Cutting cold storage energy costs with price-aware cooling and frost detection

A cold store holds -18°C around the clock and buys power at a price that changes every hour, but the plant had no way to act on that.

Two refrigeration controllers from different vendors ran the facility without exchanging data, keeping history, or knowing what energy cost at any given moment. Defrost ran on a timer, which meant defrosting clean coils and leaving blocked ones too long.

The information needed to run the plant well existed. It was just spread across systems that could not reach each other.

HB Development built a platform that reads both controllers across their incompatible protocols and combines them with day-ahead electricity prices, metered consumption, and weather data in one record.

A camera on the evaporator feeds a frost detection model trained on frames from the live installation. On top of that data sits an optimiser that plans the next 24 hours each afternoon, as soon as the following day's prices are published: pre-cooling into cheap hours, coasting through expensive ones, and defrosting when a coil needs it.

The engagement was deliberately split. The first phase only collected data, with no ability to change plant behaviour.

That history became the ground truth the thermal and energy models were fitted against: no assumptions about how the plant behaves, only records of how it did.

Every write back to a controller is validated against safe operating ranges and stays under human approval. Siemas has owned the code, the cloud accounts, and the credentials since the first day of the project.