|
The NeuroEngineering Laboratory at Ames Research Center develops flight control software that utilizes neural networks to compensate for failures resulting from control surface damage or system malfunctions. A neural network is software that works through pattern recognition, and thus is able to "learn" from sensory input, resulting in systems that can adapt to changes in their external circumstances. Neural networks have been the subject of theoretical and applied research since the early 1960s. They have been applied to a wide range of problems: from investment analysis to the control of modem equipment.
Modern commercial aircraft are among the safest transportation systems ever designed. But even with their track record of successful operation, aircraft remain vulnerable to failures of flight-control systems, whether because of accidents or equipment malfunctions. Post-accident analyses of catastrophic flight-control accidents show that stricken aircraft usually retain some working control surfaces at the time of the crash. Given enough time, a skilled pilot can determine how to compensate for the loss of a control surface, but there is usually not enough time during such an emergency for a pilot to determine the nature of the failure and learn how to compensate for it.
In 1983 a mid-air collision in the Sinai resulted in the loss of nearly all of one wing from an F-15. Even under these dire circumstances, the pilot actually managed to land the disabled aircraft successfully, demonstrating that control of such badly damaged aircraft is possible. Is it possible to develop software that is capable of relearning the flight behavior of a damaged aircraft? Can the resulting "knowledgeable system" adapt the control of the aircraft in a way that effectively compensates for changing aircraft behavior that arises as a result of serious damage? A first-generation neural flight-control system is currently being tested on a modified F-15 aircraft at NASA's Dryden Flight Research Center to answer these questions (fig. 1).
Commercial aircraft pose a different challenge than do military aircraft, a result of their inherently more stable designs and the relatively benign environments in which they operate. Commercial transports are also less maneuverable, however, which means that when severe damage occurs the results can be catastrophic. An analysis of recent transport aircraft accidents shows that in many cases, engines are still operational at the time of the crash. So with this in mind, a second-generation neural flight-control system was developed and integrated with another piece of NASA software, Propulsion Controlled Aircraft (PCA). PCA manipulates engine throttles, thereby enabling the aircraft to be maneuvered and flown without the use of the control surfaces in a limited flight envelope.
The combined Intelligent Neural Flight and Propulsion Control System (INFPCS) utilizes engine control with whatever remaining control surfaces are operational. INFPCS was successfully demonstrated in the Advanced Concepts Flight Simulator at Ames over a period of 5 weeks in FY00 (fig. 2). This achievement paved the way for this technology to advance to a flight-test phase.
In addition to enhancing aircraft safety, neural flight control offers benefits during software design and development. Neural networks can enable the development of flight controllers for new aircraft in significantly less time than it currently takes, thereby saving enormous expense during the design and development process.
Point of Contact: Guy Power
(650) 604-5989
gpower@mail.arc.nasa.gov
Back To Top
Previous Paper
Return to Revolutionary Aviation
Next Paper |
