Figure illustrates the different wake patterns (2P, 2P+2S, 2P+2S2, 4P and 4P+2S from left to right, where S and P denote a single vortex and counter rotating vortex pair, respectively) generated from variations in pitching angle and amplitude of a foil flapping in a uniform free-stream. These multi-million GPU-accelerated viscous vortex particle simulation runs were performed on a single Tesla K40 GPU with 2880 cores.
The recent thrust on miniaturization of micro air vehicles and unmanned air vehicles has necessitated a reanalysis of the conventional steady thrust generation systems and new look into unconventional unsteady ones. In 2017, Prof. Raghuraman Govardhan’s lab of Mechanical Engineering department of IISc used 6.34 million core hours of SahasraT to perform such studies. Conventional thrust generation systems rely on fixed wings for generating the thrust and lift necessary for forward flight and rapidly lose their efficacy at small sizes. In contrast, unsteady flapping wing configurations find widespread prevalence in animal locomotion. It is therefore important to investigate unconventional unsteady thrust generating mechanisms that are typically observed in nature. In this work we aim to develop an advanced understanding of the unsteady thrust generation from a biomimetic unsteady pitching foil configuration. Our overarching objective is a feasibility analysis and optimization of this configuration for eventual utilization in micro air vehicle/unmanned air vehicle propulsion.
About ten K-40 GPU cards on the Cray SahasraT system were used for the experiments. A high-resolution GPU-accelerated viscous vortex particle method was utilized for simulation of fluid-structure interaction problem associated with pitching foil configuration. The method relies on vortex blobs for flow field representation and hence problems similar to the well-known N-Body problem are encountered about 8 times per time step of flow evolution. The implementation used the Fast Multipole Method to reduce the time complexity of the N-Body algorithm from O(N2) to O(N).
The test runs indicated that compared to conventional implementation on multicore CPU architectures the lab’s GPU-accelerated implementation allowed perform computations on thousands of cores that are available on modern general purpose graphics processing units and thus achieve over an order of magnitude speed up (compared to an OPENMP implementation on 12 cores, over a ten fold speed up was achieved using a single Tesla K40 GPU with 2880 cores).
A. Das, R.K. Shukla, and R.N. Govardhan, Self-propulsion of a pitching foil, Bulletin of the American Physical Society, 2017, 70th Annual Meeting of the APS Division of Fluid Dynamics Volume 62, Number 14.