AI-Driven Development of Marine Engine Diagnostics.

Abstract Effective fault diagnosis is essential for the safe and efficient operation of marine slow-speed diesel engines. Readily accessible diagnostic tools are crucial for maritime engineers to maintain these complex systems. Traditional development of such tools often depends on extensive manual knowledge elicitation from experts, a time-consuming and resource-intensive process. This research investigates the automated generation of a fault-finding system using simulated data, leveraging the capabilities of Large Language Models (LLMs) and diagramming tools. The objective is to automate the development of a diagnostic system for marine diesel engines, eliminating the need for manual expert input. The study addresses the inefficiencies of traditional methods by demonstrating the feasibility of using LLMs and diagramming tools to generate diagnostic logic directly from simulated engine fault data. By automating knowledge extraction, logic generation, and visualization, this approach aims to create a more efficient and consistent method for developing diagnostic tools. Using the Kongsberg engine room simulator, a comprehensive dataset of fault scenarios was generated. Simulated faults—such as piston ring wear and fuel valve choking—were linked to changes in key engine parameters (e.g., SPEED, INDEX, MIP). These data were used to generate prompts for an LLM, which produced diagnostic rules and decision logic. The logic was then translated into visual diagrams using tools like Mermaid.js and PlantUML. Automated evaluation against the simulated data assessed the system’s diagnostic accuracy. The results demonstrated that the automated process could generate a comprehensive and accurate fault-finding system. The system effectively captured relationships between fault conditions and parameter changes, significantly reducing development time and effort compared to traditional methods. The consistency and clarity of the generated diagrams enhanced usability, and the high fault identification rate validated the approach’s effectiveness. This research highlights the potential of AI-assisted automation to transform diagnostic tool development in maritime engineering. The approach offers substantial benefits in efficiency, consistency, and scalability. Future work will incorporate real-world engine data and structured participant feedback to enhance robustness and generalizability. The methodology also lays the groundwork for benchmarking against expert-developed systems and integrating probabilistic reasoning and explainable AI (XAI) to improve transparency and interpretability. This technology holds promise not only for operational diagnostics but also for educational applications and broader adoption in other complex technical domains.