Research On Bearing Fault Diagnosis Based On Improved Deep Neural Network

With the development of modern industry,motors occupy an increasingly important position in the industry,and motor bearing faults are one of the most common faults in motor equipment.When a motor bearing failure occurs,it will cause huge economic losses in the slightest,and endanger personal safety in the severe case.Therefore,it is of great significance to detect and analyze the motor bearing failure.However,because the working environment of the motor bearing is relatively complex,and the signal is often accompanied by a lot of noise,how to extract useful feature information from the complex signal is a matter worthy of study now.Aiming at the problems faced by motor bearing faults,this paper proposes improved neural network bearing fault diagnosis method to conduct in-depth research on motor bearing fault diagnosis.According to the characteristics of motor rolling bearing data,the traditional method and deep learning method were used to analyze the characteristics of bearing signals from the perspectives of time domain and frequency domain,and the distribution of data features was expounded in many aspects.Then,the bearing signal is simulated by wavelet analysis and Fourier transform,and the problems in the signal are analyzed.Then,bidirectional LSTM network and deep autoencoder were used to test the bearing data,and Gaussian white noise was added to the signal to verify the noise resistance of the model.The experimental results show that the bidirectional LSTM network has a good diagnostic effect on the motor rolling bearing fault data,but when the data noise is large,there are still some problems,such as the difficulty of feature extraction and the decrease of model accuracy.Aiming at the problem that data features are not easy to extract,a fault diagnosis model combining improved sparse filtering and deep dilated gate convolutional network is proposed.First,a sliding window is used to resample the bearing vibration signals with timing characteristics.Then,by improving the sparse filtering of the objective function,the heteroscedasticity in the data is eliminated and the data features are extracted,shortening the calculation time and improving the training speed of the network model.Finally,a simulation experiment on bearing data is carried out using dilated gate convolution and the bidirectional LSTM network.Through the comparative experiment of the two bearing data,it is verified that the proposed method can effectively extract the feature information in the data,and the proposed model structure also achieves a higher accuracy rate.Aiming at the problems of large noise in the data signal and poor model generalization,a fault diagnosis model combining NAdam(natural exponential decay)algorithm and improved octave convolution is proposed.First,a natural exponential decay function is proposed to replace the exponential decay function in the optimization algorithm to update the model structure parameters.Compared with the exponential decay function,the natural exponential decay function can speed up the convergence speed of the model.Then,an improved octave convolutional network model is proposed to eliminate the redundancy in the data and classify the data.After experiments with different data sets,different loads and different signal-to-noise ratios,it is verified that the proposed method can quickly bring the model to a state of convergence,and at the same time has good anti-noise performance.To sum up,this paper studies the problems of motor rolling bearing features that are difficult to extract and strong noise,and puts forward the network model combined with improved sparse filtering and deep dilated gate convolution network and the fault diagnosis model combined with NAdam algorithm and improved octave convolution network.The experimental results show that the proposed algorithm can still achieve good diagnostic results under the conditions of variable load and strong noise,and it has certain guiding significance for the fault diagnosis of motor rolling bearings.64 pictures,8 tables,80 references.