Marine Corps prototyping AI tools for aviation supply, predictive maintenance

The Marine Corps is prototyping artificial intelligence tools to inventory aviation supplies and predict aircraft maintenance issues, officials said, a burgeoning initiative the service hopes will help shed “outdated” ways of keeping its flying fleet ready.

The effort, officials said, is meant to help maintainers and logisticians quickly identify needed aircraft parts, order those parts more efficiently and then — with an AI system the Marine Corps intends to roll out this summer — forecast replacements based on historic performance data.

“Let’s change it before it needs to be in the air, declare an emergency, land in some place we don’t want it to land, etc.,” Lt. Gen. William Swan, deputy commandant for aviation, said during a panel at the Sea-Air-Space conference on Monday. “That’s the whole idea: supply first, then maintenance, and then the op[erational] stuff pulls together.”

Aircraft maintenance is an enduring, expensive issue across the military, made difficult by aging platforms, personnel shortages and competitive, congested supply chains. In a media roundtable last week, Swan said the Marine Corps’ aviation arm was “on average” around 62-64% mission capable. Training squadrons, he said, had a lower score.

In its annual aviation plan released in February, the Marine Corps said it is making AI central to helping combat those woes, in part by fostering a “data-enabled culture” within its aircraft sustainment community and attempting a more proactive approach to fixes.

The Corps’ AI effort falls under Project Eagle, a “strategic blueprint” for its Aviation Combat Element meant to balance crisis response with modernization. A significant shift in the service’s aviation plan centered around AI, which was only briefly referenced in last year’s tome.

While the aviation plan emphasizes the new tech in comparison to previous years, it noted the AI and machine learning development effort for aviation sustainment was “isolated and underfunded.” Officials recently told reporters that is no longer the case and the service has prioritized the initiative.

Though nascent in practice, the data part of this effort began years ago. In 2022, officials said, the Marine Corps began cataloguing repair parts and “consumables” for the F-35 Lightning II after realizing the way the service approached maintenance and resupply was “outdated,” according to Col. Robert Petersen, head of the Corps’ aviation sustainment branch.

He said there were mountains of data “siloed, sitting there, really unavailable for us to harvest and use in any meaningful way” for the fighter jet. 

“That’s where we have our biggest problem currently, just because it’s a global supply chain, we compete with all of our allies and partners in that,” Petersen told DefenseScoop when asked why the service started with the F-35. 

Years later, the service has logged “every consumable” for the F-35 and recently created two “notional” parts packages using the prototyped tools, he said. The Marine Corps has already started pulling data for the KC-130J to then feed into the AI system, according to Petersen.

“This is new to us,” Swan said last week, adding that the Corps is still building the algorithms for its AI tools. “Typically, you take a part off when it’s broken, you hand it in to supply or to the vendor to fix, and they run it on your test bench or whatever, and they go, ‘Yep, it’s broken. It’s not working as advertised.’ And then they fix it.” 

Now, in a push to become less reactive to repair issues, the Corps is looking to turn those AI tools on predictive maintenance to get ahead of part replacement before equipment fails on its own. This system, dubbed the “Maintenance Assessment Tool,” aims to do just that, and will be handed to a unit at Marine Corps Air Station Yuma in Arizona this summer.

“We want to be predictive,” Swan said. With AI, the Marine Corps intends to use “the data that we have in our maintenance systems, to understand the reliability of the parts and what the environments are that we’re going to operate, and figure out when those parts are going to break to a probability.”

“If it’s 90% or so,” he added, “that’s good enough for me.”