By Dave Graff, U.S. Automotive & Industrial Equipment Industry Solutions Director at Microsoft
Gaining a foothold today in automotive and aerospace simulations, high-performance computing (HPC) offers low-cost high-speed engineering analysis in many industries.
HPC, the engineering application of cluster-based computing in complex, non-linear events, is poised to assist the auto industry by accelerating speed to market, decreasing costs and increasing flexibility in designing and testing new features, materials and models. Also, aerospace firms are using HPC to provide engineering design, testing and analysis software for spacecraft and mission analysis.
As the price of HPC technology decreases and servers become less expensive, automotive manufacturers and aerospace engineering firms can leverage this technology in a mainstream way, giving access to HPC clusters at engineers’ desktops or even on mobile devices and laptops. This technology should eliminate long waiting times for design and analysis, helping manufacturers try a multitude of real-life scenarios without breaking the bank.
HPC in automotive
HPC is giving auto manufacturers a much faster path to insight and innovation. Traditionally, making changes involved running specific simulations that could take several days. Using HPC, organizations like NASCAR racing team Chip Ganassi Racing, as well as consumer automakers, have been able to drastically slash this development time to a few hours, which saves both time and money.
Currently, HPC is employed in mechanical computer-aided design and computer-aided engineering to test structural integrity, airflow, durability of systems and crash testing. As HPC has evolved, the results of computer-aided simulations have reduced the need to perform costly physical tests.
“Windows HPC really brings tremendous computing horsepower to engineers’ desktops, thereby improving their productivity. It also gives commercial customers the best possible return on their human assets and delivers a maximum return on hardware and software assets,” says Tejas Karmakar, of the High-Performance Computing Group at Microsoft. “By making HPC easy to use and widely accessible to engineers and scientists, they can focus on science and physics to solve complex problems, not on their computer systems.”
For instance, making a small change to an airbag used to require extensive testing in crash labs with dummies representing human beings. With HPC, these types of tests can be simulated virtually on the cluster, can be repeated infinitely, and can also take into account variables for drivers and passengers of different sizes and weights. Physical testing would still be required to validate the results of changes from these simulations, but the number of iterations needed during physical testing would be reduced drastically.
Previously, cost of HPC clusters could run from $250,000 to millions of dollars. But, thanks to the commoditization of hardware and the entry of mass-market IT vendors, a server-based cluster now can cost less than $50,000.
Chip Ganassi Racing used a Microsoft Windows Compute Cluster Server 2003 along with several simulation packages from Stackpole Engineering, Services to gauge optimal starting configurations for its cars before each race.
Since its founding as a one-car IndyCar team 18 years ago, Ganassi has grown into a highly competitive racing team that competes in the NASCAR Nextel Cup, NASCAR Nationwide, Indy Racing League, Indy Pro and Rolex Grand-AM series. The team used simulation software for several years, but its usefulness was limited by the massive computing power required.
“We’re all about making cars go faster, not building fancy computer systems,” said Mark Paxton, research and development engineering manager for the NASCAR team at Chip Ganassi Racing. “At first, I thought HPC was only for big companies that can afford multimillion-dollar supercomputers. After we realized that HPC was accessible and affordable, we saw the potential competitive advantage it would bring.”
When Ganassi Racing started using Windows HPC, the team gained the ability to run simulations 38 times faster, providing enough time to run multiple simulations before each race. “Once simulation times were reduced from 24 hours to about 30 minutes, we ran multiple simulations for each race and we were better able to tune the situations for each car, track and set of track conditions,” said Paxton.
HPC in aerospace engineering
Few computational analyses are as complex and critical as those that support design of space flights. With astronauts’ lives on the line and the potential to lose billions of dollars, it’s imperative that NASA has access to the most detailed and accurate predictions and analysis of spacecraft and flight dynamics.
In the past, the enormous computer power required for these kinds of simulations could only be gained using massive, expensive Cray supercomputers. But as commodity x86 servers based on Intel or AMD chipsets have become much more affordable, they’ve also become more attractive to aerospace firms. One aerospace engineering firm A.I. Solutions’, has provided detailed predictions and analysis of spacecraft flight dynamics, optimal launch times and altitude control and navigation. But as customers like NASA and the Department of Defense began to need analyses on a much greater scale and at a much faster rate, A.I. Solutions started investigating ways to run a larger number of simulations more quickly to gain faster insight and better use personnel and technology resources.
The greatest challenge was when NASA requested that A.I. analyze the potential impact of several thousand pieces of debris from a Chinese anti-satellite test on NASA satellites in orbit over a period of twenty years.
The firm turned to Microsoft’s Windows HPC Server 2008, which allows them to pool the resources of many Server-based computers to simultaneously tackle its massive computing challenges. The solution was up and running in about three hours, required no additional training of end-users and didn’t involve any additional development of existing products.
Systems Engineer David Rand configured the first cluster in three hours, avoiding an investment of IT time by using nothing more than readily available information on the Microsoft Web site. Without any formal training, A.I. Solutions analysts were able to develop analysis solutions on their local Windows-based computers, and then use Remote Desktop Connection to push the solutions to the Windows HPC Server head node for processing by the compute nodes.
“I was pleasantly surprised by how quickly and easily we were up and running with Windows HPC Server, and our analysts didn’t need any formal training to work with it,” Rand says.
Filed Under: Aerospace + defense, FEA software, Simulation, Software