Oct 2026· Advanced Manufacturing Technology and Manufacturing System· Vol 14386, pp. 1438611 - 1438611-6· 0 citations· 16 references
Engineering
TL;DR
A decoupled two‑stage framework that replaces LSTM with four parallel dilated branches aligned to distinct physical wear timescales, and substitutes global average pooling for attention‑based aggregation, and is edge‑deployable for real‑time monitoring without tool‑specific calibration.
Abstract
Tool wear prediction under variable cutting conditions is vital for smart manufacturing, but conventional CNN‑LSTM models generalize poorly to unseen cutters. This failure arises from two limitations: LSTM recurrence confines temporal resolution to a single scale, missing both transient tooth‑passing impacts and gradual flank wear accumulation; and attention mechanisms introduce extra parameters that overfit on small industrial datasets. To address these, a decoupled two‑stage framework is adopted. In the signal stage, Adaptive‑Weight VMD (AW‑VMD) computes Softmax‑based weights from training‑set wear correlations, enhancing degradation‑sensitive frequency components without test‑data leakage. In the modeling stage, a Multi‑Scale Dilated Residual Network (MD‑ResNet1D) replaces LSTM with four parallel dilated branches (d=1,2,4,8) aligned to distinct physical wear timescales, and substitutes global average pooling for attention‑based aggregation. Leave‑One‑Tool‑Out validation on PHM 2010 achieves an average RMSE of 20.50μm across three cutters. Ablation studies show that removing AW‑VMD increases RMSE by 38%, while adding channel or temporal attention modules degrades cross‑tool generalization, confirming that attention‑free convolutional architectures are more robust under small‑sample domain shift. The model is edge‑deployable for real‑time monitoring without tool‑specific calibration.
A tool wear prediction framework that integrates Adaptive Variational Mode Decomposition with a two-stage Convolutional Block Attention Module (CBAM) to demonstrate the potential for multi-sensor tool wear estimation in the evaluated machining scenarios.
Chang-Jiang Wen, Xiao-Yu Li, Hai-Peng Su et al.· Italian National Conference...· 0 citations
Owing to the unique and complex multi-tooth structure of broaching tools and the single-pass feed machining process, direct monitoring methods suffer from inherent limitations. To address this challenge, an RC-Enhanced CNN-ViT model is proposed for tool wear state recognition in turbine disk fir-tree slot broaching. Th...
Shen-Shun Ying, Cheng Lu, Zhen-Zhong Liu et al.· Proceedings of the Instituti...· 0 citations
In mechanical manufacturing, tool wear directly affects machining quality and production efficiency. This study proposes a tool-wear-state identification method that combines continuous wavelet transform (CWT), coordinate-attention-weighted CWT (CA-CWT), a convolutional neural network (CNN), spatial and channel synergi...
Zi-Shuo Wang, Yi Jia, Shu-Ning Liang et al.· Engineering Research Express· 0 citations
A hybrid deep learning model combining 1D CNN, Bidirectional LSTM, and Multi-Head Attention mechanism to classify healthy and fractured tooth conditions using vibration signals obtained from the SpectraQuest simulator demonstrates that the model avoids overlearning and adapts to varying operating conditions in industri...
Canan Taştimur· U Porto Journal of Engineeri...· 0 citations
The presented framework can deliver accurate prediction of milling tool RUL and provide rapid response under limited edge computational resources and is validated on full-lifecycle milling vibration data collected under real machining conditions.
Jian-Liang He, Yin-Hao Zhang, Cai-Hong Han et al.· Journal of Manufacturing and...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.