mAIntenance - Investigation of AI supported maintenance and energy management

The building sector plays a significant role in achieving the Paris climate goals, as it is responsible for almost a third of global CO2 emissions. Advanced predictive control approaches have attracted a great deal of attention in the field of research in recent years and are proving to be a promising solution for increasing building efficiency. For example, data-driven load prediction models for individual building energy systems can utilize easily accessible monitoring data with the increase of low-cost installed IoT sensor technology.

The cooperative R&D project mAIntenance investigated how the use of artificial intelligence can make technical building supply systems more efficient and reliable. The aim was to evaluate sensor data collected in the building using machine learning and make the findings derived from the new data available to the FM via a dashboard. To this end, use cases were defined in the area of energy and maintenance management.

FUTUREbase was selected as the test environment for data acquisition, system monitoring, model validation and functional evaluation of the use cases. This is a four-storey research and office building at the Vienna site. In addition to the existing system monitoring, an IoT sensor network was set up to collect relevant information regarding the office climate. The mapping of a BRICK-based data model of the building topology, supply systems and their data points made it possible to process semantic knowledge in a machine-readable way.

By combining the collected data sets and the semantic data model, initial predictions were made regarding building energy consumption, thermal room comfort and fault detection within the building automation system. The generated data-based machine learning models were first trained with the data recorded and processed over several years and then validated. Furthermore, functional evaluations were carried out with regard to the prediction of thermal comfort in offices in an HVAC zone and the detected anomalies.

With regard to the prediction of building energy consumption, good results were achieved due to the availability of long, high-quality data time series. The predictions of thermal comfort in the offices proved to be less accurate. However, the evaluation with measurement series for thermal comfort considerations resulted in sufficiently accurate matches to estimate comfort violations in offices for a period of up to 3 days for facility management. With regard to fault detection, various fault classifications were carried out with the help of autoencoders by analyzing the anomalies identified. The approach of using the brick data model to select data points for the specification of individual autoencoders proved to be powerful.

A profitability analysis showed a payback period of around 2.7 years for the AI tool, leading to an increase in energy efficiency in the long term. The application of artificial intelligence in building management offers great market potential, especially for facility managers and industrial companies that can benefit from the optimization of energy consumption and anomaly detection.

Project management

Austrian Institute of Technology GmbH – Center for Energy

Project or cooperation partners

PKE Facility Management GmbH

Dipl.-Ing. Michael Schöny, BSc.
Giefinggasse 2
A-1210 Vienna
Tel.: +43 (664) 883 355 45
E-mail: michael.schoeny@ait.ac.at
Web: www.ait.ac.at/ueber-das-ait/center/center-for-energy