Research On Prediction Method Of Remaining Using Life Of Rolling Bearing Based On Model Transfer And Wiener Process

Rolling bearings are widely used in modern machinery and equipment,such as wind turbines,automobile transmission systems and aero-engines.In order to ensure the normal operation of the machinery,it is very important to accurately predict the remaining useful life of bearings.Predicting the remaining useful life of rolling bearings has industrial significance for the normal operation of the equipment and reducing equipment maintenance costs.In view of the different bearing operating conditions in the existing rolling bearing remaining useful life prediction,a new remaining useful life prediction method for rolling bearing is proposed based on model transfer and Wiener process.By using the health indicator label,and predicting the performance degradation index of the non-whole-life bearing,then Wiener process is used to construct the model of the degradation state for achieving remaining useful life prediction of rolling bearing under different working conditions.In terms of the construction of the health indicator label,when the life percentage is regarded as the label,it will be difficult to describe the bearing degradation process,so,the time domain,time-frequency domain and trigonometric function feature indexes are extracted from the original bearing vibration data in the source and target domains.Then the extracted feature indexes are input into the single-layer nonnegativity constrained autoencoder to obtain the deep features of the bearing.The strong time relevant features are selected and input into the self-organizing feature mapping network to obtain the bearing degradation trend and construct health indicators.and the frequency domain amplitude sequences obtained by fast Fourier transform are labeled.In terms of remaining useful life prediction,for the lower prediction accuracy problem of rolling bearing remaining useful life under different working conditions,a prediction method is proposed based on transfer learning.Using fast Fourier transform,frequency domain amplitude sequences of the vibration signal of the rolling bearing in the source domain can be obtained.Using the health indexes,the frequency domain amplitude sequences are labeled and regarded as the source domain data.Similarly,the whole-life rolling bearing vibration signals under other working conditions are processed and used as the target domain data.The combined network of deep NCAE network and feedforward neural network is trained by the source domain data to obtain a pre-training model.The target domain data is used to fine-tuning and the performance degradation model of the rolling bearing can be obtained,then the maximum likelihood estimation is combined with the Wiener process to build the life prediction model.Based on PHM 2012 database,the experimental results show that,using the proposed method,the average prediction error is-17.05%,the average score is 0.279.