Search Ames


Text Size


April 14, 1999

John Bluck

NASA Ames Research Center, Moffett Field, CA

Phone: 650/604-5026 or 650/604-9000


Fred A. Brown

NASA Dryden Flight Research Center, Edwards, CA

(Phone: 661/258-2663)



NASA is doing something to help pilots who find themselves in potentially disastrous situations flying severely damaged or malfunctioning aircraft. The agency is developing new "smart" software that will enable aviators to control and safely land disabled airplanes.

NASA's Dryden Flight Research Center, Edwards, CA, is now conducting flights to test the new software which is helping NASA reach its goal to reduce commercial aircraft accident rates by a factor of five over the next 10 years.

The intelligent flight control system employs experimental "neural network" software developed by computer scientists at NASA's Ames Research Center, Moffett Field, CA, and the Boeing Company's Phantom Works division, St. Louis, MO. When fully developed, the software will add a significant margin of safety for future military and commercial aircraft that incorporate the system.

Neural network software is distinguished by its ability to "learn" by observing patterns in the data it receives and processes, and then perform different tasks in response to new patterns, according to Dr. Charles Jorgensen of Ames, principal investigator for the software program at NASA. Simple neural network software has been in use since the 1960s with computer modems to enable them to receive error-free data over often-noisy phone lines, but it has never before been demonstrated in such a complex safety-related environment.

Using a highly-modified F-15 aircraft, the Dryden tests are demonstrating how a preliminary version of the neural network software that is pretrained to the F-15's aerodynamic data base, operating with an adaptive controller, can correctly identify and respond to changes in aircraft stability and control characteristics, and immediately adjust the control system to maintain the best possible flight performance under both normal and simulated failure conditions. The tests involve about 14 flights over a four-week period.

In its flight control application, the neural network software program takes data from the aircraft's air data sensors—airspeed, direction, pressure, force—and compares the pattern of how the aircraft is actually flying with the pattern of how it should fly. These patterns are based on a series of pre-programmed aeronautical equations or control laws that define how the airplane flies. If there is a mismatch due to equipment failures, combat damage or other reasons, the aircraft's flight control computer uses the new neural network programming to "relearn" to fly the plane with a new pattern six times every second.

For example, a military aircraft may sustain combat damage that disables one or more of its control surfaces, such as an aileron or flap. A commercial aircraft could sustain a major equipment or systems failure, such as the inability of using its flaps or encountering extreme icing, both of which could affect the safe performance of the aircraft.

Using its on-line learning capability, the neural net software would identify that something has changed, then reconfigure the flight control computer system to adapt to those changes, making the failure or damage almost "transparent" to the pilot. To enable the pilot to maintain or regain control, it may change the way the remaining functional control surfaces and systems are used to compensate for the loss of the inoperative or damaged surfaces or equipment.

Future versions of the software could be developed for use in new airplanes that have digital fly-by-wire flight control systems, such as the Boeing 777 jetliner, the Air Force's C-17 transport, the F-22 fighter, and other aircraft still on the drawing boards. The system also has application to NASA's proposed Mars aircraft concept. These software versions will have even faster self-learning capability.

Jorgensen noted that neural net software being developed in this NASA project could have a bearing on other aspects of contemporary life. "Once we prove neural net software can rapidly learn to fly a crippled aircraft and help pilots land it safely, then engineers will be more likely to use the intelligent software in power plants, automobiles and other less-complicated systems to avoid potential disasters after equipment failures," he said.

For additional information, please check the Neuroengineering website:

- end -


- end -

text-only version of this release

To receive Ames news releases via e-mail, send an e-mail with the word "subscribe" in the subject line to To unsubscribe, send an e-mail to the same address with "unsubscribe" in the subject line.

NASA Image Policies