Research Of Rolling Bearing Fault Diagnosis Algorithm Based On Convolutional Neural Network

As the most common parts of rotating equipment,bearings are widely used in our daily life and production.In the industry,rolling bearings have become the most widely used type of bearing due to their simple manufacturing and cheapness.Rolling bearings are generally installed in rotating machinery to play a supporting role,and its actual working environment is complicated and prone to failure.If a malfunction cannot be diagnosed and replaced in time,it may cause serious harm to the entire equipment.The fault diagnosis method based on rolling bearing mechanism model and signal decomposition theory usually compares the actual time-frequency domain characteristics with theoretical values,obtains the fault type based on expert experience,or obtains the time-frequency domain characteristics of the rolling bearing vibration signal,using the classified model to classify the fault of the rolling bearing.This fault diagnosis method requires technicians specializing in fault diagnosis to manually select fault features,which may cause the lack or redundancy of fault features,and limit the intelligent development of rolling bearing diagnosis technology.In response to the above problems,this paper proposes a rolling bearing fault diagnosis method based on two-dimensional convolutional neural network.This method first transforms the one-dimensional vibration signal into a two-dimensional form and inputs it into the two-dimensional convolutional neural network for training,then classifies the extracted features through the Softmax classifier,and finally uses the optimizer algorithm to iteratively reduce the error and get the final diagnostic model.Experimental results show that this method can directly use the original vibration signal data to complete the fault diagnosis task of rolling bearings,and the accuracy and precision of diagnosis are higher than those based on rolling bearing mechanism model and signal decomposition theory.The working environment of rotating machinery and equipment is complex and the working conditions are changeable.The existing fault diagnosis methods cannot meet the fault diagnosis requirements of multiple working conditions.To solve this problem,proposes a rolling bearing fault diagnosis method based on Multi-scale Convolutional Neural Network on the basis of 2DCNN.This method first inputs the vibration signal into two One-Dimensional Convolutional Neural Networks with different convolution kernels to extract the high-frequency and low-frequency fault features in the vibration signal,and uses a 2DCNN at the same time extract the two-dimensional fault features in the vibration signal,and then fusion one-dimensional features into two-dimensional features,and then classify the fused features by using the Softmax classifier in the fully connected layer,and finally use the optimizer algorithm to iteratively reduce The error gets the final diagnosis model.The experimental results show that the fault diagnosis method can well complete the task of self-adaptive extraction of fault features and fault type diagnosis of rolling bearing vibration signals under a variety of workloads and noise environments.