Multi-fidelity hydrodynamic analysis of an autonomous surface vehicle at surveying speed in deep water subject to variable payload

Riccardo Pellegrini, Simone Ficini, Angelo Odetti, Andrea Serani, Massimo Caccia, Matteo Diez

TL;DR

This paper presents a multi-fidelity (MF) hydrodynamic analysis of an autonomous surface vehicle (ASV) called SWAMP, focusing on how payload and center of mass variations affect its performance. By combining high-fidelity (RANSE) and low-fidelity (potential flow) simulations with a Gaussian Process model, they efficiently predict the ASV's resistance, attitude, and wave generation, offering insights for optimizing its design and operation.

What Problem Does This Paper Address?

Autonomous Surface Vehicles (ASVs) are increasingly used for various tasks, from environmental monitoring to security. However, understanding their hydrodynamic performance under different operational conditions, such as varying payloads (equipment for surveys) and the resulting changes in the center of mass, is crucial but often under-investigated. Traditional high-fidelity simulations for these analyses are computationally expensive, limiting their use for design optimization. This paper aims to bridge this gap by developing an efficient multi-fidelity approach to accurately predict ASV performance across a range of operational parameters.

Methods

The core of this research lies in a multi-fidelity (MF) approach to hydrodynamic analysis. They employ two main solvers:

  • High-Fidelity Solver: A Reynolds-Averaged Navier-Stokes (RANS) solver (CFDShip-Iowa V4.5) is used for detailed simulations, capturing complex flow phenomena.
  • Low-Fidelity Solver: A linear potential flow (PF) solver (WARP) is used, offering faster but less accurate predictions. This solver is further diversified into three fidelity levels by varying grid size and coupling strength.

These diverse simulation results are then integrated using a Multi-Fidelity Gaussian Process (MF-GP) model. This surrogate model is adaptively refined using an active learning approach to intelligently select new simulation points, focusing on areas of high uncertainty, thereby minimizing computational cost while maximizing accuracy. The analysis focuses on resistance, trim, sinkage, and free-surface elevation.

Datasets

Not discussed in the paper. The study relies on numerical simulations rather than experimental datasets for training and evaluation, though it references experimental data for validation of the solvers.

Key Insights and Ideas

  • Payload and Center of Mass are Critical: The study highlights that both the vehicle's payload and the precise location of its center of mass significantly impact its hydrodynamic performance, particularly resistance and attitude.
  • Limitations of Potential Flow: The potential flow solver, while fast, struggles to accurately predict wave elevation and pressure distribution, especially in the stern region and for the free surface between the hulls. It also fails to capture vortical structures.
  • Value of Multi-Fidelity: Combining high- and low-fidelity simulations with an MF-GP model proves effective in generating accurate response surfaces for ASV performance with significantly reduced computational cost compared to solely relying on high-fidelity methods.
  • Active Learning for Efficiency: The active learning strategy is crucial for efficiently guiding the MF-GP model's refinement, ensuring that computational resources are spent on the most informative simulation points.
  • Optimizing for Different Needs: The analysis reveals that the center of mass can be strategically adjusted to either minimize resistance (for longer range) or reduce wave elevation between the hulls (for better survey accuracy), depending on mission requirements.

Key Results

  • The RANSE solver generally provides good accuracy in predicting resistance, while the PF solver tends to overestimate it.
  • The PF solver predicts trim angles with an opposite sign compared to RANSE.
  • Both solvers overestimate sinkage similarly.
  • The MF-GP model successfully generates response surfaces for resistance, sinkage, trim, and free-surface RMS, showing how these metrics vary with payload and center of mass.
  • For reduced payloads, the center of mass can be calibrated to minimize resistance.
  • The MF approach significantly reduces the computational cost of exploring the operational space compared to brute-force high-fidelity simulations.

Limitations and Future Work

  • Calm and Deep Water: The analysis is limited to calm and deep water conditions. The authors note that the ASV is not currently intended for significant wave heights, but future work could extend the analysis to include wave effects.
  • PF Solver Limitations: The PF solver's inability to accurately predict wave elevation, pressure distribution, and vortical structures is a significant limitation, especially for tasks requiring precise free-surface interaction understanding.
  • Future Work: The authors propose extending the MF hydrodynamic analysis to variable speed conditions to provide a more comprehensive understanding of the ASV's performance across a wider range of operational scenarios.

Summary updated

BibTeX
@inproceedings{pellegrini2023multi,
  author = {Riccardo Pellegrini and Simone Ficini and Angelo Odetti and Andrea Serani and Massimo Caccia and Matteo Diez},
  title = {{Multi-fidelity hydrodynamic analysis of an autonomous surface vehicle at surveying speed in deep water subject to variable payload}},
  booktitle = {Ocean},
  year = {2023},
  doi = {10.1016/j.oceaneng.2022.113529},
}
Figure