Research On Prediction Method And Uncertainty Measurement Of Remaining Useful Life Of Rolling Bearing

As the key universal component with the highest failure rate of rotating machinery,rolling bearings are related to the safety and reliability of the entire machinery and equipment.With the development of modern industry and the improvement of intelligence and information,the traditional online diagnosis and maintenance strategy is gradually replaced by the predictive maintenance strategy.The first way to realize the predictive maintenance strategy of rolling bearings is to study the remaining useful life(RUL),which can predict the running state of the bearing in advance and discover potential safety hazards in time,which is of great significance to safe production.Aiming at the key application technology problems of the existing rolling bearing RUL prediction research,such as poor real-time performance,low prediction accuracy,and insufficient quantitative expression of prediction uncertainty.Based on one-dimensional vibration signal and data-driven technology,a bearing RUL prediction model is proposed.Finally,experiments are carried out using the public bearing data set,which proves the effectiveness and superiority of the method in this paper.The main research content of this article is summarized into the following four parts:(1)For the problem of long training time and low data utilization rate of the rolling bearing RUL prediction model based on machine learning,it cannot be updated in real time based on the obtained vibration data.A RUL online prediction model for rolling bearings based on regular extreme learning machine(RELM)is proposed,and different prediction methods are adopted for different degradation stages.Experimental results show that this method has high accuracy and speed in both time series forecasting and RUL forecasting.(2)Due to the rolling bearing degradation failure experiment under the same working conditions,the bearing degradation trend will also have certain differences,resulting in different probability distributions of the vibration signal data monitored by the sensors,thus making the prediction accuracy of the bearing RUL prediction model based on deep learning reduce.In response to the above problems,a spatial pyramid pooling convolutional migration network(SPP-CNNTL)model is proposed to realize the RUL prediction of different failed bearings.Finally,it is verified by experiments that this method has higher prediction accuracy than other existing methods in the RUL prediction of rolling bearings.(3)Due to various factors,the bearing degradation data collected by the sensor will inevitably contain noise,and the prediction error of the deep learning model predicted by the bearing RUL will bring certain uncertainty to the prediction result of the bearing RUL.Aiming at the above problems,a characterization model of rolling bearing RUL prediction uncertainty based on approximate Bayesian estimation convolutional neural network is proposed.The experimental results show that the method can describe the uncertainty of the remaining service life prediction value of the bearing while maintaining high prediction accuracy.(4)Based on the QT designer tool,the several rolling bearing RUL prediction methods proposed in this paper are systematically integrated,and an intelligent rolling bearing RUL prediction system based on the Windows platform is developed.