Development of Machine-Learning Surrogates for Hydrodynamic Performance and Wake-Field Prediction of Windships

M. Reche-Vilanova, D. Morris, H. Ward, R. Azcueta, M. Leslie-Miller, H. B. Bingham

Research output: Contribution to conferencePaperResearchpeer-review

16 Downloads (Pure)

Abstract

Current hydrodynamic modelling of Windships often uses either simplistic mathematical models or exhaustive analyses of specific hulls. This study proposes an intermediary solution using machine learning-based surrogate models to provide rapid, accurate predictions for various ship designs and sailing states. These models estimate forces, moments, and wake-fields experienced by different hull types under typical Windship conditions implying non-zero heel and leeway angles. The main surrogate model predicts hydrodynamic forces and moments, while two complementary models assess inflow speed and angle at the propeller and rudder. Results show that the surrogates match the CFD accuracy with less than 1% error for the generated hulls, capturing complex physics with minimal input data. Using a dataset from 39 systematically varied ship hulls, three Neural Networks were trained to develop these surrogates. These models offer rapid estimates without detailed knowledge of the ship geometry and expensive CFD simulations, making them ideal for early-stage design and integration into Performance Prediction Programs (PPPs).
Original languageEnglish
Publication date2024
Number of pages13
Publication statusPublished - 2024
EventWind Propulsion 2024 - International Maritime Organization's Headquarters, London, United Kingdom
Duration: 22 Oct 202423 Oct 2024

Conference

ConferenceWind Propulsion 2024
LocationInternational Maritime Organization's Headquarters
Country/TerritoryUnited Kingdom
CityLondon
Period22/10/202423/10/2024

Fingerprint

Dive into the research topics of 'Development of Machine-Learning Surrogates for Hydrodynamic Performance and Wake-Field Prediction of Windships'. Together they form a unique fingerprint.

Cite this