A novel hull form optimization framework based on multi-fidelity deep neural network

With the advancements in computer technology, the simulation-based design (SBD) technology has emerged as a highly effective method for hull form optimization. The SBD approach often employs various methods to evaluate the hydrodynamic performance of the sample ships. Although the surrogate model is applied to SBD method to replace time-consuming evaluation, many high-fidelity data are typically required to guarantee the accuracy of the surrogate model, resulting in significant computational costs. To improve the optimization efficiency and reduce computational burdens, we propose a novel hull form optimization framework utilizing the multi-fidelity deep neural network (MFDNN), leveraging multi-source data fusion and transfer learning. This framework constructs an accurate multi-fidelity surrogate model which correlates design parameters with hydrodynamic performance of the hull by blending data with different fidelity. Besides, computational fluid dynamics (CFD) evaluations based on viscous flow are served as the high-fidelity model, while potential-theory evaluations represent the low-fidelity model. Then, this framework is validated using mathematical functions to prove its practicability in optimization. Finally, the optimization design of the resistance of the DTMB-5415 ship is carried out. Our findings demonstrate that this framework can take into account both efficiency and accuracy, which is preferable in optimization tasks. The optimized hull form obtained by the framework has better resistance performance.