Scholarlink

<p>Electrochemical impedance spectroscopy (EIS) analysis, crucial for understanding electrochemical systems like batteries, relies heavily on expert-driven equivalent circuit model selection, limiting efficiency and objectivity. Here, we present AgentEIS, an integrated framework combining machine learning (ML), large language models (LLMs) and multimodal large models (MLMs) for automated EIS analysis. We constructed a comprehensive EIS database encompassing diverse electrochemical systems (e.g., lithium-ion batteries, fuel cells) with 10 distinct equivalent circuit models. Among ML classifiers, the extremely randomized trees (ET) model achieved the highest accuracy (67.4%)&#xa0;in circuit identification. By selecting the top-three probable circuits, coverage of the true model exceeded 90%, significantly reducing manual screening. We further evaluated LLMs (e.g., Qwen3, Llama3.1) and MLMs (e.g., Qwen2.5-VL), showing that fine-tuning via low-rank adaptation (LoRA) improved their performance, though accuracy remained lower than task-specific ML. AgentEIS synergizes ML’s efficiency in circuit screening with LLM’s capability for mechanistic interpretation. Validation using lithium-ion pouch cell data demonstrated that AgentEIS accurately identifies equivalent circuits, links components to physical processes and reduces reliance on expert input. This work pioneers artificial intelligence (AI)-driven integration for scalable, objective EIS analysis, accelerating electrochemical research and diagnostics.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating large language and multimodal models with machine learning for equivalent circuit analysis of electrochemical impedance spectroscopy

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Shan Zhu,
  • Gong Jia,
  • Simi Sui,
  • Kezhu Jiang,
  • Ning Wang,
  • Shijian Zheng

摘要

Electrochemical impedance spectroscopy (EIS) analysis, crucial for understanding electrochemical systems like batteries, relies heavily on expert-driven equivalent circuit model selection, limiting efficiency and objectivity. Here, we present AgentEIS, an integrated framework combining machine learning (ML), large language models (LLMs) and multimodal large models (MLMs) for automated EIS analysis. We constructed a comprehensive EIS database encompassing diverse electrochemical systems (e.g., lithium-ion batteries, fuel cells) with 10 distinct equivalent circuit models. Among ML classifiers, the extremely randomized trees (ET) model achieved the highest accuracy (67.4%) in circuit identification. By selecting the top-three probable circuits, coverage of the true model exceeded 90%, significantly reducing manual screening. We further evaluated LLMs (e.g., Qwen3, Llama3.1) and MLMs (e.g., Qwen2.5-VL), showing that fine-tuning via low-rank adaptation (LoRA) improved their performance, though accuracy remained lower than task-specific ML. AgentEIS synergizes ML’s efficiency in circuit screening with LLM’s capability for mechanistic interpretation. Validation using lithium-ion pouch cell data demonstrated that AgentEIS accurately identifies equivalent circuits, links components to physical processes and reduces reliance on expert input. This work pioneers artificial intelligence (AI)-driven integration for scalable, objective EIS analysis, accelerating electrochemical research and diagnostics.

Graphical Abstract