Replacing aircraft components before they fail is a solid maintenance strategy, but predicting when a particular part is most likely to fail and swapping it out at the optimal time can maximize maintenance efficiency, prevent flight delays and in some cases avert catastrophe.

Rather than fix aircraft after they break—reactive maintenance—or overhauling airplanes or helicopters on a pre-set schedule based on flight hours, many airlines and small fleet operators are gathering intensive data on the health of their aircraft and predicting when to service certain components or systems.

This so-called predictive maintenance uses data analysis to detect operational anomalies and possible defects so they can be fixed before they become failures.

Most modern aircraft broadcast detailed data from sensors arrayed throughout the airframe, which is then analyzed through health and usage monitoring systems, or HUMS. Those systems make data gathering analysis much easier for maintenance shops, but such advanced technology is not necessary to establish an effective predictive maintenance operation.

Take FedEx, which with 410 wide body aircraft—and hundreds of smaller contract cargo planes—is the fifth largest airline in the world.

Data analysis of maintenance logs and component failures has allowed the company to perform effective predictive maintenance on even its oldest aircraft, MD-10s and MD-11s that are decades old, Randy Provence, a data engineer with FedEx, said at the 2022 Data and Artificial Intelligence Summit in June.

FedEx uses predictive maintenance, powered by AI and machine learning to prevent aircraft departure delays by replacing components before they cause unexpected component failures.

“We have developed methods of predicting component failures without telemetry data,” Provence said. “We have telemetry data on our newer fleets—767s and 777s—but our older fleets, our 40-year-old planes do not have a lot of messages coming out of them. We have to rely more on our history of maintaining those aircraft over the last 30 years or so.”

The company evaluates 350,000 unique part serial numbers twice a week and predicts whether each is at risk to fail in the future.

“These are only serial numbers that can be removed and that we do fly to failure,” Provence said. “Something like that would be a part that is out of its family-average on-wing time. If the family of those parts flies 30,000 hours and that part’s been on there for 40,000 hours and we know they usually fail pretty quickly after that, we will set up a time when it is most cost effective to remove that part … before it fails.”

FedEx also maintains a “pain index” where it catalogs which aircraft components cause the most delays and how much it costs to repair them, Provence said. Machine algorithms automatically evaluate the 126 part families that most often cause delays in an effort to replace those troublesome components before they break and to choose when best to replace them.

“All these components, when they get removed predictively, they go to a repair vendor,” he said. “The repair vendor determines whether it was a confirmed failure, whether it really needed to be removed, and prints out their findings. When the component returns, we review that data and use it as input, a feedback loop, into our processes.”

“We would remove a lot more parts, but we are constrained by the availability of replacement parts, especially for our older fleets—MD-11s and MD-10s,” Provence said. “There’s not a lot of brand-new components out there for those.”

A single 15-minute delay to one of the more than 26,000 flights FedEx launches per year can cost the company $30,000 on average, according to David Taylor, a data scientist at FedEx, who also spoke at the 2022 summit.

“Our main goal is to predict when parts are going to fail,” Taylor said. “That way we can remove those parts on scheduled maintenance so that planes don’t have delays, which can cost us thousands and thousands of dollars.”

Because FedEx’s aircraft fleet is older, the individual airplanes are not outfitted with sensors that produce telemetry data and actively monitor the health of engine components and other flight-critical parts. Instead, the company relies on a “rich history” of maintenance data about specific aircraft components gathered and analyzed over decades, Taylor said.

“We know when the part was brand new, when it was installed, when it failed, at what age it was when it failed, where it was sent to a vendor to repair that part, where it was installed, which aircraft,” he said. “We have this rich set of features that describe the nature of this part. To analyze these parts, we are using survival analysis.”

Crunching those numbers for the past five years and 20,355 unscheduled part removals, FedEx’s predictive maintenance models caught 5,826 of them. Had those part swaps been predicted, it would have saved about $18 million over five years, Taylor said.

“Our increase in savings and delays prevented have exponentially increased, and this is just with a small fraction of our parts,” Taylor said. “Our future goal is to expand these models to more parts. Just this past year, we’ve had almost $2 million in savings from delays and prevented cancellations.”

FedEx also wants to add more sensors to its aircraft and thereby boost the amount of automated data collection into its models. The company also plans to introduce artificial intelligence to perform predictive maintenance and part removals.

“The way we’re using telemetry data right now is more of a gut-check,” he said. “We’re getting telemetry data that is going out of an upper control limit. The maintenance guy is getting these alerts. Then he will go to our smart-time models and say ‘we’re having an issue with this system’ and our predictive models will narrow it down to a specific part that could be the issue.”

“Where before we would probably have removed four parts, we’re telling him it’s probably this specific part, so using our telemetry data to fine-tune their predictions,” he said.

Military maintenance

The U.S. military has notionally been practicing predictive maintenance since the early 2000s when most of the services launched their own programs primarily for aircraft. But those maintenance programs have either not been implemented to their full extent and/or have not borne significant fruit until recently, according to a Dec. 2022 Government Accountability Office (GAO) report on the Defense Department’s predictive maintenance protocols.

The Defense Department spends upwards of $90 billion a year on weapon systems maintenance and aircraft account for a huge chunk of that recurring cost. For any given platform, whether an F-35 Joint Strike Fighter or an MQ-9 Reaper, the majority of its lifecycle cost is sunk in operations and maintenance.

“All four military services achieved relatively more progress implementing predictive maintenance on aircraft than on other weapon systems,” the GAO found. “This progress is in part because sensor technology has been available to enable predictive maintenance on some aircraft as early as 2002.”

The Army was first out of the gate when in 2005 it began installing some health-monitoring sensors on the AH-64 Apache attack helicopter. In 2012, the program took off when all Apaches and UH-60 Black Hawks were equipped with telemetry sensors and other predictive maintenance technologies. By February 2022, the Army had installed sensors on 65% of its CH-47 Chinook fleet.

The Navy and Marine Corps, which operate large fleets of both rotorcraft and fixed wing aircraft, have several platforms outfitted with condition-monitoring technologies, GAO found. The Navy issued its predictive maintenance policy in 2015, and the Marine Corps issued its predictive maintenance policy in 2020 and both are “making progress toward fuller implementation of predictive maintenance,” according to the GAO report.

The Air Force did not begin predictive maintenance efforts until 2016, when it began implementing the practice for “a smaller number of components across individual weapon systems,” the report said. Predictive maintenance for a limited number of KC-135 components began in 2019 and the program was expanded to include all Stratotanker components in 2021.

“While the military services have begun piloting predictive maintenance programs on some weapon systems, they do not replace parts or components regularly based on predictive maintenance forecasts,” GAO found. “The military services have not consistently adopted and tracked implementation of predictive maintenance. By developing plans to implement predictive maintenance, including action plans and milestones for weapon systems, the military services would be better positioned to determine where, when, and how to effectively adopt predictive maintenance.”

Mirroring the lessons FedEx has learned, officials from all of the military services recognize that predictive maintenance can and has improved maintenance outcomes, and could potentially save lives, according to GAO. Unplanned maintenance “which adversely affects costs and operations—can be reduced through greater use of predictive maintenance. Army and Navy officials also provided examples of predictive maintenance possibly preventing accidents on aircraft such as the AH-64 Apache and the F/A-18 Super Hornet.”