An Event-Driven Dual-Mode Multilingual RAG Framework for Industrial Maintenance

Industrial equipment fault diagnosis faces critical challenges in non-English operational environments. This paper presents an Event-Driven Dual-Mode Multilingual Retrieval-Augmented Generation (EDM-RAG) framework for industrial maintenance. Unlike conventional query-driven RAG systems, EDM-RAG organizes diagnostic reasoning through event-triggered pathways that coordinate diagnostic signal interpretation, equipment context profiling, historical maintenance case retrieval, and severity-aware retrieval parameter adjustment. A proof-of-concept evaluation on 45 curated maintenance cases indicates promising diagnostic behavior and bilingual report generation under controlled conditions. Additional robustness experiments under fault-code corruption and textual ambiguity, together with prompt-invariance analysis for AI-assisted report assessment, further characterize system behavior under non-ideal conditions. Taken together, these results support viewing EDM-RAG as a proof-of-concept system-level diagnostic framework for controlled industrial maintenance evaluation rather than merely a simple retrieval-generation integration.