Computational Fluid Dynamics is a wide field, where each problem has to be assesed individually in order to find the best numerical tool to tackle it. The diversity of problems found in nowaday's applications makes it impossible to develop a single strategy that could be used to improve simulations in every single application.
With this idea in mind, it was decided to take three distinct paths to cover a broader field of applications.
The results presented in the section of GPU solvers
are mostly suited for general fluid simulation applications. They represent a direct alternative to standard fluid solvers, which provide excellent computer times while remaining accurate. However the disadvantage of this method is the requirement of structured meshes.
In the Large Time Steps
section a different strategy is presented. Using unstructured triangular (or tetrahedrons) elements, it is possible to obtain completely arbritriary meshes. Using a hybrid Lagrangian-Eulerian strategy combined with streamline integration it is possible to dramatically increase the timesteps and therefore reduce the computational times.
Finally, in the Reduced Order Modelling
section a completely different approach is presented. This method is best suited when several simulations have to be run with only slight changes in the geometry or boundary conditions. The method consists on first "training" the sover running by the full simulation for a limited number of cases, and then it is possible to solve a wide range of conditions with a negligible computational cost.