Cutting tool wear prediction using terminal vibration signals and a frequency-domain adaptive residual network

Abstract

To predict the cutting tool wear, this paper proposes a frequency-domain adaptive residual network (FDAR-Net) based solely on terminal vibration signals of the machining system. To suppress field noise and high-frequency interference, FDAR-Net integrates a frequency-domain adaptive mechanism with an asymmetric residual structure for robust degradation feature extraction. Evaluated on the public PHM 2010 milling dataset without multi-source fusion, FDAR-Net outperforms benchmark models including 1D-CNN, BiGRU, BiLSTM, and Mamba in both feature extraction and monitoring accuracy. Experimental results demonstrate that FDAR-Net achieves high-precision, low-cost, and easily deployable tool monitoring with minimal hardware requirements.