Tool wear estimation in drilling of Inconel 718 via Bayesian state-space modeling

Tool condition monitoring is a difficult task due to the inability to directly observe the tool during operation, so preliminary experimental tests are essential for modeling tool degradation and sensor response. However, sometimes experiments are expensive, and limited to a few samples and cutting conditions. In this work, we study tool wear during drilling of Inconel 718 plates under different cutting speeds using the spindle motor current as the sensing signal, thus avoiding the need for installing additional instrumentation. The degradation dynamics are modeled using a nonlinear state-space model, where tool wear is treated as a latent, monotonically increasing state. For the tool-to-tool variability — which is usually neglected — we propose two hierarchical models. The statistical analysis was conducted using Bayesian inference to quantify model uncertainty, to handle missing direct wear measurements, and to provide a framework for modeling from limited experimental data. The performance of the models for filtering was assessed using leave-one-out cross-validation, and a further test on unseen cutting conditions. The HDI(95 %) of the tool wear posteriors span roughly ± 10 µm around the median, which can be improved to ± 5 µm if additional information about the tool is available.