noiseNet: A neural network to predict marine propellers’ underwater radiated noise

A dedicated neural network architecture called noiseNet has been developed to predict URN (Underwater Radiated Noise) of cavitating marine propellers. The noiseNet predicts the sound pressure level at the first three blade passing frequencies with knowing the propeller geometry, ship hull wake field and working conditions. The physical mechanism of the URN generation is firstly analyzed. Thereafter, the physical knowledge about the hydrodynamics and hydroacoustics of marine propellers are used to develop the noiseNet architecture. A dataset obtained with the boundary element method and Ffowcs Williams–Hawkings acoustic analogy is used for the training and evaluation. The evaluation conducted on fully unseen cases shows a mean absolute error of 7.34 dB. [Display omitted] •A dedicated neural network architecture for predicting marine propellers’ noise.•Hydrodynamics on multiple radius sections and their interactions are fully considered.•The architecture can be adjusted for other predictions about rotation machines.•Test is conducted on fully unseen propellers and wake fields.