Research On Remaining Useful Life Predicition Methods Of Bearing Based On Deep Learning

Rolling bearings are a key component in rotating machines.In many industries such as transportation,aerospace,high-end precision machine tools and instrumentation,they play an important function in supporting rotating shafts and components.The reliability of bearings plays an important role in the safe operation of rotating equipment or even the whole equipment.In this thesis,the theory of deep learning is utilised to conduct relevant research on its feature extraction and life prediction,and a bearing life prediction method based on attention mechanism as well as a bearing life prediction method based on contrast learning and regularised gated cycle units is proposed.The research includes.(1)Aiming at the problems that existing methods for feature extraction of bearing vibration signals rely excessively on experts’ experience and memory degradation due to too long sequences in life prediction,a method for predicting the remaining service life of bearings based on the attention mechanism is proposed.First,the initial vibration signal of the rolling bearing is converted into a frequency domain amplitude signal using the fast Fourier transform method;then,a model based on the attention mechanism is designed,in which a convolutional neural network-attention mechanism network is used for degradation feature extraction and an encoder-decoder network based on the attention mechanism is used for bearing life prediction,which further solves the problem of recurrent neural network in long-distance signal Finally,the PHM2012 bearing degradation dataset is used and validated by the bearing accelerated degradation PRONOSTIA experimental platform.The results of the study validate the accuracy and effectiveness of the attention-based mechanism model for bearing life prediction.(2)To address the problem that complex deep learning models lead to difficulties in real-time remaining life prediction of bearings,an end-to-end rolling bearing remaining life prediction method using contrast learning and regularised gated cyclic units is proposed.Firstly,a contrast learning neural network is used to obtain feature information from the vibration signals of rolling bearings;then,by combining the bearing degradation law,the problem of more health state information in the feature extraction of bearing sequence data without effective feature extraction is solved by continuous sampling of recent data,sparse sampling of long-term data,and location coding to retain temporal information;finally,the gated recurrent unit network is regularized to improve the real-time performance of model prediction while reducing the complexity of the model.The PHM 2012 bearing degradation dataset was used to validate the model using the PRONOSTIA experimental platform for accelerated bearing degradation.The results validate the accuracy and real time performance of the model for bearing life prediction.