Exploring unseen 3D scenarios of physics variables using machine learning-based synthetic data - an application to wave energy converters

Aug 6, 2026

Upload by Finn

This work applies machine learning to generate synthetic data for wave energy converters from time-expensive 3D computational fluid dynamics simulations. By using a 􀀂-VAE together with a Principal Components-based adversarial autoencoder, the study compresses simulation information while producing new samples of dynamic viscosity and velocity fields. The resulting surrogate model accelerates generation by 5–6 orders of magnitude and expands the design space, enabling improved prediction of dynamic viscosity from velocity fields. The generative framework can also represent transitions and new physical phenomena under extreme initial conditions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Abstract

1. Introduction

City Research Online City St George’s, University of London Citation: Quilodrán-Casas, C. , Li, Q. , Zhang, N. , Cheng, S. , Yan, S. , Ma, Q. & Arcucci, R. (2024) . Exploring unseen 3D scenarios of physics variables using machine learning-based synthetic data: An application to wave energy converters. Environmental Modelling & Software, 177, 106051. doi: 10.1016/j.envsoft.2024.106051 This is the accepted version of the paper. This version of the publication may differ from the final published version. To cite this item please consult the publisher's version. Permanent repository link: [https://openaccess.city.ac.uk/id/eprint/33469/](https://openaccess.city.ac.uk/id/eprint/33469/)[ ](https://openaccess.city.ac.uk/id/eprint/33469/)Link to published version: [https://doi.org/10.1016/j.envsoft.2024.106051](https://doi.org/10.1016/j.envsoft.2024.106051) [Copyright and Reuse:](Copyright and Reuse: Copyright and Moral Rights remain with the author)[ Copyright and Moral Rights remain with the author](Copyright and Reuse: Copyright and Moral Rights remain with the author)([s](s)) and/or copyright holders. Copies of full items can be used for personal research or study, educational, or not-for-profit purposes without prior permission or charge, unless otherwise indicated, provided that the authors, title and full bibliographic details are credited, a hyperlink and/or URL is given for the original metadata page and the content is not changed in any way. For full details of reuse please refer to City Research Online policy. City Research Online: [http://openaccess.city.ac.uk/](http://openaccess.city.ac.uk/ publications@citystgeorges.ac.uk)[ ](http://openaccess.city.ac.uk/ publications@citystgeorges.ac.uk)[ publications@citystgeorges.ac.uk](http://openaccess.city.ac.uk/ publications@citystgeorges.ac.uk) 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 1 38 2 39 3 40 4 4 4 4 4 4 41 41 41 51 51 52 15 53 16 54 17 55 18 56 19 57 58 59 60 61 62 63 64 Exploring unseen 3D scenarios of physics variables using machine learning-based synthetic data: an application to wave energy converters César Quilodrán-Casasa,b , Qian Lic , Ningbo Zhangc , Sibo Chenga , Shiqiang Yanc , Qingwei Mac and Rossella Arcuccia,b a Data Science Institute, Imperial College London, London, United Kingdom b Department of Earth Science and Engineering, Imperial College London, London, United Kingdom c City University, London, United Kingdom | ARTICLE INFO | | ABSTRACT | | --- | --- | --- | | Keywords:
Generative models Synthetic data
Wave energy converters Model surrogate | | This work aims to use machine learning to produce synthetic data of wave energy converters from time-expensive 3D simulations based on computational ﬂuid dynamics models. Whilst the potential to study, understand and take advantage of these converters for renewable energy is immense, the simulations to analyse the response of these systems to incoming waves are lengthy and computationally expensive to obtain. Here, we explore the use of a 􀀂-VAE and a Principal Components-based adversarial autoencoder for generating new synthetic data. The compression plus the generation of synthetic data introduces an exceptionally fast model surrogate of the original simulation and delivers more samples of either dynamic viscosity and velocity ﬁelds, enlarging the design space. The new generated synthetic samples can have a speed up from 5 to 6 orders of magnitude. The newly design space can be used to improve the prediction of dynamic viscosity given the velocity ﬁelds. The generative model has the potential to capture the transition and the new physical phenomena under extreme initial conditions. | | 1. Introduction
Due to the increasing demand for clean energy, various renewable energy

FAQs

What problem does the study address?

The study targets the high computational cost and time expense of 3D CFD simulations used to analyze wave energy converter responses to incoming waves.

Which machine learning models are used to generate synthetic data?

It uses a 􀀂-VAE and a Principal Components-based adversarial autoencoder to generate new synthetic samples.

How do the generated synthetic samples help the scientific goal?

They create a fast surrogate model, provide more samples of dynamic viscosity and velocity fields, expand the design space, and improve prediction of dynamic viscosity from velocity fields.

Aug 6, 2026

Upload by Finn