Run Time Performance |
 
|
Run Time Performance |
|
Full details of the numerical models and the benchmarking exercise may be found in the technical paper (Britton et. al., 2025). In addition to comparing solution accuracy, the stepped wind test case was used to compare the run-time performance of each OpenFAST, OrcaFlex and Flexcom. Run-time comparisons are facilitated by command prompt batch files, which ensure consistency and a valid comparison. Each dynamic simulation is performed on its own, with no other simulations competing for CPU resources, affording the solvers a good chance of achieving the quickest possible run time.
Despite some differences in the total number of elements used in the models – largely due to the independent nature of blade property interpolation from the original WindIO dataset – there is significant variation in computation times across different codes. For the stepped wind case, which has a total simulation time of 1060 s, OpenFAST with its ElastoDyn solver returns the quickest run time of just 2:09 (mm:ss), resulting in a computation time to simulation time ratio of just 12%. Notwithstanding the fact that the model does not include a support structure such as a monopile or floating platform, this is exceptionally quick. It should be noted however that the ElastoDyn solver is strictly only valid for straight isotropic blades dominated by bending, so its modelling accuracy is limited. Next quickest is Flexcom, which completes in 5:13 (mm:ss), giving a computation to simulation time ratio of 30%. Although Flexcom adopts the OpenFAST aerodynamic solver, it uses its own structural solver, a non-linear finite element formulation with fully coupled axial, bending and torsional DOFs, with greater computational effort. Next comes OrcaFlex, which completes in 8:03 (mm:ss), offering a computation to simulation time ratio of 46%. This is somewhat unexpected given that the OrcaFlex model has fewer blade structural elements than the Flexcom model (37 rather than 58) so run times would be expected to be quicker than Flexcom. Given that both aerodynamic solvers use BEMT, this suggests that the Flexcom solution technique is more efficient than that of OrcaFlex, an observation which coincidentally aligns with a previous study on mooring line modelling (Connolly et al., 2025). OpenFAST with BeamDyn returns the slowest (significantly) run time of 4:23:43 (hh:mm:ss), equating to a computation to simulation time ratio of almost 1500%. The comparatively poor performance of OpenFAST-BeamDyn is due to the small time-step of 0.001s used in the BeamDyn model (necessary to avoid numerical convergence issues); the other models use a time-step of 0.05 s. According to NREL, a new tight coupling algorithm is under development and will enable much larger time-steps and significantly improved computational performance for BeamDyn (a new release is expected in 2025).
Please note that this run-time performance exercise was performed in 2025 so the times above may not necessarily reflect the exact computation speeds of the modelling tools as of today.
