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

Since the rapid development of industry,in order to increase production efficiency,petrochemical,electric power,metallurgy and other industries constantly upgrade equipment,leading to industrial equipment and industrial systems tend to be large,intelligent,complex,and equipment maintenance costs have also increased.Monitoring and fault diagnosis of equipment operation status has become one of the current research focus.The nonlinear characteristic of the dynamics of bearing-rotor system leads to a strong non-linear relationship between the vibration signal characteristics and the operation state.The feature extraction and selection of the vibration signal often requires a large amount of prior knowledge,which makes the features not so accurate to reflect different operating conditions.Moreover,because of the interrelation and restriction among the various functional units within the complex mechanical equipment,the characteristic of the equipment running state and its signal features is time-varying.Therefore,it is difficult to completely and effectively reflect the real-time running state of the equipment by using a single static model established by offline data.Aiming at the above problems,focusing on the key problems of feature extraction and model updating in fault diagnosis of complex mechanical equipment,this paper develops a fault diagnosis method based on Convolutional Neural Networks(CNN)and Extreme Learning Machine(ELM).The core contents of the reaserch are the feature extraction and fault type recognition of rolling bearing vibration signal.The main research contents and innovations are as follows:(1)Enable to detect the operation of wind turbine in real-time and recognize its health status,a method based on vibration signal analysis using ITD(Intrinsic Time-scale Decomposition)-MSE(Multiscale Entropy)is proposed in this paper.The method is applied to ensure that the vibration signal is pretreated,and the time domain characteristics of the reconstructed signal are extracted.The ELM(Extreme Learning Machine)is then adopted to recognize the health status of wind turbine bearings.The experiments results have shown that the proposed model is capable of recognizing the health status of wind turbine bearings accurately.(2)When multi-domain features are used to characterize the equipment operating status,dimension catastrophe will occur as the feature dimension increases,causing classifier performance degradation and degrading the performance of the state monitoring model.To solve this problem,a linear local tangent space alignment(LLTSA)feature dimensionality reduction and extreme learning machine(ELM)model fault diagnosis method is proposed.Using LLTSA to extract low-dimensional manifolds from the high-dimensional feature space reduces the dimensionality of signal features,and ensures model classification performance.The monitoring experiment of the check valve of the high pressure diaphragm pump and bearing shows that the characteristic dimension of the vibration signal is reduced,and the redundancy between the features is reduced,which can improve the accuracy of the fault recognition of the ELM model.(3)In order to solve the problem of adaptive feature extraction of vibration signals,a method based on Complementary Ensemble Empirical Mode Decomposition(CEEMD)and Convolutional Neural Networks(CNN)is proposed,which adaptively extracts the sensitive features from the time-frequency diagram of vibration signals and characterizes the operation status of the equipment.The method is applied to build the bearing faults diagnosis model with different loads and fault depths.The results show that the method can effectively improve the fault recognition rate under various working conditions.(4)The single static model established is difficult to reflect the real-time running state of equipment in a complete and effective way.Therefore,a fault diagnosis method based on Convolutional Neural Networks(CNN)and Online Sequential-Extreme Learning Machine(OS-ELM)is proposed.OS-ELM is used to update the fault diagnosis model in real-time to improve the dynamic adaptability and operation efficiency of the fault diagnosis model by gradually adding new samples.It also reduces the retraining time of the model.The simulation results of bearing data sets show that the method can identify fault types quickly and effectively.This paper focuses on the key issues such as adaptive feature extraction of rolling bearing vibration signals and online updating of models.The results can be used for reference in fault diagnosis of complex mechanical equipment.