Development of an In-Process Cutting Tool Life Prediction System Using Bidirectional Long Short-Term Memory Network

This investigation presents an in-process cutting tool life prediction approach, which is used an in situ CMOS camera for continuous monitoring of cutting tool flank wear progression and hybrid bidirectional long short-term memory (BiLSTM) model for prediction of an in-process cutting tool life. During the machining process, an in situ CMOS camera acquired machined surface texture as a response of cutting tool flank wear. Subsequently, the machined surface textures features extracted using Gabor wavelet transform (GWT), and without losing the originality of the data, the most significant features were selected using principal component analysis (PCA). Finally, the significant features are fed to the hybrid BiLSTM learning model for in-process tool life predictions. While developing the BiLSTM predictive model, the hyperparameters such as number of layers and hidden units are varied and constituted a total of 50 different networks for identification of reliable and accurate predictive model. The root mean square error (RMSE) calculated for a total of 50 networks. The most accurate BiLSTM model selected by the selection of a least RMSE value of 0.00319, which is obtained from a single-layer and 100 hidden units BiLSTM network. The proposed GWT + PCA + BiLSTM is compared with the GLCM + PCA + BiLSTM and experimental results to validate the proposed model performance. The research results show that the proposed model has an excellent capability for an in-process cutting tool life predictions.