Research On Fault Diagnosis And Life Prediction Of Centrifugal Pump Bearing Based On Data Drive

Rolling bearing is an indispensable part of rotating machinery,known as the "industrial joint",its failure in the actual production not only affects the production of enterprises,and even has safety risks,therefore,the monitoring of bearing development state has practical significance.In order to accurately monitor the running state of bearings,it is necessary to obtain the relevant data characteristics of bearings and then use deep learning network model to realize fault diagnosis and life prediction.However,the deep learning diagnostic model has some shortcomings such as not obvious feature extraction and overfitting,which lead to low diagnostic accuracy and poor classification effect.In addition,the deep learning methods to achieve life prediction also have some problems,such as inobvious sign extraction,poor robustness of degradation model and poor fitting of residual service life degradation curve to real degradation curve.Aiming at the above problems,the research content of this thesis is as follows:(1)In this thesis,a fault test platform was built according to the data required by the centrifugal pump bearing fault,and test equipment required for data collection was selected.The platform mainly obtained bearing fault data under different working conditions,which provided data support for fault diagnosis based on the improved WGAN model and the realization of the system in Chapter 5,so as to verify the feasibility of the test and the validity of the data.(2)This thesis combines R-FCN network instead of discriminator with generator,and proposes an improved WGAN model fault diagnosis method.Firstly,the two-dimensional full convolution layer in R-FCN was used to extract fault features to solve the problem of not obvious feature extraction,and the deconvolution method was used to greatly shorten the time of model training.Secondly,by constructing semi-supervised learning loss function to optimize and improve the WGAN model,the overfitting problem was solved to a certain extent.Finally,the hyperparameters of the optimized network model are obtained by Bayesian optimization algorithm.Different experimental data show that the proposed method is robust and effective.(3)A life prediction method based on improved WGAN-GRU is proposed.Firstly,the multi-stage degradation law of bearing is divided.Secondly,the improved WGAN model was used to extract the characteristic information in the time domain signal,and the evaluation index was used to self-adaptively monitor the different degradation stages of bearings.Then,the improved WGAN-GRU multi-stage degraded health index(HI)curve model was constructed,and different prediction models were established according to the multi-stage.Finally,the characteristic information was input into the GRU model to obtain the RUL predicted value.The validation of XJTU-SY dataset in different methods shows the effectiveness of the proposed method.(4)The framework of the centrifugal pump bearing fault diagnosis and life prediction system was constructed and divided into different levels according to relevant functions.Then the fault diagnosis and life prediction system was built based on the combination of Unity3 D,C#,VS and other methods.Through the connection of Unity3 D front-end and back-end,UI interaction was realized,so as to realize real-time monitoring of the health state of the centrifugal pump bearing.