Dual-Threshold Denoising Convolutional Transformers for Machining Process Monitoring in Cyber-Physical Manufacturing Systems

Abstract

Chatter can severely degrade machining quality, shorten tool life, and reduce productivity, making its effective detection critical for precision and stability in cyber-physical manufacturing. Existing detection methods often struggle with significant noise from dynamic disturbances in real-world environments, limiting their industrial applicability. To address this issue, we propose a dual-threshold denoising convolutional transformer for robust online chatter detection in complex machining processes. Firstly, a dual-threshold denoising module utilizing two threshold functions effectively extracts features while adaptively purifying signals even under noisy conditions. Following noise suppression, a multi-scale convolution module enhanced with a receptive field block captures discriminative features across varied spatial scales and heterogeneous receptive fields. A channel recalibration module then optimizes feature channels through attention mechanisms. Finally, a bidirectional conformer module adeptly captures both forward and backward temporal dependencies. Experimental results demonstrate that our method achieves an average accuracy of 91.60% while maintaining robust performance even under severely noisy conditions (SNR = −6 dB), thereby underscoring its superior efficacy and practical viability in industrial applications.