Comparative Study on Health Monitoring of a Marine Engine Using Multivariate Physics-Based Models and Unsupervised Data-Driven Models

The marine engine is a complex-structured multidisciplinary system that operates in harsh environment involving high temperatures and pressures gas/fluid/solid interactions. Many malfunctions faults can occur to the efficient condition monitoring critical ensure expected performance. In this paper, test rig established its process data are recorded, including various pressures. Two data-driven models, i.e., principal component analysis sparse autoencoder, physics-based model applied for two classic faults, lubrication oil filter blocking cylinder leakage. Comparative studies discussions conducted regarding their performance terms of anomaly detection fault isolation. points collected generally times higher than thresholds set by models. model, it observed pressure falls from predicted 3.2–3.8 bar around 2.3 bar. For leakage fault, nearly four exhaust gas temperature leaked an estimated 150–200 °C about 100 °C. transferability interpretability these models finally discussed. findings present study offer insights into types provide guidance effective engines.