ThermRAG: A Multimodal Knowledge-Enhanced Agent for Power Electronics Thermal Design

Modern power converters face stringent PCB thermal management challenges driven by the continuous demand for higher power density. Conventional design methodologies are time-intensive, heuristic-dependent, and prone to human error, whereas general-purpose Large Language Models (LLMs) often lack the specific domain expertise and numerical rigor required for engineering applications. To address these limitations, this paper proposes ThermRAG, a multimodal knowledge-enhanced agent for power electronics thermal design. ThermRAG innovatively integrates a Thermal Calculation Engine with a dual-knowledge-base architecture. By establishing a hierarchical context fusion mechanism between an authoritative Core Knowledge Base and a Dynamic Knowledge Base, the agent is capable of identifying parameters from unstructured datasheets, autonomously triggering script tools for precise thermal-related calculations, and generating design recommendations via a locally deployed DeepSeek R1 model. Experimental validation demonstrates that ThermRAG significantly enhances design efficiency and delivers high-confidence thermal design consultation solutions.