cs.AINov 16, 2026

Deploying a Control-Logic LLM Assistant in Air-Gapped Power Plants: Lessons in Source Authority and Bounded Retrieval

Authors: Yiwei Fu, Nicholas Gabriel

Organizations: GE Vernova Advanced Research Center Niskayuna, NY, USA

Abstract

Deploying Large Language Model (LLM) assistants in industrial operational technology environments poses constraints largely absent from cloud applications: site policies restrict engineering and operational data from leaving the plant, on-premises compute is limited, and incorrect causal attribution can mislead incident response. We report lessons from deploying a control-logic evidence layer that underpins a broader on-premises operator assistance system at two gas power plants. Two prototypes revealed where generative models added avoidable uncertainty and cost. The first used a Vision Language Model (VLM) to reconstruct control topology from PDF printouts routinely exchanged by engineers. Across the full 6,979-page document, extraction reached 0.51 connection precision and required 3.1 hours on 4xH200 GPUs; consultation with controls engineers then revealed a native structured export from the authoring system. The second used an LLM tool-calling agent to plan repeatedly over eight predefined graph operations. In a controlled benchmark of 40 template-generated questions, replacing this agent with explicit execution plans reduced mean token use by 20x and median latency by 3x, with identical retrieved node sets across repeated runs. The deployed design ingests the authoritative export into a versioned knowledge graph, executes bounded graph operations in fixed software, and confines the local LLM to fallback intent classification and evidence-grounded explanation. At one site, the graph contains 221K nodes across seven controllers. Our experience suggests a practical division of labor for high-consequence AI: LLMs interpret natural-language requests and explain results, while deterministic software constructs and retrieves technical evidence grounded in authoritative sources. Abstention, entity disambiguation, graph versioning, and read-only integration provide additional safeguards.

Figures & tables

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Sep 28, 2026

PowerBench: A Benchmark for Agentic Retrieval and Reasoning in Power Systems

Large language model (LLM) agents offer new opportunities for automated analysis in industry. However, rigorous evaluation of such agents-for example, within power system scenarios-remains hindered: real operational data are confidential, and existing public resources fail to fully capture the chained dependencies and heterogeneous evidence. To address this gap, we propose PowerBench, comprising (1) a generation framework that derives interconnected heterogeneous operational data through a common dependency chain, and (2) a synthetic dataset generated by this framework. The dataset covers 761 devices across 100 device types, with 13.35 million hourly telemetry records spanning two years and 24,939 operational documents. Building on this dataset, we construct 300 questions across three task families that evaluate frontier LLMs' ability to complete analysis tasks that require autonomous evidence retrieval and reasoning across interconnected and heterogeneous data under restricted tool calls and time budgets. Results demonstrate that the evaluated frontier LLMs remain challenged on these tasks: the best model reaches only 74.2% joint accuracy. Our trace analysis further reveals that model performance varies across evidence discovery, content retrieval, tool use, reasoning over evidence, and answer submission. These findings provide detailed insights for evaluating LLM agents and guiding their reliable deployment in industry. The framework, dataset, and benchmark tasks are available at https://github.com/open-compass/PowerBench.
May 29, 2026

Knowledge boundary probing and demand-guided intervention for LLM-based power system code generation

Large language models (LLMs) can turn grid-analysis requests into executable programs for power-system simulation, but utilities and research laboratories often require on-premise deployment. In this setting, first-pass failures frequently arise at an API-knowledge boundary, through hallucinated functions, misused parameters, and mishandled result tables. We present PowerCodeBench, a parameterised benchmark generator released as a frozen 2,000-task suite for pandapower, and a deployment-time workflow that requires no weight updates. Documentation-driven L0-L3 probes produce per-model API profiles for diagnosis, model comparison, documentation allocation, and backend calibration. A query-side demand estimator selects layered API evidence before generation, while execution feedback routes targeted repair. Across ten open-weight LLMs (1.5B-480B) and four mid-tier APIs, the validation-enabled workflow raises scalar-match accuracy by 32-56 percentage points after up to three repair rounds relative to an unassisted first pass, for every model of at least 7B and every API. Open-weight models in the 70B-120B range reach the four-vendor mid-tier accuracy range under matched no-tool conditions. Selective injection approaches the full-layer reference using 41% of its prompt tokens. Among model-item pairs passing numerical checks under both workflows, engineering review confirms the requested analysis in 88% of full-workflow outputs versus 66% under plain repair. Round-0 pilots on OpenDSS and PyPSA motivate staged onboarding from broad retrieval at cold start to calibrated selective injection. Measured throughput, latency, energy, and allocated GPU memory establish a practical on-premise serving envelope.
May 26, 2026

Knowledge Graphs as the Missing Data Layer for LLM-Based Industrial Asset Operations

LLM-based agents for industrial asset operations show limited accuracy when reasoning over flat document stores. AssetOpsBench (KDD 2026) establishes that GPT-4 agents achieve 65% on 139 industrial maintenance scenarios, and compares LLM orchestration paradigms (Agent-As-Tool vs. Plan-Execute) on a fixed data layer. We ask the orthogonal question: how much does the data model behind the tools matter? We treat a typed knowledge graph as a grounding substrate and route each question by how it is best answered: (i) LLM-generated Cypher for structured retrieval, which lifts the same GPT-4 model from 65% to 82-83%; (ii) native graph and optimization primitives, with no LLM, reaching 99% on graph-answerable scenarios; and (iii) generation-augmented knowledge (GAK) for answers absent from the data -- the engine's agent materializes the missing facts as provenance-tagged graph nodes, then answers. A recurring theme is inverted LLM usage: we constrain the LLM to query generation or one-shot enrichment from a typed schema and let the graph execute deterministically. On the 88 real AssetOpsBench failure-mode scenarios the benchmark itself flags non-deterministic -- ten equipment types absent from the graph -- GAK lifts answerability from zero to 100% of equipment types and answers 81.8% of scenarios, every materialized fact tagged source:LLM-derived for auditability. We also contribute 40 graph-native scenarios. For structured operational domains the data layer -- not the LLM orchestration -- is the primary lever, and a typed knowledge graph serves as a grounding substrate between raw industrial data and LLM reasoning.