Computational Fluid Dynamics (CFD) nowadays provides the full capability to evaluate wind flow in a complex domain, like mountain regions or urban areas. CFD results strictly depend on domain conformation, fluid characteristics and boundary conditions. Several studies have validated the use of CFD to faithfully represent similar phenomena. But all this comes with some restrictions. In fact CFD requires onerous computational time, it produces a huge quantity of data, heavy to handle, and its numerical solutions are function of the selected boundary conditions, like wind direction or intensity, therefore its applicability is restricted to those conditions.
All those limitations imply that CFD is not suitable for real-time simulation, then to take advantage of CFD analysis, it is mandatory to overcome limitations CFD comes with.
Therefore, to overcome the long calculation time, inadequate to match the millisecond long time-frame of real-time simulation, we have decided to run the CFD numerical computation stand-alone, and save the respective result data to use them later, when the simulation is running. Moreover, to model wind flow dependence to boundary conditions (that are all the possible wind directions and velocities affecting the domain), all the conditions have been condensed to a finite set of wind directions and intensities, representative of the phenomenon. In this way a finite number of CFD numerical simulations, to be computed stand-alone, was identified.
The analysis has been performed with OpenFOAM, an open source CFD software package for Navier-Stokes equations solution, which also provided pre-processing/meshing and post-processing tools. This approach has been successfully demonstrated over a region of the Swiss Alps, about 4000 billion cubic meters wide. Indeed, to deal with the huge amount of CFD results, outcomes of about 1000 hours of computer calculation, a relational database has been set up. To handle the great quantity of data within real-time constraints, DB has been assisted by a cache system. In fact, a suitable algorithm has been implemented to extract from DB only a subset of CFD data, related to the subdomain actually needed by real-time simulation in the meanwhile. This approach optimized DB query performances with asynchronous DB interrogation, in order to meet simulation time-frame.
Besides, to include turbulent time-variant wind component in the model, a Dryden wind turbulence model was added to steady-state CFD results. Furthermore, the complete model was predisposed for man in the loop approach, for simulation validation and tuning, to be carried out by domain-expert users.