Unplanned maintenance is a $21 billion a year problem.
Planned maintenance runs on a calendar: hours flown, cycles, time since the last check. Unplanned maintenance is everything else. A sensor flags a fault, a crew reports a defect, an inspection finds something nobody expected.
That's when the costs stack up. The repair itself, a grounded aircraft earning nothing, a replacement flown in from somewhere else, compensation for passengers, and a schedule that ripples into flights that had nothing to do with it.
And it's getting worse. Fleets are growing and aircraft carry more systems than ever, and every one of them can fail.
Source: Statista (2023). 2035 is a projection.
Twins help build planes. Almost nobody uses them to keep planes flying.
Digital twins are already everywhere in aviation, but mostly on the factory floor. Airbus uses them on assembly lines, Boeing used them to cut assembly time, and GE tracks thousands of assets with them.
Once an aircraft enters service, they mostly disappear. Of everything I reviewed, only one live platform, Lufthansa Technik's AVIATAR, was aimed squarely at unplanned maintenance.
of data generated by commercial aviation in 2018. Only 1.5% of it was recorded.
Frost & Sullivan (2019)A systematic review, from 236 records down to 62 studies.
I turned four objectives into nine research questions, built a keyword map for each, and ran the search through the PRISMA 2020 method so every inclusion and exclusion was traceable.
Because there's little aviation-specific work on the topic, I also pulled in evidence from other industries that run predictive maintenance.
Two ways to put a twin to work. Both are closed loops.
Digital twin + predictive maintenance
Predictive maintenance and digital twins need the same thing: a constant stream of live data from the aircraft. Combine them and one data feed does two jobs. Take landing-gear tyres: high-speed landings and take-offs cause flat spots, flat spots wear tyres faster, and worn tyres can blow.
- Sensors stream tyre pressure, temperature and landing loads.
- The twin updates in real time against a baseline built from test data.
- An anomaly appears, like the early signature of a flat spot.
- The predictive model sizes the risk and decides if action is needed.
- Maintenance is scheduled before the failure, not after it.
- The twin confirms the fix, and monitoring carries on.
Digital twin + aircraft health management, with a data integration layer
A twin is only as good as the data fed into it, and aircraft data comes from everywhere, in every format. So before anything reaches the twin, it passes through three integration steps.
Many sensors, many units. One way to spot what's wrong.
Both frameworks depend on the same step: turning raw sensor streams into something a model can compare. An engine reports vibration in g, exhaust temperature in degrees, oil pressure in psi and fuel flow in kilograms an hour, and every one of them swings with altitude, weather and load.
The answer is the engine's own rhythm. Every flight repeats the same pattern, and vibration repeats with every turn of the shaft, so each sensor's healthy pattern can be stored as a Fourier series: a handful of spinning circles, fitted to that engine's own healthy flights.
Each new flight is compared with its pattern once the weather has been taken out. The gap, divided by how much that sensor normally wobbles, becomes a z-score. Every sensor ends up on the same scale, so one simple rule watches them all: beyond ±3 for five flights in a row is an anomaly. And because each circle stands for a physical part, a circle that grows or appears points straight at the problem.
Measured minus expected, divided by the sensor's normal wobble. Zero is healthy; beyond ±3 almost never happens by chance.
The same rule, applied to every sensorSee this working step by step, then race it against cockpit gauges and scheduled inspections on the same engine.
Engine fault detection build →Run an airline for a year. Then give it a twin.
Pick a fleet and decide how good your twin is at spotting faults early. Every fault it catches gets fixed overnight. Every one it misses hits the schedule: delayed flights, the odd cancellation, passengers stuck at the gate.
Without a twin
Every fault is a surprise- Surprise faults
- 0
- Cost so far
- $0
With a twin
Caught faults are fixed overnight- Surprise faults
- 0
- Caught early
- 0
- Cost so far
- $0
FlyingDisrupted by a surprise faultFault caught, fixed overnight
How the numbers work
- $21B a year lost to delays and cancellations from unplanned maintenance worldwide (2017). Statista (2023), from my research.
- 25,368 aircraft were in service at the start of 2017, so roughly $830k per aircraft per year. Oliver Wyman, 2017 For scale, airlines spend about $5M per aircraft a year on maintenance itself. IATA MCX, FY2024
- 12 surprise technical faults per aircraft per year, about $70k each. A 500-aircraft airline flies around 900,000 flights a year; if 0.5 to 1% are hit by a technical fault, that is 9 to 18 per aircraft. Estimate.
- A typical surprise fault costs about 6 hours of aircraft time, delays 2 flights and cancels 1 flight in 4, at 150 passengers a flight. Most are delays, not long groundings. Assumption, anchored on ~$98 a minute of aircraft time (Airlines for America, 2025), ~€100 a minute of delay (EUROCONTROL) and ~$15.7k per maintenance cancellation on a US narrowbody (masFlight, 2014).
- A fault caught early and fixed overnight costs 20% of a surprise one. Assumption.
- Not counted: the cost of building and running the twin, which my research flags as a real barrier, especially for smaller airlines. In Europe, EU261 passenger compensation would push the cost of each cancellation higher.
The technology is close. The industry around it isn't, yet.
Data at fleet scale
Messy data from many sources, huge volumes, patchy real-time links from aircraft in flight, and models that have to stay accurate as parts age.
No rulebook yet
There are no twin-specific rules. Certifying a new aircraft takes 5 to 9 years, and approving a new maintenance technology across regulators could take as long.
People and money
A cultural shift in maintenance teams, new data and IoT skills, and heavy upfront investment that smaller airlines can't easily carry.
Same idea, different subject.
A live model of a real machine, fed by its data, used to see a failure coming before it happens.
At Brox.AI we build digital twins of real people, used to predict how consumers will respond before a company spends anything.
Bütev, E. B. (2024). Use of Digital Twins in Commercial Aviation to Mitigate Unplanned Maintenance and Downtime to Increase Efficiency and Profitability. Research paper, University of Manchester. 66 pages.