Scholarlink

<p>The high gravity reactor (HGR) is a critical environmental protection equipment in the petrochemical industry. Currently, only corrective maintenance (CM) is applied to the HGR under consideration. However, this approach lacks foresight, is time-consuming, and cannot ensure smooth production or reduce costs effectively. During HGR operations, various types of data are collected, such as equipment conditions, production information, <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(NO_x\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>N</mi> <msub> <mi>O</mi> <mi>x</mi> </msub> </mrow> </math></EquationSource> </InlineEquation> concentration, and maintenance history, which can assist in making informed maintenance decisions. In this paper, we first design a hidden Markov model (HMM)-based corrective maintenance strategy to enhance the conventional approach. Furthermore, to achieve more cost-effective and timely maintenance, we develop a data-driven condition-based preventive maintenance (CBPM) strategy. A deep reinforcement learning (DRL) model is trained to determine the optimal maintenance dates and durations based on the collected data in an end-to-end manner. Experimental results demonstrate that $6,774 and $28,050 are saved by the HMM-based CM and DRL-based CBPM strategies, accounting for 4.65% and 19.26% of economic cost reductions, respectively. The results indicate that the total costs of CBPM are significantly lower than those of other data-driven strategies.</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 data-driven condition-based preventive maintenance strategy for high gravity reactors in petrochemical industry: from the perspective of reinforcement learning

  • 期刊论文

  • 发表日期：

    2026年6月20日

查看全文
  • Feng Liu,
  • Fuqiang Li,
  • Taixin Li,
  • Xinjie Xing

摘要

The high gravity reactor (HGR) is a critical environmental protection equipment in the petrochemical industry. Currently, only corrective maintenance (CM) is applied to the HGR under consideration. However, this approach lacks foresight, is time-consuming, and cannot ensure smooth production or reduce costs effectively. During HGR operations, various types of data are collected, such as equipment conditions, production information, \(NO_x\) N O x concentration, and maintenance history, which can assist in making informed maintenance decisions. In this paper, we first design a hidden Markov model (HMM)-based corrective maintenance strategy to enhance the conventional approach. Furthermore, to achieve more cost-effective and timely maintenance, we develop a data-driven condition-based preventive maintenance (CBPM) strategy. A deep reinforcement learning (DRL) model is trained to determine the optimal maintenance dates and durations based on the collected data in an end-to-end manner. Experimental results demonstrate that $6,774 and $28,050 are saved by the HMM-based CM and DRL-based CBPM strategies, accounting for 4.65% and 19.26% of economic cost reductions, respectively. The results indicate that the total costs of CBPM are significantly lower than those of other data-driven strategies.