A novel inherent information constrained unsupervised domain adaptation with feature conversion for tool wear monitoring under variable cutting conditions

TY - JOUR

T1 - A novel inherent information constrained unsupervised domain adaptation with feature conversion for tool wear monitoring under variable cutting conditions

AU - Cheng, Minghui

AU - Yan, Pei

AU - Liang, Guoxing

AU - Wang, Yong

AU - Zhao, Jian

AU - Huang, Yonggui

AU - Wang, Tao

AU - Wang, Xibin

N1 - Publisher Copyright: © 2025 Elsevier Ltd

PY - 2025/12/1

Y1 - 2025/12/1

N2 - Cutting tool as the end execution element of intelligent machine tools, its random, non-linear and time-varying wear degradation state largely determines the quality, efficiency, and energy consumption of the parts. The complexity and variability of the working conditions make the tool wear more difficult to perceive and judge. A novel framework was proposed to solve the issue of tool wear monitoring under variable cutting conditions by merging a feature conversion method, a feature extractor, and a new transfer learning paradigm. Firstly, the feature conversion method was performed on the raw feature vectors. The features unaffected by variable process parameters were retained, while magnitudes of the features were converted to ensure that the information in the feature vectors is only as closely related to the tool wear degradation as possible. Then, a new transfer learning paradigm was built by combining an unsupervised domain adaptation technique and the contrastive learning. The transferable representations were learned independently of the labelled target domain data, while preserving the inherent information within the target domain. Finally, the efficacy of the proposed framework was validated on two types of tool wear datasets characterized by diverse cutting parameters and machine tools. According to the experimental results, the generalization capability of the proposed model outperforms the non-transfer learning model and other transfer learning methods in most transfer tasks.

AB - Cutting tool as the end execution element of intelligent machine tools, its random, non-linear and time-varying wear degradation state largely determines the quality, efficiency, and energy consumption of the parts. The complexity and variability of the working conditions make the tool wear more difficult to perceive and judge. A novel framework was proposed to solve the issue of tool wear monitoring under variable cutting conditions by merging a feature conversion method, a feature extractor, and a new transfer learning paradigm. Firstly, the feature conversion method was performed on the raw feature vectors. The features unaffected by variable process parameters were retained, while magnitudes of the features were converted to ensure that the information in the feature vectors is only as closely related to the tool wear degradation as possible. Then, a new transfer learning paradigm was built by combining an unsupervised domain adaptation technique and the contrastive learning. The transferable representations were learned independently of the labelled target domain data, while preserving the inherent information within the target domain. Finally, the efficacy of the proposed framework was validated on two types of tool wear datasets characterized by diverse cutting parameters and machine tools. According to the experimental results, the generalization capability of the proposed model outperforms the non-transfer learning model and other transfer learning methods in most transfer tasks.

KW - Contrastive learning module

KW - Feature conversion method

KW - Tool wear monitoring

KW - Unsupervised domain adaptation strategy

KW - Variable cutting conditions

UR - https://www.scopus.com/pages/publications/105009633055

U2 - 10.1016/j.measurement.2025.118323

DO - 10.1016/j.measurement.2025.118323

M3 - Article

AN - SCOPUS:105009633055

SN - 0263-2241

VL - 256

JO - Measurement: Journal of the International Measurement Confederation

JF - Measurement: Journal of the International Measurement Confederation

M1 - 118323

ER -