Modelling of the cavitating propeller noise by means of semi-empirical and data driven approaches

Historically, the mitigation of the ship radiated noise in the water was a prerogative of naval ships due to quiet requirements. In the last decades, the need for merchant ships and pleasure craft to ensure high standards of comfort on board in terms of on board radiated noise and structural vibrations lead also, indirectly, towards the reduction of underwater radiated noise. Nowadays, the greater awareness about the damages to the marine ecosystem as a result of the ship noise pollution is leading governments and international institutions towards the study of possible limits to acoustic emissions, which could be applied, to different levels, to protected marine areas and to more general navigation routes. Propeller, when cavitating, is the main source of radiated noise for conventional ships together with the engines; propeller cavitation, contrarily to machinery, is not linked to single frequencies, being a broadband noise. Its reduction is thus becoming one of the objectives in new propellers design. One of the most effective and common way to assess the propeller cavitation noise is by experimental tests in model scale. This procedure is rather expensive and time consuming, thus it is rather difficult to include it in an iterative design loop. The aim of the present PhD thesis is the development of semi-empirical methods for the prediction of the propeller cavitating noise, in order to provide the designer with a tool capable of allowing prediction of underwater radiated noise at early design stages. Moreover, the same method can be applied in order to enhance the capability of prediction of underwater radiated noise from model scale tests, allowing to obtain indications also for operating conditions not directly reproducible due to scaling effects. Attention has been devoted to the most common cavitation phenomena, i.e. back sheet cavitation and tip vortex. The considered methods are derived from physical formulations available in literature and purely data driven models coming from the machine learning field, exploiting also the advantages of their combination in hybrid models. In order to build and test the noise models, a dataset of propeller cavitating noise has been collected and processed, including relevant information on the input characteristics (i.e. propeller geometry, working point, ship wake description) and corresponding radiated noise. The experimental campaigns have been performed at the cavitation tunnel of the University of Genoa, considering three controllable pitch propellers in twin screw configuration. The dataset has been exploited to build different models of increasing complexity, to predict the radiated noise spectrum. The methodologies proposed allowed to obtain encouraging results providing a valuable basis for further investigations and developments of this approach.