Wingsail profile optimisation using computationally efficient methods
Wingsail profile optimisation using computationally efficient methods
On any race yacht, having the ability to maximise boat speed is key to obtain race winning performances. To achieve this the sail or wing must be set at its optimum profile. To find the best wingsail profile the trend recently has been towards more computationally expense approaches, but can we use less intensive methods to contribute to the design and optimisation process when time and resource may be limited? With an extensive number of different flying shapes, a computationally efficient approach at accurately finding optimum wingsail profiles for any given wind speed and direction is required. Using a two-dimensional section of the wingsail, lift and drag characteristics were found using Reynolds Averaged Navier-Stokes (RANS) simulations within Star-CCM+. A modified lifting line (LL) model was programmed in Python which used the two-dimensional characteristics to give fast and accurate predictions of drive force and heeling moment for a twisted inflow. The LL code was verified using experimental data, and showed that with analytical corrections, accurate predictions of lift and induced drag could be obtained. 3D RANS simulations confirmed that the LL model with correct tuning of the root vortices could predict driving forces and heeling moments within 1% and 5% respectively for a typical range of angle of attacks (AoA) and wing shapes. LL predictions took ~8 seconds on a laptop compared to ~6 hours for 3D RANS simulations running on a High-Performance Computing cluster. A machine learning algorithm using Kernel ridge multivariate regression was trained to produce a surrogate model of the wingsail giving accurate predictions within 1% of the LL results. Using the surrogate model, performance predictions could be obtained in ~0.001 seconds showcasing the large computational savings. This method permitted an exhaustive search of different wingsail profiles, giving information on parameter trends such as AoA, camber, and twist. This provides a tool that could be adopted in a velocity prediction program (VPP) and used by sailors or designers to aid in the setup and trimming of wingsails for maximum performance.
Birch-Tomlinson, William
376924a2-9016-4156-a4bc-2a260a9b5810
Turnock, Stephen
d6442f5c-d9af-4fdb-8406-7c79a92b26ce
Prince, Martyn
b436764c-0f28-4e13-aaa8-2d3cddf5f1f0
10 June 2022
Birch-Tomlinson, William
376924a2-9016-4156-a4bc-2a260a9b5810
Turnock, Stephen
d6442f5c-d9af-4fdb-8406-7c79a92b26ce
Prince, Martyn
b436764c-0f28-4e13-aaa8-2d3cddf5f1f0
Birch-Tomlinson, William, Turnock, Stephen and Prince, Martyn
(2022)
Wingsail profile optimisation using computationally efficient methods.
The 24th Chesapeake Sailing Yacht Symposium, United States Naval Academy , Annapolis, United States.
10 - 11 Jun 2022.
20 pp
.
(doi:10.5957/CSYS-2022-007).
Record type:
Conference or Workshop Item
(Paper)
Abstract
On any race yacht, having the ability to maximise boat speed is key to obtain race winning performances. To achieve this the sail or wing must be set at its optimum profile. To find the best wingsail profile the trend recently has been towards more computationally expense approaches, but can we use less intensive methods to contribute to the design and optimisation process when time and resource may be limited? With an extensive number of different flying shapes, a computationally efficient approach at accurately finding optimum wingsail profiles for any given wind speed and direction is required. Using a two-dimensional section of the wingsail, lift and drag characteristics were found using Reynolds Averaged Navier-Stokes (RANS) simulations within Star-CCM+. A modified lifting line (LL) model was programmed in Python which used the two-dimensional characteristics to give fast and accurate predictions of drive force and heeling moment for a twisted inflow. The LL code was verified using experimental data, and showed that with analytical corrections, accurate predictions of lift and induced drag could be obtained. 3D RANS simulations confirmed that the LL model with correct tuning of the root vortices could predict driving forces and heeling moments within 1% and 5% respectively for a typical range of angle of attacks (AoA) and wing shapes. LL predictions took ~8 seconds on a laptop compared to ~6 hours for 3D RANS simulations running on a High-Performance Computing cluster. A machine learning algorithm using Kernel ridge multivariate regression was trained to produce a surrogate model of the wingsail giving accurate predictions within 1% of the LL results. Using the surrogate model, performance predictions could be obtained in ~0.001 seconds showcasing the large computational savings. This method permitted an exhaustive search of different wingsail profiles, giving information on parameter trends such as AoA, camber, and twist. This provides a tool that could be adopted in a velocity prediction program (VPP) and used by sailors or designers to aid in the setup and trimming of wingsails for maximum performance.
Text
CSYS Paper Tomlinson_Turnock_Prince
- Accepted Manuscript
More information
Published date: 10 June 2022
Additional Information:
Publisher Copyright:
© 2022 SNAME 24th Chesapeake Sailing Yacht Symposium, CSYS 2022. All rights reserved.
Venue - Dates:
The 24th Chesapeake Sailing Yacht Symposium, United States Naval Academy , Annapolis, United States, 2022-06-10 - 2022-06-11
Identifiers
Local EPrints ID: 469573
URI: http://eprints.soton.ac.uk/id/eprint/469573
PURE UUID: 7d4b9aa3-252f-458a-b2fd-0da6c0b98f17
Catalogue record
Date deposited: 20 Sep 2022 16:40
Last modified: 17 Mar 2024 02:35
Export record
Altmetrics
Contributors
Author:
William Birch-Tomlinson
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics
