Multi-Modal Agents for Power Distribution Defect Detection: An Evaluation of Foundation Models

The paper presents a Multi-Modal Agent framework for defect detection in power distribution networks, addressing limitations in traditional inspection methods. It evaluates multimodal foundation models based on their performance in perception, reasoning, and tool usage, with a focus on generating expert-level defect descriptions, diagnosing issues, and executing maintenance actions. The findings highlight the potential and challenges of deploying these models as autonomous agents in critical industrial settings, backed by a domain-specific evaluation dataset and comprehensive benchmarks.