Printed headline: Smart APU Maintenance
In which direction is maintenance that exploits big data and the Internet of Things headed? The ideal predictive or prescriptive maintenance program would turn every expensive unscheduled maintenance event into a scheduled event, with no false warnings and no premature removals of a costly piece of equipment. It would miss absolutely no problems—that is, it would have a zero false-negative rate. And it would generate no unnecessary fixes or even checks—a zero false-positive rate.
This kind of predictive accuracy is impossible in the real world. But it is a useful yardstick against which to measure the increasingly valuable predictive programs that major OEMs and MROs are offering based on connected assets and huge volumes of data. Another good place to look for progress is in auxiliary power units (APU), which are physically similar to the propulsion engines for which predictive techniques were first applied decades ago.
OEMs should be best at this game. They have the design data on which to base physical predictive models. They have test data to confirm and refine these design models. And assuming their equipment is widely installed, they should be able to gather operating and repair data from huge, multi-airline fleets to develop statistical routines to boost predictive power.
But there are advantages in having other suppliers of predictive services. First, of course, is competition. No customer wants to rely on a monopoly vendor, no matter how brilliant. With the evolution of predictive analytics still young and well short of perfection, there are virtues in having different teams trying different approaches.
Let’s start with the 1,000-lb. gorilla of APUs, Honeywell, which equips two-thirds of mainline commercial aircraft. The company is the sole supplier of APUs on Boeing 737s and 777s, Airbus A330s and A350s, and it is one of two suppliers on the A320 family, notes Bob Buddecke, Honeywell’s vice president and general manager for power systems.
As part of its new Forge nose-to-tail predictive maintenance service, Honeywell offers predictive maintenance on all these APUs, including Pratt & Whitney’s on the A320 family. In total, the OEM provides predictive maintenance for APUs on almost all 7,000 aircraft under the Forge umbrella. This service is available for APUs on 737NGs and MAXs, but not on 737 Classics, simply because “we have not focused on them,” Buddecke explains.
For APUs, the most important sensor data includes temperatures, pressures, start times and exhaust gas temperature margins. Non-APU data can also be important, including data from line-replaceable units upstream of APUs that affect them. Data on the operational environment, from the aircraft maintenance computer, quick- access recorder and flight data acquisition unit are helpful, too. It is by integrating all this APU and non-APU data that better predictions are made, Buddecke explains.
The value of APU predictive maintenance is avoiding cancellations, delays and disruptions by turning unscheduled maintenance into scheduled maintenance. Honeywell’s predictive techniques have so far yielded a 35% reduction in APU-related cancellations, delays and disruptions by advising operators about 3-5 days ahead of possible events what the risk is and what should be done about it. All this has been achieved while reducing premature APU removals by 15% and holding the fault-found rates down to 2%.
This is far short of predictive perfection, but is nevertheless a big money-saver, conserving as much as $10,000-20,000 per aircraft, per year, Buddecke says.
Honeywell’s predictive tools can work under a variety of business arrangements. In most cases, Honeywell is the MRO provider. “We alert them; they understand and follow up,” Buddecke explains. If necessary, the APU comes off wing and is sent to Honeywell or one of its partners. Or the alert may just mean help with troubleshooting a problem on-wing or removing a simpler part, like a surge valve or flow control. “We like the simplest possible actions,” he says.
Honeywell first aims to keep APUs on wing and reduce troubleshooting time, then to do the simplest maintenance possible by removing an LRU, and only last removing the APU itself in an orderly, planned fashion.
Honeywell is further refining its APU predictive tools and expanding its coverage of other components under Forge’s nose-to-tail predictive program. More sensors, more data and combinations of data, and more aircraft lie in the future.
But so do competitors.
AFI KLM E&M subsidiary EPCOR offers Prognos for APU predictive maintenance for all Honeywell and Pratt & Whitney APUs on Boeing 737NGs, 777s and 787s, as well as the Airbus A320 family, A330s and A340s and Embraer 170s and 190s—essentially all the APUs that EPCOR repairs.
All these APUs generate sufficient data for Prognos’ predictions, according to EPCOR’s APU Program Manager Niels van Hofwegen. Airlines must also have the right hardware and software to support data connections, connecting APUs to central computers and then moving sensor data to the ground. For a few aircraft, this is difficult. “But we can always find a way,” van Hofwegen says. By the end of this summer, Prognos will be covering about 800 APUs.
In looking for the right sensor data, Prognos starts with possible APU failure modes and then searches for the combination of sensors that help predict them. Common are bearing failure and carbon-seal failure, but frequent modes can differ by both APU and by airline.
“Sensor data is basic, but in order to have absolute control you need enriched-context data,” van Hofwegen says. For example, important contextual data could include sandy environments. The methods Prognos uses to collect this environmental data are part of the MRO’s confidential intellectual property.
EPCOR’s APU engineers use Prognos to predict failures and communicate these warnings to the airline’s powerplant engineers for action. “A prediction is always confirmed by physical evidence of wear and tear before the APU is removed,” van Hofwegen says. This confirming evidence can include excess wear, scratches or evidence such as contamination in oil filters. If no evidence is found, the warning is false, and the APU or part is not removed. But van Hofwegen says that with the necessary confirmation process before removal, no false removals occur.
The other kind of error—failure to predict a failure—can occur at several points, with Prognos or EPCOR’s engineers or a failure to take action by the airline. The MRO will not publish this failure rate for strategic reasons.
Typical actions based on Prognos warnings are inspections or removals of either APUs or their components.
The benefits of the program are what van Hofwegen calls “operational excellence,” the replacement of APU failures with scheduled maintenance. Unanticipated failures reduce aircraft availability and may require last-minute leases of APUs, the most expensive kind of acquisition.
EPCOR also wants to reduce costs in both scheduled and unscheduled maintenance by avoiding “consequential damage,” the damage that occurs during the last minutes of a catastrophic APU failure. “That is huge,” van Hofwegen stresses.
EPCOR, which works closely with both Honeywell and Pratt & Whitney, is certified as a licensed warranty repair station by both OEMs. The MRO has access to certain OEM data, but not all. In any case, EPCOR developed the Prognos predictive algorithms for APUs on its own.
But van Hofwegen emphasizes that smart predictions are “not about algorithms but about processes, having the right information and the right people.” His main focus is getting the right raw data and the right calculated data for Prognos.
He will not specify savings but says Prognos can now significantly improve operational reliability and reduce APU maintenance costs.
Others are also working on the challenge. Lufthansa Technik’s Aviatar offers predictive maintenance on Airbus A320 APUs, but the MRO is increasing its support of other aircraft types. Aviatar’s sales director, Frank Martin, says newer aircraft types generate more data and thus better predictions, but all APUs generate data Aviatar can use.
Aviatar can prevent APU shutdowns and improve performance. For example, data on oil temperature helps trigger maintenance to prevent shutdowns. All shutdown data is analyzed to pinpoint causes and improve troubleshooting.
To predict degradation of APU performance, Aviatar focuses on exhaust gas temperature, bleed pressure, inlet-guide vane angle and start-time data. “Analyzing those parameters in relation to others helps with finding error causes,” Martin says. Other useful inputs can include non-APU data, such as on flight schedules, airline networks and status of components related to APUs.
He acknowledges that no predictive tool foresees every APU failure. But Aviatar is adjusting its APU predictions with every operator that joins, so predictions are constantly improving. Even now, users are reporting significant improvements in APU availability and reduced times to troubleshoot and fix problems.
Aviatar algorithms combine with OEM technical documents and Luft-hansa Technik engineers to provide the warnings. The process can be customized for each new airline. Martin believes that automated troubleshooting at airlines will fit well with Aviatar tools, further reducing troubleshooting costs.