From detection to action: Using LLM agents for Fault-Tolerant Control

Abstract

We propose an agentic Large Language Model (LLM) framework for active Fault-Tolerant Control (FTC) that transforms fault detection outputs into constraint-aware recovery actions grounded in plant-specific knowledge. The approach couples (i) a multi-agent workflow that decomposes operator duties into monitoring, planning, action synthesis, simulation, validation, and reprompting; (ii) a Digital Process Plant Twin (DPPT) that exposes plant data, models, and a simulation service for pre-execution testing; and (iii) a Graph Retrieval-Augmented Generation (Graph RAG) layer built on the CPSMod ontology, which organizes plant knowledge (structure, function, hybrid dynamics, control context, and fault semantics) into a graph that supports relation-aware, multi-hop retrieval for the agents. 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 any actuation. If no acceptable plan is found within a bounded time window, control is handed to a deterministic fallback policy. The framework is evaluated exclusively in simulation on two representative benchmarks: a discrete batch Mixing Module and a Continuous Stirred-Tank Reactor (CSTR) under closed-loop PID regulation. The evaluation isolates the detection-to-action step under a matched-model assumption: the DPPT used for pre-execution validation is identical to the model generating the simulated plant behavior and uses the same injected fault configuration. Robustness to plant-model mismatch and sensor-noise variation or misspecification is not evaluated in this study. Results with lightweight LLMs (GPT-4o-mini and GPT-4.1-mini) show that semantically grounded agents can derive recovery decisions that satisfy the configured validation criteria within latency budgets compatible with the respective process dynamics, demonstrating a pathway from detection to validated corrective action under this matched-model simulation setting.