Bearing residual service life prediction method based on deep mutual learning and dynamic feature construction

The invention belongs to the field of bearing life prediction, and discloses a bearing residual service life prediction method based on deep mutual learning and dynamic feature construction, and the method comprises the steps: selecting a more stable bearing feature RRMS, employing a DML improved convolutional neural network to automatically extract features in a first stage, and indicating the health condition of a bearing; when the bearing degenerates to 50% (output is smaller than 0.5), the last 50% is predicted through a long-short-term memory network, finally, results of the two stages are combined to obtain a service life degeneration curve of the bearing, model output obtained through the method can be directly used for RUL calculation, selection of a bearing failure threshold value is avoided, and the service life degeneration curve of the bearing is obtained. In the whole prediction process, the CNN and the LSTM are respectively used for different stages of bearing degradation, and the existing full-