An efficient and accurate overset grid technique applied to maritime CFD

Abstract

Ship manoeuvring and other maritime applications pose complex challenges involving the motion of multiple bodies. The overset method enables simulations of such cases using computational fluid dynamics (CFD) by allowing arbitrary movement of any number of meshes. However, this versatility comes at a cost in terms of performance and accuracy. This research, therefore, aims to provide a modern, efficient and accurate overset method and study its reliability using advanced Verification techniques. To this end, a novel overset method that includes a wide variety of interpolation schemes has been implemented and is the basis for quantitative and qualitative error analysis studies performed on several test cases, as well as an investigation on computational performance.

Quantifying local and global errors, convergence orders, and mass imbalance for different interpolation schemes on several test cases allows to draw guidelines for both overset method users and developers. Moreover, the range of test cases used, such as a realistic 3D unsteady RANS manufactured solution of a recirculation bubble or a rudder-propeller flow, allows to extend the conclusions to many maritime applications. The results show that the choice of interpolation scheme has a significant impact on accuracy. First order schemes lower the overall convergence order of the discretisation error, increasing the amount of artefacts visible in the field and resulting in larger high-frequency temporal oscillations. Higher order schemes such as second and third order accurate ones help to limit the error production and maintain second order accuracy of the solver's discretisation scheme. However, the limitations of first order schemes do not come with a significant reduction in computational overhead. In fact, the second order nearest cell gradient is even cheaper than the first order inverse distance scheme. Finally, the third order least squares scheme, though more expensive, is still a viable option as it shows only a 8% performance overhead after several sequential and parallel programming performance tuning operations. In addition to providing guidelines and analysis methodologies, this work has also produced two opensource tools aimed at helping Verification, with the automatic generation of manufactured solutions and the computation of statistical uncertainties.

More information

Identifiers

ORCID for Sebastien Lemaire: orcid.org/0000-0002-9959-2100

ORCID for Stephen Turnock: orcid.org/0000-0001-6288-0400

Catalogue record

Export record