Machine learning for shaft power prediction and analysis of fouling related performance deterioration

Improving operational performance and reducing fuel consumption is increasingly important for shipping companies. Ship performance degrades over time due to hull and propeller fouling; therefore assessing when fouling effects are significant enough to warrant cleaning is critical. Advancements in onboard data logging systems, combined with machine learning techniques, unlock the potential to predict fouling effects accurately and determine when to clean. This study evaluates five models for shaft power prediction: Multiple Linear Regression, Decision Tree (AdaBoost), K - Nearest Neighbours, Artificial Neural Network and Random Forest. The importance of pre-processing is highlighted, contributing to the creation of a model with lower errors than previous studies. The significance of environmental parameters was explored, with the novel integration of wave statistics to the operational dataset, and simulated power-speed curves created from predictions to identify performance deterioration due to fouling. The Random Forest model was most effective in predicting shaft power, with an error of 1.17%. The addition of 'Days Since Clean' and 'Significant Wave Height' increased prediction accuracy by 0.07% and 0.12% respectively. Simulated power-speed curves revealed a 5.2% increase in shaft power due to fouling. This study provides operators with a method to determine when to conduct hull and propeller cleaning.