Skip to content

Appearance

Language

Editing

Research and development

Research at sinnograte

With HYDRA we develop physics-informed AI for heat pumps: models that detect faults and wear early, in the cloud and directly on the device.

Status as of

The FFG has approved HYDRA. The first step is the data foundation on which everything else builds.

  • HYDRAResearch project on physics-informed AI
  • HVACopilotthe same AI in the cloud and on the device

What it is about

Heat pumps report a lot of measurements, but hardly anything about their condition. A limit value only triggers once something is already broken. HYDRA investigates how fouling, leaks and sensor faults can be detected earlier, before an installation stops.

Physics, not just statistics

A purely data-driven model can reproduce a curve well and still violate an energy balance that cannot be violated. HYDRA gives the models the thermodynamic relationships that apply anyway. This makes them leaner in data and more reliable at operating points that rarely occur, which is exactly where they are needed.

In the cloud and on the device

In the cloud, an anomaly can be compared with many installations of the same type, and only that comparison turns it into a finding. On the device, data is analysed without raw data leaving the building. The data foundation for this is HVACloud, our platform for manufacturers of HVAC equipment.

What we do not claim

Research is research. Whatever passes from HYDRA into a product does so when it is robust. The AI narrows things down and makes suggestions, decisions are made by people.