Tools of the trade
Two of the leading software tools for simulating complex crash environments are Altair's Radioss and ESI Group's PAM-Crash. GM currently uses LS-Dyna from Livermore Software Technology Corp., originally developed at Lawrence Livermore National Laboratory.
According to Pradeep Srinivasan, a senior technical specialist at Altair, automakers that use Radioss incorporate millions of data elements -- such as tire pressure, car weight and road conditions -- into a single simulation. One Japanese manufacturer he declined to name used 10 million to 12 million data elements in an especially detailed crash simulation, Srinivasan says.
With such detailed simulation and analysis, it's not surprising that processing power is an important factor. For many common computer simulations, such as one vehicle crashing into another, carmakers have the supercomputing power they need in-house. Altair has publicly demonstrated that even a complex simulation of a full crash test with 1 million elements can take just five minutes to render using a cluster of Intel Xeon 5500 processors.
American Honda Motor Co. (which includes Honda and Acura cars, as well as Honda motorcycles, motors and power equipment) has more than 3,000 processors dedicated to crash analyses, according to Eric DeHoff, manager and principal engineer for vehicle structure research and computer-aided engineering. "We use high-power computer clusters -- load-balancing computers with many processors that share the computational workload -- to process many different standardized regulatory and consumer information crash modes," he says. "We perform structural deformation analyses and occupant injury mode analyses, which require the modeling of restraint system parts like the seat belts, airbags and the actual crash dummies."
Stuttgart, Germany-based Mercedes-Benz performs thorough simulations on new vehicle designs, says Richard Krüger, manager of safety communications. "We run approximately 5,000 crash simulations with complete car models during the whole development phase [for each car]," he says.
Krüger says Mercedes uses LS-Dyna for the simulation solver for crash tests, Medina from T-Systems for car modeling and Animator4 from GNS for crash visualization to see how the models perform in the rendered environments. "A typical turnaround time for [the complete crash-test simulation] rendering is approximately 15 hours," he adds.
Simulation limitations
Despite today's sophisticated testing capabilities, Bhatti says one of the roadblocks to the zero-fatality car is researchers' understanding of human anatomy and physiology -- for example, they don't know enough about how the brain responds to a head injury to simulate a model that can distinguish between light and severe injuries.
Currently, virtual dummies help measure only displacements, velocities and physical force, Bhatti says. Work is under way -- mostly by a group of U.S. and international partners led by the Global Human Body Models Consortium -- on building a human model that can measure tissue damage, brain trauma and other accident damage beyond what the typical crash test dummy can show.