Artificial Intelligence (AI): how far have we come in aviation?
The incredible achievements in Artificial Intelligence make it seem like anything is possible…In aviation however, AI is confined to non-critical ground operations. For more complex tasks, such as air traffic control or flying the aircraft, AI has yet to overcome aviation-specific challenges in terms of dependability, cybersecurity and the real-time factor.
AI grounded until further notice
In the aviation sector, artificial intelligence has proven to be an effective tool for extracting useful information from image databases, and for performing repetitive, time-consuming tasks requiring many operators. In this vein, the French firm Donecle uses its autonomous UAVs, equipped with a visual recognition system, to inspect aircraft and detect faults after an automated learning phase using databases. Artificial Intelligence can also be of great use in helping assess data collected during flights to enhance troubleshooting activities and to provide quick analyses of in-service events experienced by the aircraft.
Artificial Intelligence is also attracting the interest of airport operators, for its potential in terms of air traffic management and optimisation of airport capacity. The challenge is to improve the way we predict changes in the number of aircraft entering and leaving an airport, with a view to keeping take-off and landing queues as short as possible, while optimising the allocation of parking spaces for passenger disembarkation. A technology which could become critical, given the new sanitary constraints, distancing measures and cleaning procedures which are becoming the new normal for today’s air travellers.
« Eventually, the benefits of connectivity and the integration of AI into information and decision-making systems will be two-fold: optimised flight paths and, crucially, lower fuel burn leading to reduced CO2 emissions and a lighter workload for flight crews. »
Stéphane Viala, Senior-Vice President Engineering at ATR.
AI in the cockpit: coming soon?
It is true that a host of flying applications currently in development or test phase involve the use of AI for autonomous flight planning, without any human intervention. However, for the time being, they are confined to UAV flights or tests on “flying taxis”, for short urban flights.
All issues become more complex when it comes to aircraft, given that a plane’s flight path depends on weather conditions and air traffic. By constantly recalculating the impact of these two parameters on the route, it would be possible to optimise the flight path, thereby reducing the flight time and fuel burn. What if Artificial Intelligence could help us reduce our environmental footprint even further?
With this in mind, engineers are working on how to integrate artificial intelligence combined with miniature quantum computers into the next-generation Flight Management Systems (FMS). Expected to be on the market within the next 5 years, these applications will enable instant correlation of on-board data and data from the external environment. For example, if it receives real-time information about storms in the flight zone, the system will be able to recalculate the ideal flight path, avoiding the turbulence zone, for a smoother flying experience.
“Eventually, the benefits of connectivity and the integration of AI into information and decision-making systems will be two-fold: optimised flight paths and, crucially, lower fuel burn leading to reduced CO2 emissions and a lighter workload for flight crews. The possibility of having planes fly with just one pilot on board, the co-pilot being replaced by a range of virtual assistants, could then start to look more convincing. However, the other side of the coin is that the upsurge in external data in the cockpit requires zero-failure cybersecurity.” says Stéphane Viala, Senior Vice-President Engineering at ATR.
Humans still have the edge over machines
Cybersecurity is one of the significant challenges facing aviation that constitute obstacles to the development of Artificial Intelligence in the sector. Needless to say, aircraft manufacturers cannot make do with the systems provided by the Web giants, because the stakes and data involved are in a whole different league, and the level of reliability required exceeds by far all mainstream uses.
Stéphane Viala adds: “The other major challenge is related to the certification of critical systems developed using empirical methods for AI (in other words, based on learning algorithms), which still tend to make a lot of mistakes.”
Certification will be the final essential stage to be validated before autonomous planes flown by Artificial Intelligence take to the skies. That implies achieving perfect trust between operators and the machine, since it goes without saying that humans will always remain in the loop. If we look at it from that angle, there is still a long way to go…
“Technically, the industry is probably more prepared for autonomous aircraft than for self-driving cars, notably because the environment is simpler. There are no pedestrians or cats crossing the road, no signs, etc. The biggest hurdle, beyond data security aspects and complex failures management, is likely society coming to terms with man being replaced by machine!” observes Leopold Sartorius, Head of Data Analytics at ATR. “In the short term, decision-support tools will flourish in both cockpits and control rooms. It will still be a human decision, of course. However, the decision-making process may be eased by tools having potentially a much greater capacity for processing information than ours!” he concludes.
Share this article
