Research On Fault Diagnosis And Life Prediction Of Rolling Bearing Based On Machine Learning

With the increasing application of rolling bearing in recent years,it is an indispensable standardized part of various complex mechanical equipment.Higher requirements are put forward for the stability and safety of rolling bearing operation in engineering.Rolling bearing works in complex environment for a long time,and it is easy to be impacted by external load.It is very easy to break down in the process of operation,and then endanger the life of mechanical equipment and even personal and property safety.Therefore,the fault diagnosis and residual life prediction of rolling bearing is very meaningful.The traditional health analysis of rolling bearing is mainly for signal processing and manual extraction of signal features,which requires a high performance of the algorithm and relies on the experience and professional level of experts and scholars,which brings great difficulties to the health analysis.In view of the above problems,this paper analyzes the health status of rolling bearing mainly considering two aspects.The first is fault diagnosis of rolling bearing,and the second is life prediction of rolling bearing.For the fault diagnosis of rolling bearings,a kind of gating recurrent unit structure based on residual network is proposed.The algorithm can reduce the loss of timing information and the degradation of network performance caused by deep network.This method extracts information from the collected bearing vibration signals by using the powerful feature extraction ability of convolution neural network.In order to avoid losing time sequence information,the extracted information is input into GRU network in turn.Finally,a residual module is used to solve the problem of deep neural network.The model includes two CNN layers,two GRU layers,one residual block and one output layer.The experimental results show that this method can diagnose the faults of different positions and sizes of various bearings at one time,and the classification results of fault diagnosis under the same experimental conditions are better than those of deep learning networks such as CNN and CNN-GRU.For the remaining life prediction of rolling bearing,firstly,the refined network structure of Transformer is analyzed,and the whole process of multi attention mechanism is analyzed.An adaptive network Transformer is proposed.The method also extracts the information from the collected bearing vibration signal by using CNN,and then inputs it into the adaptive position coding module to retain the time sequence information of the vibration signal,and then inputs the extracted information into the coding network of transformer,and finally obtains the output through full connection.The model consists of two CNN layers,one location coding layer,one transformer coding layer,two fully connected hidden layers and one output layer.The experimental results show that this method can effectively predict the residual life of rolling bearing,and the fault diagnosis classification results under the same experimental conditions are more stable than deep learning networks such as CNN and CNN-LSTM.In order to improve the convenience of using the algorithm,this paper develops a rolling bearing health status research system,integrates the algorithm into the system interface,which is more convenient and fast for practical use.