bimkernel · 07

    Concepts · standards · software

    Digital
    twins
    on site.

    A model is not a twin until data flows back on its own from the real asset.

    The problem

    The model is handed over and frozen.

    Construction lasts a few years and the building, decades. On handover day the model stops changing, but the asset does not.

    Life cycle of a building designbuild operation · 50 years or more handover: the model freezes
    Service life 50years set by the CTE (Spanish building code) for an ordinary building
    What changesUsesEquipmentRefurbishments
    What happensTemperatureOccupancyFaults
    What is missingA model that keeps up
    Definition

    Three parts and a round trip.

    A physical asset, its digital replica and the connection that keeps them in sync. That connection is what makes the model a twin.

    Physical · connection · digital datasyncdecisionsPHYSICAL ASSETDIGITAL REPLICA
    The concept 2002 introduced in industrial product lifecycle management
    What it is notA pretty renderwithout real data there is no twin
    The testDoes it update itself?as often as its use requires
    Model · shadow · twin

    It depends on how the data travels.

    Manual both ways is a model. Automatic one way only, a shadow. Automatic both ways, a twin.

    01 · Model P D Modelmanual · manual
    02 · Shadow P D Shadowautomatic · manual
    03 · Twin P D Twinautomatic · automatic
    How to read it
    P = physical assetD = digital replicasolid line: automaticdashed: manual
    Components

    Six layers, from asset to decision.

    Data is born in a sensor, travels over a network, is stored with its context, is matched with the model and ends up in an action.

    The stack of a twin 1 · ASSET 2 · SENSORS AND CONTROL 3 · CONNECTIVITY 4 · DATA WITH CONTEXT 5 · MODEL + ANALYTICS 6 · DECISION building, bridge, networktemperature, flowcable, radio, gateway
    The layer that fails most4 · Contexta reading without «where» or «of what» is useless
    The layer that sells most5 · The 3D viewereye-catching, but only the tip of the iceberg
    Maturity

    From scan to autonomous building.

    Six steps, from the point cloud to the building that runs itself. Each one rests on the one before.

    Maturity spectrum · 0 to 5 012345 capturemodel+ staticdata + realtimetwo-waylinkautonomous The leap that turns a shadow into a twin lies between 3 and 4
    Frequency and fidelity

    As live as the decision it serves.

    Each kind of data has its own pace: geometry changes over years, occupancy in minutes and a vibration in milliseconds.

    Typical update interval
    Yearsgeometry and refurbishments
    Monthsscanned construction progress
    Daysmaintenance work orders
    Mintemperature, CO₂, occupancy
    Secenergy meters
    msstructural vibration
    Indicative scale Inferred
    The ruleFrequency and fidelityare set by use, not by technology
    Life cycle

    Born in design, matures in operation.

    In design it simulates, during construction it compares as-built with as-designed, and in operation it monitors and learns.

    Design Simulateenergy, evacuation, airflow
    Construction Comparedesigned versus scanned
    Operation Monitorsensors, alerts, maintenance
    The thread that ties it together
    Design model→Construction model→Asset information model→Twin
    Use cases

    Six questions it answers.

    A twin earns its place through the questions it answers better than a spreadsheet or a site visit.

    MaintenanceWhat will fail?
    EnergyWhere is it lost?
    SpaceWhich rooms are surplus?
    ComfortIs the air good?
    StructureIs it moving more than usual?
    ScenariosWhat if…?
    Real cases

    From a building to a country.

    The most cited examples range from an instrumented bridge to the living model of an entire city. All of them started from a specific question.

    Railway · LondonCrossrailasset information planned from day one
    City · SingaporeVirtual Singapore3D model of the whole country, announced in 2014
    Bridge · ScotlandQueensferry Crossingstructural health monitoring since 2017
    Airport · VancouverOperations in a twinpassengers, gates and baggage
    City · MadridMunicipal urban twin€5.1 M of 3D mapping until 2028
    Risks

    More connection, more attack surface.

    A twin joins the office network to the building services network. You have to decide who sees what, who owns the data and how much it costs to keep it running.

    01 · Security Building services network exposed
    02 · Ownership Who owns the data?
    03 · Cost Keeping it alive costs money
    04 · Showcase A viewer with no real data
    Map of standards

    There is no single twin standard.

    There is one per layer: vocabulary, information management, model, data meaning and transport. The twin stitches them together.

    One standard per layer VOCABULARYISO/IEC 30173 · Gemini PrinciplesMANAGEMENTISO 19650-1, -2, -3 · COBieMODELIFC 4.3 · ISO 16739-1:2024SEMANTICSBrick · Haystack · REC · SOSATRANSPORTBACnet · MQTT · OPC UA · SensorThings
    The stitchingA common keythe element identifier, the same in every layer
    ISO 19650

    From the project model to the asset model.

    The standard separates the model used to design and build from the one handed over for operation. The twin is fed by the second.

    Requirements → models OIR AIR EIR PIM AIM twin ← handover
    The relevant part19650-3operational phase of assets
    IFC

    The model already knows what a sensor is.

    The open standard has entities for sensors, actuators and time series, and a unique identifier for every element.

    What measuresIfcSensorwith a type: temperature, CO₂, flow…
    What actsIfcActuator · IfcControllervalves, dampers, controllers
    The keyGlobalId22 characters that never change
    What it does not doStore live datatime series in IFC are a history, not a stream
    Semantics

    Let the data say what it is and whose it is.

    A point called “AI-3.07” says nothing. An ontology turns it into “air temperature of zone 104, supplied by unit 1”.

    A graph of relationships CL-01 VAV-104 Zone-104 P01-104 TT-104 feeds feeds hasPart isPointOf
    Four vocabularies
    BrickHaystackRealEstateCoreSOSA/SSN
    converging: Brick already aligns with the others
    Gemini Principles

    Nine principles in three groups.

    The UK published them so that twins owned by different parties could one day be connected. They are still the best checklist.

    Purpose Public goodValue creationInsight
    Trust SecurityOpennessQuality
    Function FederationCurationEvolution
    The underlying idea
    Twin of one asset→Connected twins→National twin
    Software map

    Each vendor pushes from its own ground.

    BIM authoring vendors stretch their model towards operation. Building services and industrial vendors climb up from the data. They meet in the middle.

    Buildings ↔ infrastructure · model ↔ data ↑ from the BIM model↓ from building services dataBUILDINGSINFRASTRUCTURE AUTODESK TANDEM NEMETSCHEK DTWIN BENTLEY ITWIN 3DEXPERIENCE WILLOW SIEMENS BUILDING X IOT PLATFORMS Indicative positioning Inferred
    Autodesk Tandem

    From the handed-over model to the building in use.

    It takes the project models, keeps the assets that matter to the operator, gives them parameters and connects them to sensor streams.

    Four steps
    Revit · IFC · DWG→Classify assets→Parameter template→Sensor streams
    Who it is forOwners and maintenance teams
    What it is notAn authoring toolyou do not model in it
    Bentley iTwin

    Infrastructure, with its change history.

    It synchronises models from many programs into an iModel that records every change, like a code repository, and attaches sensor and inspection data to it.

    One iModel, many sources DGN RVT IFC iModel change by change
    Since 2024Cesiumthe 3D map of the world, inside Bentley
    Visualisation

    Game engines are the shop window.

    They provide the realistic image and real-time simulation. The data and its meaning still live in another layer.

    Unreal EngineRealistic imagereal-time visualisation and simulation
    UnityCustom applicationsthe Vancouver airport twin
    NVIDIA OmniverseOpenUSDscenes shared by many programs
    The trapPretty is not reliablea twin’s quality lies in the data, not in the render
    Platforms

    The cloud brings the graph and the scale.

    The big clouds offer a twin service: a graph of entities and relationships connected to millions of readings.

    Azure Digital TwinsGraph in DTDLrelationship queries and event routes
    AWS IoT TwinMakerEntities and components3D scenes and data connectors
    Eclipse DittoOpen sourcewithout depending on a cloud
    Cesium ion3D Tileshuge models over the internet
    The riskServices changeAWS retired IoT Events in 2026
    The graph

    A twin is a graph, not a file.

    Every room, piece of equipment or sensor is a node with properties and telemetry. The relationships are what let you ask questions.

    Building → storey → room → sensor Building P00 P01 Room 001 Room 002 Room 104 23.4 °C · now
    Modelling language DTDL JSON-LD: properties, telemetry, relationships and commands
    What it lets you ask“Rooms on level 1 above 26 °C with the air handling unit running”
    On the map

    A twin without coordinates cannot be connected.

    To join a building’s twin with that of its street, its network or its city, they all have to speak the same reference system.

    Building · network · city buildingnetworkcityEPSG:25830
    How it is doneShared coordinatesArticle 01 →
    Interoperability

    Three worlds that need stitching together.

    The BIM model is static, the semantic graph explains relationships and the building services network speaks in readings. Each has its own format.

    StaticIFC · COBiewhat is there and where
    SemanticBrick · DTDLwhat serves what
    LiveBACnet · MQTTwhat is happening
    The seam IfcSensor brick:TT-104 AI-3.07 · 23.4 same identifier in all three
    Identifiers

    If the GUID changes, the twin breaks.

    The sensor in the model and the one on the network are matched by an identifier. Deleting and recreating an element changes it and leaves the reading orphaned.

    IFC0wkdTQadv0cuaIsExdNkQZsensor GlobalId · the model’s one
    TagTT-104the code on the nameplate
    Control networkAI-3.07the controller point
    The table nobody wants to buildMapping tableone row per point, with all three codes
    Protocols

    From the controller to the cloud.

    Building services speak industrial protocols. A gateway translates them into lightweight messages that the cloud stores as time series.

    Field → gateway → cloud BACnet Modbus KNX · LoRaWAN gateway MQTT OPC UA time series
    Handover

    The twin is specified in the contract.

    If the tender documents do not require assets with code, manufacturer and link to their space, the model arrives without them and nobody adds them later.

    From tender to twin
    AIR in the tender→IDS that checks it→IFC + COBie at handover→Point map→Twin
    ExpensiveSurveying the data afterwardson-site inventory, item by item
    CheapAsking for it from the starteach party fills in its own part
    Quality control

    Four questions before trusting it.

    Is everything that was requested there? Can it be linked? Do the relationships make sense? Is the incoming data good?

    01 · CompleteAssets with their dataIDS on the IFC
    02 · LinkableOne identifier per pointmapping table
    03 · ConsistentRelationships that make senseSHACL on the graph
    04 · ReliableHealthy readingsgaps, jumps and drift
    IDS

    The specification, turned into a test.

    One file requires every sensor to have a tag and a room, and every piece of equipment to come with manufacturer and serial number. The IFC passes or fails.

    Tested5test elements
    Pass2sensor and pump complete
    Fail3tag, room, serial number
    What is required
    Tag = AA-000Inside an IfcSpaceManufacturerSerialNumber
    SHACL

    Rules for the graph.

    What IDS does with the IFC, SHACL does with the graph: it checks that each sensor hangs from what it should and measures in the right unit.

    The built-in shapesGuard the structureclasses, units, deprecated terms
    Your own shapesGuard the project“CO₂ goes in ppm”
    Test result
    Built-in0 project errors
    Own3 errors: unit and two “points” that are not
    Sensor data

    A sensor can lie too.

    It freezes, drifts out of calibration, loses readings or gets the time wrong. Check the data before you believe the twin.

    Four typical faults in a temperature series gapfrozenspikedrift
    In summary

    Four rules.

    01A purposethe questions set the frequency
    02A contractAIR + IDS from the specification
    03A keythe same identifier everywhere
    04A checkIDS, SHACL and the live data