How a Data-Driven Approach to Pilot Training will make Aviation Safer while Saving Airlines Money
New rules and new technology are paving the way for adaptive training
The aviation industry is in the midst of a shift towards more evidence-based approaches to pilot training. The European Aviation Safety Agency (EASA) has proposed amending its regulations and drafting new rules requiring pilot training organizations to adopt a competency-based approach:
… to ensure that aviation personnel have the right competencies and training methods to cope with new challenges. This is one of the most significant systemic issues in the European Plan for Aviation Safety (EPAS) 2018–2022.
The basic premise is that total flight hours and proficiency checks aren’t the same thing as competency. Deep knowledge and understanding, along with procedural knowledge, are what we’re supposed to be training for.
If things move according to schedule, the new rules around evidence-based training (EBT) will be decided, along with acceptable means of compliance, in Q3 of 2020. That means that training organizations will soon need to change they way they train pilots.
For decades, the aviation industry has made use of sensors and data to make flying safer for everyone. Many complex systems aboard aircraft use data to see through clouds, navigate, avoid collisions, and provide a smooth journey to passengers. Data from the Quick Access Recorder gets downloaded and analyzed to make flying safer and more efficient. And in the worst case scenarios, black boxes help us understand want went wrong and how we can avoid disaster in the future.
Data from aircraft are starting to inform how pilots themselves are trained. For instance, decisions about pitch and throttle settings during the takeoff and climb can have a huge effect on overall fuel burn, which has an impact on both the environment and the economics of the airline.
But there is a larger revolution happening in training, also involving data. While the idea of EBT is generally to exploit operations data to improve pilot training, the idea of competency-based training (CBT) is to exploit observations about the individual pilot’s knowledge, skills, and attitudes to tailor the training to the individual needs of the pilot.
At Paladin AI, we call this approach adaptive training, because every person learns at a different rate and comes with different experiences and background. A one-size-fits-all approach was the default in the past, but technology is enabling more personalized approaches. And the benefits could be significant.
In this piece, we’ll be describing how adaptive training could significantly improve the economics of pilot training in ways that benefit everyone.
The various kinds of pilot training
Let’s quickly recap how pilots are trained. When first starting out, the prospective pilot does their ab initio training (Latin: “from the start”). This is typically done at a dedicated flight school or at a college/university. There are thousands of flight schools all over the world, but most of them are private and the trainee is typically the one paying the tuition.
A pilot must be certified to fly a particular type of aircraft. Graduating from a flight school does not automatically qualify you to fly a commercial jet. The pilot must obtain an initial “type rating,” which qualifies them for a single type of aircraft. This also means that holding a type rating on an Airbus A320 does not qualify the pilot to fly a Boeing 747. If the airline wants the pilot to fly a different type of aircraft, the pilot must go through a “type conversion”.
Once employed by an airline, every pilot undergoes regular recurrent training on an annual basis. During recurrent training, the pilot travels to an approved training organization (ATO) where for several days they are subjected to a battery of exercises in a flight simulator to test and reinforce their knowledge and procedures. These grueling training sessions are supervised by a flight instructor, who must sign off on each pilot’s proficiency. Assuming they pass, the pilots return home and resume flying for their employer.
The costs of training
It comes as no surprise that pilot training is expensive. To pass all ab initio training, from the private pilot’s license straight through to an Airline Transport Pilot Licence (ATPL) or equivalent, can cost the pilot as much as $250,000 on the high end. The whole process can also take years.
When it comes to obtaining new type ratings or going for recurrent training, it is the airline that is paying for training, and the costs are significant. The pilot must be taken “off the line”, then travel to the training center (which may be in another country or even on another continent), then spend up to a week on site, staying in hotels, before returning home.
There are many factors influencing cost, but recurrent training can cost as much as $20,000 per pilot per year. Large airlines have thousands of pilots and thus spend millions annually keeping them current. In order to help control costs, an airline may build or purchase a dedicated training organization. This only works for large airlines, who can maximize training center utilization. A single full flight simulator costs about $10 million. Smaller airlines usually engage third-party training centers.
By one EASA estimate, the total annual costs of recurrent training for a medium/large airline with 1000 pilots are approximately EUR 13.4 million (about $14.5 million USD).
Cost/benefit analysis of evidence-based training
On the surface, switching from the current training approach to evidence-based training will have a lot of up-front costs. For a typical airline, EASA estimates that adoption will cost EUR 1.1 million in the first year, then EUR 10,000 per year on an ongoing basis.
Training centers implementing EBT/CBT are worried that they’ll have to throw out all their training materials, design completely new courseware, and retrain all their instructors in the new approach. This is a major headache and potentially quite expensive, but the hope is that the training organization will be more competitive.
Airlines can expect to see economic benefits, eventually. As the airline demonstrates compliance and improved safety, they can be granted special privileges: Line checks and ground training requirements reduced from once every year to once every two years.
Regulators also anticipate that with the implementation of EBT, the need for costly remedial training will decrease, as more pilots successful pass their regular proficiency checks.
In the final cost/benefit analysis, the medium/large airline could be expected to save EUR 900 per pilot every year.
But those savings are small, given that an airline might spend 10 times that amount on the jet fuel required for a single flight. Can we do better?
Adaptive training through artificial intelligence
The guiding document for implementing evidence-based training is the ICAO Manual of Evidence-based Training (Doc 9995). Its recommendation places a significant burden on flight instructors:
Competency-based training programmes, such as EBT and ab-initio MPL courses, are highly dependent upon the analytical and assessment skills of the instructor cadre. Furthermore, it is important that only those individuals who
possess a good understanding of the learning process and how to positively influence human behaviour are considered for instructor positions. Prospective instructors should be selected, trained and qualified in accordance with the provisions in Chapter 6, paragraph 6.1.2 of the Procedures for Air Navigation Services — Training (Doc 9868).
This implies that flight instructors will need to be retrained in compliance with the new training paradigm and taught how to assess pilot competencies. In effect, transitioning to EBT is going to be difficult and costly.
There’s also the risk that transitioning away from a very concrete approach of checks and procedures to something where the instructor must correlate observations of behavioural indicators (like “Identifies and manages threats to the safety of the aircraft and people.”) to the outcomes of the training session results in overloading the instructor.
The temptation to default to simply checking boxes instead of assessing competency will be strong.
What if artificial intelligence could be used to detect and measure competency?
Artificial Intelligence (AI) excels at pattern recognition. Show a deep neural network enough examples of some complex pattern in data, and it will learn to see that pattern wherever it occurs.
Assuming that more competent pilots respond differently to challenging circumstances, we can train an AI to recognize that difference.
Aircraft and flight simulators are already covered in sensors and generate magnificent amounts of data. What if this data already contained the signatures of pilot competency? If we could extract that signal from the noise, it would change everything.
ICAO, in their manual, has already done us the favor of identifying all of the core competencies a pilot should have, as well as the behavioral and performance indicators that should be present.
Core competencies are things like situational awareness, workload management, and problem solving.
By interfacing directly with flight simulators, our powerful machine learning systems can analyze hundreds of flight parameters and detect signatures of pilot competency. We use a combination of deep neural networks, Bayesian inference, Markov chain Monte Carlo sampling, and classical supervised and unsupervised learning algorithms to understand what’s happening in the simulator and create a composite profile of the pilot.
Rather than asking instructors to throw away their lesson plans and grading sheets, we simply ask instructors to train using our dashboard, which monitors the pilot in real time and then provides training analytics and competency feedback after the session.
Evidence-based training can be adopted for much less than EASA or other regulators are estimating, but only if we use the tools of modern data science. By offloading the detection and measurement of dozens of behavioural indicators to machine learning algorithms, instructors can focus on the instilling the human qualities that make for better pilots: leadership, teamwork, communication, discipline and professionalism. Those are qualities that no algorithm can deliver. Good teachers embody those qualities.
Adaptive training using the tools of AI will make the industry safer while making training more affordable. The investments made by airlines and training centers into this technology will pay dividends for a long time to come as training needs are reduced 30–50%.
Using the tools of AI, compliance with the new regulations can be achieved in much less time and at much less cost.
Paladin AI is creating the tools and technologies that will enable adaptive training at low cost. It will allow training centers to begin adopting evidence-based training methods without having to completely overhaul their training syllabus or purchase expensive new equipment.
While this article has mainly focussed on the costs of training existing pilots, we expect that adaptive training will also make it easier than ever to become a pilot. Removing the financial barriers that prevent many from launching a career in aviation will not only create economic opportunities for many more pilots, it will also make aviation safer by alleviating the pilot shortage.
As of February 2020, Paladin AI is trialling InstructIQ™, our training analytics platform, with multiple partners around the world. We believe in a future where human and machine intelligence take flight together. Together it will make for a safer world.