Sample-Efficient and Surrogate-Based Design Optimization of Underwater Vehicle Hulls

Overview

Peer-reviewed journal publication demonstrating Bayesian optimization and deep neural network-based surrogate modeling for UUV hull design. This is the first study applying both BO and DNN surrogates to underwater vehicle hull optimization.

Key Results

• DNN surrogate achieves 1.85% mean absolute percentage error compared to full CFD
• Two-orders-of-magnitude speedup over traditional CFD-in-the-loop optimization
• Bayesian optimization shown as the most sample-efficient method compared to genetic algorithms, random search, and grid search

Approach

1. Generate training data via OpenFOAM CFD simulations over a design space
2. Train a DNN surrogate to predict drag force from hull geometry parameters
3. Use the surrogate within a Bayesian optimization loop for rapid convergence
4. Validate optimal designs with full CFD simulation