LLM Agents Automate Fault Recovery for Industrial Control Systems

A new paper by Javal Vyas and colleagues introduces an agentic Large Language Model (LLM) framework for active Fault-Tolerant Control (FTC) in industrial plants. The system combines three core components: a multi-agent workflow that decomposes operator tasks into monitoring, planning, action synthesis, simulation, validation, and reprompting; a Digital Process Plant Twin (DPPT) that provides plant data, models, and a simulation service for pre-execution testing; and a Graph Retrieval-Augmented Generation (Graph RAG) layer built on the CPSMod ontology. This ontology organizes plant knowledge (structure, function, hybrid dynamics, control context, and fault semantics) into a graph that supports relation-aware, multi-hop retrieval. Corrective actions are generated as minimal-risk state-machine recovery paths and corresponding discrete commands or continuous setpoint adaptations, then validated deterministically against interlocks, envelopes, and dynamic feasibility before actuation. If no acceptable plan is found within a bounded time window, control is handed to a safety fallback.

The framework was evaluated in simulation on two representative benchmarks: a discrete batch Mixing Module and a Continuous Stirred-Tank Reactor (CSTR) under closed-loop PID regulation. Using lightweight LLMs (GPT-4o-mini and GPT-4.1-mini), the agents derived valid recovery decisions within latency budgets compatible with the respective process dynamics. Results demonstrate a practical pathway from fault detection to validated corrective action across both discrete and continuous FTC tasks, showing that semantically grounded LLM agents can handle industrial control challenges without requiring massive models or extensive retraining. This work points toward safer, more autonomous industrial operations where AI can manage unexpected failures in real time.