Scholarlink

<p>Root cause analysis (RCA) of abnormal events in smart manufacturing remains slow and inconsistent when diagnostic knowledge is fragmented across equipment logs and expert experience. This study introduces a rule-grounded diagnostic knowledge system for traceable decision support, which converts equipment events and log data into auditable diagnostic rules and enables evidence-backed retrieval through natural-language queries. The system integrates association-rule mining to construct a diagnostic rule base with rule identifiers and supporting evidence, semantic retrieval with controlled alignment to map user queries to candidate rules, and an abstention mechanism that withholds outputs when evidence is insufficient, thereby reducing unsupported recommendations. Retrieved rules are organized into RCA-oriented rule groups, such as fishbone-style structures, to support prioritized inspection and interpretation without implying causality. The approach was evaluated using six months of autonomous mobile robot (AMR) operational data from a smart factory. On previously unseen diagnostic queries, the system achieved 87.5% Top-1 diagnostic decision accuracy and a mean reciprocal rank (MRR) of 0.91, while maintaining an unsupported-output rate of 2.5% (95% CI: 1.1–6.4). During field deployment, the average abnormality-analysis meeting time decreased from 180 to 65&#xa0;min (− 63.9%), indicating faster convergence in human-in-the-loop diagnostics. These results demonstrate that rule-grounded, traceable retrieval can improve diagnostic consistency and the reuse of shop-floor knowledge while controlling output risk in smart manufacturing operations.</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

Rule-grounded diagnostic knowledge for traceable decision support in smart manufacturing AMR operations

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Chien-Chih Wang,
  • Ming-Nan Tsai

摘要

Root cause analysis (RCA) of abnormal events in smart manufacturing remains slow and inconsistent when diagnostic knowledge is fragmented across equipment logs and expert experience. This study introduces a rule-grounded diagnostic knowledge system for traceable decision support, which converts equipment events and log data into auditable diagnostic rules and enables evidence-backed retrieval through natural-language queries. The system integrates association-rule mining to construct a diagnostic rule base with rule identifiers and supporting evidence, semantic retrieval with controlled alignment to map user queries to candidate rules, and an abstention mechanism that withholds outputs when evidence is insufficient, thereby reducing unsupported recommendations. Retrieved rules are organized into RCA-oriented rule groups, such as fishbone-style structures, to support prioritized inspection and interpretation without implying causality. The approach was evaluated using six months of autonomous mobile robot (AMR) operational data from a smart factory. On previously unseen diagnostic queries, the system achieved 87.5% Top-1 diagnostic decision accuracy and a mean reciprocal rank (MRR) of 0.91, while maintaining an unsupported-output rate of 2.5% (95% CI: 1.1–6.4). During field deployment, the average abnormality-analysis meeting time decreased from 180 to 65 min (− 63.9%), indicating faster convergence in human-in-the-loop diagnostics. These results demonstrate that rule-grounded, traceable retrieval can improve diagnostic consistency and the reuse of shop-floor knowledge while controlling output risk in smart manufacturing operations.