Machine Learning-based Tool Wear Prediction using Bolt type Piezo Sensor

In this study, we propose a method for predicting the flank wear of cutting tools using bolt-type piezo sensors (Piezobolts) and experimentally validate its effectiveness through simultaneous measurements with a three-axis dynamometer, which serves as the laboratory standard for accurate cutting-force acquisition.During machining, the reaction forces transmitted from the workpiece through the clamping interface are measured by the Piezobolts and converted into approximate three-axis cutting forces, which are then used to estimate tool flank wear.The analysis results demonstrate that the wear progression trends derived from the Piezobolts closely follow those obtained from the dynamometer throughout the entire wear process.A support vector regression (SVR) model that preserves the temporal order of the measured data successfully predicts the later-stage tool wear using only early-stage training data.Notably, when trained solely on features extracted from the Piezobolt signals, the model captures both the gradual wear evolution and the accelerated wear behavior near the end of tool life.These results demonstrate the feasibility of using bolt-type piezo sensors as surrogate sensors for tool-wear monitoring without direct cutting-force measurement or structural modification of the machine tool.Because the present validation was limited to one tool, one workpiece material, one machining operation, and one cutting condition, the findings should be interpreted as an experimental feasibility study.Nevertheless, the proposed framework provides a basis for extending Piezobolt-based tool condition monitoring to broader machining conditions in future work.