Industries / Industrial equipment and services / IoT and industrial automation / Digital twins
Use in maintenance
In maintenance the twin follows each unit's own use and wear, so that inspections and repairs can be set by that unit's history rather than by one schedule for the whole fleet.
4
Milestones achieved, from 1958 to 2012.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1958
Each aircraft's fatigue counted
On 13 March 1958 two US Air Force B-47 bombers failed from metal fatigue, the cracking caused by repeated loads, in separate incidents. General Curtis LeMay gave informal approval on 12 June 1958 to an "Aircraft Structural Integrity Program" proposed by the Wright Air Development Center, and a directive of 19 November 1958 formalised it, by the account of the Air Force engineer John W. Lincoln. The Air Force later made individual aircraft tracking part of the program: records of each aircraft's use "adjust the crack-growth-based damage rates and inspection intervals" for each critical area, in the words of a 1997 National Research Council report.
LessonThe B-47 had been designed on the premise that overload was the only threat to its structure, and Lincoln writes that its fatigue failures "demonstrated the fallacy of this premise".
Credit
US Air Force; Wright Air Development Center
1969
One flaw, one aircraft
On 22 December 1969 an F-111A crashed after about 100 hours of flight when the lower plate of its left wing pivot fitting failed; a manufacturing defect had started a fatigue crack. The Air Force grounded the fleet and from 1970 proof-loaded each aircraft at −40 °C, where steel is more brittle, so that any crack of dangerous size would break on the ground. By February 1972, 340 aircraft had been tested, and there were later 11 major failures in proof testing, from defects that "could have caused an in-flight catastrophic failure". Lincoln writes that these failures "proved the lack of validity of the safe life methodology", which set one safe life for every aircraft of a type from a fatigue test.
LessonA life worked out for a type says nothing about a flaw in one airframe; each one's own condition has to be known.
Credit
US Air Force; General Dynamics
2008
A shared test, but simulated
Abhinav Saxena, Kai Goebel, Don Simon and Neil Eklund presented simulated run-to-failure data for a fleet of turbofan engines, made with NASA's C-MAPSS engine model, at the 2008 Prognostics and Health Management conference in Denver, and NASA Ames' Prognostics Center of Excellence published it. In 2021 Manuel Arias Chao, Chetan Kulkarni, Kai Goebel and Olga Fink called it "one of the most widely used prognostics datasets", but wrote that its flight conditions are "restricted to six snapshots during a standard cruise phase" and that "there is a fidelity gap in the dataset as the simulated degradation trajectories lack important factors of complexity that are present in real engines".
LessonMethods tuned to a simulated fleet may not carry over to real engines.
Credit
Saxena, Goebel, Simon and Eklund; NASA Ames Research Center
2012
A twin for each vehicle
Edward Glaessgen of NASA Langley Research Center and D. S. Stargel of the Air Force Office of Scientific Research wrote that inspection schedules "are typically determined from experience with similar vehicles", often "the fleet leader, the vehicle in the fleet that experienced the most flights or the worst degradation". They proposed instead a twin: "an integrated multiphysics, multiscale, probabilistic simulation of an as-built vehicle or system that uses the best available physical models, sensor updates, fleet history, etc., to mirror the life of its corresponding flying twin". The paper was presented at the AIAA structures conference in Honolulu in April 2012.
Credit
Edward H. Glaessgen; D. S. Stargel
Watch for
Engine maintenance by analytics
GE Aerospace says engineers in Bengaluru are "advancing Analytics-Based Maintenance (ABM) with the ABM.AI tool and aircraft engine health monitoring that combines visual inspection, AI, and physics-based insights to support condition-based maintenance", to help extend time on wing, the time an engine stays installed between shop visits. The release gives no figures for these tools.
Achieved whenan airline or outside group publishes unscheduled engine removals or time on wing for engines managed this way, compared with similar engines that were not.
Claimed by
GE Aerospace, August 2026
Open
01
Real pumps run to failure
A public run-to-failure dataset from a fleet of identical real machines, for example 24 small centrifugal pumps run on water carrying fine sand to speed up wear, published with a physics model of the pump so that twin-based and data-only predictions of remaining life can be compared on the same units.
The briefCredit · unclaimed02
Each unit's own inspection date
A twelve-month trial on a fleet of identical machines, such as the fans of 40 air-handling units on one campus, in which half are inspected on the usual calendar and half on dates set by a model of each unit fed by its run hours, motor power and vibration.
The briefCredit · unclaimed
Open 01 · the brief
Real pumps run to failure
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public run-to-failure dataset from a fleet of identical real machines, for example 24 small centrifugal pumps run on water carrying fine sand to speed up wear, published with a physics model of the pump so that twin-based and data-only predictions of remaining life can be compared on the same units.
Why it is still open
The most used public engine dataset is simulated, and its successor, N-CMAPSS, still imposes simulated wear on real flight profiles; its authors added artificial sensor noise because "there were no real data available to characterize true noise levels".
Built with what exists
off-the-shelf pumps fitted with flow, pressure, power and vibration sensors; a Raspberry Pi logging over MQTT; a pump model written in Modelica and exported as an FMU; scikit-learn and PyTorch for the data-only models; a 3D scanner to measure impeller wear at teardown.
First step
Run two pumps on sand-laden water for a month and check that their head falls by a measurable amount.
Done looks like
At least 20 pumps run to a defined end of life, such as a 15% fall in head at the best-efficiency flow, with all sensor data and teardown scans published, and a scored comparison of at least two twin-based and two data-only predictions of remaining life made at 50% and 80% of each pump's life.
Open 02 · the brief
Each unit's own inspection date
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A twelve-month trial on a fleet of identical machines, such as the fans of 40 air-handling units on one campus, in which half are inspected on the usual calendar and half on dates set by a model of each unit fed by its run hours, motor power and vibration.
Why it is still open
Glaessgen and Stargel describe inspections set by experience with the fleet leader rather than by each unit's history. GE's August 2026 release gives no figures for its condition-based tools, and the site's Note "Maintenance inside the joint" found no independent measurement of MIPS's claim that its Actus chip runs predictive-maintenance models inside a robot joint.
Built with what exists
the campus building automation system's trend logs; clamp-on wireless vibration sensors; a wear model for each unit written in Python; the campus maintenance management system for work orders.
First step
Pull two years of work orders and run hours for 40 identical fans on one campus and see how much wear differs between units.
Done looks like
For both groups, the inspection hours, faults found per inspection and unplanned failures over twelve months, published with the rules used to set the dates.