Recently [during what little television I actually watch] I’ve been noticing a Toyota commercial where a pair of kids excitedly press a big button [repeatedly] to send cars hurtling down a test track under the gaze of a man clad in a white lab coat. Tonight, I noticed a Subaru ad while flipping through Cooking Light magazine, proudly displaying the freeze-frame results of a NHTSA crash test.
Yes, I read Cooking Light. It’s a good magazine!
Meanwhile, I’ve been spending my days learning the ways that automotive designs are rigorously tested before they ever leave the digital safehaven of an engineer’s workstation. There are requirements reviews, design reviews, compliance reviews, simulations, and reviews of the reviews. Let’s just say that everything is reviewed pretty damn thoroughly – and that’s why we (rightly) expect higher reliability and safety from our cars and trucks than we do from, say, certain Microsoft products.
Here’s where it gets confusing: how much cheaper would our cars be if the manufacturers didn’t have to turn over X number of each and every model for crash testing? The automotive graveyard for crash test victims must be huge. Even if the cars were crushed – a storage method that seems unconducive to potential civil suit evidence – they’d still take up a lot of room. Thousands and thousands of carcasses, crashed into brick walls, steel plates and each other in the name of safety.
“But Dave!” you say, “How else could we learn how safe the cars are?”
Let me take a brief break here. We’ll get back to crashing cars in a minute. For now, let’s talk computers. Computers like the ones at Pixar Animation Studios. There, smart people use a warehouse full of networked Linux computers to create top-grossing blockbuster hit movies like Toy Story and Finding Nemo. Meanwhile, Boeing engineers use clustered computers to model the next generation of stealth aircraft, or predict the behavior of orbiting spy satellites. I won’t even get into the ways scientists use high performance computers to model the structure of our DNA, design a new cancer drug or predict the results of nuclear fusion experiments.
While we’re on the topic of science, let’s also touch briefly on the science of materials. We engineers have a pretty good handle on material properties. While it’s not exactly my discipline (electrical, if you didn’t know), I’m pretty confident that those mech-Es out there know a good bit about what happens to molded plastic parts, drop-forged bolts, panes of tempered glass and chunks of steel when various forces are applied to them. Just like I have computer models for the behavior of transistors and capacitors, they’ve got nifty software that tells them how much strain a cable will withstand without breaking, or how fast a good old American ’454 can turn without throwing a rod.
I’m wondering if it’s occurred to anyone that perhaps we might save some money and come out with – gasp! – better results if we let the computers handle the whole car crash thing.
Consider first that once you’ve crashed a car, it’s done. You get one angle, one speed, one target, and then it’s over. You get your results, download the data from the force sensors and the test dummies, evaluate the damage, then the car goes out to the bone yard. Another one bites the dust.
Every part in that car can be modeled, and every attribute of every part can be turned into a collection of bits sitting on a cluster of computers. The assembly of the car already lives in vast collections of CAD data – right down to the torque on every bolt and the coordinates of every crumple zone. Add to this the aforementioned behavioral models for all those parts, and you’re rewarded with the ability to simulate every aspect of a crash.
You can crash this car as many times as you want. You can crash it from any angle, into any object, with any number of occupants, at any speed. Crashing it again is as simple (and inexpensive) as resetting the test with a few keystrokes. After you model other vehicles, you can even crash them into each other! Thanks to Monte Carlo analysis techniques, you could give the computer a range of crash parameters and let it statistically analyze the results from hundreds of crashes at different speeds and angles … then plot out the car’s weaknesses for you after crunching numbers for a few days.
Update 2005-08-09: What I’d intended to include here was that – following the identification of the specific crash parameters that do the most damage, this set of parameters could be used to set up an actual test crash – allowing data to be collected in an event that’s as severe as possible.
Want to take it a step further? Try going for distributed computing, like the folks at SETI@home. Thanks to tens of thousands of SETI enthusiasts whose PCs crunch SETI numbers while in screensaver mode, SETI doesn’t have to invest in clusters of supercomputers or pay for mainframe time. Plenty of people already have screensavers that flash expensive cars at them – wouldn’t it be nice for those same folks to gaze at your products while their PCs do your crash test dirtywork?
Given all the other ways that computer simulations provide us with accurate predictions of real-world events, and given the huge range of variables that can affect the outcome of a car crash, I really don’t see just why automotive manufacturers weren’t among the first to adopt computer simulation to make safety predictions.
If computer sims are good enough to test birth control pills, sub-atomic particle physics and fighter jets before people entrust their health or their lives or their livelihoods to them … why aren’t they good enough for minivans?
Loading ...