Cutting tool prognostics enabled by hybrid CNN-LSTM with transfer learning

Abstract

An effective strategy to predict the remaining useful life (RUL) of a cutting tool could maximise tool utilisation, optimise machining cost, and improve machining quality. In this paper, a novel approach, which is enabled by a hybrid CNN-LSTM (convolutional neural network-long short-term memory network) model with an embedded transfer learning mechanism, is designed for predicting the RUL of a cutting tool. The innovative characteristics of the approach are that the volume of datasets required for training the deep learning model for a cutting tool is alleviated by introducing the transfer learning mechanism, and the hybrid CNN-LSTM model is designed to improve the accuracy of the prediction. In specific, this approach, which takes multimodal data of a cutting tool as input, leverages a pre-trained ResNet-18 CNN model to extract features from visual inspection images of the cutting tool, the maximum mean discrepancy (MMD)-based transfer learning to adapt the trained model to the cutting tool, and a LSTM model to conduct the RUL prediction based on the image features aggregated with machining process parameters (MPPs). The performance of the approach is evaluated in terms of the root mean square error (RMS) and the mean absolute error (MAE). The results indicate the suitability of the approach for accurate wear and RUL prediction of cutting tools, enabling adaptive prognostics and health management (PHM) on cutting tools.