Research On Remaining Useful Life Prediction Of Rolling Bearings Based On Deep Learning

As our country’s development of industry becomes more and moer intelligent,people put forward higher requirement for safety and reliability of equipment.Rolling bearings are the key components of many rotating equipment.The research on remaining useful life(RUL)prediction of rolling bearings is beneficial to ensuring equipment performance and reducing accidents.However,the traditional data-driven methods of RUL prediction almost ignore the influence of the environment noise.To boost the performance of bearing’s health indicator(HI)and improve the RUL prediction accuracy,this thesis based on deep learning focus on reducing the impact of noise.The specific research contents of this thesis are as follows:Aiming at the problem of the bearing feature’s poor performance due to the impact of environment noise,this thesis proposed a bearing feature extraction method based on Convolutional Autoencoder(CAE).At first,the CAE is designed to a extract multidimensional features from bearing vibration data,and then reduce the dimension by the method of Principal Components Analysis(PCA)to build one-dimensional HI with denoising ability.What’s more,using the extracted features and multiple statistical features,the initial degenerate point(IDP)of the bearing is determined by the method of self-zero space projection analysis.Finally,the experimental results show that,the proposed method can extract bearing degradation with better performance in a noisy environment are carried out using real data.Compared with the time-domain statistical features and fusion features,the monotonicity is increased by 5.71% and the robustness is increased by 6.84%.Aming at the problem of the low RUL prediction accuracy due to the time-variant noise,this thesis proposed a RUL prediction model based on feature fusion and Temporal Convolutional Network(TCN).At first,to reduce the impact on the RUL prediction adaptively,soft threshold and attention mechanism are used in the TCN’s residual network.What’s more,some artficial features are fused in the RUL prediction stage of the model,which can further improve the prediction accuracy.Finally,the experimental results show that,the method proposed can predict the RUL of the bearing more accurately.Compared with the typical RUL prediction model,the proposed model’s Score is increased by 4.22%.Based on the researched methods and models,a bearing management system is designed and implemented.The system includes three parts: beaing monitoring unit,system data unit and system client.On this basis,the function verification of the important modules of each part is carried out.The verification results show that,compared with the existing bearing health management system,the designed system can accurately determine the IDP of the bearing in the noise environment,and the RUL prediction results are more accurate.It is extended to the actual industrial environment of rotating equipment,and it can be widely used in enterprise health management equipment.