An LLM-Agentic Framework for Intelligent Work Order Management

An LLM-Agentic Framework for Intelligent Work Order Management in Industrial Maintenance


Authors: Nianjun Zhou, Dhaval Patel

arXiv: 2611.25070 [cs.AI] — Submitted 18 November 2026

Subject: Artificial Intelligence (cs.AI)

DOI: https://doi.org/10.48550/arXiv.2611.25070


Abstract


This paper introduces an AI-driven framework—the CodeReAct agent—designed to enhance industrial equipment maintenance through natural language (NL) interactions. The framework leverages an advanced AI agentic approach to interpret human instructions, delivering knowledge extraction, asset status summarization, and work order recommendations. It identifies patterns and trends and predicts work orders by standardizing industrial data into business objects (BOs) and invoking analytical function APIs.


The agent operates through a Thought–Action–Observation process that mirrors human problem-solving, automating complex analytical tasks typically reserved for data scientists. Notably, the Action phase generates executable code, replacing manual coding effort.


A case study on chiller maintenance demonstrates the framework's ability to analyze multiple years of work orders, as well as alerts and anomalies. Results show that the agent effectively processes human utterances to analyze data, extract actionable insights, and support informed decision-making—reducing reliance on data scientists. By recontextualizing NLP agent frameworks within industrial engineering, the approach enables autonomous function generation and task execution across a range of domain-specific maintenance scenarios, with chillers serving as a representative case study.


Key Contributions


  • CodeReAct agent architecture: An LLM-agentic framework for natural language interaction with industrial maintenance systems.
  • Thought–Action–Observation reasoning loop: Automates analytical workflows traditionally handled by data scientists.
  • Executable code generation: The Action phase produces runnable code, eliminating manual coding for analytical tasks.
  • Business object standardization: Industrial data is normalized into BOs to support consistent analytical API calls.
  • Chiller maintenance case study: Validated over multi-year work order, alert, and anomaly data.

Implications for Industrial AI in 2026


As industrial operators increasingly adopt LLM-based agents for operations and maintenance (O&M), frameworks like CodeReAct reflect a broader 2026 shift: moving from passive analytics dashboards toward conversational, action-capable agents that can both reason over asset data and generate their own analytical tooling. This aligns with growing enterprise interest in agentic AI for asset management, where natural language interfaces lower the barrier to complex predictive maintenance workflows and reduce dependence on specialized data science teams.


Citation


@article{zhou2026codereact,
  title   = {An LLM-Agentic Framework for Intelligent Work Order Management in Industrial Maintenance},
  author  = {Zhou, Nianjun and Patel, Dhaval},
  journal = {arXiv preprint arXiv:2611.25070},
  year    = {2026},
  doi     = {10.48550/arXiv.2611.25070}
}

via ArXiv AI

Related