Selected work

Representative Papers

Narrow and Deep: An Ontology Tower as the Knowledge of an LLM Agent for an Industrial Equipment System

Oct 08, 2026

This study addresses the misjudgments in industrial equipment control caused by LLM agents lacking domain knowledge, proposing a "narrow but deep" ontology tower architecture. This approach integrates physical quantity derivation with operational logs to transform tacit experience into explicit knowledge nodes. Through ontology engineering and physical relational reasoning, it achieves streamlined entity representation and semantically rich knowledge modeling, injecting real-time PLC data and historical knowledge into LLM agents. Experimental results demonstrate that a 9B-parameter model performs comparably to a 750B-parameter counterpart, reducing critical misjudgment rates by approximately 20%. Furthermore, in physical system evaluations, the method successfully regulated controlled variables into target ranges in 12 out of 14 trials.

0 citationsRead paper

Automated generation of experimentally validated digital twins for desiccant-based low-dew-point air-conditioning systems from declarative topology specifications

Aug 09, 2026

This study addresses the long-standing reliance on expert knowledge in constructing digital twins for low dew-point air conditioning systems, which hinders industrial deployment. The authors propose an automated modeling framework based on declarative topological specifications that leverages a library of physical components to generate dynamic models, solvers, and telemetry interfaces automatically. By integrating adsorption–heat transfer coupling mechanisms with parameter identifiability analysis, the framework achieves calibration using only three humidity measurement nodes. It enables rapid generation and automatic calibration of digital twins directly from natural language descriptions and characterizes unknown commercial desiccants via equivalent adsorption isotherms, ensuring high-fidelity prediction rather than empirical curve-fitting. Experimental results demonstrate a 15-fold acceleration in model generation compared to manual methods, with dew-point temperature prediction errors below 0.1°C under bypass conditions, regeneration heating power errors under 5%, and accurate reproduction of step-response dynamics.

0 citationsRead paper

LLM-Guided Planning for Multi-hop Reasoning over Multimodal Nuclear Regulatory Documents

Jun 28, 2026

This study addresses the challenge of multi-hop reasoning across tens of thousands of pages in nuclear regulatory document review, where evidence is highly dispersed. The authors propose a state-aware planning framework based on large language models (LLMs) that formulates reasoning as dynamic navigation within an unvectorized document tree. An agent progressively constructs and updates an internal dynamic knowledge graph through browsing, reading, and search actions until sufficient evidence is gathered. A novel auditable edge-reasoning module is introduced to enhance decision traceability without requiring offline indexing. Evaluated on the NuScale FSAR 200-question benchmark, the method achieves 81.5% accuracy and a RAGAS faithfulness score of 0.93, substantially outperforming existing approaches such as PageIndex (+38.0 percentage points), LightRAG, HippoRAG, and GraphRAG.

0 citationsRead paper

LLM agent safety, multi-turn red-teaming, jailbreak benchmarks, adversarial robustness, safety-critical systems

Jun 18, 2026

This work addresses the insufficient robustness of large language model (LLM) agents in controlling safety-critical systems under persistent, adaptive adversarial attacks. To this end, the authors introduce NRT-Bench, a novel benchmark that simulates a nuclear power plant control room staffed by a five-member LLM operator team. The framework evaluates agent resilience through multi-channel, multi-turn red-teaming attacks coupled with an adversarial feedback mechanism. Crucially, it defines objective harm via the loss of critical safety functions grounded in actual system states—rather than textual judgments—and employs a fixed attack pairing replay protocol. Experiments across four state-of-the-art models reveal that 8.7%–12.1% of attack sessions result in safety function loss. While none of the 149 attacks compromised all models, approximately one-third succeeded against at least one, highlighting highly heterogeneous vulnerabilities and strong model-dependent defense efficacy.

0 citationsRead paper

Recent publications

Latest Papers

Narrow and Deep: An Ontology Tower as the Knowledge of an LLM Agent for an Industrial Equipment System

Oct 08, 2026

This study addresses the misjudgments in industrial equipment control caused by LLM agents lacking domain knowledge, proposing a "narrow but deep" ontology tower architecture. This approach integrates physical quantity derivation with operational logs to transform tacit experience into explicit knowledge nodes. Through ontology engineering and physical relational reasoning, it achieves streamlined entity representation and semantically rich knowledge modeling, injecting real-time PLC data and historical knowledge into LLM agents. Experimental results demonstrate that a 9B-parameter model performs comparably to a 750B-parameter counterpart, reducing critical misjudgment rates by approximately 20%. Furthermore, in physical system evaluations, the method successfully regulated controlled variables into target ranges in 12 out of 14 trials.

0 citationsRead paper

Automated generation of experimentally validated digital twins for desiccant-based low-dew-point air-conditioning systems from declarative topology specifications

Aug 09, 2026

This study addresses the long-standing reliance on expert knowledge in constructing digital twins for low dew-point air conditioning systems, which hinders industrial deployment. The authors propose an automated modeling framework based on declarative topological specifications that leverages a library of physical components to generate dynamic models, solvers, and telemetry interfaces automatically. By integrating adsorption–heat transfer coupling mechanisms with parameter identifiability analysis, the framework achieves calibration using only three humidity measurement nodes. It enables rapid generation and automatic calibration of digital twins directly from natural language descriptions and characterizes unknown commercial desiccants via equivalent adsorption isotherms, ensuring high-fidelity prediction rather than empirical curve-fitting. Experimental results demonstrate a 15-fold acceleration in model generation compared to manual methods, with dew-point temperature prediction errors below 0.1°C under bypass conditions, regeneration heating power errors under 5%, and accurate reproduction of step-response dynamics.

0 citationsRead paper

LLM-Guided Planning for Multi-hop Reasoning over Multimodal Nuclear Regulatory Documents

Jun 28, 2026

This study addresses the challenge of multi-hop reasoning across tens of thousands of pages in nuclear regulatory document review, where evidence is highly dispersed. The authors propose a state-aware planning framework based on large language models (LLMs) that formulates reasoning as dynamic navigation within an unvectorized document tree. An agent progressively constructs and updates an internal dynamic knowledge graph through browsing, reading, and search actions until sufficient evidence is gathered. A novel auditable edge-reasoning module is introduced to enhance decision traceability without requiring offline indexing. Evaluated on the NuScale FSAR 200-question benchmark, the method achieves 81.5% accuracy and a RAGAS faithfulness score of 0.93, substantially outperforming existing approaches such as PageIndex (+38.0 percentage points), LightRAG, HippoRAG, and GraphRAG.

0 citationsRead paper

LLM agent safety, multi-turn red-teaming, jailbreak benchmarks, adversarial robustness, safety-critical systems

Jun 18, 2026

This work addresses the insufficient robustness of large language model (LLM) agents in controlling safety-critical systems under persistent, adaptive adversarial attacks. To this end, the authors introduce NRT-Bench, a novel benchmark that simulates a nuclear power plant control room staffed by a five-member LLM operator team. The framework evaluates agent resilience through multi-channel, multi-turn red-teaming attacks coupled with an adversarial feedback mechanism. Crucially, it defines objective harm via the loss of critical safety functions grounded in actual system states—rather than textual judgments—and employs a fixed attack pairing replay protocol. Experiments across four state-of-the-art models reveal that 8.7%–12.1% of attack sessions result in safety function loss. While none of the 149 attacks compromised all models, approximately one-third succeeded against at least one, highlighting highly heterogeneous vulnerabilities and strong model-dependent defense efficacy.

0 citationsRead paper