Research On Remaining Life Prediction Of Rolling Bearing Based On Vibration Signal And Deep Learning

Rolling bearings are very common parts in industry,Due to the special working environment,overload and other factors,rolling bearings are more prone to failures than other parts.Rolling bearings usually reflect their operation condition indirectly through vibration signals,indicate fault information,and predict the operation condition in a certain period in the future through vibration signals to achieve preventive maintenance of mechanical equipment and prevent major material losses and accidents.Therefore,the online monitoring and prediction of the remaining life of rolling bearings is of great importance to prospects for the future of the industry.In this article,Rolling bearings are taken as the research object,and a research of remaining life prediction based on vibration signal and deep learning is proposed by taking the advantage of deep learning network.The main research contents are as follows.(1)To address the problem that the noise reduction effect is not satisfactory due to reduce the inappropriate adjustment of VMD parameters when VMD reduces the noise of rail vehicle vibration signals,an improved sparrow search algorithm is proposed in order to determine the decomposition parameters of VMD through the envelope entropy search.Firstly,the previous sparrow search algorithm is enhanced by the following the improved follower position update and introducing the CorsiGaussian variation strategy,and the improved Sparrow search algorithm is combined with VMD to obtain the optimal search parameters;then,the optimal search parameters are input into VMD to decompose the multiple modal components;secondly,the modal components containing more fault information are filtered by CK indicators and Finally,this method is compared with the conventional VMD or EMD method using simulated signals and experimental data.The experimental outcomes demonstrate that the method submitted in this paper is significantly more effective in terms of noise reduction.(2)To address the problem of insufficient degradation features in one-dimensional time series data,a convolutional neural network is proposed for obtaining the degradation information in time-frequency maps.First,a one-dimensional vibration signal is transformed into a two-dimensional time-frequency map using a wavelet transform;then,a health factor prediction model based on a convolutional neural network,and the time-frequency map is input into the constructed model to construct a health factor curve,and the rationality of the constructed health factor curve is verified by the evaluation index of the health factor curve;finally,based on the experimental data,the method of this paper is compared with RMS,PCA and BP Finally,based on the experimental data,the method is compared with RMS,PCA and BP.The experimental results show that the proposed method is more effective in extracting health factors.(3)A method for predicting the life of rolling bearings based on LSTM neural networks is proposed to address the problems of long-term dependence of neural networks on input data and gradient disappearance.First,a model that is the basis of an LSTM neural network is established and the model is trained;secondly,the test set was fed into the network to predict the health factor over time and the feasibility of the prediction component was verified by comparison with the true health factor;finally,the method of this paper is compared with RNN-HI and SOM-HI through the experimental analysis of remaining life prediction.The results of the experiments demonstrate that the approach proposed in this paper is more advantageous in remaining life prediction.