Military aerospace is a driver of innovation, with a long history of adopting and adapting new technologies. However, questions remain around how AI capabilities can be introduced while meeting the sector’s famously exacting requirements for safety, reliability, security and certification.
These questions are at the heart of a recent episode of The Innovation Crowd, a podcast hosted by innovation specialist Helen Dawson, featuring Jonathan Burnip, Head of Advisory Services at Marshall Aerospace.
In the first of three articles unpacking Helen and Jonathan’s discussion, we explore what effective AI adoption could look like in military aerospace, from regulation and data quality to security, collaboration and the realities of working across complex programmes.
Jonathan Burnip is Head of Advisory Services at Marshall Aerospace and a Fellow of the Royal Aeronautical Society.
The Advisory Services team draws on Marshall Aerospace’s experience gained through complex engineering and integration programmes to help innovators address new technical and organisational challenges.
Debunking the “regulation vs. innovation” myth
Aerospace is often portrayed as a uniquely challenging environment for developing and proving any technologies that originate outside the sector. After all, aircraft systems are subject to rigorous safety and reliability standards, and new technologies also have to fit within established approaches to certification and assurance. Military programmes add further layers of considerations around security, intellectual property and export control.
Many people therefore assume that certification authorities are instinctively reluctant to accommodate AI.
In his discussion with Helen, Jonathan takes a different view, noting that regulators are actively studying the latest developments in AI and considering how their existing frameworks may need to evolve as applications evolve.
“There’s a perception that certification authorities either don’t allow or are against AI adoption. That’s not really the case.”
For aerospace, the core issue is whether there is enough evidence to support confidence in a new application. The same principle applies beyond AI, extending to any new capability introduced into a safety-critical environment.
That can certainly make adoption look different from sectors where the consequences of failure are less severe, and it is true that some applications may develop more gradually while methods of assurance and certification catch up.
However, the regulatory approach is cautious, rather than dogmatic and closed-minded: innovation is perfectly possible and encouraged when there is a clear understanding of performance, reliability and risk.
Can aerospace provide the data AI needs?
Aerospace generates a large amount of data through design, testing, operation and maintenance. At face value, that makes the sector particularly fertile ground for AI.
The difficulty, as Jonathan explains, is that aircraft and their systems are deliberately engineered to be highly reliable: structures are designed to be fatigue-resistant and damage-tolerant, while safety-critical systems may be designed around extremely low probabilities of failure. Aircraft can also spend most of their working lives operating within only part of the conditions they were designed for.
As a result, the events engineers most want to understand are, paradoxically, some of the rarest in the dataset. A large volume of information gathered over many years may contain only a handful of examples of a particular event, and even those examples will only depict a partial selection of the operating and environmental conditions in which it could occur.
“We need quality of data and assured data to make sure we can trust the outcome.”
This places greater emphasis on whether data is representative, traceable and sufficiently well understood to support the application being developed.
As Jonathan points out, simulation and synthetic data can help fill some of the gaps. Aerospace already makes extensive use of modelling and simulation to explore conditions that would be difficult or impractical to recreate repeatedly through physical testing. Confidence in synthetic data, however, depends on confidence that the underlying model represents the relevant real-world behaviour well enough. As a result, established disciplines around testing, verification and validation remain central to deciding whether the resulting outputs can be trusted.
AI in a world of controlled information
One especially pronounced challenge facing military aerospace is the difficulty in sharing useful information freely between organisations. Significant amounts of data that could potentially be used for, or by, AI may be classified, export-controlled or commercially sensitive.
Individual pieces of information can also become more sensitive when combined with other data. Jonathan gives an example of information relating to the fuel capacity of a military platform and separate information about the efficiency and weight of its engine. Either dataset might be individually appropriate for a particular organisation to access, but together they could potentially allow someone to infer sensitive information about aircraft performance.
“It’s about having the right ways of sharing the data, the right systems to host that data, and trusted people accessing that data.”
No single organisation is likely to hold all of the specialist knowledge, technology and data needed to understand every part of a sophisticated platform or programme. Jonathan describes engineering as a “team sport,” with the team increasingly extending across organisational boundaries.
Data sharing does, therefore, take place across defence programmes, but it may require assured systems, appropriately trusted people and carefully defined boundaries around access.
Sometimes information can be brought together within a secure environment. In other cases, suitably cleared specialists can work where the data already resides. Collaboration with customers, partners and end users can take different forms depending on the sensitivities involved.
These requirements still tend to add cost and complexity, particularly where specialist infrastructure is needed. For military aerospace organisations, the challenge is to provide the access needed for a particular task without weakening the protections around sensitive information.
Jonathan sees security as another constraint with which aerospace engineers are already accustomed to working.
“Good engineering is all about optimising a solution within known constraints.”
Performance, weight, maintainability, certification, cost and safety already shape engineering decisions. Military programmes add security, IP and export control. While AI introduces transformative new technical possibilities, it does has need to be considered within that wider environment. Jonathan’s argument is that the most useful applications of AI will be those that address a genuine engineering need while being supported by the right data, infrastructure, expertise and assurance.
In short: for AI to be used responsibly and successful in aerospace, the technology has to fit the problem, rather than the other way around.
The next article in this series looks at what happens once AI becomes part of a safety-critical engineering process, including the role of human judgement, accountability and the ability to learn when something goes wrong.
In the meantime, you can listen to Jonathan’s full conversation with Helen Dawson on The Innovation Crowd.
Images used in this article were created using generative AI.