MEMS Microphone-Driven Near-Sensor Reservoir Computing for Lightweight Tool Wear Classification in Milling

This paper presents a real-time tool wear monitoring method using a near-sensor reservoir computing system with a MEMS microphone. The system processes audio signals generated during the milling of S50C steel with a three-flute cutting tool through a nonlinear Mackey-Glass circuit that serves as the reservoir. The reservoir transforms raw audio signals into high-dimensional representations, and a linear regression classifier identifies the tool wear state. Compared to traditional frequency-domain methods using Fast Fourier Transform (FFT), this approach achieves higher accuracy, reduces computational complexity, and decreases latency. Experimental results validate the effectiveness of the method and practical potential in machining applications.