.png){kind=icon}
.png){kind=icon}
Return
Cross-condition tool wear prediction via feature decoupling and domain-adaptive representation learning
.png){kind=icon}
PRE
AI
J
X
J
李
J
X
Y
DOI：10.1016/j.ymssp.2026.114783.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Tool wear prediction across varying operating conditions is challenging because wear-related degradation information is usually entangled with condition-dependent variations, which degrades prediction accuracy and cross-domain generalization. To address this issue, this paper proposes a cross-condition tool wear prediction framework based on feature decoupling and domain-adaptive representation learning. A multimodal adversarial autoencoder is first developed to learn a structured latent space, in which wear-related and condition-related information are explicitly separated. This design suppresses condition interference and yields more transferable degradation representations. A domain-adaptive prediction model with domain-aware contrastive learning is then constructed to align wear-relevant features across domains while preserving discriminative structure, thereby mitigating negative transfer caused by feature entanglement and distribution shift. Experiments on the Jilin University turning dataset and the NASA milling dataset demonstrate that the proposed method consistently outperforms representative methods under cross-condition settings and achieves superior prediction accuracy, robustness, and generalization. The results indicate that the proposed framework provides an effective solution for tool wear prediction in complex manufacturing environments.
Keywords:
Tool wear prediction
Feature decoupling
Domain-adaptive
Cross-condition
Journal
IF:
8.9
Papers:
1.2W
Citations:
6.6W
Organization
Cited Papers