ANSYS ICEM CFD в проектировании гоночных машин.
Running ANSYS ICEM CFD On Supercomputer Helps Aletheon ‘Cut Corners’ In Race-Car Design
It's not just how fast you go that makes winners in race car design, it's also how fast you go around the curves. In large part, that is a function of how much "downforce" is created by air flowing over the car's surfaces.
Thus much of the computational fluid dynamics (CFD) analysis of race cars has to do with making sure the airflow pushes down on the car as it does with how smoothly the car slips through the air. Top-end speed is often limited by a particular racing series "sanctioning bodies." Their highly detailed rules are ultimately framed by a countervailing force of nature, the tolerance for risk in the insurance companies that underwrite race coverage.
One of the leaders in racecar CFD is Aletheon Technologies LLC, since early 2002 a unit Pratt & Miller Engineering, New Hudson, Michigan. Pratt & Miller designs, develops, builds, races "all the way from a blank sheet of paper to the victory circle" for General Motors, Ford and DaimlerChrysler, among many others. Formerly known as BRD Motorsports, Aletheon is based in Mooresville, North Carolina.
At Aletheon much of this work is done by Dr. David F. Robinson, an alumnus of the Department of Defense in-house programs in CFD and its technological first cousin, finite element analysis (FEA). A major benefit: Cutting down on high-cost physical testing in wind tunnels and on race tracks.
Modeling for CFD and FEA really differ in just one major respect and that is CFD's need to model the layer of surface flow that is as thin as 0.000l inch. But with all the vortexes and friction, this flow is so complex-over 30 million points if an entire race car is modeled*-that solving it needs the equivalent of computing's race track, a supercomputer.
"At the current state of development for CFD, the more points the better," Dr. Robinson observed. "Computationally speaking, what happens with each element effects all of its neighbors and, in fact, everything else in the model. Dr. Robinson has built his own supercomputer, nicknamed Hannibal.
He also built his own solver, Raven CFD, soon to be marketed by Aletheon.** This software was specifically designed to efficiently distribute and solve the CFD points, nearly 200,000 at a time, among his supercomputer's 210 CPUs.
For the remainder of his CFD work-tedious and time-consuming gridding and pre-processing-Robinson relies on ICEM-CFD from ANSYS Inc., Canonsburg, Pennsylvania, USA (NASDAQ: ANSS).
"The goal of our CFD analyses is not just reducing the drag, and hence improving the straight-away speed," Dr. Robinson pointed out. "In reality, we put most of our emphasis on improving downforce and making sure this downforce is balanced around the vehicle CG," or center of gravity. He emphasized that "distribution of both drag and downforce around the CG is often times as important, if not more so, than the magnitude of these values."
"The aerodynamic performance envelope for any given race series is strictly controlled by that race's sanctioning body," he added. "We use ICEM to expand this envelope."
The Modeling Challenge
In its analyses for fluid flow, heat flux and what Dr. Robinson calls "full-scale CFD." Aletheon models the entire car body-tires, wheel tubs, cockpit, underbody, shock absorbers, exhaust manifold, radiator-anything that air passes over, under or through
"The main reason we model the entire car all at once instead of a single component at a time, is due to the coupled nature of the solution," he pointed out. "Since the flow is subsonic, information can propagate in all directions. This means changes in the rear-wing can, and do, have strong influence on the aerodynamics of the front of the vehicle. Therefore, you cannot do an analysis of the wing independent of the other components.
"Software and hardware limitations makes this type of fully-coupled analysis infeasible for most race teams," he continued. "We do full-design iterations of the entire vehicle. Most racing outfits can manage at most a small number of full-scale CFD runs, often only one or two. At Aletheon, we will run in excess of a hundred full-scale CFD runs in 2003 alone."
"For example," he said, "radiator air flows are modeled to make them more efficient and thus smaller. That affects the car's CG [center of gravity], which in turn affects cornering and overall handling. We can put the weight we saved in the radiator somewhere else closer to the CG."
Modeling virtually every surface in a race car is what generates the enormous files. A rule of thumb in solving models of this complexity is that every million data points requires a gigabit of random access memory (RAM), also known as "core" memory, hence the supercomputer. In mid-2003, Aletheon upgraded that machine with 70 additional CPUs, Intel Corp.'s workhorse Xeon processors.
"All this," he pointed out, "is to deal with a layer of air turbulence thinner one one-hundredth of an inch." Dr. Robinson estimates that Aletheon has about $250,000 invested so far in its Mooresville machine. "That's about a factor of fifty cheaper" than the multimillion-dollar proprietary Cray and SGI supercomputers he worked with in the Defense Dept. The peripheral machines are far less costly, too. For compatibility, the SGI supercomputers required using a several SGI workstations that cost $50,000 to $60,000 apiece. "Now we use Dells that cost $6,000 or $7,000."
For this work, ICEM CFD is the pre-processor "and a very good one. It lets us do models in an unstructured way that could never be done as structured models. They would have taken months to model and solve." For Aletheon, other big benefits of ICEM gridding include:
Speed And Power
The speed and power of ICEM also lets Aletheon build better models. "ICEM lets us model directly on the CAD data rather than rebuilding the geometry model in the pre-processor," Dr. Robinson said. "You get an accurate representation rather than an approximation. Human error is eliminated along with inaccurate interpolations for lofted curves. And ICEM helps use make sure we have all the right elements in all the right places."
Aletheon builds geometry files from data supplied by customers, usually in Unigraphics or CATIA formats, and in-house modeling, usually in Rhino3D. These files are translated with IGES or STEP into SolidWorks and from there by direct translator into ICEM.
"The direct translator for SolidWorks makes ICEM very easy to use," he said. The anticipated addition of a direct STEP translator in ICEM CFD Version 4.3 "will probably let us eliminate the SolidWorks intermediate step," he added.
"With ICEM and the Raven solver, we can do a full suite of CFD runs-dozens of cases-for a racing program for $150,000 to $250,000," Dr. Robinson pointed out. "The people who buy and use supercomputers need annual budgets of $5 million to $20 million. That pretty much limits the market to the big money, Formula One racing teams and research labs runs by the federal government."
In addition to savings in hardware costs, Aletheon Technologies has realized significant software cost savings by developing its own CFD solver, Raven. "One of the widely used commercial CFD solvers recently was quoted to us at a quarter of a millions dollars a year per user," he noted. At this level, it makes it difficult to be cost competitive with either track or wind tunnel testing. It prices CFD out of the reach of all but the most highly funded race teams.
"Having our own in house solver does more than save us money up-front," he continued. "We also have complete control over the future development of the tool. We are specifically tuning the Raven code to improve performance for racing applications. For example, in addition to the racing applications, Aletheon Technologies has a number of contracts with the DOD. Specific capabilities were required for this work and having our own in-house code allows us to tailor the code for these unique needs." The ANSYS ICEM CFD organization has already written a direct translator for the Raven solver.
The Raven software does not, of course, do everything that general-purpose commercial CFD solvers can do-combustion, multi-phase and particulates to name three big ones. But to Dr. Robinson, that is beside the point. "The commercial packages that supposedly do all the CFD analyses don't really do any of them very well," he said. "By specifically tuning our tool for a particular problem, efficiencies can obviously be obtained over existing solvers. We are staying away from the one-size fits all approach to code development, and focusing instead on a small niche in the market."
In addition, Raven was designed from the ground up to be highly scalable. "We have found that it scales up better than linearly, one for one, in speed per additional processor. On 100 CPUs, it's more than 100 times faster than on a single processor. On 200 CPUs, it's more than 200 times faster. We tested it up to a thousand CPUs and performance stayed better than linear." In the computer hardware business, this is called "superlinear scalability," he pointed out. "Everyone strives for it. Not everyone gets it."
Aletheon's analyses are used mainly to cut down on wind tunnel tests and building and testing prototypes. "Just to build a one-half scale model and all of the test parts needed for a year of wind tunnel tests might cost upwards of $500,000," said Dr. Robinson. "Rent for a wind tunnel for one day can be north of $20,000. On top of that you still need to bring in half a dozen test engineers" to set up, monitor and interpret the results.
"Wind tunnel tests give you the total drag, the drag of the airflow over the entire vehicle and the total downforce," he noted. "This does not accurately simulate a race car's operating environment unless you also model the physics of the road rolling under the car. To do this, a 'rolling-road' wind tunnel must be used." In addition, to the global vehicle data provided by a wind-tunnel, CFD also provides the drag and downforce of every single component, even the left rear tail-light, if you need that information.
"Each CFD run may take several days, but you get significantly more data from each run," Dr. Robinson pointed out. "The entire flowfield is captured and can be viewed using a post-processing tool. With wind-tunnel testing's global data results, this can't be done very effectively.
Starting the design process using CFD lets Aletheon learn enough about the problem to put bounds around the amount of additional testing that will be required. "We use CFD to narrow our choices to only those that have the highest likelihood of improving the performance of the vehicle," he continued. "We choose maybe five to ten parts to test in the wind tunnel and optimize their design. After that, we might have just two or three things to test on the track with a prototype racer. Building a prototype car costs $400,000 to $600,000," Dr. Robinson noted. "Testing it means a lot of other costs" such as renting a track, expert personnel, travel, insurance, etc.
"In the 'old' days with the Navy's Cray and SGI systems, a big aerospace problem would take three to six months to model and grid and two weeks to a month to solve," he recalled. "Even then we could only do individual components because the models were so big. In designing and engineering a racecar, breaking up a big model into components just won't work. Too many things affect too many other things. For example, the rear wing design and downforce has a big influence on what you do with the under-wing at the front of the car."
Moreover, CFD analyses can be carried on while the car is being designed, long before any metal is cut or components assembled.
One thing not done with ICEM CFD or the Raven solver is optimizing the design of the entire racecar. "That's because of the complexity and interactions of the design rules laid down by the sanctioning body*** for the races," Dr. Robinson noted. "Design optimization really only works when there are at most a handful of rules to accommodate, things like overall size or mass or simple volume constraints rather than exactly what it looks like and how it has to work.
"Racing rules are very detailed," he explained, "very constraining on the design and, worst of all, very subjective. The vehicle shapes mandated by the rules are often too subjective for an optimization approach. Often the vehicle has to 'look' like the production vehicle upon which it was based. This is a difficult, if not impossible constraint to impose on an optimization tool. Optimization and verification possibly could be done on a wing, something that fits into a definable design envelope; but not for the whole car. The current state of the art doesn't allow for the removal of the human interaction from the design process. At least not yet.
On the other hand, Dr. Robinson said that the ICEM CFD claim of "turning bricklayers into sculptors" is true at Aletheon. He adds: "Nobody else does the kind of design optimization and parametric studies that we do. They don't have the capabilities or staff expertise with modeling, gridding and solvers."