AI-Powered Predictive Maintenance Is Changing Defence Aerospace MRO

A modern jet engine generates one terabyte of data per flight. A fleet of 200 fighter aircraft produces more telemetry in a week than most Fortune 500 companies generate in a year. Human maintenance crews, working with spreadsheets and paper logs, monitor a fraction of it. The gap between the data available and the data actually used to keep aircraft flying is the single largest unaddressed readiness problem in Western defence today.

Defence maintenance today works on a schedule. Every 400 flight hours, inspect the engine. Every 800, replace the oil. Every 1,200, change the turbine disc. This calendar-based approach was designed in the 1960s, when an engine's black box was a cockpit gauge and a grease-stained logbook. It catches failures that happen on schedule. It misses everything else.

The shift to condition-based and predictive maintenance is happening now. Inside the engine nacelles of America's training fleet. In the cable arrestor beds of its aircraft carriers. On the factory floor of its largest propulsion manufacturer.

The T-38 got there first

The US Air Force's T-38 Talon is a supersonic trainer that first flew in 1959. Its engine, the J85, powers an aircraft that has trained every American pilot since Vietnam. The J85 is reliable. It is also predictable in the wrong way: its failure modes are well understood, but its maintenance schedule has been calendar-driven for six decades.

In early 2024, GE Aerospace and Palantir piloted a new approach. They connected the J85 engine's maintenance data — parts demand, supply chain status, repair history — to Palantir's Artificial Intelligence Platform (AIP). The system learned to predict which parts would be needed, where shortages would appear, and when a component was approaching failure.

The pilot covered 6,000 J85 engine parts. It worked. By March 2026, the two companies had expanded the partnership to the entire GE Aerospace production system: sustainment, MRO, and new engine production for both military and commercial fleets.

"By integrating data across the enterprise and applying AI to predict demand and identify constraints earlier, our collaboration with Palantir is helping our customers keep more aircraft available," said Amy Gowder, president and CEO of Defense and Systems at GE Aerospace.

It operates an installed base of approximately 50,000 commercial and 30,000 military aircraft engines. A GE engine takes off somewhere in the world every two seconds. Scaling the predictive maintenance model across this base is an infrastructure problem of the first order.

The Navy follows

In May 2026, Odysight.ai signed a Cooperative Research and Development Agreement (CRADA) with the US Navy's Naval Air Warfare Center Aircraft Division. The initial focus: carrier arresting cables, the steel ropes that catch landing aircraft on a flight deck. They are mission-critical, notoriously hard to inspect, and failure is not an option.

Odysight.ai embedded ruggedized high-resolution visual sensors in hard-to-access locations around the arresting gear. Machine learning models analyse the imagery in real time at the edge, detecting micro-cracks, wear patterns, and material fatigue before they become visible to the human eye.

The Navy sees this as a launchpad. The CRADA explicitly targets expansion into fixed-wing and rotary aircraft, ground vehicles, and broader defence platforms. The same sensor-plus-AI pipeline that monitors arresting cables can monitor landing gear, rotor blades, engine inlets, and weapons bay doors.

Robots doing the labour

Predictive maintenance is not just about software watching sensors. The physical labour of repair is also being automated. In January 2026, GrayMatter Robotics received a $1.8M AFWERX SBIR Phase II contract to develop an autonomous robotic system for aircraft canopy sanding. This is one of the most demanding precision tasks in aviation maintenance.

The traditional process: a technician hand-sands an acrylic canopy for hours, applying consistent pressure across a curved 3D surface. One mistake, and the canopy's optical clarity is ruined. Replacement cost: $200,000 per canopy. GrayMatter's robot, guided by its proprietary GMR-AI platform, adapts in real time to geometry and defect patterns, restoring optical clarity with repeatable precision.

"The same adaptive intelligence that enables our systems to handle thousands of unique parts in manufacturing environments will now support the exacting requirements of optical surface restoration," said Dr Satyandra K. Gupta, chief scientist and co-founder of GrayMatter Robotics.

What is growing

Three signals define the trajectory of this convergence.

First, the mandate. The DoD's Condition-Based Maintenance Plus (CBM+) program now requires AI-enabled diagnostics across all major weapons systems. This is not optional. By 2028, any sustainment contract for a new platform must include a predictive maintenance component. Lockheed Martin, Northrop Grumman, and BAE Systems are all integrating advanced analytics into their sustainment solutions.

Second, the stack. Agentic AI moved from pilot to production in 2026. The early experiments — one-off ML models trained on isolated sensor streams — are being replaced by platform architectures. Palantir's AIP, GE Aerospace's data pipelines, and the emerging digital twin standards from the Air Force's Advanced Battle Management System create an infrastructure layer that did not exist two years ago.

Third, the numbers. The AI Digital Aerospace & Defence Technology Market is projected to reach $23.2 billion in 2025 alone, with predictive maintenance as the fastest-growing segment. US aerospace and defence spending on AI is expected to reach $5.8 billion by 2029, 3.5 times 2025 levels.

What is falling away

The traditional scheduled-maintenance model is structurally unsuited to the data environment it now operates in. Calendar-based inspections ignore variance between aircraft. Two T-38s of the same age, flown by the same squadron, can have radically different wear profiles. One has a cracked turbine blade. The other is fine. A scheduled interval catches both, or neither.

The spreadsheets-and-tribal-knowledge approach that still dominates most military depots is the real bottleneck. Deloitte's 2026 Aerospace & Defense Outlook notes that "most A&D organizations still run on spreadsheets, tribal knowledge, and maintenance schedules designed in the 1990s." The shift to predictive maintenance is slowed not by the technology but by the organisational inertia of base-level maintenance units operating on legacy IT systems.

The scarcity of skilled maintenance technicians is another structural constraint. The US Air Force reported a shortfall of approximately 4,000 aircraft maintenance personnel in 2025. AI-driven predictive maintenance does not replace these technicians. It changes what they do. Fewer unscheduled repairs, more data-driven inspections, less time spent on paperwork.

Who else is in the arena

The convergence of AI and defence maintenance is attracting a widening set of entrants beyond the primes. C3 AI has deployed its AI-powered reliability platform across multiple US Army aviation units, analysing engine vibration data and hydraulic system telemetry to predict component wear. The company's defence contracts have grown 40% year over year since 2023, reflecting service-level demand for off-the-shelf predictive maintenance software rather than bespoke prime contractor solutions.

Uptake, an industrial AI firm originally built for rail and energy, has pivoted into defence aerospace. Its platform ingests sensor data from maintenance logs and flight records to generate remaining-useful-life estimates for specific components. The US Air Force's Agile Combat Support directorate has tested Uptake's system on C-130 and KC-135 fleets since late 2025.

Honeywell Aerospace, a traditional MRO provider, is embedding its own Forge analytics platform into engine repair contracts. In early 2026, Honeywell secured a T55 engine repair and overhaul contract with the US Army that includes mandatory AI-based predictive diagnostics. Even two years ago, that contractual requirement would have been unthinkable. The shift from optional analytics to mandated AI in sustainment contracts is the fastest adoption signal in the entire market.

The startup layer is also moving. GuardKnox, an Israeli defence-tech firm, raised a $45 million Series C in March 2026 for its AI-driven maintenance platform targeting legacy platforms that lack native sensor infrastructure. That is the gap the F-16s and C-130s represent. These platforms generate little telemetry natively, so GuardKnox's approach layers retrofit sensor packs onto existing aircraft and trains predictive models on the resulting data stream.

Key signals to track

Sources