AI in aviation: how machine learning is keeping airliners flying

Kent German explains how companies are using artificial intelligence to perform vital visual inspections of aircraft and even check jet engines for missing parts

You don’t buy a commercial aeroplane and never take it in for a tune-up. Expensive (the sticker price for a Boeing 777 is around $350 million) and enormously complicated, they need constant attention to ensure they’re in top condition. You don’t take risks when flying hundreds of people through the sky.

Any routine maintenance starts with a visual inspection. Similar to walking around a car you just hired to look for any problems, a visual inspection is the process of closely scanning a plane’s fuselage, engines and components for scratches, dents or other defects. 

It’s critical work – even tiny blemishes can affect an aircraft’s aerodynamics – and it’s changed little for more than a century. Whether it’s a pilot performing a walkaround inspection before a flight or mechanics in a hangar conducting a weeklong check, the human eye is the most common tool.

But some see room for improvement. By using drones to photograph every centimetre of a plane and machine learning to identify potential problems, a Dutch firm promises to make the process faster, more accurate and more consistent. It’s not about replacing the human eye, but rather about giving it a (literal) boost.

Another tool in the box

Michael Sprehe, the Head of Marketing and Communications for Mainblades, calls drone inspections another tool in a mechanic’s toolbox. The small flying machines take over scanning an aircraft, but mechanics still evaluate what the drones see, verify potential issues and decide on possible fixes.

“[A visual inspection] is a highly manual procedure… You want to minimise any kind of crack on the fuselage and maintain operational efficiency,” Sprehe said. “If you regularly do it, you foresee where damage might become an issue.”

Based in The Hague, Mainblades was formed in 2017 by three co-founders, one of whom remains an active KLM pilot. In addition to the flag carrier of the Netherlands, the company also counts Lufthansa Technik Philippines and Delta Air Lines as customers.

Modelling a process

The Mainblades solution begins with making a 3D model of an aircraft for drones to use as a reference point. All types of airliners, from a small Embraer regional jet to a giant Airbus A380, can fall under Mainblades’ aerial eye.

Once a drone is loaded with the model’s specs, it flies autonomously around and on top of the entire aircraft taking high-resolution images. Mainblades uses drones from DJI but augments them with a small module that holds a company-developed guidance computer.

After a flight, Mainblades’ machine learning software then analyses the images to identify areas that need attention. Sprehe describes the AI technology as an additional layer of support.

“It suggests critical photos and provides focus areas that really matter,” he said. “What kind of damage is there, and where exactly on the fuselage is it?”

Related reading: What is machine learning?

Speed and consistency

It takes about an hour to inspect a narrowbody plane and roughly two hours to buzz around a widebody. Conventional inspections can take between six to eight hours or up to several days.

Drones also replace the need to continually move a ladder or cherry picker to reach the top of a fuselage or tail assembly. But while saving time and labour is important – a plane not carrying paying passengers is a depreciating asset – Sprehe says the benefits go far beyond that. 

“A person may carry out an inspection one way, but then a couple of months later another mechanic carries out an inspection in a slightly different way,” he said. “A drone will always cover an aircraft model in the exact same way every time. That leads to consistency.”

For now, Mainblades’ inspections are conducted in or just outside of an airline’s hangar, but Sprehe said the company eventually hopes to perform them at an airport gate. Also in the works are using drones and AI to check aircraft for damage after lightning strikes and for peeling paint, which is an issue on aircraft made of composite materials like the Boeing 787 and Airbus A350.

“We are experts in robotics, we’re not experts in aircraft maintenance,” Sprehe said. “We need the input from our customers to tell us what makes sense.”

AI inside an engine

Though every part of an aircraft is essential, its thrust-producing engines are at the top of the list. At the Paris Air Show in June, jet manufacturer Pratt & Whitney announced an AI-based tool to streamline the engine inspection process.

Called Percept, it lets mechanics photograph an engine using a phone or tablet. It then analyses the photos for missing components and pulls a real-time report of part availability from a cloud-based interface.

Percept Tool scanning a PW V2500 engine on a mobile device
Scanning a PW V2500 engine on a mobile device

Powering Percept is a computer vision and artificial intelligence platform built by Awiros, a startup based outside of Delhi. Pratt & Whitney says the tool can cut the required time for an inspection by as much as 90%.

“Currently, the engine inventory process is performed manually. It can be labour intensive and time-consuming,” O Sung Kwon, Pratt & Whitney’s Vice President for Customer Support, told us in an email. “Percept automates much of the manual effort required for the inventory process.” 

To teach Percept about an engine, Pratt & Whitney captures 360-degree videos from a variety of camera angles. The videos are then separated into images, which helps the tool detect and identify individual parts.

PW and Awiros team using the Percept Tool to scan a V2500 engine
Using Percept Tool to scan a V2500 engine

Sung Kwon said the tool is designed for someone leasing an engine and needs to verify it’s in top shape when the lease is over. Saving both time and money is the goal.

“The Percept tool empowers our customers to scan engines on their own. So they know what needs to be done to properly return an engine before it’s redelivered.”

Percept currently does not analyse an engine’s parts for damage. However, the company promises it will continue to develop the tool and add capabilities. 

Mainblades says it will keep enhancing its tool, as well. But however it progresses, Sprehe says the point always will be helping people do their jobs. 

“It’s not meant to be completely reinventing the wheel,” he said. “Humans and robots should work hand in hand.”

How AI is shaping every industry

AI is now present in almost every industry. Below are more stories on how it’s being deployed.

Kent German
Kent German

Kent German is an award-winning journalist who spent almost two decades at CNET as a senior editor. Kent is based in San Francisco but covers technology all around the world.