The case for AI in flight test

It doesn’t sound right, does it? A test pilot advocating less time in the air. After all, pilots are there to fly, but as a commercially-aware test pilot, I understand the business of delivering a product on time and on budget. I also think AI presents a perfect opportunity to step outside the narrow, siloed test matrix approach allowing a broader application of professional skillsets, better utilisation of airborne time and hopefully, consequently better designs. And if we can do that by minimising the exposure to costly test activities then everyone’s a winner.

Digital-twin (DT) technology has been implemented in aerospace and defence for over 4 decades (Duke et al., 1988), and validating models through testing is commonplace in aviation. Confined mainly to the ‘as designed’ phase, DTs highlight vital production changes early where “electrons are cheaper than atoms” (Digital Twin: Definition & Value, 2020). But recent advances in computing and the growth of AI applications will have an increasingly greater impact on how testing is conducted. I believe that the area with the greatest potential for AI now is in the ‘as tested’ stage where combining DTs, Machine Learning (ML) and simulated aircraft (‘constructives’) to augment physical tests through modelling and simulation would accelerate the airborne flight test stage.

So, it’s therefore worth understanding where these tools can help. What type of technology you want to use to gather test data, needs to be considered early in the programme, because it could affect the way your test programme is planned — and costed. Implementing AI from scratch is not going to be cheap and may well be in excess of a traditional flight test campaign. However, the use of a DT, for example, should be seen as an investment in the and that future savings will be exponentially greater once you’ve added up all the flight tests required for all the modifications over the aircraft’s life. Taking the flight test analogy about data: “Test once, use many times” with AI becomes, “Use once, test many times and ways”.

Taking the flight test analogy about data: “Test once, use many times”, with AI this becomes, “Use once, test many times and in many ways”.

Depending on what you use and what is available, changes may be required in your process to ensure that the selected AI solutions are built into the plan and not dovetailed in at a later date. The digital twin idea — a virtual model of your product from the nuts and bolts to the windscreen thickness — needs a thorough evaluation to enable trust proper utilisation as a trusted replacement for actual flight tests. UK’s next fighter jet, Tempest, will be among the 1st procurement projects to be run on Agile project management principles (as opposed to the typical Waterfall approach) and using AI applications, in a bold attempt to bring the fighter jet into service in record time (well, 15 years but its all relative!) BAE systems planned on using the DT from its inception which has already proven to speed up the design. Conceptual shapes were virtually designed and tested, with high-performance computers able to calculate the aerodynamic performance of different aircraft features before test pilots flew the predicted models in a ground-based simulator — and all of this before a prototype or scale model had been built. After digitally testing, scale models were 3D printed and put through their wind tunnel where more data was gathered to feed back into the DT model to improve its predictions and importantly, building trust in its capabilities. So it is clear: for any flight test project using AI technology, the philosophy needs to be planned into your programme from inception.

Traditional flight test campaigns are almost always flown sequentially, analysing each system’s data independently before progressing to the next, culminating in a series of integration tests. Conversely, DTs could rapidly run concurrent evaluations of the myriad interactions taking place in a ‘system of systems’ providing hitherto unavailable, invaluable insight.

Supervised Learning could also be used to instantly predict the most efficient flight test profile (cost and time benefits) based on any given changes (priorities, weather, airspace, vehicle performance and even the characteristics of the individual test pilot). The resulting ML/DT combination would allow for the rapid execution of these variables at specific test points, especially pertinent when testing edge cases (sudden flight path changes due to aerodynamic or control limitations).

Combining these technologies would decimate the time taken by gathering the right amount of data most effectively, efficiently and more safely than traditional methods, resulting in more targeted physical testing to calibrate the models and close the knowledge gap.

In-flight, ML could track the behaviour patterns of measurement sensors through anomaly detection in real time (Cooke, Melia and Grayson, 2017), affording crews the option of repeating test points before discovery on the ground requiring further investigation flights.

Constructives meanwhile, would allow testing of onboard systems without putting the test crew at risk (Livermore and Leonard, 2020). For example, airborne collision-avoidance tasks could be performed by flying towards a constructive which appears “real” to the pilots. Because the systems onboard behave identically as if the approaching constructive was a real aircraft, the risk of a mid-air collision would be mitigated while simultaneously permitting ‘carefree’ human testing.

The technology now available — especially AI technology — is growing rapidly and faster than we are seemingly able to keep pace with. Some test programmes in the UK military already use DT technology to help reduce the amount of flight test needed, keeping timelines and projects on budget. Despite regulators’ best efforts in being as proactive as they can be for the development of autonomous air systems, these AI concepts are only just coming over the civilian regulators’ horizons. Acknowledging that AI technology will continue to lead regulation in the near future and while evidence can be gathered using the latest advances in digital technology, consideration must be given to the quality and provenance of the data you gather to ensure that the regulators can, or are willing to, accept the data you provide.

While not eliminating the need for physical test flight measurements, ‘digital evaluations’ would significantly reduce the number and dependency on them.

Importantly, routine tasks which can be offloaded to machines would improve the pilot’s situational awareness and increase mental capacity, bringing deeper introspection and insights into the human-machine interface aspects.

Get in touch to find out how an AI strategy could work for you.


2020. Digital Twin: Definition & Value. [PDF] American Institute of Aeronautics and Astronautics, Inc. Available at: <> [Accessed 12 March 2021].

Cooke, A., Melia, T. and Grayson, S., 2017. The Application of Machine Learning Techniques in Flight Test Applications. Curtiss-Wright.

Davies, E., McMaster, J. and Stark, M., 1994. The use of Generic Algorithms for Flight Test and Evaluation of Artificial Intelligence and Complex Software Systems. [PDF] NAWC-AD Patuxent River. Available at: <> [Accessed 12 March 2021].

Duke, E., Hewett, M., Brumbaugh, R., Tartt, D., Antoniewicz, R. and Agarwal, A., 1988. The use of an automated flight test management system in the development of a rapid-prototyping flight research facility. [PDF] NASA. Available at: <> [Accessed 13 March 2021].

Livermore, R. and Leonard, A., 2020. Test and Evaluation of Autonomy for Air Platforms. [PDF] United States Air Force. Available at: <> [Accessed 13 March 2021].

Royal Aeronautical Society. 2021. Revolutionising flight test and evaluation — Royal Aeronautical Society. [online] Available at: <> [Accessed 11 March 2021].



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Matthew Ward

Matthew Ward

Experienced helicopter test pilot and Flight Test Instructor; Human-Centred Design champion with User Experience & Testing in aviation background