Explainable AI for tool condition monitoring using Explainable Boosting Machine

Abstract

Machining is one of the most critical sectors in the manufacturing industry, but the quality of the parts produced is highly dependent on the condition of the cutting tools. Poor management of tool replacement can lead to increased production costs and reduced product quality. Various methods exist to monitor tool wear and optimize their replacement. However, most of these methods rely on ”black-box” AI models, which significantly limit their practical use. In this article, a “glass-box” method called Explainable Boosting Machine (EBM) is used to monitor the degradation of cutting tools for the turning operation. This method aims to be as accurate as ”black-box” models while being fully interpretable. A comparison between the EBM method and an Artificial Neural Network (ANN) approach is presented to evaluate the performance differences between these two models. The results indicate that although the EBM model’s performance is slightly lower than the ANN’s, it remains adequate for monitoring tool wear, with an average R² score only 2% lower. The global and local explainability of the model is also presented. The global analysis demonstrates that the model uses coherent features for estimating tool wear, proving that it has successfully understood the wear phenomena being monitored. The local explainability highlights the contribution of each input to the tool wear estimation. These two explainability analyses show results that are consistent with the physical phenomenon of wear, for example, the model identifies that an increase in cutting force implies an increase in tool wear. This provides explainability for its use, improving trust in the monitoring method.