Building Energy Digital Twin 2026: EPBD, FDD, Baselines

Q: What does the recast EPBD require for building automation and control systems?

Directive (EU) 2024/1275 entered into force on 28 May 2024 and had to be transposed into national law by 29 May 2026. Article 13 requires building automation and control systems in non-residential buildings whose heating, air-conditioning or combined ventilation systems exceed 290 kW, and extends that to systems over 70 kW by 31 December 2029. The systems must be capable of continuously monitoring, logging, analysing and allowing adjustment of energy use; benchmarking the building's energy efficiency, detecting losses in efficiency of technical building systems and informing the person responsible about improvement opportunities; communicating and interoperating across proprietary technologies and manufacturers; and, by 29 May 2026, monitoring indoor environmental quality. National laws implement the detail, so check your member state's transposition for exemptions, alternatives and enforcement.

Q: Why call a building energy twin a counterfactual?

Because everything valuable it produces is a comparison against what should have happened. Savings are metered consumption against a baseline adjusted for weather and occupancy. A fault is equipment behaviour against the sequence of operations the engineer specified for the current mode. A retrofit's value is measured use against what the building would have used without it. The building management system already stores what happened; it does not store what should have happened. A twin that only renders live data on a 3D model has no reference and therefore cannot detect a loss in efficiency, which is exactly the capability the EPBD's Article 13 asks for. Building the baseline and the machine-readable control intent first, and the viewer last if at all, follows directly from that framing.

Q: What is the Brick schema and why does a building twin need it?

Brick is an open-source ontology, maintained by the Brick Consortium, that describes the physical, logical and virtual assets in buildings and the relationships between them using Semantic Web technology. In practice it lets you say that a BMS point named AHU3_SAT is a supply air temperature sensor on an air handler that feeds a particular set of zones. That mapping is what allows fault-detection rules and baseline models to be written once and run across a whole portfolio regardless of which vendor's BMS is in each building, which is also what the directive's interoperability clause is reaching for. Without a semantic point map every building is a fresh integration project with rules written against proprietary point names. RealEstateCore is a common alternative or companion; the important decision is to use an open model rather than a vendor's internal one.

Q: Which HVAC faults does fault detection and diagnostics usually find?

Mostly ordinary ones that waste a lot of energy for a long time. Simultaneous heating and cooling in the same air handler, where both valves are open and fighting each other. Economisers stuck closed, so compressors run on mild days when outside air could cool the building for free. Schedules overridden for an event and never restored, so systems run nights and weekends. Stuck or leaking valves that drive zones to their limits. Sensors that drift, so a control loop chases a value that is not real. Each is detectable by a rule that compares what the equipment is doing with what its sequence says it should do in its current mode, over a few points at 5 to 15 minute resolution. The value comes from ranking faults by estimated cost and having a named person close the top ones every week, verified by the data rather than by ticket status.

Q: How do you prove energy savings from a building twin?

Against an adjusted baseline, consistently with the International Performance Measurement and Verification Protocol. IPMVP's options are all counterfactuals: isolate a retrofit and measure it against its prior behaviour, regress whole-facility metered energy on weather and other drivers and compare actual use with the model's prediction, or compare against a calibrated simulation where no usable baseline period exists. In practice an existing portfolio uses a whole-building regression trained on at least a year of interval meter data and weather, with the training period recorded, non-routine events such as occupancy changes adjusted for, and an uncertainty band reported with every savings figure. Unadjusted year-on-year comparisons are not proof: a mild winter makes a programme look brilliant and a cold one gets it cancelled, and neither says what operations did.

Q: Do I need a 3D model for a building energy twin?

Not to save energy in an existing building. Geometry is useful for space management, for showing a technician where on a floor a fault is, for communicating with stakeholders, and for physics simulation, where envelope, zoning and constructions matter, for example when designing a deep renovation or checking a zero-emission new build with a calibrated EnergyPlus model. It contributes nothing to detecting simultaneous heating and cooling or a stuck economiser, which are time-series rules over a handful of points. Vendors lead with the viewer because it demonstrates well. If a proposal begins with a scan of the building and defers analytics to a later phase, the order is wrong, and the programme is likely to stall before it produces any measurable saving.

Q: When should a building twin start writing setpoints back to the BMS?

Late, and only with guardrails. Supervisory optimisation, where the twin adjusts setpoints, schedules or reset strategies, is where the largest savings claims come from, but its results can only be measured against a baseline, so the baseline has to have been trusted for a full year first, and the fault-detection layer has to have cleaned up the broken equipment an optimiser would otherwise work around. When you do switch it on, set hard limits on what it can change, keep a manual override at the building, feed comfort complaints and indoor environmental quality readings back into its objective, and measure its effect as a separate line in the savings report. An optimiser with unlimited write access and a wrong baseline will save energy on paper and lose the occupants' trust in the first heatwave.

Q: When is a building energy twin not worth building?

When the building or portfolio is too small or too simple for the counterfactual layers to pay back. A single small building with one packaged rooftop unit and a utility meter needs a commissioning visit, a sensible schedule and a monthly comparison of the bill against degree days, not a platform. A portfolio below the directive's thresholds, with no retrofit programme and no green financing to report against, may get more from one retro-commissioning exercise. And any organisation that cannot name the person responsible for working the top faults each week should not buy fault detection yet, because the most common failure of building analytics is a platform producing hundreds of ignored faults. Build the trend store and the point map first in any case; they are cheap and every later decision depends on them.

Satyam Kumar

Back to Blog

architecture

Building Energy Digital Twins in 2026: The Twin Is a Counterfactual, Not a 3D Model, So Build the Baseline Before the Viewer

By Satyam KumarOctober 24, 202617 min read

building energy digital twin epbd 2024/1275 building automation and control systems bacs epbd article 13 hvac fault detection and diagnostics fdd building analytics brick schema ipmvp measurement and verification energy baseline model energyplus calibrated simulation smart readiness indicator zero-emission buildings 2030 commercial real estate energy bms analytics supervisory hvac optimisation building energy management

Building Energy Digital Twins in 2026: The Twin Is a Counterfactual, Not a 3D Model, So Build the Baseline Before the Viewer

Frequently Asked Questions

Share this article

Twitter LinkedIn WhatsApp

Satyam Kumar

Founder & AI Architect, AppScale LLP

AI & Cloud Architect. Helping teams build systems that scale to millions.

LinkedIn GitHub

Building Energy Digital Twins in 2026: The Twin Is a Counterfactual, Not a 3D Model, So Build the Baseline Before the Viewer

Frequently Asked Questions

Share this article

Comments

Leave a comment