|
This article was first published in 2001.
|
Some people describe Computational Fluid Dynamics (CFD) as "a wind tunnel in the computer." CFD is a tool which, in the proper hands, can completely transform the capability of a racecar constructor for aerodynamic research and development. CFD enhances that capability because it allows the design engineers to exercise many, many more testing options during the design cycle.
To gain an idea of the potential impact of CFD, imagine possessing the ability to put an aerodynamic concept through a very thorough test process, without actually having to build it. In terms of time saving alone, the benefits are vast.
CFD is a branch of Fluid Mechanics that has a complex mathematical basis. This is due to the fact that the physical laws which govern fluid flows can be described in terms of rigorous mathematical relationships which form the basis for any analysis. In the case of CFD, the mathematical basis is a differential equation. There are relatively few people in the world who voluntarily attack problems that are guaranteed to give them a headache. As of this moment, you can count in this group the scientists at Fluent Europe. Fluent Europe was founded in 1983 by Dr. Ferit Boysan (its present managing director). Among its products are three CFD codes known as FLUENT, RAMPANT and NEKTON.
Design
The implementation of CFD into the design cycle starts at the concept stage with CAD (Computer Aided Drafting) - or, in this case, it could start with RAMPANT's preprocessor, GeoMesh. In addition to constructing the geometry to be modelled and a computational grid, the preprocessor also imposes the flow conditions which are determined by parameters such as density, viscosity and boundary conditions (walls, thin surfaces, pressure boundaries and others). The problem is actually defined by the boundary conditions.
Typical applications would be the analysis of a rear wing assembly, the cooling system, the front section of the racecar, and even the entire racecar. Incidentally, the last could take a few days (or even weeks) to achieve a solution, depending on the complexity and size of the model. It is likely that the files for the aforementioned problems have been drawn using EDS Unigraphics, which is the de rigueur CAD software at some racecar teams. RAMPANT can accept the IGES standard files which EDS Unigraphics is capable of creating.
So, let us assume that a member of a race team aerodynamic staff has (say) a cooling system 'constructed' on his computer screen in one hour. At this point, the model would normally consist of line-and- point entities. A depiction of this stage (showing the entire racecar) is shown here.
Discretisation, also known as 'gridding', is the next phase. Gridding involves dividing up the surface of a model into tiny elements. The analyst selects each line entity, applying on each a finite number of the intersections of cell corners - called 'nodes'. The node density would normally increase towards areas of high fluid dynamic gradients, corners, surfaces and areas of interest, as shown here.
At the end of this process, a command is given to tell GeoMesh to place nodes throughout the entire geometry, using the nodes set by the user as reference. In GeoMesh, there is a choice of the type of grid that can be used for the computation. Essentially, it lies between structured and unstructured grids. The pictures above show unstructured grids.
It used to be the case that structured grids - as shown here - were the first and last word in CFD. This is because they are computationally efficient. (It doesn't take long for the computer to number-crunch them.) But, as with most situations, there is an obligation accompanying this computational privilege. That obligation, for complex geometries, could demand that the user takes long periods of time to create a grid for a problem. Some problems have taken years to attain a grid, and only hours to run! This is because CFD and CAD engineers do not always communicate as they should...
Unstructured grids can be a dream come true. A close look at this diagram reveals that triangles slice through the basic elements, which are tetrahedrals. Unstructured grids can have just about any elemental shape including quadrilaterals, tetrahedrals, and hexahedrals. The basic element shape selected (the 'primitive') is subjective: it is determined in part by the overall shapes that are involved in the simulation.
The benefit of an unstructured grid lies in its convenience. Such elements can map the most complex geometries with ease. However, they can also be computationally inefficient. In part, this inefficiency has been addressed over the last decade or so by the mathematical treatment of the elements and through faster computers. Indeed, unstructured grids have long been an acceptable technique, and have become an indispensable part of CFD.
At some stage, the boundary conditions of the problem have to be imposed. This has to be done because they define the problem - without them, a simulation would be meaningless, as well as impossible! Let us say that the user at a race team scurries his mouse across the screen, selecting various entities of the model, and defines an inlet for a radiator duct intake for his cooling system. This is then assigned a theoretical velocity in 3-D space. Other similarly treated parts of the model might be wall entities (for the duct walls), heat sources with temperatures (for the radiator faces), and pressure boundaries (for the duct exhaust). The properties of the fluid or fluids are also entered at this stage.
Crunching Numbers
With the use of complex mathematical schemes, RAMPANT is now told to crunch numbers. It discretises the governing equations - the Navier-Stokes Energy and Continuity Equations - and implements the Turbulence Model, which is arguably the most important part of the calculation. An understanding of the concepts is indispensable - CFD is not something that your average mechanic-cum-wind-tunnel-operator-cum-designer is realistically going to pick up as he or she goes along. If a race team has no one who understands these concepts, and is not prepared to hire some one who does, then they should stick to wind tunnels - because abusing a CFD code is just as bad as not using one, and may even be a whole lot worse. The aerodynamicist who is experimental in his or her nature is duty bound to understand and master the concept of scale-effects, boundary layers and turbulence - for starters - if he or she is to make good use of a wind tunnel. Such knowledge will be a sound foundation upon which to interact with his or her numerical counterpart.
When the simulation is completed, it can be viewed with a postprocessor. RAMPANT is capable of analysing the completed flow simulation in several ways. They include velocity vectors, streamlines/streaklines, line or filled contours, profiles, iso-surfaces and x-y plots.
Examples of some of these are shown here in the diagrams.
It is also possible to examine the forces on individual components, provided the grid has been constructed so that the software can distinguish between these components. It is in situations like this that a multiblock grid becomes indispensable. In a multiblock grid, adjacent components will share the same nodes, but the characteristics of each block can be markedly different with regard to direction of propagation, density and number of cells.
For example, the front wing mainplane/endplate junction on the racecar could be part of a multiblock grid where the mainplane is formed of one block, and the endplate formed of another, while sharing the same nodes where they abut. Assuming that such a grid was constructed for the geometry depicted in the wing above, then the user could simply select mainplates, flaps and end plates, and direct RAMPANT to display the forces on each component, or the resultant force acting on a group of components.
Information of this kind would be useful, say, in the design of front wing mounting posts. These posts could then be made as stiff as possible, but only as strong (and therefore as heavy) as they need to be. The analysis suite provides a formidable array with which to determine 'the way to go' or, as the case might have it, 'the way not to go'.
The author thanks Dr. Matthew Wheeler of the University of London Computlng Centre for his assistance in the preparation of the post processing images for this article. This article is abridged version of a paper originally entitled: "Running RAMPANT: Computational Fluid Dynamics in Formula 1 Design", Copyright © 1999 Fluent Inc. 10 Cavendish Court Lebanon, NH 03766-1442.
|