Scholarlink

<p>Maintenance is critical to ensuring the reliability of industrial systems. In practice, diverse customer usage habits lead to significant variations in system degradation paths, while the heterogeneity of maintenance personnel can also significantly impact maintenance outcomes. In this paper, we introduce a Bayesian-reinforcement learning framework to optimize maintenance policies by accounting for both customer and maintenance heterogeneity. The framework leverages sequential Bayesian inference for real time parameter updates and reinforcement learning to dynamically adjust maintenance actions, balancing exploration and exploitation. The purpose of balancing exploration is to collect more data to enhance parameter estimation accuracy. Numerical examples illustrate that increased data collection enhances parameter accuracy, and maintenance costs also tend to converge. Sensitivity analysis reveals that accurate prior information is crucial for policy convergence speed. Moreover, the optimal maintenance policy is more sensitive to maintenance costs than repair effects, emphasizing the importance of cost considerations in maintenance decision making.</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

A Bayesian-reinforcement learning approach for maintenance optimization under usage and maintenance heterogeneity

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Siqi Qin,
  • Xiujie Zhao,
  • Le Gao,
  • Min Zhang

摘要

Maintenance is critical to ensuring the reliability of industrial systems. In practice, diverse customer usage habits lead to significant variations in system degradation paths, while the heterogeneity of maintenance personnel can also significantly impact maintenance outcomes. In this paper, we introduce a Bayesian-reinforcement learning framework to optimize maintenance policies by accounting for both customer and maintenance heterogeneity. The framework leverages sequential Bayesian inference for real time parameter updates and reinforcement learning to dynamically adjust maintenance actions, balancing exploration and exploitation. The purpose of balancing exploration is to collect more data to enhance parameter estimation accuracy. Numerical examples illustrate that increased data collection enhances parameter accuracy, and maintenance costs also tend to converge. Sensitivity analysis reveals that accurate prior information is crucial for policy convergence speed. Moreover, the optimal maintenance policy is more sensitive to maintenance costs than repair effects, emphasizing the importance of cost considerations in maintenance decision making.