Marine propeller optimisation through user interaction and machine learning for advanced blade design scenarios

The complexity of the marine propeller design process is well recognised and is related to contradicting requirements of the stakeholders, complex physical phenomena, and fast analysis tools, where the latter are preferred due to the strict time limitations under which the entire process is carried out. With all this in mind, an optimisation methodology has been proposed and presented earlier that combines user interactivity with machine learning and proved to be useful for a simple blade design scenario. More specifically, the blade designer manually evaluates the cavitation of the designs during the optimisation and this information is systematically returned into the optimisation algorithm, a process called interactive optimisation. As part of the optimisation, a machine learning pipeline has been implemented in this study, which is used for cavitation evaluation prediction in order to solve the user fatigue problem that is connected to interactive optimisation processes. The proposed methodology is investigated for two case studies of advanced design scenarios, relevant for a real commercial situation, that regard controllable-pitch propellers for ROPAX vessels, and the aim is to obtain a set of optimal, competent blade designs. Both cases represent scenarios with several design variables, objectives and constraints and with conditions that have either suction side or pressure side cavitation. The results show that the proposed methodology can be used as a support tool for the blade designers to, under strict time constraints, find a suitable set of propeller designs, some of which can be considered equal or even superior to the delivered design.