Rapid Prediction of Surface Flow Fields for Marine Propellers via Deep Learning

Existing surrogate models for marine propeller hydrodynamics efficiently predict integral coefficients but are limited in providing detailed flow field information, restricting their use in cavitation assessment and refined optimization. This study proposes a conditioned deep learning framework for the rapid prediction of blade surface flow fields and open-water performance of marine propellers under varying geometric and operating conditions. The model takes local geometric information of a single blade as input, while the blade number and advance coefficient are jointly encoded into a condition embedding vector. This vector dynamically modulates convolutional features to enable flow field prediction on blade surfaces. A dataset comprising 2710 operating cases across 560 propeller samples with 3-, 4-, and 5-bladed configurations is constructed for model training and evaluation. Results demonstrate that the proposed model achieves high accuracy in predicting both surface flow field and derived hydrodynamic coefficients, while significantly improving computational efficiency compared to CFD simulations. Furthermore, the model yields reasonable predictions for new propeller geometries within the covered design space, demonstrating interpolation generalization capabilities. The proposed framework provides a promising tool for rapid hydrodynamic assessment and refined propeller optimization.