A Lightweight Network Based on Adaptive Knowledge Distillation for Remaining Useful Life Prediction Under Cross-Working Conditions

• Published in: 2023 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD) pp. 1 - 6
• Main Authors: Chen, Jiaxian, Li, Dongpeng, Huang, Ruyi, Chen, Zhuyun, Li, Weihua
• Format: Conference Proceeding
• Language: English
• Published: IEEE 02.11.2023
• Subjects: Adaptation models; Adaptive systems; aero-engine; Employee welfare; Fault detection; knowledge distillation; Knowledge engineering; Neural networks; prognostics and health management; RUL prediction; Transfer learning
• DOI: 10.1109/ICSMD60522.2023.10490721

The application of deep learning (DL) methods has undergone comprehensive investigation and proven efficacy in the field of remaining useful life (RUL) prediction. However, the current DL-based RUL prediction methods have two limitations: 1) Many DL networks improve RUL prediction results by increasing the complexity of the model, which makes it difficult to deploy in practical industrial engineering. 2) DL methods exhibit excellent prediction performance when large amounts of run-to-failure data are available, which is also not satisfied in cross-working conditions. To solve the above problems, a lightweight network based on adaptive knowledge distillation is proposed to execute the RUL prediction under cross-working conditions. First, a teacher network based on a three-layer neural network is constructed where the dropout technique is adopted to prevent overfitting. Second, a student network is built with a more lightweight network. Third, the maximum mean discrepancy algorithm is employed to achieve domain adaptation. Finally, the N-CMAPSS 2021 Challenge dataset was employed for experimental validation, aiming to assess the impact of the proposed approach. Comparative findings demonstrate that the proposed method is superior to other RUL methods in industrial engineering.