In short
- A twin is a virtual model synchronised with live data: geometry, assets, power, cooling and IT load.
- A BIM model alone is the foundation (level 1), not yet a twin.
- Five layers: geometry, live data, simulation, optimisation, control. Skipping one is the most common reason twins stall.
- Clean tags and naming matter more than the software choice.
What a digital twin is
ISO/IEC 30173:2023 sets out the concepts and terms. In a data center the twin combines the building's geometry and assets with live readings from the building management, power monitoring and IT systems. Because it reflects the hall as it is now, it can answer questions about the real hall: where the next racks can go, what happens if a cooling unit fails, how far supply temperatures can rise.
A model is not yet a twin
| As-built model | Digital twin | |
|---|---|---|
| Geometry and assets | Yes | Yes, kept current |
| Live data | No | BMS, EPMS, DCIM, sensors, IT telemetry |
| Simulation | Occasionally, one-off | Calibrated against measured data |
| Answers | What is where | What is happening and what if |
An as-built model, often from scan to BIM, is the right starting point. See scan to BIM on allinSCAN.
Five layers under one model
- Geometry: as-built model of halls, racks and plant, updated with every change.
- Live data: BMS, EPMS, DCIM, sensors and IT telemetry with consistent tags.
- Simulation: CFD and electrical models calibrated against measured data.
- Optimisation: analytics and machine learning that recommend, with the reasoning shown.
- Control: write-back to setpoints within limits, with override and audit.
Each layer reads from the one below. The twin stack on the home page shows the same order; maturity levels are in the levels guide.
What feeds the twin
| Source | What it gives | Update |
|---|---|---|
| As-built model | Geometry, rooms, racks, plant | At changes |
| Asset register · CMMS | Equipment, ratings, maintenance history | Daily |
| BMS | Cooling plant states, setpoints, temperatures | Seconds to minutes |
| EPMS · meters | Power, breakers, UPS, generators | Seconds |
| DCIM | Racks, power chains, connectivity | Minutes |
| Environmental sensors | Rack inlet temperature, humidity, pressure | Minutes |
| IT telemetry | Server and GPU power, fan speed | Seconds |
| Weather · grid | Outdoor conditions, prices, grid signals | Hourly |
Start with the sources you already have and fix the naming before adding new sensors.
What a twin does
- Capacity: prove power and cooling headroom before placing racks; release stranded capacity.
- Energy: raise supply temperatures and fan setpoints safely; track PUE live.
- Resilience: trip a CRAH, chiller or UPS in the model first.
- Change and maintenance: check works against redundancy; predict failures from trends.
- Handover: commissioning results become the twin's starting point.
- AI clusters: model CDUs, flow and heat rejection for liquid-cooled racks.
- Training: rehearse incidents on the twin instead of the live plant.
Standards that keep a twin portable
| Area | Standard |
|---|---|
| Concepts | ISO/IEC 30173:2023, digital twin concepts and terminology |
| Framework | ISO 23247, a reference structure from manufacturing |
| Information | ISO 19650, information management with BIM |
| Model format | IFC · ISO 16739 |
| Semantics and tags | Brick schema · Project Haystack |
| Industry | Digital Twin Consortium; NVIDIA Omniverse DSX blueprint for AI factories |
Questions about digital twins
Is DCIM a digital twin?
DCIM covers part of the live-data layer: racks, power chains and capacity. A twin adds geometry, simulation and, at higher levels, optimisation and control.
Is a 3D BIM model already a twin?
Not on its own. It is level 1, the foundation. It becomes a twin when live data flows into it and it can answer questions about the real hall.
What data is needed first?
An as-built model, an asset register, BMS and power meter points with consistent tags, and rack power.
How common are twins today?
In an Eaton and 451 Research survey, 33 % of operators had or planned a digital twin.
Sources
- ISO/IEC 30173:2023; ISO 23247; ISO 19650; ISO 16739 (IFC); Brick schema; Project Haystack; Digital Twin Consortium.
- Eaton / 451 Research, 2023 (adoption).