Is It Real, or Is It Cray?

BOB DICK IS DROPPING CANS. Aluminum casings filled with beer or soda, they fall a foot or two to the floor, and Dick looks at the bottoms for damage. It is all part of a service that his employer, ALCOA, offers to can manufacturers-checking out various can designs to see which ones hold up best under abuse. They are not real cans, however. It is not a real floor, and the beer and soda are imaginary, too. Only Dick is real. The rest of the scenario exists only as a pattem of electrical currents flashing through the circuits of a Cray Y-MP/832 supercomputer at the Pittsburgh Supercomputing Center, one of five such centers funded by the National Science Foundation. Trained as a structural engineer, Dick is one of a new breed ofexperimental scientists from disciplines as diverse as biology and aerophysics who locate their laboratories inside a computer. The power of today's supercomputers lets these researchers mimic a variety of physical experiments using computer hardware and software in place of lab ware. But not only can computp er experiments substitute for many of the studies normally done in a lab, they also are allowing scientists to gather data and test hypotheses in __ ways that were closed to them before. And the field is stlll in its infancy-as computers become more powerful and as more scientists become aware of their potential, computer experiments are likely to change the way research is done. The promise of computational science has led some researchers to suggest that the field will eventually grow into a third domain of science, coequal with the traditional domains of thecomputational experiments, Dick says, for two simple reasons: time and money. Andy Trageser, Dick's supervisor at ALCOA Laboratories in Pittsburgh, recalls that in the early 1980s, before the company got into computer modeling, "it would take 6 months to a year to come up with a promising design." A can manufacturer would come to ALCOA with an idea for a new shape, and the ALCOA engineers would take on the challenge ofmodifying this basic form so that it would best stand up to such things as drops and internal pressure. To do this, they would manufacture dozens of cans with slightly differing forms, fill them with liquid, then drop, bop, or otherwise abuse them. Today, Dick simply keys into his supercomputer the numbers that describe how he wants to modify the can-he might make the dome-like indentation on the bottom of the can a little deeper, for instance, or widen the base of the can-and then lets the Cray do the work. The simulated drop test is -9 a marvel of number crunching. The computer knows the shape of the can, the speed at which it hits the ground, the mass and pressure of the liquid inside the can, and the physical characteristics of the aluminum. It knows the equations that describe the complicated interplay of mass, velocity, acceleration, pressure, and other relevant physical properties. To calculate the effects of the can striking the ground, it looks at the can as a collection of tiny elements, each of which acts upon and _______ is acted upon by the other pieces around it, as well as by the ground and the liquid in the can. The computer calculates the position of _ ~ each of these elements at 1 millisecond after impact, 2 milliseconds, 3 milliseconds, and so on, until the can stabilizes, usually after about 8 milliseconds. Besides the drop test, Dick and Trageser have also used the supercomputer to check how well cans resist the intemal pressure that can cause the bottom of a can to pop outwards, and to model the denting of cans as they bump into something. The beauty oftesting cans on the computer is that it is so fast and simple. With the simulated drop test, ALCOA can shoot a design back to a manufacturer in 2 weeks, sometimes in as little as 2 days. And the cost of developing a new can design with computer modeling, Trageser says, is only about $2,000, compared with about $100,000 using traditional laboratory methods. For the same reasons that ALCOA models cans on a computer, aerodynamics researchers are increasingly swapping wind tunnel work for numerical experiments. Ron Bailey, manager of the Numerical Aerodynamic Simulation Systems Division at NASA's Ames Research Center, says that in the past engineers would go through the slow, tedious process of building different wings and testing 30 or 40 designs at a time in the wind tunnel. "With the computer we can look at many more wings in a shorter time," he says. They still end up putting four or five wings through the wind tunnel, but by using the supercomputer for an initial evaluation, "we use our wind tunnel smarter." The convenience ofcomputer simulations is not their only advantage over wind tunnel experiments, however. Once researchers began to play around with aerodynamic tests on computer, they found that one of the great strengths of computer experimentation is that it can tell you things you could never see in real life. The space shuttle is a good example. "It is almost impossible to get a hot plume in a wind tunnel," says Terry Holst, chief of the applied computational fluids branch at NASA-Ames. Engineers also find it quite difficult to model the shuttle's behavior during its separation from the launch vehicle, he