Scholarlink

<p>Faults, uncertainties, and noise represent significant challenges to the stability, safety, and reliability of complex systems data-driven control systems. Faults unintentionally occur in different components of the system, resulting in a decline in both system performance and reliability. The development of artificial intelligence (AI)-based fault diagnosis system (FDS) for enhancing safety and reliability in data-driven control environments has recently shown promising results; however, their limited interaction with dynamic environments and insufficient decision-making capabilities under varying conditions pose significant challenges to their widespread adoption. To address this challenge, the present study employs reinforcement learning (RL), which is grounded in perception and continuous interaction with the environment, offering a promising solution to the problem. In this study, adaptive sliding mode controller (ASMC) is employed to contrast model uncertainties, and adaptive sliding mode observer (ASMO) is employed to reduce system noise for fault detection. Furthermore, the RL algorithm under fractional-order difference (FOD) observations is integrated to enhance the accuracy of fault estimation. The simulation results on the discrete data-driven system demonstrate improvements of 13% under normal conditions and 17% in the presence of system noise achieved by the proposed approach, compared to integer-order difference (IOD)-RL.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust reinforcement learning fault diagnosis for complex data-driven nonlinear systems

  • 期刊论文

  • 发表日期：

    2026年7月14日

查看全文
  • Mohsen Shiri,
  • Hadi Delavari

摘要

Faults, uncertainties, and noise represent significant challenges to the stability, safety, and reliability of complex systems data-driven control systems. Faults unintentionally occur in different components of the system, resulting in a decline in both system performance and reliability. The development of artificial intelligence (AI)-based fault diagnosis system (FDS) for enhancing safety and reliability in data-driven control environments has recently shown promising results; however, their limited interaction with dynamic environments and insufficient decision-making capabilities under varying conditions pose significant challenges to their widespread adoption. To address this challenge, the present study employs reinforcement learning (RL), which is grounded in perception and continuous interaction with the environment, offering a promising solution to the problem. In this study, adaptive sliding mode controller (ASMC) is employed to contrast model uncertainties, and adaptive sliding mode observer (ASMO) is employed to reduce system noise for fault detection. Furthermore, the RL algorithm under fractional-order difference (FOD) observations is integrated to enhance the accuracy of fault estimation. The simulation results on the discrete data-driven system demonstrate improvements of 13% under normal conditions and 17% in the presence of system noise achieved by the proposed approach, compared to integer-order difference (IOD)-RL.