Tool wear prediction method based on bidirectional long short-term memory neural network of single crystal silicon micro-grinding

Although many efforts have been made on off-line tool wear prediction, the on-line intelligent prediction of tool wear prediction based on indeterminate factor relationship has not been addressed. In this paper, a tool wear prediction model is built based on bidirectional long short-term memory neural network (BiLSTM) to deal with these challenges. The acoustic emission (AE) signal and tool wear image are selected as indicators to characterize the wear behavior of micro-grinding tool. The BiLSTM model is constructed with the input of 4-dimensional feature vector, which is composed of medium frequency energy ratio of the AE signal, initial tool cross-sectional area, micro-grinding depth and micro-grinding length, and the output of the loss of tool cross-sectional area. Two derived models of genetic algorithm-optimized BiLSTM (GA-BiLSTM) and long short-term memory neural network (LSTM) are developed to compare the accuracy of the BiLSTM model. Two machine learning models of back propagation neural network (BP) and algorithm optimized BP neural network (GA-BP) are developed to compare the stability and superiority of the BiLSTM model. The micro-grinding experiment is conducted by the electroplated diamond micro-grinding tool to verify the feasibility using the proposed methods and the results show that the average prediction accuracy of the BiLSTM model reached 92.08% while the accuracies of other models from GA-BiLSTM to GA-BP are separately 87.2%, 88.6%, 84.4%, and 85.8%. The BiLSTM model provides a novel wear characterization and prediction method that combines AE signals and visual images using small-sample and multi-sourced heterogeneous data. It undoubtedly promotes sustainable manufacturing and provide theoretical basis for independent decision-making in precision intelligent manufacturing.