A

Verified LLM-assisted capability- and skill-based process planning framework for modular plants

at - Automatisierungstechnik, vol. 74, pp. 787–802

Abstract

Abstract Capability- and skill-based engineering describes modular-plant functions, but it does not decide how an ISA-88 General Recipe becomes executable in a current plant state. This paper presents ModPlant-LLM , a verified LLM-assisted capability- and skill-based process planning framework for modular plants. It distills symbolic teacher plans into a student proposer and validates proposed steps with a finite-state automaton (FSA) checker plus symbolic fallback. On a 48-sequence linear-recipe benchmark, the fine-tuned 3B model reaches 97.9 % operation-sequence correctness with a planning time of 36.92 s on the GPU host. Compared with the prior CPU-based symbolic workflow (157.5 s average in batched runs), this corresponds to a planning-time reduction of about 76.6 % and supports fast recipe-to-plan generation with checked admissibility. The framework also shows how symbolic teachers can supply high-quality industrial corpora for fine-tuning compact models in resource-constrained settings, while fallback recomputation protects process-plan reliability when generated steps are inadmissible.

Authors 4

  1. RWTH Aachen University

    Affiliation as printed

    Chair of Information and Automation Systems for Process and Material Technology, RWTH Aachen University , Aachen , Germany

  2. RWTH Aachen University

    Affiliation as printed

    Chair of Information and Automation Systems for Process and Material Technology, RWTH Aachen University , Aachen , Germany

  3. RWTH Aachen University

    Affiliation as printed

    Chair of Information and Automation Systems for Process and Material Technology, RWTH Aachen University , Aachen , Germany

  4. RWTH Aachen University

    Affiliation as printed

    Chair of Information and Automation Systems for Process and Material Technology, RWTH Aachen University , Aachen , Germany

Cited by 0 stored of 0

References 22