Multi-source thermal and performance data for thermal fault diagnosis in marine diesel engines
2026-08-22
This paper presents a multi-source parameter dataset for the predictive health management (PHM) of marine diesel engines, addressing the scarcity of fault data arising from the prohibitive cost of destructive fault experiments aboard real vessels. Covering load conditions from 25% to 100% of full load, the dataset comprises four core subsets: noise-free calibrated simulation data serving as a thermodynamic baseline, bench-measured data incorporating industrial background noise, real-vessel telemetry data characterised by wave-induced load fluctuations and broadband engine room interference, and superimposed fault data generated via virtual-real fusion, representative of the dynamic impacts of severe sea states. Rigorously calibrated against real-vessel telemetry, the dataset encompasses 24-dimensional high-frequency transient gas-path parameters and in-cylinder steady-state performance indicators across one healthy state and ten representative thermal fault modes, involving turbochargers, intercoolers, intake and exhaust manifold networks, and fuel injectors. Validation confirms strong physical consistency and feature separability, whilst a zero-shot virtual-to-real transfer experiment achieved 70.18% accuracy, demonstrating effective handling of domain discrepancy challenges. This dataset is well-suited for generalised multi-source information fusion, noise-robust weak-drift feature extraction, and unsupervised domain adaptation algorithm development and benchmarking. The paper details the test equipment, simulation model, fault cases, noise extraction and superposition strategy, and technical validation procedures.