Generative Design of Ship Propellers using Conditional Flow Matching

TL;DR

Our work tackles inverse design of ship propellers by learning a conditional flow mapping $q(\mathbf{p}\mid \mathbf{l})$ that, given target performance $\mathbf{l}=(\eta^*,J^*,k_T^*)$, generates diverse propeller geometries $\mathbf{p}$. We integrate a fully-parametric propeller dataset derived from CAESES geometry and OpenProp/VLM flow labels to train a Conditional Flow Matching model, a continuous normalizing flow that transports samples from a base distribution to the conditional target. The results demonstrate high fidelity between generated and target labels and substantial geometric diversity, with surrogate-based data augmentation enabling effective learning under limited data. This approach offers a practical pathway for rapid, data-efficient inverse design of propellers, while highlighting avenues for extension to more complex physics and hull interactions under realistic operating conditions.

Abstract

In this paper, we explore the use of generative artificial intelligence (GenAI) for ship propeller design. While traditional forward machine learning models predict the performance of mechanical components based on given design parameters, GenAI models aim to generate designs that achieve specified performance targets. In particular, we employ conditional flow matching to establish a bidirectional mapping between design parameters and simulated noise that is conditioned on performance labels. This approach enables the generation of multiple valid designs corresponding to the same performance targets by sampling over the noise vector. To support model training, we generate data using a vortex lattice method for numerical simulation and analyze the trade-off between model accuracy and the amount of available data. We further propose data augmentation using pseudo-labels derived from less data-intensive forward surrogate models, which can often improve overall model performance. Finally, we present examples of distinct propeller geometries that exhibit nearly identical performance characteristics, illustrating the versatility and potential of GenAI in engineering design.

Figures (13)

• Figure 1: Variation of Chord and Pitch with $P$ and $W_c$ (left) and $W_{rc}$ and $W_{rp}$ (right)
• Figure 2: Camber variation of NACA 66 profile
• Figure 3: An exemplary propeller created with CAESES
• Figure 4: Radial distribution functions of the propeller shown in \ref{['fig:prop']}
• Figure 5: Open Water Diagram of the propeller shown in \ref{['fig:prop']}