D A T A A N A L Y T I C S
BUSINESS AIRPOR T INTERNATIONAL J U L Y 2 0 1 9 | 19
applications in the business aviation sector.
Trout attributes this lack of uptake to a
couple of key factors. The first is that, besides
data collected from the airframe, the data sets
generated within business aviation are usually too
small to qualify as big data. He says that since
machine-learning techniques require large volumes of data
to learn “AI could struggle to produce results you would
really want to trust”.
“Certainly, companies can and should invest in data
collection and analysis – you don’t need AI or big data to
gain a lot of value from your data. But the truth is many
companies struggle to do this effectively because they are
very resource constrained or don’t have the competence
internally to start programs in these areas.”
European charter jet company GlobeAir has introduced
AI into its booking protocols. In response to an increase
in online booking enquiries the company introduced an
automatic pricing system that leverages more than 100
separate parameters. After running a data analysis on
the requests it found that of two million, only about 6,500
translated into bookings. The company realized that “we
couldn’t accommodate all these communications from a
staffing perspective, especially with such a low conversion
rate,” said GlobeAir’s CEO and founder Bernhard Fragner,
speaking at the EBACE conference in Geneva, Switzerland
in May. “We needed to find the needle in the haystack – the
high potential customer that we could use.”
To help find these customers GlobeAir successfully
developed a machine-learning algorithm that uses data
about the customer to assess whether or not their request
is worth following up. However, despite the algorithm’s
success, staff still found that they were being inundated
with phone calls and email communications from potential
customers. “This industry is still very much about personal
contact,” says Fragner.
Maintenance
Another potential application of AI within business aviation
is predictive maintenance. This involves combining historical
data, sensor data from aircraft and control system data
into a data set. Data analytics and AI can then be used
to predict upcoming maintenance issues with aircraft
fleets so you can pro-actively address them. However, the
implementation of such systems in business aviation is
some way off. Aircraft OEMs and maintenance, repair and
overhaul (MRO) firms are still overcoming technological
issues including challenges with legacy aircraft, data
Data safety
Avinode Group’s chief
technology officer Noel
Trout has three basic tenets
to successfully handle data,
no matter how large the
amount is.
Compliance – “Are you
legally compliant in your
collection and use of the
data? To be compliant
you will need to deeply
understand what data you
have, how sensitive it is and
what exactly you do with it.”
Access – “Ensure that only
authorized users have
access to the data. This
includes staff and external
parties. Ideally, have an
audit log of all access and
intelligent alerting around
suspicious activity.”
Retention – “As a company,
if you are collecting lots of
sensitive data you will have
to weigh the value you gain
from having that data versus
the risk you carry if you were
to have an incident with
regards to that data. Having
a well thought through
retention policy can be a
way to balance those two.”
“Companies should invest in data collection
and analysis; you don’t need AI or big data
to gain a lot of value from your data”
“Big data refers to data sets of such
large size and complexity that they
present a challenge for traditional
computer systems to cope with in
regard to storage and analysis”
Noel Trout, chief technology officer at the Avinode Group
transmission and the development of the tools for
accurate predictive analysis.
Another issue hanging over the development
of these predictive systems is the question of who
owns the aircraft data. In applications with airlines
so far deployed, most agree that airlines own the
data their aircraft produce, but a OEM who takes
the raw data and runs it through their analytics
tools owns the intellectual property of the output.
Operations
Introducing AI into aircraft systems also raises
even more thorny questions, not least of which
is how to ensure the AI doesn’t make mistakes.
Kurt Doughty, senior manager of prognostics,
health management and data analytics at Collins
Aerospace, believes there are several implications
to transferring the machine-learning technologies
currently used in smart phones to running aircraft.
“If you think of Alexa, when you ask it to play
a song it might or it might not guess the right
song,” said Doughty, speaking at the EBACE
conference.
“From the user’s perspective they just ask
again until the right song plays. But in the aviation
industry you can’t have a situation where the AI
picks the wrong thing in flight, or deals with some
data and misconstrues the data and ends up
causing some kind of issue with the aircraft.”
To counter this problem developers in the
field are currently looking at a concept known as
“humble AI”.
Doughty says, “You make sure the algorithm
knows its own bounds and when its outside those
bounds it pushes it back to the user to make the
decision. You teach the AI to understand that it
knows that it doesn’t know everything.”