustic emission signals of bearing no fault,cage wear fault fault,rolling element wear fault,inner ring wear and outer ring wear fault fault through the QPZZII rotating machinery vibration and fault simulation experimental platform,extracts the signal characteristics through simple preprocessing,obtains the data set required for neural network training,builds a suitable bearing fault diagnosis model,and successfully applies the fault sample to the fault diagnosis model proposed in this paper.The first is a bearing fault diagnosis model based on a convolutional network.The network obtains the optimal weight by setting certain hyperparameters and then using a backpropagation algorithm.Through the adjustment of the network structure,a comparative test is carried out to obtain the optimal diagnostic model.Experimental results show that the bearing fault diagnosis model based on convolutional network has an average identification accuracy of up to 96.80%,which can complete the bearing fault diagnosis task well.Secondly,the bearing fault diagnosis model based on The Conv-gru network is constructed,and the problem of insensitivity of CNN network to time characteristics is solved by introducing the gate control unit in the recurrent neural network,and the optimal diagnostic model is obtained by conducting comparative tests on the network structure.The accuracy rate of bearing fault identification of the Conv-gru network used in this paper has reached 99.13%,which proves that the Conv-gru network can better classify bearing faults and improve the accuracy of bearing fault identification,and effect is better than non-deep BP networks and CNN networks.