This study introduces a real-time punch wear monitoring system tailored for the sheet metal trimming process using acoustic emission and a data-driven interface. The approach uses burr height as a key indicator of tool degradation and categorises punch wear into three distinct states: freshly ground, partially worn, and fully worn. The process acoustic signals are denoised and converted into Mel-frequency cepstral coefficients (MFCCs). These features are subsequently fed into a feed-forward artificial neural network (ANN) to accurately classify the punch wear condition. The approach is inspired by experienced machine operators who can intuitively discern punch wear from the sound emitted during trimming operations. On a collected dataset the model achieved 99.26% accuracy during training and 97.45% accuracy during testing. The system tracked progressive punch wear, demonstrating robustness to process noise and repeatability across runs. The system enables continuous, non-invasive tool monitoring, reducing manual inspection, unplanned downtime and improving product quality.
Badgujar et al. (Wed,) studied this question.