The Cutting Edge: COMPUTING / TECHNOLOGY / INNOVATION : Designing Airplane Wings as Well-Trained as Pilots
Pilots say that changing the shape of a fighter aircraft wing to match the movement of the plane would produce more efficient flight than using the same-shape wing for different types of maneuvers. By embedding “smart” wires into wings, researchers at the University of Illinois at Urbana-Champaign believe they have found the simplest means yet for changing a wing’s shape.
The smart wires are created from a nickel-titanium alloy known as nitinol. Nitinol has the ability to “remember” two shapes and can be made to change into these shapes at different temperatures.
Because the metal changes shape forcefully, the wire can, in effect, do the work of a motor.
Nitinol is trained by being stretched when cold. The deformed metal is clamped so that it cannot move. It is heated and then allowed to return to the original temperature of the training session--approximately room temperature. This cycle is repeated a number of times.
The repetition changes the crystalline structure of the wire and makes the stretched nitinol a “remembered” state to which the composite will forcefully return at cold temperatures. At higher temperatures, the metal will return the shape it was in when first formed.
Until now, embedding smart materials in a plastic matrix was a problem: bubbles formed around the metal wires as the plastic set, thus diminishing the ability of the wires to change the shape of the plastic.
But the University of Illinois researchers found that by using a unique curing and surface pre-treatment process, they could prevent bubbles from forming around the embedded wires--and thus vastly expand the possible uses for smart wires.
Probably Maison Means House: The statistical techniques that allow a speech recognition system to turn speech into text are now being used by a group of IBM researchers to turn text in one language into text in another.
The idea behind statistically based machine translation is to assess a large number of matched sentences in the source language and the target language--say, English and French--which have already been translated by humans. The computer inspects these matched sentences one by one, accumulating statistics for the occurrence of each French and English word pair.
For instance, when the computer sees that the French maison and the English house appear in a large number of matched sentences, it assigns a high probability to translating maison as house , for example.
But the IBM system, called Candide, does much more than word-for-word translation. In the same way that it learns word associations, it also learns probabilities regarding word order in both languages and the way words may appear or disappear in translation. Once all the probabilities have been calculated, Candide could be used to translate previously unseen French sentences.
Currently the system achieves about 65% accuracy when tested on a mix of financial reports and general-purpose news stories. But even with this low level of accuracy, the program could be used to aid a human translator speed up his work.
