Edge computing in smart buildings: HVAC and occupancy control

Commercial buildings rarely operate as one system. Lighting, ventilation, lifts, access control, solar generation and tenant comfort each have different owners, data formats and priorities. Edge computing in smart buildings gives local controllers enough intelligence to react quickly while sending selected insights to the building management system and wider facilities platform.

For Australian operators, the payoff is practical: lower peak demand, steadier comfort, fewer faults and better use of installed plant. The approach works best when occupancy data is treated as an operating signal rather than a surveillance system, and when cloud analytics remain connected to dependable on-site control.

Why local edge intelligence matters

At the edge, a gateway or controller combines readings from air-quality sensors, room controllers and occupancy devices. It can infer the state of a zone and issue a local command without waiting for a round trip to a remote cloud service. That matters when a meeting room fills up, a loading bay becomes busy or a chiller trip changes the load.

Local processing reduces raw data traffic and keeps essential automation operating during unstable internet connections. It also limits exposure by keeping detailed room-level data on site. Over time, aggregated information can reveal persistent inefficiencies, such as a conference wing running at full capacity each morning while the adjacent zone remains empty until lunch.

Edge systems also create a useful record of how spaces are used over time. That record can support maintenance planning, capital replacement and future automation projects, provided operators can interpret the data. A common naming standard and data dictionary prevent each new sensor vendor from creating a separate set of occupancy definitions.

A practical control architecture

A useful architecture separates three layers. Sensors and actuators provide measurements and execute commands; an edge layer performs rules, sensor fusion and local optimisation; cloud or enterprise systems handle historical trends, fleet-wide comparison and longer-term forecasting. This division prevents a loss of internet access from stopping heating, cooling or ventilation.

The edge layer should connect to the building management system through stable interfaces such as BACnet/IP, Modbus, MQTT or OPC UA. It can also expose selected metrics to a digital twin, maintenance system or analytics platform. Clear tags for rooms, equipment and alarms are essential, otherwise local intelligence simply creates a faster way to produce confusing data.

Commissioning belongs in the design, not after handover. Engineers should verify sensor placement, command limits, override behaviour and failure states with the facilities team. Manuals must explain when the system optimises, when it holds a safe state and who can change a schedule or setpoint.

Choosing occupancy signals that respect privacy

Reliable occupancy sensing combines complementary signals. mmWave, infrared or thermal sensors can detect movement and direction; acoustic and pressure sensors can support space-use estimates; Wi-Fi or Bluetooth device counts can be useful in some offices. CO2 readings act as a ventilation proxy, not a perfect count of people, so the system should combine context rather than trust one feed.

Convert the raw stream into a simple state such as occupied, unoccupied or uncertain, then retain that state instead of identities or movement trails. Set short retention periods, remove cameras and facial recognition, restrict access by role and publish the purpose of collection. Confidence scores and sensor health belong in the data model, allowing a controller to make a conservative decision when a device fails.

Accuracy depends on calibration and context. Daylight can overwhelm passive infrared detection, open doors can distort acoustic counts, and CO2 readings shift quickly when ventilation changes. Mount sensors at suitable heights, avoid direct sunlight and supply air, compare outputs with manual observations, and schedule cleaning and recalibration as part of the maintenance plan.

Turning occupancy into HVAC savings

Occupancy data becomes useful when it changes how plant is sequenced. An empty meeting room can move into a setback band, while a busy area can receive more airflow and an earlier cool-down. The objective is to remove simultaneous heating and cooling, avoid over-ventilating vacant zones and reduce the hours for which full capacity is unnecessary.

The edge controller can combine occupancy, temperature, CO2, humidity, equipment status and weather forecasts to adjust variable-air-volume boxes, fan speed, chilled-water flow and zone setpoints. Demand-controlled ventilation, static-pressure reset and heat recovery offer further gains. Free cooling can be enabled only when outdoor conditions, humidity and indoor humidity limits make it worthwhile.

Measure results against a baseline rather than assuming every algorithmic change is an improvement. Track energy per occupied hour, peak demand, temperature complaints, CO2 exceedances, override frequency and equipment faults. Normalise for weather and tenancy changes, and retain a manual mode so operators can investigate unusual conditions without undoing the entire programme.

Optimisation can also reveal mechanical weaknesses. A valve that never reaches full stroke, a fan running at full speed with little airflow or simultaneous heating and cooling may produce distinctive control patterns. Tag alarms for review rather than allowing the algorithm to mask them, and feed confirmed faults into maintenance work orders.

Security, standards and operational resilience

Because edge devices sit close to mechanical plant, they create a direct path into building operations. Place them in dedicated network segments, require unique credentials and encrypted connections, disable unused services, and monitor for abnormal commands. Signed firmware, controlled updates, tested backup configurations and rapid isolation of a compromised controller should be part of the operating standard.

Privacy and cyber security are connected in shared spaces. Under the Australian Privacy Act and Australian Privacy Principles, some occupancy datasets may not be personal information, but device identifiers, individual tracking or sensitive locations can change that assessment. Avoid unnecessary granularity, document retention and access, and align controls with frameworks such as the ACSC Essential Eight where they suit the organisation’s risk profile.

At an organisational level, create an asset register that names the controller, firmware owner, support contact and expected service life. Procurement should include vulnerability disclosure, patch access and secure configuration handover. When a vendor ends support, the organisation needs a documented replacement or isolation plan instead of leaving an active pathway into plant.

Building the case for Australian sites

Australian sites add distinctive operating conditions. Melbourne’s rapid temperature changes make free-cooling windows short and reliable scheduling essential. Sydney’s warmer summer afternoons and humidity can push chiller demand, while Brisbane’s subtropical conditions require careful humidity and ventilation control. These local patterns make generic schedules less convincing than controls driven by live conditions.

Building stock is equally varied: a Sydney CBD tower may have rooftop chillers and variable-air-volume terminals, while a Melbourne school, warehouse or strata apartment may use packaged units and split systems. In everyday Australian facilities language, “the aircon” may cover everything from a single room unit to a central plant. A solution should therefore support both BACnet-connected equipment and simpler controllers with secure gateways.

The commercial case should align with Australia’s energy-efficiency framework. The National Construction Code influences requirements for new and renovated buildings, while NABERS operational ratings can benchmark performance for commercial and institutional property. Electricity tariffs, demand charges and network conditions differ across states and retailers, so a Victorian office should not inherit a schedule designed around Queensland peak pricing. Test savings against the site’s actual tariff and tenancy arrangements.

Australian building services are often split among owners or strata managers, facility teams, controls contractors and service providers. A project can stall when nobody owns the gateway, data or setpoint changes. Define responsibility before installation, include handover training in the contract, and make performance information available through channels the operating team already uses.

A staged deployment blueprint

A measured pilot can prove the operating model before an organisation commits to a building-wide rollout. Choose a site with a clear problem, reliable plant data and an engaged facilities team, then expand only after the controls and safeguards have been accepted in daily use.

• Select a representative site with clear ownership, a repeatable energy or comfort issue, and accessible historical plant data.
• Establish a baseline for electricity, peak demand, indoor conditions, complaints, overrides and equipment faults before changing the controls.
• Instrument a small number of meaningful zones with temperature, CO2 and occupancy sensors, checking placement rather than simply adding more devices.
• Place optimisation and safety rules on an edge gateway with documented schedules, setpoint limits and manual override paths.
• Run an eight-to-twelve-week trial that accounts for weather, holidays and occupancy patterns, then compare results with the baseline.
• Assign named owners for facilities, digital systems and cybersecurity, and document data retention, device updates and rollback procedures.

After the trial, review comfort and energy together. An apparently small energy gain is not worthwhile if occupants report stale air, temperature swings or slow response. Tune the rules, train operators and communicate what the system does before adding more sites or sensor types.

Building teams can turn the findings into an internal business case and an industry briefing. Organisations preparing a workshop or awareness campaign can adapt the Edge Computing Association’s edge media kit for consistent background material and event promotion.

Choose one high-occupancy building, identify two controllable zones, and begin a measured 90-day trial with baseline data, named operational owners and a documented rollback plan.