Cutter remaining service life prediction method based on spatio-temporal feature extraction

The invention relates to a method for predicting the remaining service life of a cutter based on spatio-temporal feature extraction, which comprises an end-to-end architecture comprising a signal segmentation module, an information extraction module and an information aggregation module, and comprises the following steps of: modeling a cutter service life degradation process by utilizing multiple paths of monitoring signals collected from a cutter cutting process; constructing an augmented feature matrix containing spatio-temporal related information, performing self-extraction of time domain and space domain depth features by using a bidirectional deep convolution architecture, and realizing relation mapping from spatio-temporal features to residual life by using a multi-layer perceptron; dynamically updating a network parameter value around a smooth mean absolute error between a real life value and a predicted life value by using a back propagation algorithm to obtain an optimal solution enabling the error