Machine-learning algorithms for predictive response surfaces for propulsion propellers

Through algorithms and statistical models, machine learning enables the analysis of large datasets and allows for predictions or decision-making based on recognized patterns. This thesis explores the use of machine learning to construct and analyze response surfaces related to the performance coefficients of marine propulsion propellers, particularly the thrust coefficient and the torque coefficient. These two variables, fundamental in evaluating propeller efficiency, depend on various geometric parameters associated with the airfoil profiles of a propeller. The research, conducted within the context of Fincantieri aims to determine the response surfaces associated with the efficiency coefficients, predicting how geometric variations affect them. Through learning techniques, the study explores non-linear relationships between the variables, utilizing regression models and radial basis functions. The work, structured into five chapters, covers the fundamentals of machine learning, mathematical modeling, propeller geometry, and the implementation of response surfaces, concluding with a case study that demonstrates the practical application of the developed models on a test propeller. This research aims to develop accurate predictive models and evaluate their effectiveness, offering an innovative and strategic contribution to the company, potentially supporting the preliminary phase of propeller design.