23 / 2025-05-06 19:02:35
Automatic Discovery of System Degradation Laws Using Reinforced Symbolic Learning
Symbolic Regression,Reinforcement Learning,Physics-guided Optimization,Degradation Law Discovery,Interpretable Machine Learning
全文待审
Pei Li / Zhejiang University
Yiming Zhang / Zhejiang University
Jinfa Zhang / Zhejiang University
Shuyou Zhang / Zhejiang University
Accurate and interpretable degradation formulas are critical for reliable life prediction of engineering systems. This paper presents the Reinforced Symbolic Learning framework that automatically constructs explicit system degradation equations from experimental data. By integrating symbolic regression with reinforcement learning and embedding dimensional and logical constraints, our method yields concise, physically consistent expressions with minimized manual tuning. We validate the approach on GH4169 nickel‐based superalloy under two temperature conditions (25 °C and 650 °C), where the discovered formulas achieve over 90 % of predictions within a twofold error margin. Comparative evaluations against 6 classical empirical models and 5 standard machine‐learning methods demonstrate that our framework delivers superior predictive accuracy while preserving full formula interpretability. The resulting degradation laws not only reveal underlying fatigue mechanisms but also facilitate real‐time deployment in monitoring systems, offering a generalizable path for automated, data‐driven extraction of system degradation laws.
重要日期
  • 会议日期

    08月01日

    2025

    08月04日

    2025

  • 06月15日 2025

    初稿截稿日期

主办单位
中国机械工程学会设备智能运维分会
承办单位
新疆大学
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询