Tool wear evaluation and its evolution in micro-EDM using radio frequency (RF) signals

Tool wear evaluation in micro electrical discharge machining (micro-EDM) involves primarily empirical and data-driven models. The empirical models typically lack accuracy and exhibit poor adaptability under unstable machining conditions. The external monitoring equipment employed in the data-driven approaches consists of different sensors for electrical signals, and it is often invasive, costly, and prone to interference with the process. To mitigate these drawbacks, this research introduces a cost-effective data collection method in which radio frequency (RF) signals are used to estimate the tool wear in micro-EDM. First, the feasibility of employing RF signals to predict tool wear was examined from the perspective of the material removal mechanism. Then, in total 12 signal features pertinent to tool wear length were extracted and selected through multi-criteria correlation analysis. These features are associated with the amplitude and distribution of the RF signal and are intricately linked to discharge states. Finally, a data-driven model with machine learning was developed for tool wear evaluation. It was a linear-regression-based method which demonstrated a superior performance, achieving mean absolute percentage error ( MAPE ) and R 2 values of 1.19% and 0.9989, respectively, with a maximum relative error of only 3.64% within the dataset. The proposed model also exhibited an outstanding performance in comparison with other empirical models. This approach was further applied to analyze tool wear evolution. It was observed that the dynamic characteristics of tool wear length predicted by the model aligned with discharge state variations, thereby further confirming the model’s accuracy and efficacy. In future applications, this method could be integrated into a servo control system to facilitate an online compensation of tool wear and closed-loop control for enhanced dimensional accuracy.