Printed headline: Big Data Outcomes
Building a results-generating airline predictive maintenance analytics system is much like constructing a spider web.
It starts as a loose but connected pattern that fills in with a tighter weave as the architect continues to add elements. Soon, the silken threads form closer patterns and the web’s density increases, capturing tangible results—food for the spider and actionable data for an aircraft operator.
The aviation industry has produced a preponderance of data for years but often has not done a good job of weaving it together to form patterns and connect the dots. Information—from pilot reports to maintenance logs to the aircraft communications addressing and reporting system (ACARS)—and other data long remained siloed in separate parts of the web. Now the threads are joining to capture patterns—via algorithms, analysts and machine learning.
While there is much to do to further link data sources together, big data analytics are producing increased aircraft availability, faster turnaround times, fewer maintenance delays and cost savings.
Aircraft generate plenty of data—and the newer the aircraft, the more there is. But even older-generation aircraft and engines generate data from key systems that can be gathered, analyzed and used to drive maintenance programs.
Delta TechOps works with OEMs for analytics but also uses an internal team of 18 people to monitor various sources of data to predict when a component is getting close to failure—from older Boeing 757s to its latest Airbus A350. “We’ve learned that just about every component or system has a signature footprint,” which could be electrical, current or resistance-related, or a frequency in which it operates, says Gary Hammes, Delta TechOps vice president for engineering and quality assurance, planning and logistics.
To help monitor certain components, the MRO has added electrical, mechanical or sound-generating sensors to certain parts to watch for those slight changes. It also basically created an Airbus A320 flying testbed by adding, in conjunction with Airbus, “as many sensors as possible, to learn as much as possible,” says Hammes.
These efforts combined have prevented 3,000 cancellations or delays in the Past two years, says Hammes. Of equal importance: Pending failure predictions have been correct 95% of the time, based on shop tests, meaning TechOps is not simply replacing equipment for no reason.
Delta, which is incorporating Airbus’ Skywise platform into its day-to-day business to track and analyze operations and performance data on its A320 and A330 fleets, also seeks to “build some key joint venture agreements or partnerships with sensor technology companies as well as Airbus and/or Boeing—or both—to continue to drive long-term predictive technologies that will ultimately result in a new way to generate maintenance programs for aircraft,” says Hammes.
In contrast, FlightWatching, a Toulouse-based company launched six years ago, was created to “get the best out of existing fleets without any aircraft modification, by tapping existing data sources,” says co-founder and CEO Jean-Philippe Beaujard. Chief among them: ARINC communication protocols and ACARS, the real-time air-to-ground communication infrastructure used by hundreds of airlines.
The startup specializes in digital analytics, visualization and predictive maintenance. “The FlightWatching software platform is enhancing the usage of this old ACARS protocol by making it very easy to use: What sounded so complex in the ACARS A620 standard now becomes extremely simple,” he says.
The company, purchased by Revima earlier this year, has been monitoring Airbus A300s for more than five years and using its onboard avionics system to “monitor many different ATA chapters useful for maintenance, including hydraulic, engines, auxiliary power units and flight controls,” says Beaujard.
Insights can come from pairing different data sources together, but first you have to gather the information—and do it well. “Data quality is the key issue in predictive analytics,” says Jan Stoevesand, Lufthansa Technik senior director for analytics and data solutions. “The old wisdom from the recording studio ‘garbage in, garbage out,’ is true here as well.” The situation gets more challenging when data comes from different sources.
American Airlines’ legacy infrastructure means “the environments that house the sensor data are different from those that have the operational data” such as delays, cancellations and deferred maintenance items, “which are different from the environments that contain our workload scheduling data or our manuals,” says Stacy Morrissey, managing director of fleet engineering. To rectify this, American has made a concerted effort over the years to store operational data in one place. This creates “one source of the ‘truth’ in our data that is accessible to many,” says Morrissey. “For other data that just doesn’t make sense to move or store in that data warehouse, we leverage tools that allow us to blend and visualize the data so that it appears seamlessly integrated to the end users,” she adds.
While the airline has seen a number of benefits since it started its data analytics journey, Morrissey says American is moving beyond predictive maintenance and delay-prevention “to help troubleshoot chronic issues, identify trends across the fleet and make better operational decisions.”
The work extends to improving the maintenance process. “We use text mining to reduce the amount of time analysts spend manually reviewing data, as well to identify repeat or chronic items on aircraft to aid in troubleshooting,” says Morrissey. By combining data from various sources, the airline has been able to develop indicators of future component-level failures and empower its “maintenance control and planning teams to make better, real-time decisions,” she says.
Lufthansa Technik’s Stoevesand agrees and points out the empowerment piece does not just come from a tool. “The positive outcomes at the end are generated by our ability to master the whole data-processing and fulfillment chain, (i.e., data gathering, cleansing and quality) followed by the actual analytics and, most important, the ability to turn a prediction into an actionable task—directly triggered in the fulfillment system, as in our customer’s organization,” he says. “To achieve this, we had to find a suitable set of tools for the different tasks within this processing chain and integrate these tools for a smooth operation on a 24/7 basis.”
Aviatar, the open platform Lufthansa Technik created, includes an “analytics stack” used by engineers, data scientists and developers to help find these outcomes.
One example is predicting integrated drive generator (IDG) failure modes. “[By] constantly analyzing certain sensor readings, we were able to not only predict failures way in advance but we were also able to suggest simple maintenance tasks such as an oil exchange” to prevent the component’s removal, Stoevesand says.
Etihad Airways Engineering says big data and predictive maintenance are major areas of focus for the group. It uses satcom, VHF, quick-access recorder data, aircraft condition monitoring system (ACMS) fault data, flight schedules, maintenance logs and shop-visit data to glean insights. So far it has relied on OEM monitoring tools to limit unscheduled maintenance and dispatch delays. However, it “is in the process of developing an in-house platform, while in parallel collaborating with major OEMs in regard to predictive maintenance tools for specific components,” says a representative.
Like Delta, EasyJet also is using Skywise—but for its entire fleet. It signed a five-year predictive maintenance partnership program in March 2018 with Airbus to forecast aircraft technical faults before they occur. By using the Airbus platform, EasyJet engineers proactively replace parts before they fail. Delays from technical issues have declined from 10 per 1,000 flights in 2010 to just more than three per 1,000 flights today on the airline’s newest aircraft. Not surprisingly, EasyJet’s goal is to get that number to zero.
One way the airline plans to get there is by installing Collins Aerospace’s flight operations and maintenance exchanger (FOMAX), which EasyJet says can collect 60 times more data than existing systems, on its fleet by this summer. After the installation, EasyJet expects to collect 800 GB of data from up to 24,000 parameters each year.
AirAsia also is using Skywise for predictive maintenance to reduce service disruptions, and it has its own digital transformation team reporting directly to CEO Tony Fernandes, says Nantha Kumar, head of group aircraft engineering. The group also has an internal team looking at using blockchain to reduce aircraft-on-ground downtime.
As a major supplier of both components and communications services with significant content on most large-aircraft platforms, Collins is arguably as well-positioned as any company to satisfy operators’ appetite for actionable data. The company’s Ascentia prognostics and health-monitoring offering, launched in 2018, uses three core data-analysis methods to drive reliability improvement: physics-based insights, statistical analysis and machine learning.
“Aircraft systems are complex, and our deep product expertise enables us to lead with the physics behind our equipment and, in many cases, how the equipment interacts with other systems on the aircraft,” says Collins Associate Director for Digital Programs Shiv Trisal. “This allows more focused, smart-data approaches to assess equipment performance and health. These deep methods are also complemented by data science approaches such as machine learning and statistical analysis to maximize results. It’s not just the quantity of data, but the quality and application of the data.”
Ascentia is platform-agnostic. The current focus is on Boeing 787s and 777s as well as Airbus A320s and A380s—each of which contain significant Collins content. Created through last year’s merger of Rockwell Collins and United Technologies’ UTC Aerospace units—both major players on their own—Collins Aerospace has deep expertise in designing components with sensors, transmitting data to and from aircraft and analyzing it. Ascentia’s early returns show Collins is delivering. Aggregated performance data shared by the company include a 30% reduction in delays and cancellations, and a 20% dip in unscheduled maintenance for fleets covered by the service.
In most cases, the wins come issue by issue. Collins’ predictive models were able to identify deterioration on one 787 cabin air compressor outlet check valve and alert the operator, providing it with a proactive maintenance recommendation. An inspection confirmed the issue, and an operational interruption was avoided.
IDG-wear data from one operator’s twin-engine fleet detected uneven wear between the left- and right-side IDGs, leading the carrier to modify taxi procedures. An IDG temperature difference on another operator’s twinjet was traced to the fuel recirculation system, avoiding unnecessary maintenance.
“While predictive models are one part of the overall equation, the other is better customer support,” says Trisal. “We can place dedicated data analytics expertise on-site, embedded with our customer’s operations. We make specific recommendations on when to remove equipment ahead of a failure. This has led to real and actionable results.”
Collins relies on data available via main avionics buses, using its IntelliSight aircraft interface device. While it could add more sensors, “there is plenty of data that hasn’t been fully mined yet,” Trisal says.
Vibration and Engine Debris
Customers of Calgary-based flight-data streaming specialist Flyht use the company’s Automated Flight Information Reporting System (AFIRS) hardware and FlyhtHealth software service to monitor airframe and engine data and reduce unpleasant maintenance-related surprises. The most common tasks: real-time engine vibration and exceedance monitoring, says Gino Davoli, Flyht’s customer support manager.
The engineering manager from one operator that has used Flyht’s services for 11 years shared some specifics with Inside MRO. “Using the AFIRS trend data has helped us to identify engine degradation over the years, thus enabling us to schedule engine changes before it becomes an on-wing issue,” says the engineering manager, whose fleet includes multiple models of Boeing widebody aircraft powered by either GE or Pratt & Whitney engines. “We have had several engines over the years where we have had N1 [fan speed] exceedances that have alerted us to perform a fan blade lube or even fan balance prior to it becoming an issue,” he adds.
Flyht’s triggered alerts instantly provide data snippets that include chunks of time just before and after the flagged event. A basic setup monitors about a dozen parameters on each engine, including common indicators such as engine gas temperature margins as well as N1 and N2, or shaft speed, percentages. Instant alerts provide 21 sec. of data, including 10-sec. snapshots just before and after the triggering event.
Customers also use Flyht’s services to monitor OOOIs, short for “out, off, on, in,” or when an aircraft pushes back, takes off, touches down and arrives at its destination gate.
“We have 14 flavors of outs,” Davoli says. “Some of our operators use gate-to-gate [for out and in], some want ‘out’ to be when the door closes. We can customize activity for the customer.”
Automated OOOI data-delivery into many common software systems that Flyht supports appeals to both flight operations and maintenance. “[Airline] operations will sign up and select OOOIs, while maintenance is using OOOI data in another system to monitor their parts-wear times and other key metrics,” says Davoli. “When maintenance figures out we can deliver [OOOI data] automatically, they want it, too.”
Flyht’s AFIRS is in use with about 90 airlines, including more than 20 in China, where a satcom device that connects aircraft and operations centers within 4 min. will soon be mandatory. AFIRS uses proprietary software to gather and send data to the ground in real time. It is processed and delivered to the operator using its UpTime server network. Besides health monitoring, the setup supports real-time aircraft tracking and flight-record data streaming that meets new International Civil Aviation Organization Annex 6 standards. The AFIRS unit also serves as a quick-access recorder.
While connectivity is Flyht’s calling card, the company sees significant opportunity in expanding its real-time monitoring services. It has developed a highly customized event-alert profile for one Bombardier CRJ operator that provides alerts on more than 80 parameters, ranging from engine-related trends such as fuel and oil consumption to passenger-door anomaly messages.
“We start with an out-of-the box-package, then things can be added,” says Davoli. “The operator can select which parameters they want, and have pushed. It’s a lot of data, but they can see what’s going on with each aircraft. Over time, they begin to spot trends.”
Ottawa-based Gastops is helping operators spot trends in another crucial area: detecting wear in engines and other mechanical components before failures occur. A long-time sensor supplier—its oil-debris detector is standard on Pratt & Whitney’s F119 and its PW1000G geared-turbofan lines— the company also can retrofit in-service engines and provide analysis. One of its newer products is ChipCheck, a portable fluid-sample analyzer that instantly identifies debris in engine oil that can signal major failures before they occur.
“Inspection of particles in oil is typically subjective,” says David Lefebvre, Gastops senior accounts manager and engineer. “Chipcheck identifies the particle size, count and what the alloy is made of. It takes away the subjectivity.”
Such accuracy is typically available only through a lab analysis—and even then, certainty is not guaranteed.
Lefebvre recalls a time when a customer found a single particle on an engine’s magnetic chip detector. ChipCheck identified the bit as 52100 material common to bearings. “Typically, such a failure mode would result in many particles being generated, thus leading to an immediate grounding of the aircraft and teardown of the engine,” he says.
Determining whether the material was 52100—thus pointing to an imminent bearing failure or something more innocuous—was critical. Dispatching the engine would be too risky, while an unnecessary teardown would be costly.
The customer needed a second opinion and sent the sample to a lab. Two days later, the lab said it could not classify the material. As a last resort, the sample was sent to the engine manufacturer, which confirmed what Chipcheck reported: The material was 52100. The engine was torn down, revealing that a critical component that shed the debris was failing.
Another airline operating a GE-powered widebody on a transatlantic flight received an engine-debris warning. The crew diverted to Paris-Charles de Gaulle Airport. The airline consulted with GE, which knew a ChipCheck customer had a unit at the airport. Within an hour, the verdict was in: The engine was not about to fail, and the flight could safely continue. “Without a ChipCheck on-site, this operator would have been forced to send the debris off to a laboratory and wait several days for the results,” Lefebvre says.
While instant, accurate analysis is useful, the system’s longer-term value is monitoring engine condition over time. ChipCheck can store samples by serial number, meaning operators can perform periodic tests and track changes over time, essentially developing an in-house predictive maintenance analytics program for any engine platform. Among the product’s converts: the U.S. Air Force, which has ordered 75 ChipCheck units to help support its GE F110 engine fleet.
The combination of new equipment with everything needed to feed a data analytics program plus retrofit services that can tap into data sources on older models means most operators can leverage predictive maintenance data. The results are getting so predictable that MSG-3 could be obsolete within 10 years, says Delta’s Hammes. He foresees that the fuselage and structure will stay on hard maintenance times, but the components and systems will switch to on-condition, driven by predictive maintenance and machine learning.