Bearing Fault Research Based On Neural Networks And System Realization

Rolling bearing is the joint of rotating equipment,used in automotive,electrical,medical equipment,aviation field and so on.Rolling bearing's fault will reduces production efficiency,threatens the safety of workers.Rolling bearing is prone to has a fault.Therefore,the health diagnosis of rolling bearing is an important issue.This article will introduce the state-of-the-art method in artificial intelligence to conduct research.First,introduce the components of rolling bearing,analyze the causes of faults,vibration mechanism,and characteristic frequencies,to make a foundation for subsequent research.Secondly,understand vibration signal feature extraction methods,compares their advantages and disadvantages to choose a better method.Feature extraction of the original vibration signal is the key to next task.Signal processing methods can be divided into time domain,frequency domain,and time-frequency domain.This article gives analyzes for these three methods.Considering the characteristics of the bearing signal,choose wavelet packet decomposition of the time-frequency analysis method to extract data features.The research data in this paper obtained from the bearing data center of Case Western Reserve University.The layer of wavelet packet decomposition for each sample is five.Gets 32 features for each sample finally.Then,research on several commonly used neural network structures and summarize their characteristics respectively.In order to use the time attribute of bearing data,choose to use long short term memory neural networks.Deep learning framework tensorflow is used to perform fault classification on the feature data extracted by wavelet packet decomposition.However,the internal structure of the standard long short term memory neural network is too complicated,with too many network parameters which result in longer running time of the network relatively.LSTM was designed for the natural language problem originally which has complex logic structure.But the logic structure of the bearing data is simple.To solve this problem,a simplified strategy for the internal structure of the long short term memory neural network is proposed.Single Gated Unit(SGU)neural network is designed.In order to use the information of each time step effectively,this paper uses Bi-directional Single Gated Unit(Bi-SGU)to diagnose bearing faults,proving that the structure has good performance in accuracy and time efficiency.In addition,this paper introduces Transformer for bearing fault diagnosis.The parallel feature is the advantage of Transformer for this task.It can use the time attribute without being limited by time step.At the same time,the concept of dense connection is introduced to strengthen the information transmission between the network layers which form a Dense Transformer structure for bearing fault diagnosis.Finally,through the analysis of the required functions of the bearing fault diagnosis system,based on the pyqt5 framework,a simple bearing fault diagnosis system is designed to realize the interfacial and visual operation of the bearing fault diagnosis,which is convenient for the operation of the monitor and the management of the bearing data.