NEXUS - AI for Next Generation Smart Buildings

he NEXUS project develops a novel AI-based framework for scalable fault detection and predictive maintenance of HVAC systems in buildings. NEXUS enables early fault detection without the need for large labeled datasets. The project aims to significantly reduce energy consumption, extend system lifetimes, and contribute to the decarbonization of the building sector.

Short Description

Background/Motivation

Buildings account for around 35% of total energy consumption in Austria and therefore contribute significantly to CO₂ emissions. A large share of this consumption is caused by heating, ventilation, and air-conditioning (HVAC) systems.

Studies show that operational faults in HVAC systems can increase energy consumption by up to 30%. Although automated fault detection and diagnosis methods offer substantial potential for energy savings, existing approaches have so far seen limited adoption in practice, as they are often not sufficiently scalable, highly data-intensive, and only partially interpretable.

Contents and goals

The goal of NEXUS is to close this research gap by purposefully combining unsupervised and semi-supervised machine learning methods to enable robust and scalable fault detection as well as predictive maintenance for building systems. The focus is on solutions that do not require large amounts of labeled fault data and that can be transferred across different buildings and system configurations.

Methodical Approach

Within the project, deep learning models and representation learning techniques are employed to learn normal operating behavior and to reliably identify deviations. This enables early fault detection even when no explicit fault data are available.

To improve the transparency and acceptance of the results, classical machine learning methods are deliberately combined with deep learning approaches in order to provide meaningful and understandable diagnoses.

Expected Results

The project aims to achieve a significant performance improvement in fault detection compared to the state of the art, as well as a reduction in maintenance-related energy consumption of at least 10%.

The results form the basis for practical applications in building management and make a substantial contribution to improving energy efficiency and decarbonizing the building sector.

Project Partners

Project management

TU Wien; Institute of Building and Industrial Construction

Project or cooperation partners

DiLT Analytics

Contact Address

TU Vienna
Dr. Adil Mukhtar
Karlsplatz 13
A-1040 Wien
E-mail: adil.mukhtar@tuwien.ac.at
Web: https://github.com/tuw-isab