Research And Design Of Rolling Bearing Fault Monitoring System Based On Convolution Neural Network

Rolling bearing is one of the key components that affect the safety and security of vehicle service.With the accumulation of the number of trains and the service life of existing vehicles,at the same time,under the high-strength operation of high-speed,high-volume and high-density trains,its rolling bearings are often prone to various faults.If the bearing failure occurs,it is easy to cause major safety accidents,including derailment,clutching,shaft breaking and so on.Therefore,it is of great significance to monitor the fault state of rolling bearings in the process of train operation.At present,the fault detection of the bearing is mainly through the sensor to collect the vibration signal of the bearing,and then use the signal analysis method to extract the fault features,and then further analyze the extracted features,the process of feature extraction is more tedious.At the same time,the train actually runs under the complex working conditions of strong noise,variable speed and variable load,which leads to the collected vibration signals containing a large number of complex external factors such as random vibration and noise,so it is often difficult to obtain fault characteristics by using the signal processing method directly,and the fault diagnosis performance is poor.Therefore,this paper takes the rolling bearing as the research object,puts forward the method of combining resonance demodulation and convolution neural network to improve the effect of fault diagnosis,and realizes the design of rolling bearing fault monitoring system according to the actual application requirements.The fault of the rolling bearing will produce periodic fault impact in the process of train operation and cause resonance of accelerometer.In this paper,the rolling bearing fault monitoring system is divided into two parts: acceleration sensor and bearing fault monitoring host to realize the fault state monitoring of multiple rolling bearings.The acceleration sensor is responsible for collecting the vibration signal during the operation of the rolling bearing and transmitting it directly to the host computer through the vibration signal acquisition channel.Then the host collects and diagnoses the received vibration signals.The main machine of bearing fault monitoring is realized by the hardware structure of ARM+FPGA,and the method of combining software and hardware of resonance demodulation circuit and convolution neural network is used for fault diagnosis and analysis.Firstly,the host uses analog devices such as operational amplifiers to build a hardware resonance demodulation circuit to preprocess the vibration signal,which can effectively filter and suppress the interference and noise in the bearing signal which has nothing to do with the fault information.the circuit further improves the efficiency of fault feature extraction by improving the signal-to-noise ratio of the signal.Then ARM controls FPGA to collect resonance signals after filtering.Finally,the collected characteristic data is sent to ARM for fault diagnosis,analysis and storage,and the diagnosis results and characteristic data are sent to the safety monitoring center on the train through Ethernet.The bearing fault diagnosis algorithm completes the whole process of automatic fault feature extraction and classification by building a one-dimensional convolution neural network model,replacing the tedious feature engineering of the traditional fault diagnosis algorithm to realize "end-to-end" rolling bearing fault diagnosis.Through the rolling bearing fault contrast diagnosis test,the recognition rate of the model on the original data set is 86.45%,and the recognition rate on the resonance demodulation data set is 96.59%.The experimental results show that the model can effectively diagnose the fault of rolling bearing,and through the pretreatment of resonance demodulation,the recognition rate of the model for fault diagnosis is obviously improved.Finally,this paper completes the test and verification of the system through the static and dynamic debugging of the motor rolling bearing vibration test experiment.the experimental results show that the functions of data acquisition,diagnosis,storage and Ethernet communication of the system run normally,and a good fault diagnosis effect has been achieved in practical application.