On December 5, the MEETT in Toulouse hosted the Future Intelligence, an event dedicated to AI for aerospace industry.
Bringing together professionals, academics and decision-makers, this session explored the multiple facets of AI and its transformative potential for the aerospace industry. The presentations covered various topics, from image and signal analysis to simulation, decision support and innovative document management processes.
In this article, we invite you to discover how AI is transforming the aerospace sector. We will be revisiting conferences addressing :
- compliance and safety,
- sustainability,
- hybrid AI and simulation,
- predictive maintenance,
- the development of AI-related skills.
Compliance & Security: The challenge of certifiability
“Every takeoff is optional. Every landing is mandatory”
Rule #1 for the beginner pilot
A tradition of software engineering innovation for reliability
The aeronautics and aerospace industries are distinguished by their unmatched level of complexity and certification requirements. In these sectors, where every failure can have major consequences, safety and reliability are absolute imperatives.
From the creation of the ADA programming language in the 1970s, through software certification techniques based on formal proof, to more recent simulation validation approaches, the aerospace industry has always positioned itself as a pioneer in developing innovative methods to enhance the reliability of semi-autonomous systems. France, in particular, has played a key role in these advances thanks to its talent pool and the excellence of its scientific research.
The crash of Ariane 5 Flight 501 in 1996, caused by a few corrupted bits, left a lasting impression. It was a turning point in recognition of the importance of applying the same rigour in reliability to software as for physical components. Since then, new approaches have been developed to address these challenges. However, the emergence of AI for aeronautics is rekindling discussions. It highlights the need to design new validation methods, as current tools are not adapted to these new types of software.
The challenges posed by artificial intelligence in critical systems
Regulatory bodies, such as the EASA and the FAA in the United States, are committed to developing normative frameworks to integrate specific criteria related to AI systems’ robustness, resilience and safety. However, the path remains strewn with pitfalls. Everything remains to be designed: the necessary tools and the methodology that will accompany them.
The integration of AI in aerospace presents specific challenges. Embedded systems must meet rigorous performance requirements while ensuring impeccable transparency and traceability of AI decisions. These aspects are essential to enabling engineers and regulators to understand and validate algorithms’ functioning, especially in critical situations requiring rapid human intervention.
Explainability, reliability and cyber risks: the new pillars of certification
Many interventions highlighted the challenges associated with connectionist models, contrasting with traditional approaches, often favoured for their explainability. David Sadek, from Thales, also emphasised:
“Artificial intelligence must be explainable to generate trust, particularly in contexts where security is paramount.”
Patrick Fabiani of Dassault Aviation highlighted the importance of designing AI systems that perform well and whose prediction reliability can be rigorously qualified.
This includes their resistance to cyber risks, such as adversarial attacks. To meet these challenges, it appears essential to set up Red Teams. They specialise in offensive security and can assess and strengthen the robustness of models. These approaches are becoming necessary to ensure the resilience of systems integrating AI.
It is also worth noting that warranty requirements vary across applications. Critical functions like trajectory management, autonomous navigation or predictive maintenance require high trust. In contrast, other areas, like logistics, can adopt AI with greater flexibility and speed.
Sustainability: AI for a sustainable future
“The ONLY time you have too much fuel is when you’re on fire.”
Rule #5 for the beginner pilot
The aerospace industry is leading the transition to a greener future. Bruno Dahan of Aerospace Valley states:
“The goal is to eliminate carbon emissions by 2050.”
More than in any other area, the belated awareness of the need for environmental protection and emission reduction will have an unprecedented impact on this industry. The time when we could burn tons of kerosene without a guilty conscience is over.
When AI in aerospace charts the course for a greener, smoother sky
AI for green approaches attempts to use AI to benefit the environment. ML models developed by OpenAirlines, for example, offer advice to airlines on optimising their flights and reducing fuel consumption.
Some approaches, such as continuous descent operations (CDO), can optimise fuel consumption. However, commercial flights must follow predefined trajectories within air corridors to ensure separation between aircraft and facilitate air traffic management while flight levels structure airspace. Thus, during the approach and landing phases, aircraft generally cannot perform a continuous descent but must follow a step descent, which increases fuel consumption. Experiments are currently underway to optimise this approach.
To this day, air traffic management still relies entirely on controllers—humans. Automating these systems, with the potential integration of artificial intelligence, would make managing more complex situations in three dimensions possible. This advance would help smooth traffic flow while ensuring an optimal level of safety.
Perform non-destructive testing and extend equipment life
TESTIA, led by Guillaume Ithurralde, CTO, is redefining quality control standards by combining 3D tomography and artificial intelligence. TESTIA is specialising in non-destructive testing and manufacturing process monitoring. They use techniques from industry and the medical field, such as radiography and tomography, to accurately detect anomalies, particularly those produced by additive manufacturing methods.
The analysis is based on astronomical volumes of data processed by ML models to identify defects invisible to the naked eye. At the same time, TESTIA is working closely with researchers and manufacturers to establish new qualification standards. The objective is ambitious: to introduce automated assistance systems by 2027, leading to fully autonomous solutions by 2028.
At a conference, attention was focused on InDRa (INcreasing electrical Drone RAnge), an innovative project aimed at increasing the autonomy of electric drones by 20% through the strategic use of frugal artificial intelligence and advanced mathematics. This project, led by a consortium bringing together ADAGOS, a specialist in AI and deep learning, the Toulouse Institute of Mathematics and the drone manufacturer DELAIR, proposes an innovative approach based on creating a digital twin. This virtual model, combining the characteristics of the battery and the drone, is used to identify the ideal configurations to reduce energy consumption while optimising the management of flight controls.
“Our goal is to show that our tools can achieve the 20% increase in autonomy announced”
Mohamed Masmoudi, CEO of ADAGOS
Funded as part of the MAELE (Light and Environmentally Responsible Air Mobility) call for projects and supported by the Aerospace Valley competitiveness cluster, InDRa is part of an ambitious vision of decarbonising light aviation by tackling concrete challenges with cutting-edge technological solutions.
The ecological emergency and the need for radical transformation
The ecological impact has become a central concern for all players in the aeronautics industry. The sector is at a critical turning point when facing the urgent need to reduce greenhouse gas emissions. Without concrete and sufficiently ambitious solutions, it is plausible that this industry, as we know it, will disappear in the coming decades. The need to reinvent ourselves is therefore vital. Not only for the planet but also for the very sustainability of aeronautics.
As one of the speakers explains with great lucidity:
“It is no longer a question of applying incremental optimisations, allowing us to pick up a few performance percentages. Doing what we have done for the last 20 years will not allow us to continue.
Aeronautics is THE field where almost everything has already been optimised to the maximum. The hope of AI is to bring an entirely new perspective to our processes and our production lines. We do not need an optimisation AI. We need a disruptive AI that allows us to rethink our processes from scratch and imagine tomorrow’s aircraft. Without these developments, it could well be that our grandchildren will never be able to travel by plane. It is a fantastic technology, and it would still be a shame.”
This observation calls for a systemic transformation based on emerging technologies capable of redefining existing paradigms and offering disruptive solutions to reduce air transport’s environmental impact drastically.
Predictive Maintenance, AI and Simulations: A Successful Hybridization
“The propeller is just a big fan in front of the plane that keeps the pilot cool.
When it stops, you can watch the pilot start sweating.”
Rule #6 for the beginner pilot
Analysing technical problems with a simple smartphone
Akawan, nicknamed the “Shazam of acoustics,” is revolutionising leak detection using artificial intelligence and enriched data. Presented by Gabrielle Arnault-Lazard, this innovative solution achieves an impressive 98% accuracy in anomaly analysis, offering a fast and efficient method to ensure door quality. Flight attendants can now perform a diagnosis using a simple smartphone application, simplifying checks and reducing the number of interventions required.
This performance is based on an ingenious strategy: Akawan started with a modest volume of data, only 500 megabytes, before enriching its model with synthetic data to improve its robustness. The goal is to provide reliable detection in varied and complex environments.
True to its ambition for innovation, Akawan is currently working to integrate multimodal data, combining acoustics and vision, to increase diagnostic accuracy further. This proactive and technological approach is part of a logic of continuous improvement, positioning Akawan as a key player in optimising control and maintenance processes.
Automated inspections using drones
Autonomous drones transform predictive maintenance, enabling precise inspections in diverse and often hostile environments. Alexis Pradille of Delair and Mohamed Masmoudi, CEO of ADAGOS, presented solutions that leverage the synergy between Delair’s expertise in long-endurance drone design and ADAGOS’ advances in frugal artificial intelligence. These devices, capable of travelling up to 100 km autonomously and flying for 6 hours, are used for critical missions, such as power line inspection.
Robustness is at the heart of their design, allowing them to withstand extreme environments like deserts, icy waters, or conflict zones like Ukraine. Integrated technologies also ensure enhanced operational security: the drones remain functional despite attempts to use GPS spoof or radio jamming, using autonomous position estimation algorithms.
With these capabilities, predictive maintenance drones provide an innovative technological response to safety and airworthiness challenges while minimising human intervention in demanding contexts. These advances are part of an ambitious industrial vision in which reliability, autonomy, and efficiency converge to optimise operations and reduce costs.
Computer vision, at the service of logistics optimisation
Daher, represented by Maxime Vicaire, Logistics Innovation Manager, presented a breakthrough in supply chain optimisation with its “Palet AI” platform. This innovative solution uses computer vision to fill data gaps, transforming logistics flow management. Using six strategically positioned cameras, the system analyses pallets and packages in real-time, automatically identifying priority items.
The information collected is displayed in augmented reality on screens. It provides a clear visualisation of the pallets with key indicators such as their priority level, waiting time or their precise position in the warehouse. This approach eliminates the “blind spots” of traditional processes, particularly the critical moments when a pallet can be delivered but not immediately processed. According to Maxime Vicaire, “AI allows us to look for data where there is none“, an essential approach to achieving high efficiency.
By leveraging computer vision, Daher also optimises on-time delivery (OTD). Accurate tracking of pallet presence time and automated flow analysis provides valuable support to operators, simplifying decision-making and reducing inefficiencies.
Anticipating breakdowns, thanks to hybrid AI
At Airbus Helicopters, artificial intelligence transforms maintenance from a traditional corrective approach to revolutionary predictive logic. The “Helico Health Monitoring” solution is based on advanced analysis of mechanical vibrations, combining hybrid models and symbolic AI. This dual approach improves aircraft availability and operational safety and significantly reduces maintenance costs.
The process begins with analysing vibration patterns using connectionist models, which can detect anomalies based on a normality model. These results are then interpreted by a symbolic AI, applying expert rules and fuzzy logic to determine the failures’ origin. Ammar Mechouche, representing Airbus Helicopters, highlights the impact of these innovations:
“Thanks to these methods, we prevent failures and gain in efficiency.”
Historically, Airbus Helicopters relied on vibration analysis and signal processing. Today, integrating new data sources and hybrid systems enables proactive detection of weak signals, anticipating failures before they occur. However, these advances require rigorous safeguards to ensure the reliability of the analyses. Particularly in a context where connectionist models pose significant challenges in terms of certification due to their statistical nature.
ANITI: a unique research ecosystem on hybrid AI
“Learn from the mistakes of others.
You won’t live long enough to make all of them yourself.”
Rule #9 for the beginner pilot
Presented by Romaric Redon, its operational director, the Artificial and Natural Intelligence Toulouse Institute (ANITI), is a cutting-edge interdisciplinary artificial intelligence (AI) institute, born from the Future Investments Program (PIA3) and located in the Toulouse metropolitan area, the cradle of European aeronautics.
It brings together more than 200 researchers from universities, engineering schools and research organisations (CNRS, INRAE, CNES, ONERA, etc.) and around thirty industrial partners (Airbus, Thales, Renault, Continental, etc.). Supported by the Occitanie Region and Toulouse Métropole, ANITI is part of the national dynamic, alongside the three other French 3IA institutes (Grenoble, Nice, Paris), to position France among the world leaders in AI.
A strategic focus on hybrid AI and its aeronautical applications
ANITI’s originality is based on a hybrid AI approach that integrates, in a tightly coupled manner, machine learning from data (machine learning, deep learning) and symbolic models based on rules, constraints and logical reasoning. In the aeronautics sector, this integration meets several challenges:
- Reliability and Robustness: AI systems must meet extremely strict safety standards. Hybrid approaches ensure the consistency and resilience of algorithms, thanks to the formalisation of critical properties and constraints that guarantee safe and predictable operation.
- Explainability and Transparency: Decisions’ interpretability is crucial in a field where trust is imperative. Symbolic models give meaning to predictions, which facilitates the understanding and verification of AI systems’ choices by engineers and certification authorities.
- Compliance with Physical Laws, Integration with Simulations and Surrogate Models: Aeronautics relies heavily on simulation to design and validate complex aircraft. Hybrid AI, rather than supplanting these established approaches, complements them. Symbolic models integrate fundamental physical laws, ensuring that predictions do not stray from the basic principles of aerodynamics. In addition, using “surrogate models” – reduced models – accelerates aerodynamic simulation and embedded system design by reducing the need for intensive calculations. This complementarity allows for more efficient design and validation space exploration without giving up traditional simulation and numerical validation techniques.
High-impact research programs
ANITI is structured around three programs (IP):
Acceptability, fair representative data for AI
Developing methods to ensure ethical, fair AI that complies with international standards is a major challenge in a globally regulated sector such as aviation.
Certifiable AI towards autonomous critical systems
Developing hybrid AI tools for certifiable autonomous systems adapted to aeronautics’ safety and security constraints.
Assistants for design, decisions and optimised industry processes
Design intelligent assistants that jointly exploit physical simulations and machine learning in order to accelerate the design process and improve predictive maintenance, traffic monitoring and industrial optimisation without supplanting traditional approaches but by reinforcing their effectiveness.
In short, ANITI embodies a new generation of AI institutes. They focus on hybridising approaches, integration with physical simulations, and complementarity with proven methods to design reliable, explainable, and sustainable systems. Thus, it will meet the major challenges of the future aeronautics.
Industrial AI: a balance between innovation and know-how
“Good judgment comes from experience.
Unfortunately, the experience usually comes from bad judgment.”
Rule #20 for the beginner pilot
The rise of artificial intelligence (AI) requires major changes in skills, technologies, and collaborative strategies. The discussions at the Future Intelligence event revealed three key areas:
- training
- technological innovation
- and the complementarity between know-how and innovation
Training to meet the challenges of tomorrow
AI is profoundly transforming professions and requiring a renewal of skills. Patrick Fabiani (Dassault Aviation) insists: “Training is crucial to combine industrial expertise and the potential of AI“. This issue goes beyond industry boundaries and concerns public authorities and the education sector.
As Agnès Plagneux Bertrand (Toulouse Métropole) points out:
“The public sector is responsible for anticipating future needs and preparing new generations“
This renewal of skills is based on a hybridisation of knowledge. The speakers are unanimous about the importance of combining business expertise, understanding of physical phenomena, and data-driven approaches. Far from eliminating traditional know-how, AI must strengthen it by relying on adapted training and innovative educational tools.
“We must not lose the richness of physical models in this transition to data-driven AI“, they insisted. The future is emerging not as a technological breakthrough but as an evolution based on complementarity.
Innovating through hybrid approaches
Hybrid approaches, such as surrogate models, are transforming industrial practices. These innovations help accelerate processes and reduce risks and costs while meeting the specific needs of critical sectors such as aeronautics.
Jayant Sen Gupta (Head of Airbus AI Research) explains that the transition from “artisanal” AI to “industrial” AI is essential. This involves the development of standardised frameworks, reliable software libraries and appropriate certifications. These technical solutions are accompanied by an opening to the academic ecosystem thanks to open data and open source initiatives. This cooperation promotes a faster and more relevant orientation of research.
Building Trusted AI Through Collaboration
As Marc Sztulman, Regional Councillor Delegate for Digital Affairs in the Occitanie Region reminds us, AI is a subject of enthusiasm and concern among the population. It is no longer just a technical field but also a social issue that will have a very real impact on how our society functions.
It is, therefore also up to public services to serve as catalysts and regulators, through financial support for this type of event, to promote the development of this French ecosystem, which, on its own, will be able to allow us to have a trustworthy AI that we will know how to use in such a way that the future of artificial intelligence does not resemble HAL’s AI in A Space Odyssey, with the navigator becoming shipwrecked and accelerating our fall, but rather an R2-D2, a friendly and helpful companion.