Remaining Useful Life Of Rolling Bearings Based On The Multivariate Feedback Extreme Learning Machine

Rolling bearing is a key component in mechanical equipment,and its working condition directly affects the overall performance of mechanical equipment.Once the rolling bearing fails,it will lead to chain failures and may even cause huge economic and personnel accidents.Therefore,it is important to monitor the health status of rolling bearings and predict their remaining useful life to ensure the reliable and safe operation of the whole mechanical system.This paper focuses on three key issues of data-driven methods in remaining useful life prediction,namely feature extraction,health state division and remaining useful life prediction.The main research works are as follows:Aiming at the problem that the traditional degradation features are not sensitive to the degradation trend of rolling bearings,this thesis proposes a feature extraction method based on cumulative transformation.The time domain and frequency domain features are extracted from the rolling bearing vibration acceleration signal,then the features are smoothed by locally weighted smoothing scatterplots,and finally the cumulative transformation algorithm is introduced to convert the smoothed features into the corresponding cumulative transformation features.The features processed by cumulative transformation enhance the fault sensitivity and may more precisely depict the rolling bearing deterioration trend.To address the problem that the initial failure point is difficult to determine,this thesis proposes a rolling bearing early degradation detection method based on a continuous trigger mechanism.Firstly,the Pearson coefficient is used to screen out the cumulative transformation features that are sensitive to the degradation trend,and then the first principal component is obtained using principal component analysis.The first-order difference series of the first principal component is used as the health indicator.In order to determine the precise beginning failure occurrence point,the health indicator is finally separated into health states based on the continuous triggering mechanism.In order to improve the accuracy of rolling bearing remaining life prediction,this thesis proposes multivariate feedback type limit learning machine network.The extreme learning machine has the advantages of high learning efficiency and strong generalization,and the feedback mechanism can adjust the network adaptively,based on both of them this thesis proposes the multivariate feedback extreme learning machine.Using multiple cumulative transformation features in the degradation stage as input data,the paper constructs a prediction model for the remaining life of rolling bearings with the feedback extreme learning machine.To assess the suggested prediction model’s efficacy and accuracy,it is used to PRONOSTIA carrying accelerated life test data.