At home, we have been promised a revolution. Internet-enabled devices will transform our lives much as washing machines and vacuum cleaners once did. Soon, refrigerators will order food online when empty; coffee-makers will start brewing as their owners wake up; and lighting systems will respond to moods.
In the air, however, the Internet of Things has been up and running for some time. Modern aircraft and engines are packed with sensors that relay huge volumes of data, both in- and post-flight, to ground teams that can sometimes react even before wheels are on the tarmac.
The biggest concentration of sensors in any aircraft is in its engines. New powerplants such as the Leap, PW1000G, GEnx and Trent 1000 deliver about 1 terabyte of readings per flight. This equates to the storage capacity of the average home computer, and with a growing fleet incorporating ever-more advanced technologies the volume of data generated is surging. CFM’s Leap engine, for instance, transmits about 3.5 times more information than its predecessor, the CFM56. This is thanks to a bigger array of sensors such as eight on the Leap combustor—double the count of the CFM56’s.
Engineers can use the information in different ways. The simplest analysis might detect a single spiked reading from a vibration sensor, for instance, and conclude that a component is, or is close to, failing. A more complex task, however, is to pick out trends from thousands of hours of data in order to identify risks weeks in advance.
“We are good at catching problems after they have happened the first time, but we want to use our tools and data to make sure we understand something before it even happens that first time,” says Vijayant Singh, general manager for data analytics at GE Aviation.
Although it is difficult to say how prophetic such analysis might become, or how much engine data is still sitting on servers awaiting analysis, GE has made inroads. Singh cites an analytic developed to track a customer’s oil consumption over time. A trend of increased consumption was traced back to three root causes that need to be addressed to avoid, potentially, an inflight shutdown at some point. “So we told them if they didn’t take action in the next seven to 10 days it could cause an event,” says Singh.
This advanced troubleshooting is one of GE’s three goals for engine data analytics. Another is improving customers’ operational performance in areas beyond engine reliability; the third is to smooth maintenance cycles and ensure that when engines do come off wing, it is not in an emergency situation.
Along with the OEMs, maintenance companies are also developing their data analysis capabilities. Exhaust-gas temperature (EGT) margins, vibration, fuel flow, shaft speed, oil pressure and the position of variable stator vanes are all important measurements for MRO providers, which can use them to support customers during operations and speed up maintenance in the hangar—for example, by determining which parts of an engine need inspection.
Only relatively small amounts of data can be transmitted during flight, so the majority is downloaded via the aircraft condition monitoring system (ACMS) once a flight stops at its gate. For inflight transmissions, ACMS data is broadcast either on VHF3 or by satellite via the aircraft communication and recording system (ACARS). However, fleet and ground-based infrastructure is still too limited to allow for anything beyond the highest-priority sensor readings to be sent inflight.
Once data has been received, MRO providers—such as Air France Industries KLM Engineering & Maintenance (AFI-KLM E&M)—use software algorithms to analyze it and trigger any necessary alerts. These alerts are then reviewed by engineering teams that decide on appropriate further action. Alerts are often triggered by simple exceedances—that is, when readings are outside thresholds—but the company also uses an in-house tool called Prognos for data mining, statistical analysis and long-term engine health monitoring.
Prognos includes machine-learning capabilities that allow for the type of dynamic modeling necessary for early warning of potential problems. AFI-KLM E&M says the program is superior to legacy tools OEMs have provided to airlines, as these “were mainly based on software containing hidden equations of engine performance, using one engine model for the worldwide fleet.”
Lufthansa Technik uses both in-house and OEM-provided tools to analyze data in different ways. Any overstepping of preset threshold values is first detected at its computer center in Frankfurt.
“In the case of important changes to parameters, an alert will be generated which informs an engine-condition monitoring analyst to watch the conditions of the affected engine. The analyst will issue a work order for the engine if he confirms a problem,” says Dirk Hirner, director of Lufthansa Technik’s technical competence center.
All engine manufacturers have invested heavily in data analysis to protect and enhance their brands by providing better reliability and operational improvements for customers.
In early 2016, Pratt & Whitney unveiled its big data project to enhance engine maintenance planning.
“Predictive analytics . . . help us examine how different environments—altitude, climate and pollution—affect the performance of our engines. This allows us to better forecast shop visits, better manage our customers’ fleet maintenance and improve our engineering in the long term,” says Matthew Bromberg, president of Pratt & Whitney Aftermarket.
Rolls-Royce, meanwhile, has teamed up with Microsoft to improve its analytic capabilities. It will use the latter’s Azure cloud-computing platforms to collect and aggregate engine data from around the world, and Microsoft’s Cortana Intelligence Suite to crunch the numbers.
GE Aviation, in contrast, uses an in-house platform developed by its parent conglomerate. Called Predix, it incorporates machine-learning capabilities to perform multifaceted analysis and to streamline the alert process.
“We teach it that this is the way to look at a hot and harsh flight, whereas this is how you look at a normal situation, so when it alerts next time it alerts after taking that into consideration,” says Singh.
Of course, analysis means nothing without follow-up, so GE has one team to continuously monitor data for exceedances and trends, and another to act upon their conclusions. This operations team checks any alerts generated, decides how quickly work needs to be performed and provides troubleshooting recommendations to the customer.
Singh is proud of the big strides made in connecting the dots between analysis and fleet management. “With earlier systems, it would take us weeks or months to operationalize new analytics—now we do it in a matter of days,” he observes.
Data as an Asset
Airlines own the data their engines generate but usually share much of it with OEMs under specific agreements. GE, for instance, has access to snapshot data for 90% of its fleet, and uses this to monitor 35,000 commercial engines per day.
“The data in itself is of no value; it’s the smarts that are associated with it: the analytics, the platform, the outcomes—that is what makes the complete system invaluable,” says Singh.
Both Lufthansa Technik and AFI-KLM E&M confirm that they also work closely with OEMs to better understand and predict engine behavior. “However, certain parameters are considered airline-operator proprietary and [are] not fully shared with OEMs,” notes an AFI-KLM E&M representative.
But as Singh points out, most data is useless without the right tools to analyze it. This raises the question of whether the aviation industry or third-party software companies are best placed to develop the solutions that incorporate advanced technologies such as big-data analysis, machine learning and cloud computing.
Rolls-Royce is collaborating with Microsoft to improve its capabilities in these areas, but at GE Aviation Singh is skeptical about nonspecialist solutions.
“I cannot use iOS [Apple’s mobile operating system] for my aircraft engines because the analytics are very different [from] what iOS is designed for. Unless you have the domain [aircraft engine] expertise I don’t think you can design an off-the-shelf solution that can work,” Singh says.
In Prognos, AFI-KLM E&M acquired analytic software from a nonaviation source but says it is now developing further big-data-processing capabilities in-house. “We see the huge potential for using this massive amount of data; however, the exact amount of useful data from the generated volume [still needs] to be fully understood,” says a representative.
Indeed, the staggering volume of data now being generated by modern engines represents a huge reserve of untapped potential—though not for long, according to Lufthansa Technik’s Hirner. Yet even if the whole terabyte of data from a Leap or PW1000G cycle is soon subject to some form of analysis, there will always be scope for better interpretation and programs that crunch the numbers in new and more insightful fashions.
“We are still not fully predictive [to] where I can give a customer 30 days’ lead time [and] say a thing is going to happen,” says Singh. “I’m proud of what we have done so far, but there is still a lot to do on the predictive side—making sure that first event never happens,” Singh adds.