AI4FM - Artificial Intelligence for Facility Management

AI-based anomaly and fault detection in buildings. Digital twins of buildings with simulation models for testing and optimizing rule-based fault detection methods. Mining of the recorded time-series data from existing Building Management Systems to train Machine Learning models for fault detection.

Short Description

Starting point / motivation

The project partner Flughafen Wien AG manages more than one hundred buildings at the airport site. In 2023, internal consumption amounted to over 78 GWh of electrical energy (incl. heating/cooling units), over 33 GWh of heating energy and 25 GWh of cooling energy. The installed sensors and the building management system allow the systems to be monitored in real time and faults in the systems and components to be detected through manual analyses.

Faults or non-optimal operating modes of the building services systems often remain undetected due to the limited resources of the operating staff and can therefore cause unnecessary energy consumption and thus increased operating costs. The detection of anomalies and optimization potentials makes it possible to initiate countermeasures promptly and thus save energy and costs.

In the Smart Airport City (SAC) preliminary project, an analysis system was developed that detects anomalies by means of monitoring rules for technical building systems and their collected trend data.

Contents and goals

In the existing fault detection system false alarms are possible due to the sometimes high complexity of the rules and the large number of different technical building systems. In order to reduce the rate of false alarms and to check the rules and regulations before going live, a digital twin of the systems and subsystems is being developed as part of the AI4FM project.

A further aim of the AI4FM project is to develop a fault detection system based on artificial intelligence (AI) as an extension and further development of the system based on expert rules. The AI methods will be trained to recognize faults and anomalies without hand-crafted rules, where meaningful representation of the data will be learned by the training of neural networks.

Methods

The developed digital twin will contain a building HVAC system simulation tool that allows the rule set to be tested in advance by simulating various anomaly cases, thus enabling the precise development of accurate and complex fault detection rules.

The existing fault detection rules will be extended with the help of the new digital twins. Alternatively, the training of AI methods will rely on the Vienna Airport's existing building management data pool, which enables the use of modern data mining methods to improve the existing technical building equipment anomaly detection system.

The cleaning and preparation of the existing dataset will be the essential basis for the subsequent machine learning. Data-driven approaches will be applied to the airport database and tested to train AI models to automatically detect (and classify) anomalies and failures.

Expected results

The methods developed within the project are intended to improve the fault detection accuracy, quickly find malfunctions and inefficiencies in the heating/cooling system and detect faulty components in real time. The methods developed have the potential to make a significant contribution to increasing efficiency and sustainability in the operation of large buildings such as those at Vienna Airport.

Project Partners

Project management

AIT Austrian Institute of Technology GmbH

Project or cooperation partners

Flughafen Wien Aktiengesellschaft

Contact Address

Dr. Adam Buruzs
Giefinggasse 4
1210 Vienna
Tel.: +43 (664) 889 043 16
E-mail: adam.buruzs@ait.ac.at
Web: www.ait.ac.at