Research On The Rolling Bearing Performance Degradation Assessment Based On Deep Denoising Autoencoder

With the rapid development of various fields,the use of mechanical equipment has been increasing in recent years.And rolling bearings have been widely studied due to their wide application and fragility.With the development of artificial intelligence,the equipment has changed from mechanization to intelligence.How to obtain the fault characteristic information of bearing effectively and automatically has become a research hotspot in the field of fault diagnosis and health management.Therefore,the characteristic information in complex vibration signals is extracted automatically by the deep denoise auto-encoder in deep learing,and the one-class support vector machines is used to evaluate the performance degradation of bearings.According to the characteristics of the autoregressive model,the autoregressive coefficients and residuals of the model are taken as the extracted feature matrix.The euclidean distance is often used as the degradation index in the performance degradation assessment distance model,while the mahalanobis distance has the characteristics of simple and could not consider the influence of sample units.Therefore,the mahalanobis distance is used as the model of bearing performance degradation in this paper,and the degree of membership function is combined to optimize the degradation curve,which is more convenient for degradation analysis.However,the Mahalanobis distance model needs to use fault samples for training,and the deep denoise auto-encoder has the feature extraction advantages.So we combine the mahalanobis distance with it.And using the bearing data of the University of Cincinnati to verify the validity of the model.The one class support vector machine was used as the evaluation model of bearing performance degradation.Combined with the characteristics of single classification,the data of XJTU-XY dataset under different working conditions were used to verify the effectiveness of the model,and the initial failure frequency point of the bearing was verified by envelope spectrum analysis.In view of the problem that the traditional feature extraction method cannot automatically obtain the feature information,this paper proposes to use deep denoise auto-encoder to automatically extract the deep features of the original vibration data,and explains the parameter selection process of deep denoise auto-encoder.In order to prove the effectiveness of extracting deep features and the practicability of the fusion model,the commonly used time-frequency domain,autoregressive model and wavelet packet features are used for experimental comparison,which indicates that the features of deep denoise auto-encoder have strong robust performance and can reduce the loss of feature information.The fusion model method was verified using the experimental data of the University of Cincinnati and the experimental data of Xi'an Jiaotong University.It was proved that the proposed method can effectively analyze the performance degradation state of each stage of the bearing,and it has certain engineering practicality.