Abstract
Estimating the sailing resistance of Unmanned Submersible Vehicles (USVs) is a critical step in hull profile design. Since USVs operate entirely submerged, they experience no wave resistance; instead, drag is dominated by friction and viscous pressure resistance. This paper presents a resistance optimization system that replaces computationally intensive Computational Fluid Dynamics (CFD) simulations with surrogate models developed using the Response Surface Method (RSM). By evaluating five polynomial function forms, the study identifies that second-order (quadratic) response surfaces with cross-terms provide the optimal balance of accuracy and computational efficiency.
Introduction
The design of efficient underwater vehicles necessitates precise modeling of hull resistance. For USVs, the primary components of drag are friction resistance and viscous pressure resistance, influenced by parameters such as bow length, stern length, middle body length, molded breadth, depth, and rudder aspect ratio. Traditionally, optimizing these parameters requires iterative CFD simulations, which are resource-heavy and time-consuming.
To overcome this bottleneck, the authors propose a mathematical "surrogate model" capable of predicting resistance values nearly instantaneously. This approach decouples complex fluid analysis from the optimization process, allowing for rapid exploration of the design space.
Methodology
The core of the proposed system is the Response Surface Method (RSM), an approximation technique that maps the relationship between design variables and objective functions based on a limited set of data points.
The researchers utilized a dataset generated for the AUTOSUB AUV, consisting of 16 distinct geometric models. CFD simulations were performed on these models to establish a baseline for resistance. Five polynomial forms were tested to determine the best fit for the physics of the hull:
- Linear equations
- Quadratic equations without cross-terms
- Quadratic equations with cross-terms
- Cubic equations without cross-terms
- Cubic equations with cross-terms
The optimization framework uses the Admiralty coefficient as the primary objective function, subject to constraints like buoyancy and hydrodynamic torque.
Results and Discussion
The study revealed significant differences in the predictive accuracy of the various RSM models. A key finding was the "Goldilocks" complexity of the mathematical models: higher-order equations did not necessarily yield better results.
As shown in the performance metrics, the quadratic equation with cross-terms emerged as the top performer. It achieved the highest correlation coefficient ($R^2$) and the lowest Root Mean Square Error (RMSE). Interestingly, cubic equations without cross-terms exhibited the lowest fitting accuracy, highlighting that the interdependencies between variables (e.g., how bow length interacts with molded breadth) are essential for accurate modeling.
Furthermore, the response surface for the hull's "wet area"—a primary driver of friction—showed exceptionally high fitting accuracy across all models, suggesting that geometric properties are more easily approximated than complex fluid behaviors.
Conclusion
The application of the polynomial response surface method effectively streamlines the USV resistance optimization process. By identifying the second-order response surface as the most accurate surrogate, this research provides a foundation for more complex multi-disciplinary optimization. Future work may involve expanding the dataset beyond 16 points to further refine higher-order models and testing the framework on unconventional hull geometries.
Summary updated