Research On Remaining Useful Life Prediction Method Of Rolling Bearing Based On Convolution Neural Network

As an important part of mechanical equipment,rolling bearing failure will have a significant impact on the safety and stability of the equipment,and the bearing failure problem has become an important factor limiting the life and reliability of mechanical equipment.As one of the key technologies of prognostics and health management,accurately predicting the remaining useful life of rolling bearings is of great importance for the normal operation and maintenance of mechanical equipment.In recent years,deep learning has been widely used in the task of remaining useful life prediction of rolling bearings because of its powerful characterization learning ability.In this paper,the research method of remaining life prediction of mechanical rolling bearings will be studied based on convolutional neural network in deep learning algorithm,and the specific work is as follows.To solve the problems of incomplete feature extraction and reduced prediction accuracy caused by complex manual feature extraction and screening process in traditional prediction methods,in this paper,a multi-scale convolutional rolling bearing prediction model with dynamic connection layer is proposed.Firstly,the time-frequency map of the original vibration signal is used as the input of the network model to reduce the manual feature extraction process,and secondly,the multi-scale convolutional module is used instead of the common convolutional module to realize the comprehensive extraction of rolling bearing degradation features.Finally,the global average pooling and dynamic connection layer with dynamic Gaussian deactivation are used instead of the fully connected layer to enhance the model generalization capability,and the effectiveness and superiority of the proposed model are experimented.For the problem that the actual operating conditions of rolling bearings are complex and harsh,which is very likely to cause the weak bearing degradation signals to be submerged in strong noise,resulting in the degradation of model prediction accuracy or even prediction failure,this paper proposes a temporal feature processing module with attention fusion mechanism,which achieves the fusion of global and local temporal features by constructing two temporal feature processing sub-modules with different forms of attention mechanism,and enhance the bearing key degradation information,suppress the useless noise,and improve the robustness of the model.After that,the module is embedded in a convolutional neural network to build a convolutional neural network prediction model with temporal attention fusion mechanism.The model learns bearing degradation characteristics directly from the original vibration signal and performs end-to-end rolling bearing remaining useful life prediction,further reducing the complexity caused by data preprocessing.