Abstract: Objectives To address the urgent need for the intelligent transformation of equipment support in coast guard vessels, this paper systematically reviews the research progress of large language model-based ship fault detection and diagnosis (FDD) systems and clarifies their technical pathways and development directions. Methods A systematic literature review method was adopted to analyze 60 relevant publications from 2022 to 2025. Optimization strategies of large language models in industrial FDD were summarized from four dimensions: input-side enhancement, model-side optimization, output-side refinement, and architectural design. In response to the characteristics of shipboard data heterogeneity, limited resources, and high reliability requirements, a hierarchical, decoupled, and multi-agent collaborative reference architecture for shipboard FDD systems was proposed. Results The number of research publications on large language models in the industrial FDD field increased from 4 in 2023 to 35 in 2025, with approximately 65% focused on input-side enhancement and model-side optimization. For shipboard scenarios, the proposed architecture supports the localized deployment of lightweight large language models, enables multi-source data fusion through vector knowledge bases and knowledge graphs, and realizes automatic task decomposition and closed-loop decision-making for fault diagnosis based on multi-agent collaboration. ConclusionsLarge language models provide a feasible pathway for transforming shipboard FDD systems toward an intelligent mode driven by "data and knowledge synergy." Future research should focus on lightweight deployment, trustworthy output, adaptive capabilities, and system integration to enhance equipment support effectiveness and navigation safety.