Artificial intel-based system 'to detect flaws in aircraft'

A team at Swinburne University says that its inspection system has the potential to increase aeronautical safety and speed up component safety checks.

Aircraft made mostly from composite materials such as carbon-fibre-reinforced polymers are already on the drawing boards of major aeronautical manufacturers, which seek lighter planes able to carry more passengers, cargo and fuel.

While these ultralight materials are currently available, their widespread use is problematic as the process of scanning for potential flaws is expensive, time-consuming and gives less confidence than similar processes used for checking and certifying metals.

But, the scientists claim to be tackling this challenge by developing an automated approach to processing data from scans of composite materials. The goal is a process based on AI technology that enables analysis to be carried out with much greater speed and accuracy than a human technician could achieve.

"There is a lot of pressure on the technicians who analyse the scans of composite materials for certification," said Dr Mark Hodge, CEO of the Defence Materials Technology Centre based at Swinburne's Hawthorn campus in Australia.

He added: "Getting it wrong could cost lives and a lot of money. The risk of those consequences means there is a tendency for the technician to be conservative and not certify parts that have any potentially threatening flaw."

Defects can be introduced into a composite material during manufacture or while the plane is in service.

The difficulty in detecting these is one reason why composite parts are currently used only on non-load-bearing aircraft parts such as aerolons (the flaps that descend from the wing to control side-to-side movement), and even then only after a rigorous and time-consuming certification process.

Interiors of composite panels are currently examined using non-destructive inspection technologies, such as ultrasound. The panel is scanned with an ultrasonic probe attached to a robot that sends raw data displayed as squiggles akin to those produced by an electrocardiogram, says the team.

"The AI inspection system developed at Swinburne mimics human intelligence to examine a sensor signal and draw out valuable information. The signal information can then lead to the identification of any defects existing in the component," said team leader Prof Romesh Nagarajah.

Now, that this stage of development is complete, the next step is for software coding engineers to refine the system for commercial use, say the scientists.