Building Digital Twins for UK Commercial Properties

From NASA to your plant room

The digital twin concept originated with NASA's Apollo programme in the 1960s — physical replicas of spacecraft on Earth fed with real-time mission telemetry. During Apollo 13, engineers ran rescue scenarios in those simulators in real time. That bidirectional, data-driven relationship between a physical system and its virtual replica is exactly what defines a digital twin today.

The term was formally defined in 2002 by Dr Michael Grieves at the University of Michigan, and coined as "digital twin" by NASA engineer John Vickers in 2010. By 2028, the market is projected to reach USD $110 billion, driven by nearly 21 billion connected sensors globally. In UK smart buildings, the tools are ready and the business case is increasingly undeniable.

The five maturity levels — where are you?

Level	Type	Capability	Human Role
ML1	Status Twin	Real-time data capture and visualisation	Performs all analysis and decisions
ML2	Informative Twin	Historical + real-time data structured for operators	Interprets structured data
ML3	Predictive Twin	Forecasts future states using physics or ML models	Validates predictions
ML4	Optimisation Twin	Explores what-if scenarios, recommends optimal actions	Selects from options
ML5	Autonomous Twin	Deduces and enacts optimal control independently	Monitors; twin controls

Most building digital twins in the UK today sit at ML1 or ML2 — dashboards with live data. ML3 and ML4 are commercially available and deployable now. ML5 exists in specialist HVAC optimisation platforms and is approaching the mainstream.

What autonomous actually looks like — the HVAC example

Armstrong Fluid Technology's platform is the closest real-world example to a Level 5 twin in the HVAC domain. It builds evolving equipment performance maps — data-driven models correlating plant inputs and outputs with varying heat loads. The algorithm continuously reviews those maps, evaluates all valid control combinations ranked by total power consumption, and implements the lowest-consumption option — but only when a statistically significant percentage of results outperform the current physical settings. Any behavioural deviation triggers a predictive maintenance flag.

This is not a dashboard. It is a self-learning control loop that replaces static PID tuning with continuously evolving models. For a medium-sized commercial building, this typically delivers 12–22% HVAC energy reduction beyond what a well-commissioned conventional BMS achieves.

The DTaaP architecture — security built in

The Digital Twin as a Proxy (DTaaP) model, developed at Luleå University of Technology, defines four layers: physical devices, a communication layer, a proxy layer (the twin itself), and an application layer. External systems — dashboards, control logic, analytics — interact only with the digital twin, never directly with physical hardware. The twin performs all security and authorisation checks before passing commands through.

Open-source tools you can deploy today

• Eclipse Ditto — open-source digital twin middleware managing millions of device twins via HTTP, MQTT, WebSocket, Kafka. Proven in the DTaaP reference implementation. EPL 2.0 licence.
• OpenModelica + Modelica Buildings Library — physics-based simulation of HVAC, thermal, and hydraulic systems. Pre-built models for VAV, chiller plants, and hydronic heating.
• Fledge (LF Edge) — industrial edge data integration connecting BACnet, Modbus, OPC UA and sensor data into your twin. Apache 2.0.
• Grafana — operational visualisation. A floorplan background with real-time metrics delivers a functional ML1/ML2 twin immediately, with zero proprietary lock-in.
• OpenTwins (University of Malaga) — full-stack open-source platform covering real-time state, simulation, and predicted data.

From BIM to live twin — the pipeline

Building Information Modelling (BIM) is the natural starting point. The pipeline is: BIM model (IFC format) → geometry and asset data extraction → integration with live sensor feeds (BACnet/Modbus/IoT) → twin middleware → analytics and control layer. For UK projects already delivering BIM Level 2, the path to a live operational twin is shorter than most clients realise.

Where to start

For most UK contractors and building owners, the right entry point is ML2 — a structured operational twin aggregating BMS data, IoT sensor feeds, and energy meter data into a single coherent view. Achievable in 8–14 weeks on an existing building with an accessible BMS. The ML3 predictive layer typically follows within 6 months once the data pipeline is stable.

The critical architecture decision: choose open protocols and open middleware from day one, and every subsequent layer is replaceable. Choose a proprietary platform and you are locked in for the life of the asset.