快猫短视频

CRASHES caused by ice building up on an aircraft鈥檚 wings could be prevented
by a system that senses the aerodynamic effect of the ice and compensates for
it.

The ice management system is based on a neural network鈥攁 program that
mimics the way the human brain learns. It picks up the tiny changes ice makes to
a plane鈥檚 manoeuvrability, and adapts the flight controls so that the pilot can
continue to fly it in the normal way.

鈥淲e can change the relationship between the pilot and the controls so that
the plane seems to the pilot to act as if it is not iced,鈥� says Michael Bragg,
who runs the 鈥渟mart icing systems鈥� research project at the University of
Illinois at Urbana-Champaign. Such a strategy could save pilots from making
potentially disastrous errors when they suddenly find that the plane is not
responding to the controls as they expect.

Many aircraft are rigged with ice-detectors on the wings, nose and other
surfaces. But variations in air temperature and pressure can mean that ice
builds up in unexpected places where there are no sensors. So alarms don鈥檛 sound
and pilots often don鈥檛 realise what鈥檚 happening until it鈥檚 too late to activate
on-board de-icing systems.

To get round this, Bragg and his team used data from both flight tests and
simulators to generate models of the aerodynamic problems that ice causes. Then
they trained a neural network to recognise these performance defects such as
subtle changes in vertical acceleration or in the plane鈥檚 ability to make a
turn. However, it takes months to train the neural network to recognise the
numerous ways each type of aircraft responds to ice, so it will be at least five
years before planes are fitted with the system.

While the technology is promising, many practical issues still have to be
tackled, says Eugene Hill of the US Federal Aviation Administration. 鈥淚f this
system improves flight safety, then we should take a serious look at it, but we
must also consider the cost and complex training that the pilots must
耻苍诲别谤迟补办别.鈥�