Physics-guided machine learning for ship biofouling assessment in support of maritime decarbonization

Ship biofouling significantly increases hull resistance and propeller loading, resulting in increased fuel usage and greenhouse gas emissions. This study presents a physics-guided machine learning approach for detecting performance degradation caused by biofouling. Physics-informed neural networks are first used to generate warm-up data that represent ideal, unfouled ship states under calm water conditions. A baseline model is built from this data and subsequently refined through an incremental learning scheme with new data collected in sliding temporal windows. The resulting incremental models are applied under reference conditions to quantify biofouling-induced performance changes, expressed as key performance indicators. Validation against conventional retraining approaches and the ISO 19030 standard shows that the proposed method more accurately captures both gradual degradation and rapid post-cleaning recovery. By delivering reliable and timely assessments of fouling impact, the framework supports optimized hull and propeller maintenance planning and contributes to improved energy efficiency and emission reduction.