Research On Remain Life Prediction Of Rolling Bearings Based On DS-CNN-AM And Transfer Learning

Rolling bearing is one of the most critical parts in rotating machinery,and their operating conditions are closely related to the performance of the equipment.The prediction of remaining useful life(RUL)of bearings plays a vital role in ensuring the safe operation of machinery and reducing maintenance losses.With the rapid development of information technology,many artificial intelligence-based bearing RUL prediction methods have emerged,but most of them have two shortcomings.On the one hand,there is a lack of attention to the time sequence information and characteristic frequency information in the bearing vibration data;on the other hand,the training and testing data is the bearing data under the same working condition,with the same data distribution,which can not well predict the RUL of the bearing under other working conditions.In this paper,a rolling bearing RUL prediction method based on deep learning is proposed to solve these problems.The main contents of this paper are as follows:(1)By analyzing the failure form and vibration mechanism of rolling bearing,the evolution law of bearing vibration signal with degradation state is revealed,which provides a theoretical basis for bearing vibration signal as degradation data.The signal marginal spectrum was obtained by Hilbert Huang transform of the raw vibration signal of the bearing,which was preprocessed and used as the training data of the RUL prediction model.(2)Aiming at the problem of weak feature extraction capabilities of traditional deep learning prediction models,a prediction model that combines depth wise separable convolution(DS-CNN)and attention mechanism(AM)is proposed.The model has multi-channel input layer,which is used to input the bearing data of several continuous times.Compared with the traditional convolution operation,the DS-CNN reduces the number of model parameters and the cost of operation.Combined with the residual network structure,the depth of the model is deepened and the capability of feature extraction is effectively improved.The AM is composed of channel attention and spatial attention,which can strengthen important information related to the degraded state and suppress useless information.The channel attention mechanism is used to extract the timing information between bearing sequence data in the time direction,and the spatial attention mechanism is used to extract the frequency information related to the degradation state in the frequency direction,which improves the performance of the prediction model.(3)Aiming at the low accuracy of bearing RUL prediction under different working conditions,a RUL prediction method based on transfer learning is proposed.The labeled bearing data were taken as the source domain data,and some unlabeled bearing data of other working conditions were taken as the target domain data.The method of combining global and local domain adaptation is used to optimize the model parameters in an adversarial manner.This method reduces the difference of bearing data distribution between source domain and target domain,and can effectively predict the RUL of bearings under different working conditions.Compared with the prediction results of the original model,the proposed method obtains higher prediction accuracy,which makes the model have better generalization and robustness.