Domain-specific ReAct for physics-integrated iterative modeling: A case study of LLM agents for gas path analysis of gas turbines

TL;DR

The paper probes the feasibility of deploying LLMs with callable, domain-specific tools for gas-path analysis in gas turbines, employing a dual-agent ReAct-based workflow to fuse expert knowledge with solver tools. It compares multiple models, finding that near-100B parameter models—when fine-tuned and prompted effectively—offer the best potential for professional engineering tasks, though complex multi-component reasoning remains difficult for current systems. The methodology uses four domain tools to compute compressor, burner, turbine, and nozzle outcomes while enforcing thermodynamic consistency through guided reasoning. While results are encouraging for tool-augmented LLMs, the study notes persistent challenges in handling highly integrated gas-turbine problems, underscoring the need for further model refinement and advanced prompting to achieve robust AI-driven analysis.

Abstract

This study explores the application of large language models (LLMs) with callable tools in energy and power engineering domain, focusing on gas path analysis of gas turbines. We developed a dual-agent tool-calling process to integrate expert knowledge, predefined tools, and LLM reasoning. We evaluated various LLMs, including LLama3, Qwen1.5 and GPT. Smaller models struggled with tool usage and parameter extraction, while larger models demonstrated favorable capabilities. All models faced challenges with complex, multi-component problems. Based on the test results, we infer that LLMs with nearly 100 billion parameters could meet professional scenario requirements with fine-tuning and advanced prompt design. Continued development are likely to enhance their accuracy and effectiveness, paving the way for more robust AI-driven solutions.

Figures (2)

• Figure 1: Schema of gas path analysis for gas turbines
• Figure 2: Dual-agent tool-calling process