Andrew Doyle/LONDON
Aircraft, which suffer major equipment failures or explosions, could be landed safely using software developed jointly at NASA Ames Research Center and McDonnell Douglas (MDC).
The new research envisages that in less than 1s a damaged aircraft's computers would be able to "relearn" to fly the aircraft using "neural-network" (artificial intelligence) software. Preliminary flight tests are under way at NASA Dryden Flight Research Center using early versions of the software installed in a modified MDC F-15.
The computers compare data on what is actually happening to the aircraft with data on how the aircraft should fly. If there is a mismatch, the software, which uses basic aeronautical equations, attempts to modify the aircraft's control laws.
"If sensor data shows that a rule is being violated and the aircraft is turning too abruptly, the neural network can rapidly learn to assist the pilot in use of the stick, engines, flaps, rudders and other control surfaces, in ways that may be very unconventional but possibly successful," says NASA Ames computer scientist Charles Jorgensen.
The Neural Net Group at Ames is working with MDC aerospace engineers to develop the software and modify aircraft to enable them to recover during flight after severe damage or unexpected events. "We are looking at technologies that are five to 15 years away from widespread use," admits Jorgensen.
In the project's next phase, MDC will validate the latest version of the software in a high-fidelity simulator. "They'll have F-15 test pilots fly the simulator through the full performance envelope. Then the software will go into an F-15 at Dryden, probably in 1997," says Jorgensen. Later versions could be used in a hypersonic passenger-transport, he adds.
"You may need to fly parts of those aircraft neurally because conditions change so quickly that humans cannot react rapidly enough to complete tasks such as emergency engine restarts," he says.
Source: Flight International
