Innovative Digitalised Services for Future Heat Pump Systems

Background

Heating and cooling in the residential sector accounts for about 80% of the EU’s final energy use [1]. This energy is predominantly supplied by fossil fuels, resulting in high CO2 emissions. Because of this there has been a significant policy push to adopt heat pumps as they represent one of the most promising technologies for the decarbonisation of the heating sector in Europe. To reach environmental targets it is projected that a doubling of the heat pump stock is needed by 2030 [2]. However, there is a major barrier preventing this: a shortage of trained installers and servicemen. The IEA predicts that roughly 70,000 new installers will be needed in Europe by 2030 to keep up with demand [3].

The idea behind this project is to develop a digital tool that will serve as an aid to these new installers, streamlining the installation process as well as offering support when needed. In literature, it has been found that machine learning methods can be used as virtual sensors to fill gaps in collected datasets, allowing for more information about the heat pump to be known will little added cost and no retrofitting needed [4]. Similarly, it has been found that machine learning methods can be used as fault detection and diagnosis models, helping installers pinpoint the issues during maintenance checks [5].

However, a barrier to the deployment of machine learning models in industry is low trust due to their black-box nature. Therefore, in this project, the concept of physics informed machine learning will be investigated. Physics-informed machine learning is a field of machine learning where physical knowledge is implemented into machine learning algorithms to improve their interpretability as well as extrapolation abilities [6].

Combining this technical approach with insights from industry partners, the digital tool will become a one-stop-shop for the installation and maintenance of heat pump systems.

Aim and objectives

• Development of physics-informed machine learning models for accurate modelling of heat pump systems, acting as virtual sensors for missing information
• Development of fault detection and diagnosis methods for conducting health checks on the heat pump system, ensuring efficient operation
• Development of a digital tool to aid installers and servicemen

Project partners

AIT Austrian Institute of Technology, GmHb

Funding is provided by AIT Austrian Institute of Technology, GmHb.

Timeframe: 2025-2029

Researchers

Contact person

Adam Buruzs

Bernd Windholz

Publications

Physics-informed neural networks as virtual sensors for domestic heat pumps by Ellen Louise Nicholson, Yang Song, Adam Buruzs, Bernd Windholz, Davide Ronaldo, Michael Lauermann, Hatef Madani in 15th IEA Heat Pump Conference 2026

References

[1] European Commission, 2025. Energy performance of buildings directive [accessed 5 June 2026]

[2] European Heat Pump Association, 2026. Heat pump sales testify to government action [accessed 8 June 2026].

[3] IEA (2022), The Future of Heat Pumps, IEA, Paris, Licence: CC BY 4.0

[4] Song, Y., Rolando, D., Avellaneda, J. M., Zucker, G., & Madani, H. (2023). Data-driven soft sensors targeting heat pump systems. Energy Conversion and Management, 279, 116769.

[5] Z. Liu, Y. Liu, D. Zhang, B. Cai, C. Zheng, Fault diagnosis for a solar assisted heat pump system under incomplete data and expert knowledge, Energy 87 (2015) 41-48.

[6] Karniadakis, G.E., Kevrekidis, I.G., Lu, L. et al. Physics-informed machine learning. Nat Rev Phys 3, 422–440 (2021).

Keytags: Heat Pumps, Physics-Informed Machine Learning, Fault Detection and Diagnosis, Modelling