United Airlines executive Tom Romanowski envisions a day when the voluminous sensor data generated by modern aircraft flows in real time to his team's data-crunching computers.
At that point – perhaps only then – airlines may finally begin achieving the broad, long-sought-after operational benefits promised by the so-called "big data" revolution.
"We are not there yet," says Romanowski, United's director of technical operations performance and analysis. "We are getting there with a couple of systems, but this is going to be a much longer process than most people realise."
Most commercial aircraft already transmit some data directly to the ground and airlines already reap real operational benefits from data they collect.
But the bigger promise – to analyse real-time sensor data and weave the findings into an active airline network – remains a prospect challenged by technology, fleet age, culture and data-access difficulties, say airline maintenance executives.
"The measure of success, or Holy Grail of predictive maintenance, is to find the problems before that plane and pilots are impacted," says Ross MacArthur, Southwest Airlines' fleet chief within the carrier's technical operations department. "There's still a lot of opportunity out there for the 737."
FlightGlobal asked more than 20 airlines about their use of aircraft data. Some failed to respond; others declined, citing the need to focus on other priorities during the busy summer travel season.
But several did respond, providing insight into their data capabilities. They discussed the ways in which they already use data to pre-emptively address maintenance problems and reviewed challenges to achieving broader goals.
Brazilian carrier Gol collects data from maintenance logs, pilot reports and 1,500 aircraft system parameters.
Its data monitoring team feeds that information into statistical analyses to study key engine readings (turbine temperature and fan vibration, for instance), avionics faults and events such as hard landings, Gol says.
Gol uses those findings to troubleshoot potential problems, including early degradation of aircraft systems, and to perform pre-emptive maintenance.
Data has particularly helped Dallas-based Southwest to address recurring maintenance issues, such as those involving Boeing 737NG pneumatic systems (which power air conditioning and pressurisation) and flap and slat sensor systems, says MacArthur.
Several years ago, pneumatic problems were among Southwest's top maintenance concerns. So, the company added sensors to 737NGs to collect real-time data on how well the air conditioning systems were working, MacArthur says.
That data allowed Southwest to fix problems before they occurred. As a result, pneumatic problems are no longer among Southwest's top 20 maintenance concerns.
"That's a dramatic improvement," MacArthur says.
Southwest receives most its aircraft data via aircraft communications addressing and reporting systems (ACARS), paying for transmission "just like a cell phone plan", he adds.
Chicago-based United also receives data in flight via ACARS, but the messages typically include only limited "snapshots" of information, says Romanowski.
The company's 777-300ERs and 787s can transmit "rich sensor data" over wi-fi while at the gate, but technicians also manually download data from onboard quick access recorders (QARs) to disks.
By analysing that data, United can notify technicians of potential maintenance problems and use it to write new ACARS reports that include data sets of specific concern, Romanowski says.
The carrier uses machine learning to "categorise and classify" maintenance log data, enabling it to understand which issues affect which aircraft, and to predict potential problems.
For instance, the technology can help United determine the cause of recurring fault messages, possibly avoiding the lengthy process of troubleshooting by a technician, Romanowski says.
AWASH IN DATA
Today's widebody jets generate about 30 times more data than the aircraft they replace, and the worldwide fleet of such aircraft continues to grow, notes a 2017 report from consultancy Oliver Wyman.
But airlines face notable barriers to fully using all that information.
Gol cites ACARS transmission costs and "limited computational power" as hindrances.
MacArthur notes that different aircraft types generate vastly different volumes of data.
The company's 737NGs (it has 715) spit out far less information than its new 737 Max. Because the Max retains some older systems, it cannot match the data output of new widebodies, he says.
Southwest has therefore needed to create its own algorithms and reports, an intricate and lengthy process, says MacArthur.
United faces familiar challenges.
Its 737NGs, 757s and 767s (457 aircraft in total) generate far less data than newer types, Romanowski says.
The company also faces delays receiving sensor data, much of which comes from QAR downloads that occur at five- to 10-day intervals.
In addition, United's pilot contract requires it to consult the union before it can access QAR data, and the data is scrubbed of flight-identifying information, Romanowski says.
"We don't get QAR data for a couple of weeks after the flight," he says. "It's not terribly useful… for operationalising big data or machine learning models."
Also, some workers – particularly those extremely familiar with aircraft – can be "philosophically" sceptical of data science, preferring instead to draw their own conclusions, Romanowski says.
More broadly, much of United's maintenance information lies not in databases but in PDF documents.
"The technology infrastructure… has never existed," he adds.
Challenges aside, Romanowski has not lost sight of the goal: for rich sensor data to stream off aircraft in real time.
"Once we have that, it's almost game over," he says. "We are sprinting, at that point, to come up with as many predictive models as we can."