导航

ACTA AERONAUTICAET ASTRONAUTICA SINICA ›› 2021, Vol. 42 ›› Issue (10): 524879-524879.doi: 10.7527/S1000-6893.2020.24879

• Article • Previous Articles     Next Articles

Method for accurate prediction of tool wear under varying cutting conditions based on domain adversarial gating neural network

WAN Peng, LI Yingguang, LIU Changqing, HUA Jiaqi   

  1. College of Mechanical and Electrical Engineering, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, China
  • Received:2020-10-15 Revised:2020-11-03 Published:2020-12-08
  • Supported by:
    National Key Research and Development Program of China (2018YFB1703203); National Science Foundation of China for Distinguished Young Scholars (51925505); National Natural Science Foundation of China (51775278)

Abstract: The accurate prediction of tool wear plays an important role in ensuring the quality of parts, improving production efficiency and reducing manufacturing costs. In the actual machining process, cutting parameters, tool geometric parameters, tool materials and other cutting conditions are complex and variable. The coupling effect of cutting condition and tool wear on monitoring signals brings great challenges to accurate prediction of tool wear. To address the above issue, this paper proposes a method for accurate prediction of tool wear under variable cutting conditions based on Domain Adversarial Gating Neural Network (DAGNN). A cutting condition classification network is trained using unlabeled samples, and key signal features, which can represent tool wear and are not sensitive to cutting conditions, are adaptively extracted from the original monitoring signals of different cutting conditions through domain adversarial and gated filtering mechanisms. The signal feature extraction network and tool wear prediction network are iteratively optimized to achieve accurate prediction of tool wear under variable cutting conditions. Experimental results show that compared with existing methods, the method proposed can achieve higher prediction accuracy even when only a few wear labeled samples of target cutting condition are available.

Key words: tool wear prediction, varying cutting condition, feature extraction, domain adversarial gating neural network, few-shot learning

CLC Number: