[4Yin-A] Poster 6

[4Yin-A-04] AI for Anomaly Response Support:Explanation Generation and Action Recommendation via Multimodal RAG

〇Yusuke Yamashina¹, Keisuke Niimi¹, Caio Cesar Pinheiro de Moura¹, Kazuma Shiomi¹, Yoshihiko Ichikawa¹, Tsukasa Kamo², Atsushi Kuboya², Yuji Ayusawa², Tatsuya Yamamoto², Keigo Yoshida² (1. Insight Edge, Inc., 2. SCSK Corporation)

Keywords:

XAI,Anomaly Detection,Image Processing,LLM

This study proposes an AI framework for anomaly response support that integrates an image-based anomaly detection model with a large language model (LLM) to generate both explanations and action recommendations. Although anomaly detection is widely used in manufacturing and medical domains, its outputs are often numerical and abstract, making it difficult to understand anomaly causes and appropriate responses. To address this issue, we extend our previously proposed language-driven explainable AI by incorporating multimodal Retrieval-Augmented Generation (RAG), enabling the use of textual and visual materials such as manuals and diagrams. By referencing external knowledge sources, the framework produces grounded explanations and context-aware action suggestions. Experimental results demonstrate that the proposed method significantly improves transparency and practical applicability, enabling evidence-based action recommendations and supporting real-world operational decision-making.