Computational Steering of CFD Simulations

In computational fluid dynamics (CFD) analyses, the tasks of simulation and visualization are usually done as independent steps - visualization following numerical simulation. In a typical procedure, the researcher first performs a calculation for some fixed amount of computational time, and then views the solution in a separate post-processing step, using a stand-alone visualization system. Based on the solution at this intermediate stage, a decision may be made to change simulation parameters, regrid, modify the modeled geometry, or simply continue the computation unchanged from that point. This process is repeated until the solution of a steady problem has converged, or until the solution of an unsteady problem is evolved to a given point in time.

We believe that it is more efficient, and more enlightening, to perform both steps - simulation and visualization - at the same time. A special-purpose visualization system that can be directly coupled with a running simulation, e.g., via a network connection, and that allows us to visualize "live" data as it is produced by the simulation, can be used to observe the simulation progressing towards its final result, instead of just visualizing that result. Such a system is not only a valuable tool for debugging a simulation under development, but it can also help in understanding the phenomenon being simulated, by showing how the simulation arrived at the final result. Even when a simulation code is already well-established, a specific simulation can still go awry, for example if inappropriate simulation parameters are specified. Instead of having to wait for a run to finish, and then finding out that the result is not as expected, it is often possible to detect that a simulation will yield suboptimal results early on and interrupt it, or sometimes even to improve its solution by changing parameters on-the-fly. In situations like this, a coupled simulation-visualization system can reduce the overall computation time significantly.
Figure 1: Parabolic grid with exponential cell size growth used in shock wave reflection simulation. The finest-resolution central grid area is located at the intersection of the horizontal and vertical white bands.

Project Goals

The main goal of this project was to design an interactive visual exploration program capable of handling "live" data that could be used to steer a running CFD simulation code. This included the following detail goals:

Project Status

The steering framework was developed in collaboration with Allen M. Tesdall, then a post-doctoral researcher at the Center for Image Processing and Integrated Computing (CIPIC) researching the focusing and reflection of "weak" shock waves. After implementing the generic visualization program, specific data set structures to represent rectilinear and parabolic curvilinear grids were developed to visualize live data from his simulation code, and a simulation-side data export module was integrated into his code. As main steering technique, the visualization program allows to change the refinement parameters of a simulation grid to follow moving features and always resolve them at the highest possible level of resolution. Development reached a stable state in late 2001, when a paper was written and submitted to the Joint EUROGRAPHICS-IEEE Symposium on Visualization. The paper was later presented at the VisSym 2002 conference in Barcelona, Spain.

Related Publications

Pages In This Section

Screen Shots