In order to achieve the national and European climate neutrality targets for 2050, CO2 emissions must be significantly reduced. The building sector plays a central role in this: in the EU, it accounts for 40% of final energy consumption and 36% of emissions. However, continuous, systematic monitoring of building operation is rarely carried out due to the complexity of heating, ventilation and air conditioning (HVAC) systems, insufficient data and tools or a lack of personnel, even though ongoing monitoring of operating parameters has great potential. Studies show a great potential for energy savings of up to 30 % through optimised operational management and intelligent monitoring in non-residential buildings. However, software available on the market for optimising operations requires highly qualified staff, is complex to set up and often requires the measurement technology of the existing HVAC to be upgraded in advance. The possibilities and advantages of intelligent fault detection and automatic fault correction in buildings have not yet been realised in practice.
SELF²B demonstrates an AI-based self-learning & self-diagnosing fault detection and diagnosis solution (FDD) in the building portfolio of the Bundesimmobiliengesellschaft at the site of the University of Veterinary Medicine Vienna.
The solutions developed in the project are to be demonstrated in real operation in the form of a real-time online FDD prototype and the benefits evaluated using an assessment matrix (technical, economic, ecological). In addition to the HVAC systems, the PV system at the site will also be continuously monitored. Furthermore, a technology concept for "self-learning, self-optimising" existing buildings will be developed for the next generation of efficient building operation.
The innovations planned in the SELF²B project go beyond the international state of the art: the combination of semantic data and ontologies, heuristics and machine learning guarantees scalable and robust solutions for HVAC and PV systems. The combination of semi-supervised machine learning models with autoencoders in combination with automated clustering and classification models planned in the project also represents an innovation in machine learning that can potentially be transferred to other areas.
The planned user integration during development and the focus on explainability and user-friendliness address the market hurdle of technological scepticism among the relevant stakeholder groups, which is relevant for fully automated software solutions. Important research work comes mainly from China and the USA, i.e. the planned pilot project is one of the first real-time implementations in this form in Europe.
