Intelligent control system learns to fly

NASA and the Air Force last week unveiled an experimental aircraft that will test an intelligent flight control system which could play a critical role in the development of high-speed aircraft in the defense civilian and eventually commercial aviation markets.

The new flight control system is based upon so-called "neural network technology" - a variation of artificial intelligence in which a computer system learns from experience by "observing" human reactions.

In this month's Low-Observable Flight Test Experiment (LoFlyte) NASA and the Air Force will test whether the neural network-based flight control system on remote-piloted experimental aircraft might assist pilots at the controls of future hypersonic aircraft that can fly more than four times the speed of sound.

"At that kind of speed if you would get some kind of anomaly - either atmospheric disturbance or engine [failure] - human response time is so relatively slow [that] the pilot would not be able to handle it so you've got to have [an advanced] control system" that reacts more quickly said David Reubush the acting deputy director of the hypersonic vehicles office at NASA's Langley Research Center Hampton Va.

LoFlyte is a remote-piloted vehicle that is more than 8 feet long and can reach speeds of only about 100 miles per hour. However it is based on a design called waverider: a hypersonic aircraft designed to ride on top of its own shock waves while exceeding speeds of Mach 4 or four times the speed of sound according to NASA.

Although LoFlyte cannot reach such speeds NASA and the Air Force will use the vehicle to test the challenges of taking off and landing an aircraft of hypersonic design in which the aerodynamics are much different - and trickier - than aircraft designed for conventional speeds Reubush said.

If tests this month justify continued trials the agencies hope to deploy the flight control system into a larger aircraft eventually testing it at speeds of about Mach 1 NASA said.

Computerized flight control systems have been around for decades and have become increasingly more sophisticated. In fact conventional control systems could be used in future hypersonic aircraft NASA said.

But the present model of flight controls however sophisticated is still limited by human programming Reubush said. That means such a system only reflects the information a person puts into it.A neural network on the other hand learns to mimic human behavior by monitoring through the flight controls how the pilot flies the aircraft.

To borrow a concept from developmental psychology the system then is able to generalize what it learned from one situation and apply it to many others.

The Air Force considering battle space applications rather than civilian ones also is concerned about the potential for using the technology to compensate for slow human reaction an Air Force spokesman said.The Air Force would use the technology to fly an aircraft despite battle damage the spokesman said.

"Now you have a new situation [in which] the flight controls may not respond in an intuitively predictable way. But if you have a neural flight control system you could react to that " he said.

The same application could be used as a safety feature for commercial aircraft.

For example several years ago a DC-10 crashed in Iowa after many of its flight controls failed in flight.

Many passengers survived that crash because the pilot was able to cushion the plane's descent by firing its differential thrusters.

Conceivably a neural network-based flight control system could have lessened the impact of the crash more because the system would have reacted more quickly than the pilot according to Reubush.