Research On Fault Diagnosis Method Of Rolling Bearing Based On Convolution Neural Network

As an important part,bearing is widely used in automobile,engineering machinery,aerospace and other fields.The performance of bearing has a direct impact on the working condition of mechanical equipment.The fault of bearing may leads to equipment working unnormally,it will cause huge economic losses and even serious accidents.Therefore,the study of bearing fault diagnosis technology is of great significance in improving the reliability of equipment.The traditional technique,by analyzing time and frequency information of the signal,has higher requirements for experience and lower intelligence.In this thesis,the rolling bearing is decided to be the target of experiment.And the convolution neural network is used to summarize and learn the regular pattern of signal,so as to fulfill the intelligent fault diagnosis technique.(1)This paper introduces the mechanical structure and characteristics of rolling bearing.Starting from the common failure modes and failure mechanisms,it discusses the self factors and environmental factors that cause vibration when rolling bearing works,as well as the internal relationship between vibration characteristics and failure modes,and introduces the calculation methods of natural frequency and failure characteristic frequency of rolling bearing structures.(2)The problem is that the traditional signal analysing method only takes time and frequency domain as the unit of analysis,makes it harder to extract information.This thesis suggests a model based on wave transformation and neural network.In this technique,wavelet transform is taken as a tool to fetch the information of the signal,which overcomes the problem that the short-time Fourier transform is not sensitive to the frequency component of the signal.One dimensional time-series vibration signal can be transformed into two-dimensional image data for two-dimensional convolution neural network training.The experiments prove that this technique has the ability to distinguish the fault types of rolling bearing and other rotating machinery with obvious vibration characteristics correctly.(3)This thesis thinks of a fault diagnosis model combining vibration signal and one-dimensional convolution neural network to solve the problem of complicated data preprocessing in signal analysis.The method directly uses one-dimensional vibration signal training model,and uses the strong feature extraction characteristics of convolution neural network to realize end-to-end intelligent fault diagnosis.To settle the over fitting problem of convolutional neural network,this thesis optimizes the model structure and improves the generalization performance of the network by applying the global average pooling theory.The experimental results show that the method achieves good results improves the efficiency.