Ford Slips A Neural Net Under The Hood
FORD MOTOR CO. HAS A BETTER IDEA FOR DIAGNOSING ENGINE TROUBLE: an advanced neural network program running on chips designed by NASA's Jet Propulsion Laboratory. Starting in 2001, the chips will continuously monitor Ford car engines for misfires caused by such problems as damaged piston rings.
Today, onboard computers can't detect misfires under all operating conditions. So they are built to be hypersensitive: They catch misfires, but with an unacceptable level of false alarms. Ford's neural net program does a better job because, like the human mind, it learns by doing. When fed data on deliberately induced engine misfires simulating all operating conditions, the program learns to spot problems with precision. JPL designed the hardware, compressing the electronic elements onto a chip. Ford has the right to license the chip to other carmakers, and JPL holds the rights for nonauto uses.