General Motors - Simulation to Identify and Resolve Underhood/Underbody Vehicle Thermal Issues

The most complete simulation to date of vehicle underbody and underhood components has helped to identify and resolve vehicle thermal issues early in the design phase. This recent computational fluid dynamics (CFD) analysis was one of the first to include all important fluid and thermal phenomena -geometry of all components, thermal as well as flow analysis, heat rejection from the radiator, thermal radiation from the exhaust pipe, a multi-reference model of the fan, and air density variation with temperature. The flow predicted in the simulation correlated with test results to within 5%, while temperatures correlated to within 10% and 15%. A key aspect of the simulation was the use of a software package (FLUENT) that made it possible to use planar conduction models for all components, dramatically reducing modeling time by eliminating the need to build prism layers in the mesh to represent component thickness.

Challenges in underhood/ underbody simulation

component thickness. While there's nothing particularly new about modeling the underhood or underbody of a truck or automobile with CFD, until now these simulations have been of limited use because they lacked one or more elements needed to provide the design information that is most critical at this stage. First of all, the front-end, underhood, and underbody are usually modeled and analyzed separately. While the results of each analysis are used as input to the other as boundary conditions, this approach reduces accuracy because of the lack of more accurate interaction between the two separate models. In most models, air density is assumed to be constant, rather than vary with temperature, which typically involves an error of about 10%. The majority of models also do not include thermal analysis, so they provide flow information only. The in-plane conduction feature that was recently introduced made the task of modeling planar conduction in the component walls easier.

In most models the exhaust skin temperatures are provided as inputs to the model. These temperatures are based on either experience or past experimental results. Prediction of these temperatures would be an important improvement to the model.

Preliminary modeling and analysis

The starting point for this analysis was a full vehicle model of the sport utility vehicle design that was created in the Unigraphics computer aided design system. GM engineers used Automatic Net-generation for Structural Analysis (ANSA) software from Beta CAE Systems, Thessaloniki, Greece, to produce a full vehicle surface model that includes all significant underbody and underhood components.

A constant speed and grade vehicle simulation model with the type of engine, fan, air conditioning losses, transmission, torque converter and shift maps that are characteristic of the design specifications was created to provide boundary conditions for the analysis. This model generates the engine speed, fuel flow rate, air/fuel (A/F) ratio and gear ratio needed as inputs for a power train cooling model. The engine cycle-average fuel+air flow rate is an input to the inlet of the exhaust manifold. Based on the fuel flow rate, the inlet temperature to the exhaust is determined using a power train cooling model. The power train cooling model produces exhaust gas temperatures, flow at manifold inlet values, and radiator and condenser heat rejections that are required for use as boundary conditions in the CFD model. Vehicle speed and fan rpm from the vehicle simulation model were also used as inputs to the analysis.

CFD simulation

A CFD simulation provides fluid velocity, pressure values, and temperature through the solution domain for time-dependent problems with complex geometries and boundary conditions. As part of the analysis, a designer may change the geometry of the system or the boundary conditions such as the inlet velocity or fan RPM, and view the effect on fluid flow patterns. CFD is an efficient tool for generating parametric studies with the potential of significantly reducing the amount of experimentation required to optimize the performance of a design.

In the past, it sometimes took as long to build a complex CFD model as it took to build a prototype. The reason is that the CFD codes available used a structured mesh approach that required that a block structure be defined before a volume mesh was generated. It took weeks or months to produce a grid for the analysis of complex underhood and underbody components using this approach. The modeling tools available were also too difficult to be used by design engineers who generally do not have the chance to gain the familiarity of a dedicated user. This meant that analysis had to be delegated to specialists inside the company or in outside consulting firms, which further increased the lead-time requirements. As a result, it typically took as long to build a CFD model as it took to construct a prototype.

Advantages of unstructured mesh

Recently, General Motors adopted an unstructured-mesh CFD tool with an intuitive graphical user interface that made it possible for the design engineer to perform the entire modeling process in much less time than before. The mesh generation software, TGrid from Fluent Inc., Lebanon, New Hampshire, generates an unstructured tetrahedral volume mesh from a triangular surface mesh. While tetrahedral meshes are useful for complex geometries, prism elements are more suitable for resolving boundary layers. The prism elements, extruded from surface triangles, allow a more accurate solution with fewer elements. TGrid can create a hybrid mesh consisting of prism layers in near wall regions and tetrahedral cells in the remainder of the domain. It can also layer prism elements at the walls. The result is better accuracy without the time-consuming task of building an all-hex mesh.

GM engineers imported their full body surface mesh into the TGrid preprocessor. The preprocessor created the volume mesh while correcting the complicated mesh geometry to avoid skewness problems that would have otherwise affected the accuracy of the analysis results. The tetrahedral generation scheme in TGrid is completely automated and provides automatic mesh refinement, requiring no intervention from the user. The mesh density was controlled to resolve important flow features locally while retaining a cost-effective coarse mesh in more uniform regions of flow. Sometimes the automatic procedure fails due to issues regarding the surface mesh and the surface mesh has to be modified manually. TGrid has specialized tools for this purpose.

Multiple frame of reference model

GM engineers modeled the fan using the multiple reference frames (MRF) model, a steady-state method that simplifies the analysis. This method evolved from an earlier CFD practice in which the entire fluid flow problem was solved in a single frame of reference attached to the rotating part. In this rotating frame, the velocity of the rotating part is zero and the boundaries are assigned a rotational speed opposite that of the reference frame. Unfortunately, this rotating frame model is not valid for complex geometries such as a vehicle underhood area. With the MRF model, separate or multiple reference frames for the fan and the stationary underhood components are used. The frame of the fan blades and hub is a rotating one; that of the engine compartment is a stationary one. The solution proceeds with a steady transfer of information across a pre-defined interface between the two frames.

The analysis treated all three modes of heat transfer -convection, conduction and radiation -simultaneously. A critical factor in creating accurate thermal analysis results is modeling the heat radiated from the engine exhaust system. In previous analyses, these temperatures were often imposed as boundary conditions, which reduced the accuracy of the analysis. One reason is that the requirement of most CFD software for using prism conduction elements around the surface of a heat source increases the grid size and substantially drives up the solution time for what is already a very computationally intensive analysis. FLUENT CFD software provides a unique solution to this problem by providing planar conduction elements that eliminate the need to extrude a prism layer to account for conduction characteristics. Instead, the user simply specifies the actual thickness of the planar element for use in the solver, without needing to adjust the grid. The implementation can model any element thickness without affecting the tetrahedral mesh of the fluid region. The actual thickness of the component can be modeled without having to adjust the component thermal conductivity. It can be applied at boundary walls (walls with cells attached to one side) and internal walls (walls with cells attached to both sides). The implementation can also model multiple surfaces joined at a junction. The discrete ordinates (DO) radiation model was used to account for radiation. The DO radiation model solves the radiative transfer equation for a finite number of discrete solid angles. The number of solid angles can be specified by the user. The DO radiation model can deal with an entire range of optical thicknesses, and can be applied to a wide range of applications, such as surface-to-surface radiation and participating radiation in combustion problems. It also allows the solution of radiation in semi-transparent media. Computational cost is moderate for typical angular discretizations, and memory requirements are modest. The implementation is simple. It does not require any additional files for storage of temporary data. The angular discretization can be changed with a touch of a button and the solution based on a pervious coarser discretization can be used.

The net result from combining critical underbody and underhood design issues in a single analysis was a reduction in analysis time and an improvement in accuracy. The single analysis takes much less time than the multiple analyses that were performed in the past: three to four weeks vs. seven to eight weeks previously. Despite the fact that no test inputs were used as boundary conditions, the analysis results correlated closely with test results, providing far greater confidence in their validity than the traditional fragmented approach. This made it possible to generate thermal design specifications prior to beginning testing, thereby reducing time to market by allowing underhood components to be finalized at an earlier stage. GM engineers plan to expand on this approach in future vehicle programs by simulating actual hill profiles in order to provide even more realistic temperature results with time. They are also planing to extend this model to include soaking and idling conditions.

Fluent Inc. - Automotive Division, 220 E. Huron, Suite 470, Ann Arbor, MI 48104 . Tel: 734-213-6821; Fax: 734-213-0147.