Author(s): Irfan Harun and Jia-Hui Phang

Abstract: Since the early days of numerically controlled machining, tool condition monitoring has been pursued as a route to higher productivity and tighter dimensional tolerances, yet most shop-floor lathes in small and medium enterprises still rely on operator experience to judge when a cutting insert should be replaced ^[1]. This research describes the design, assembly and experimental validation of a low-cost acoustic-based monitoring system that classifies tool wear into five stages fresh, initial, steady, accelerated and severe during the dry and wet turning of AISI 1045 steel on a conventional engine lathe. Two MEMS microphones, mounted 120 mm from the cutting zone, captured airborne acoustic signals at a sampling rate of 44.1 kHz. The raw waveforms were segmented into overlapping windows and transformed using the short-time Fourier transform (STFT) to produce time–frequency spectrograms from which twelve statistical and spectral descriptors were extracted. A random-forest classifier trained on 840 labelled segments achieved an overall accuracy of 91.7% across the five wear stages when using STFT-derived features, compared with 83.1% when the same classifier was fed conventional FFT-based features. Flank wear (VB) progressed from 0 to 0.507 mm over 40 minutes of dry cutting, crossing the ISO 3685 end-of-life threshold (VB = 0.3 mm) near 28 minutes; the monitoring system detected the transition to the accelerated-wear stage within 1.3 minutes of that crossing. The total hardware cost, excluding the lathe and the host computer, was approximately USD 38. The results show that commodity MEMS microphones paired with STFT-based feature extraction can provide industrially useful wear-stage classification at a fraction of the cost of piezoelectric acoustic-emission or dynamometer-based systems.

DOI: https://www.doi.org/10.22271/27078043.2026.v7.i2a.140

Pages: 65-70 | Views: 92 | Downloads: 40

Download Full Article: Click Here