Industries / Industrial equipment and services / IoT and industrial automation / Digital twins
Keeping it current
Keeping a twin current means updating the model when the real asset changes, and checking again and again that it still matches what is measured.
5
Milestones achieved, from 1960 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1960
Estimates corrected by measurement
In 1960 Rudolf Kálmán published "A New Approach to Linear Filtering and Prediction Problems". His method, now called the Kalman filter, blends a model's prediction with each new measurement to keep a best estimate of a system's state. Its "first publicly known application was made at NASA Ames Research Center in the early 1960s", in studies of navigation for the Apollo spacecraft, by the account of Leonard McGee and Stanley Schmidt of NASA Ames.
Credit
Rudolf E. Kálmán; Stanley F. Schmidt and colleagues at NASA Ames
2001
A simulator must keep matching
In its 2001 rule on simulation facilities, the NRC dropped routine reports to it but kept "Continued assurance of simulator fidelity": a plant must conduct performance testing and keep the results for four years, and "correct modeling and hardware discrepancies and discrepancies identified from scenario validation and from performance testing".
Credit
US Nuclear Regulatory Commission
2002
A pass mark for models
ASHRAE's Guideline 14, in its 2002 edition, declared an energy model of a building calibrated if its errors against measured data stayed within set limits. The US Department of Energy's Federal Energy Management Program reprints the limits from the guideline: a mean bias error within ±5% and a CV(RMSE), the typical size of the errors relative to the mean, within 15% against monthly data, or ±10% and 30% against hourly data. It adds that calibration is "time-intensive" and "should be performed by an accomplished building simulation specialist".
Credit
ASHRAE
2023
Drawings that may be stale
The US Air Force had two B-1 bomber airframes taken apart at Wichita State University's National Institute for Aviation Research and "pretty much all the structural parts" scanned twice to make CAD models. In August 2023 Air & Space Forces Magazine reported that 4,000 of a planned 51,000 models were done, and that engineers would check them against "the drawings from the 1980s that may or may not have ever been updated".
LessonRecords not kept up to date can cost tens of thousands of models rebuilt from the physical parts.
Credit
the US Air Force's B-1 division; the National Institute for Aviation Research
2024
A legal duty for records
The Higher-Risk Buildings (Keeping and Provision of Information etc.) (England) Regulations 2024 were made on 13 January 2024 under the Building Safety Act 2022. Government guidance says the client, principal designer, principal contractor and accountable persons for a higher-risk building (at least 7 storeys or 18 metres high, with at least 2 homes, or a hospital or care home) "must keep a digital record of information about a building", known as the golden thread, and keep the information up to date.
Credit
UK Parliament and the UK government
Watch for
A model synced to controls
IES says its twin of Queen Margaret University was created "by synchronising physics-based simulation with BMS data", so that the model follows the live readings of the building management system (BMS), and that it identified potential savings of about £64,000 a year.
Achieved whenthe university or an outside party publishes, after a year, how closely the twin's predictions matched metered energy use, by ASHRAE Guideline 14's measures.
Claimed by
IES, May 2026
Open
01
Notice when the building changed
A tool that compares a regular walk-through with a 360-degree camera, or a scan by a patrol robot, with an existing building's model and lists every element that has moved, gone or appeared, each with a picture, for a person to accept into the model.
The briefCredit · unclaimed02
Models that recalibrate themselves
A pipeline that builds an energy model for each of 100 buildings in the Building Data Genome 2 set from its public description alone (use, floor area, location), calibrates it to the 2016 meter data automatically, and then tests it, unchanged, against 2017.
The briefCredit · unclaimed
Open 01 · the brief
Notice when the building changed
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A tool that compares a regular walk-through with a 360-degree camera, or a scan by a patrol robot, with an existing building's model and lists every element that has moved, gone or appeared, each with a picture, for a person to accept into the model.
Why it is still open
H. Tran and K. Khoshelham of the University of Melbourne wrote in 2019 that "BIMs are generally not updated to reflect changes in the environment" and that manual survey checks are "very time-consuming, error-prone, and laborious". Their automatic method was tested on a synthetic dataset and one benchmark dataset, not on working buildings over time.
Built with what exists
a consumer 360-degree camera such as an Insta360 or a Ricoh Theta; COLMAP and gsplat to rebuild each walk-through in 3D; Open3D to compare it with the model; IfcOpenShell to read the IFC model; a multimodal model such as Claude to describe each change; Boston Dynamics' Spot with a Leica BLK ARC scanner where a site already has one.
First step
Film one office floor with a 360-degree camera twice, a week apart, move ten objects in between, and test whether a plain scan-to-model comparison finds them.
Done looks like
In three working buildings over six months, with 50 changes planted and logged by staff (furniture, equipment, partitions) as well as any real ones, the tool finds at least 90% of the planted changes with no more than one false alarm per 100 m² per walk-through.
Open 02 · the brief
Models that recalibrate themselves
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A pipeline that builds an energy model for each of 100 buildings in the Building Data Genome 2 set from its public description alone (use, floor area, location), calibrates it to the 2016 meter data automatically, and then tests it, unchanged, against 2017.
Why it is still open
The ORNL team behind the Autotune method wrote that "Manual calibration for only dozens of parameters could require months". A recent paper led from the University of Utah says manual calibration "is time-consuming, heavily relying on expert judgments" and reports one case cut "from >600 hours to ~1.5 hours" with its method. Both tests were on single buildings.
Built with what exists
Building Data Genome 2, hourly readings from 3,053 meters in 1,636 non-residential buildings for 2016 and 2017, under a CC BY 4.0 licence; the US Department of Energy's prototype building models for EnergyPlus; OpenStudio; Bayesian calibration with a fast surrogate model in PyTorch; ASHRAE Guideline 14's error limits.
First step
Pick ten office buildings from the set, build their prototype models, and calibrate one by hand to learn the steps.
Done looks like
For 100 buildings, the share that meets Guideline 14's monthly limits (±5% bias, 15% CV(RMSE)) on 2017 data the model never saw, with the time spent per building, published with all code.