NASA TESTS `SMART' JET SOFTWARE; LEARNING PROGRAM AIMS TO ADJUST FLIGHT, REDUCE CRASHES.

Byline: Jim Skeen Daily News Staff Writer

Looking to improve flying safety for military and commercial jets, researchers are wrapping up the first series of flight tests of the ``smart'' software they believe will someday help pilots land disabled airplanes.

Part of the National Aeronautics and Space Administration's aim to significantly reduce commercial aircraft accidents, the Intelligent Flight Control System uses neural network software that modifies its responses by observing patterns in the data it receives and processes - in other words, it learns.

Researchers are trying to develop a system that would adjust when those patterns change, such as when an airplane is damaged in flight.

``If an accident occurs, the aircraft will seek usable control surfaces like flaps, rudders or ailerons that would successfully compensate, restoring control to the pilot,'' said Charles Jorgensen, principal investigator for the program at NASA's Ames Research Center at Moffett Field in Mountain View, Calif.

The flights are being conducted at NASA's Dryden Flight Research Center at Edwards Air Force Base using modified F-15 fighter jets. The flight tests involve a preliminary version of the neural network software that is pretrained to the F-15's aerodynamic database.

The software can identify the airplane's stability and control characteristics and immediately adjust the control system to maintain the best possible flight performance.

``Overall, the system is working pretty well, but it's not what I'd call a fully mature system,'' said Dryden test pilot Dana Purifoy. ``We haven't noticed anything dramatically different than what we had anticipated.''

Flight tests have shown that the system tends to be sloppy in the pitch control - pitch means the up or down motion of the plane's nose - and leaves a slight bobbing motion. That's a flying qualities issue that can be handled through a tweak of the software, researchers said.

The next step is to fly neural network software that ``learns'' as it flies. Because the learning software requires more computer power, researchers at Ames are now wrestling with finding a way to integrate the learning software into the airplane.

It is expected that the software will be flight-tested in late summer or early fall, said Mike Thomson, Dryden's chief engineer on the project.

In addition to making military and commercial aircraft safer, the software could be used to improve the odds of success of the proposed Mars airplane, a data-collection aircraft NASA researchers envision flying through the thin atmosphere of the Red Planet in 2003.