Tool Wear Prediction in Milling: A comparative analysis based on machine learning and deep learning approaches

Abstract

The milling machine's cutting tool is a vital asset; its breakdown results in unplanned downtime, which reduces industrial efficiency. Tool-Wear Monitoring (TWM) is one of the primary goals of the manufacturing industry due to the manifold benefits it provides, such as optimizing production efficiency, improving performance, and increasing the life of the tool. Most of the work carried out in this domain involves statistical-based techniques, which require expert domain knowledge in formulating degradation models of the tools. Data-driven machine learning and deep learning models have recently been used to analyze tool wear data and make efficient predictions about its remaining useful life. This paper presents a comparative approach to tool wear monitoring using the clustering machine learning technique of K-Nearest Neighbour (k-NN) and deep learning technique of Convolutional Neural Network (CNN) and hybrid Autoencoder-LSTM (AE-LSTM) models. The CNN and AE-LSTM techniques out-perform k-NN by achieving a higher degree of separability of around 93% and 87%, respectively, as per the ROC-AUC values. The techniques provide improved outcomes in terms of precision, recall, and f1-score, indicating that the models are more accurate at detecting false positives