Bayesian framework for process-aware tool wear predictive modeling of 1.4301 steel

Abstract

Abstract Tool wear is inherent in material removal processes. It results in increased production costs, poor surface finish, and reduced dimensional accuracy of the workpieces, particularly when machining difficult-to-cut materials. Currently, industries rely on manual tool inspection and general tool life estimates. Recent research has focused on experimental data, mathematical modeling, and the integration of signal-based analysis. However, despite the availability of a large volume of research focused on investigating tool wear. Most of them limited the flank wear to the ISO 3685 (International Organization for Standardization), which is less suitable for modern manufacturing systems. The objective of this study is to develop Bayesian framework and model flank wear estimation in turning of 1.4301 austenitic stainless steel. The first model estimates wear progression from the cutting parameters and cutting time. Cutting speed was the dominant driver of flank wear, while the feed effect could not be resolved within the tested range. From this model, a Taylor speed–life relation was derived with credible intervals (CrI), giving a predicted tool life of approximately 57 min at 250 m/min and 15.5 min at 300 m/min, indicating that tool life is highly sensitive to cutting speed. The second model is process-aware wear estimator, the model was derived by integrating the process parameters – cutting forces and surface roughness, and to quantify the uncertainty in the wear estimate and the derived tool-life relation. This model estimates wear from instantaneous measurements of cutting parameters and process indicators; when applied retrospectively, it flagged two of four catastrophic failures as anomalies. Leave-one-insert-out cross-validation yielded R 2 values of 0.690 for the Taylor model and 0.608 for the process-aware wear estimator, indicating meaningful but improvable predictive performance for unseen inserts. The analysis suggests that flank wear alone does not conclusively indicate the condition of tool and real-time process data are also necessary for tool condition monitoring. The model is designed to integrate real-time sensor inputs, indicating potential applicability to real-time manufacturing systems, though deployment validation remains to be conducted.