Investigators searching for the cause of the fiery crash of TWA Flight 800 know that computer-simulated mechanics and virtual reality can't always generate the digital hiccups that represent the turbulence of real life.
"They couldn't get the temperatures and pressures [in the fuel tanks]," said Shelly Hazle, spokeswoman for the National Transportation Safety Board.
Instead, the NTSB took to the air to recreate the conditions around the central fuel tank that likely existed onboard the Boeing 747 on the fateful flight a year ago. From the cavernous insides of a rented cargo plane, investigators peppered the tank and its environs with probes and sensors to gage the vibrations, measure the temperature, and sample the gaseous cocktail of vapors - all in an effort to determine what scenario, or combination of scenarios, could have created the energy to cause the tank to explode. A laptop computer was receptacle for the data.
Nothing about these types of tests are new; aerospace engineers say this is part and parcel of examining the cause of a crash or testing how a craft - and its components - will stand up in the line of duty. Still, in a world that looks increasingly to 3-D renderings of structures and machines for answers, the testing methods used at John F. Kennedy Airport seem outdated. But they are a sign of the shortcomings of simulations - and of the machines designed to run them.
"Those limitations [NTSB tests] can be traced to a lack of computing power," said Charles Peskin, professor of mathematics at the Courant Institute of Mathematical Sciences at New York University.
We live in an age governed by Moore's Law - where computing power nearly doubles every 18 to 24 months. This tendency has been matched - even surpassed - by increasingly ambitious applications. Automotive engineers want to examine and test, for example, how a person's hips, legs, and feet can be better secured in a car to reduce injury. Pharmaceutical researchers want to take a peek at how their medications react with the enzymes and proteins at the cell level to see if a drug will be successful in attacking a disease.
Instead of feeling sated by all of the computing power available, researchers like Peskin who build bigger and bigger simulations are famished. Supercomputing simply isn't super enough.
"There's a general belief, even among scientists, that computers are already powerful enough to do what you want," Peskin observed. "And for most of the people doing word processing, they are. But to simulate the heart, aircraft, and tasks like fluid flow, they're barely powerful enough and not nearly as powerful as we need them."
Peskin's life work, designing an artificial valve for the heart, is a profile of the evolution of the supercomputing and the benefits that came with it. Peskin began his project when Seymour Cray was merely getting his feet wet in the world of high-charged computing. Armed with one of Cray's early machines, the CDC 6600, Peskin built a model of his valve. This model, while showing where the valve would pivot and how blood would flow, still didn't prove to be an accurate representation of how it would react to all the pressures produced by the heart chamber. For this, Peskin student David M. McQueen recognized that the project had to evolve into a simulation of the heart.
The power of the CDC 6600, however, limited this simulation to only a 2-D view of half the heart - the left chambers. Peskin could see what would happen to the valve when blood flows into the heart but still had no idea how it would stand up inside the entire circulatory system. And lab testing in a handheld model of the heart would yield similar results.
"There is information you want - like the flow pattern of blood around a valve - that you can't get from a physical test," Peskin said.
Later models of the Cray have made it possible for Peskin and McQueen, now a research scientist at the Courant Institute, to build a 3-D simulation with all four chambers, valves, and the nearby vessels that can reproduce the right pressures that exist in and around the heart.
Peskin said he and McQueen are able to build only the heart - and not the entire circulatory system - because information such as the pressures that exist in and around the heart are known quantities. But there are some systems for which the data is not known or where the systems are so complex that building a computer model won't yield an accurate representation of how something worked or will work - no matter how much processing muscle is thrown at it. This is the case with aerospace giant Boeing Corp.
"Obviously, we have significant capabilities to simulate physical systems - instrumentation to train flight crews," said Barry Latter, Boeing chief engineer of airplane performance, safety, and certification for the 737 and the 757. "But also, when we certify an airplane, we need to gather data we can use to understand the parameters of a model."
Latter says the decision of when to create a simulation and when to test is primarily philosophical. For example, engineers may have tested temperature and pressure on a plane when it is on the ground and therefore have data looking at the effect of hot and freezing air on fuel tanks and engines. But if they want to use that data to represent what happens to the plane once it's in the air, they leave the realm of known properties based on testing data and begin making assumptions.
"You have to have a high level of confidence of the physics of a device to move across the boundaries of a set of parameters," Latter said. "If you don't have this assurance, then it's probably a sign that you need to do more testing."
Several other factors contribute to the decision to test, including whether a client will accept a simulation - or prefer the cold, hard data, which Latter says is the case many times. There's a also a matter of whether a system such as an engine is in a static or dynamic state when a model is built. After all, a simulation is only as good as the information that is fed into it.
Or, in the case of the investigation into TWA Flight 800, it's as good as the data captured by the laptop.
