Colloquia

Samenvatting

The Royal Netherlands Navy is undertaking a major fleet renewal; however, the design authority lacks accurate operational data and failure statistics from current platforms. This information asymmetry limits the effectiveness of predictive maintenance and constrains optimal configuration decisions for future vessel designs. The present thesis fills this gap by developing a digital shadow for the ocean-going patrol vessel powertrain, with the objective of generating reliable operational insights through the integration of data-driven analytics and first-principles simulation methods.

The research methodology merges sensor data from the integrated platform management system with a thermodynamic model of the main diesel engine. The study evaluates the effectiveness of support vector machines for classifying sea states when meteorological data are unavailable and applies ordinary least squares regression for residual-based fault monitoring. Additionally, four anomaly detection methods: Rolling Window, Consecutive Runs, M-of-N, and CUSUM, are evaluated on modelled ramp and pulse faults to assess their robustness.

The results demonstrate that the SVM algorithm with a radial basis function kernel achieves 96.69% success in predicting sea states. Residual analysis using OLS regression is highly effective, detecting approximately 67.51% of ramp faults and 97.56% of pulse faults, which makes additional filtering largely unnecessary for abrupt anomalies. The Rolling Window method, however, performs better at identifying incipient faults. The study also identifies a measurable decline in engine fuel efficiency, as Brake Specific Fuel Consumption increased by 2.38% and 3.57% for the Starboard and Port engines, respectively, relative to Factory Acceptance Test values. Overall, the findings indicate that a digital shadow integrating simulation and sensor data can effectively bridge the gap between design assumptions and operational reality, provided the simulation remains adaptive to the vessel's evolving state.