CFD analysis of vehicle underhood flows is not new at GM, however, this is the first time GM engineers have modeled the full vehicle in a single simulation. Previously, the front-end, underbody, and underhood portions of the vehicle were modeled seperately with the results from each simulation specified as boundary conditions for other simulations. Such an approach is practical, but introduces errors because key interactions between the components are neglected.

Furthermore, the current analysis models several components directly which were previously specified only through static boundary conditions. These include the exhaust system, radiator heat rejection, radiator fan, and air density variation with temperature.

One important component, the engine exhaust system, was completely modeled starting just after the engine exhaust ports. The engine exhaust gas temperature and flow rate was specified from a proprietary GM code. From there, the exhaust system skin temperature was calculated as a coupled system with the underhood flow and included convection, conduction, and radiation heat transfer modes. This effort was aided by the CFD solver's planar conduction model which allows for thin-wall conduction without having to actually model the wall thickness.

The radiator fan, which was previously approximated as a constant swirling flow boundary condition, included the blades and hub geometry in a rotating frame of reference and utilized a sliding interface between it and the underhood flow.

In a testiment to the interoperability of today's design software, the GM engineers utilized Unigraphics to generate a solid model of the vehicle, structural analysis software ANSA to generate a full vehicle surface mesh, TGrid from Fluent to generate the volume mesh, and the Fluent CFD solver to generate the solution.

The CFD solver was run on 8 processors of a 64 processor SGI Origin computer. The solution used approximately 8GB of RAM and was completed in 14 hours.

The payoff for modeling the entire system in a single analysis, was reduced reliance on test data for boundary conditions, a reduction of analysis time by 50%, and an increase in solution accuracy.

The predicted flowfield matched experimental results to within 5 percent and the predicted temperatures correlated within 10 to 15 precent.

The information gathered from the full vehicle thermal load analysis allows GM engineers to generate thermal design specifications before testing, thereby reducing overall time to market.

This application demonstrates several aspects of the future of CFD analysis. First is the use of several software packages from different companies during the course of the analysis. Good software interoperability lets the design engineer pick the right tool for the job without incurring loss of data when moving along the analysis stream. Second, the analysis demonstrates the convergence of analysis models into complete systems. By modeling the system as a whole, key interactions are captured and accuracy reducing assumptions are avoided. Next is the inclusion of simplified models coupled with the overall analysis. In this model, the thin-wall conduction in the exhaust system was modeled without having to resolve the wall thickness. Such approximations are valid and, because they are coupled with the rest of the system, do not break the overall conservation balance. Lastly, while not being a new technology, the increasing use of parallel processing in CFD solvers is paving the way for whole-system analysis such as this one.