Errors and inefficiencies in building technology systems cause unnecessary CO₂ emissions and operating costs. The resources available to operators are often insufficient for continuous monitoring of building automation systems. An automated fault detection and diagnostics system offers the advantage of early detection of faults and inefficiencies in heating, ventilation, and air conditioning (HVAC) systems.
AI-Based Fault Detection for HVAC Systems
The goal of the project is to develop fault detection systems for HVAC systems based on Artificial Intelligence (AI). Project partner Flughafen Wien AG (FWAG) contributes an existing database of long-term monitoring data from their building automation systems to the project. These data will be used to train AI algorithms for anomaly and fault detection. Additionally, simulation models for HVAC systems will be created to enable gray-box modeling. Digital twins based on simulation models will allow for the testing of existing fault detection rules and the development of new ones.
Project Approach and Methodology
- Data Quality Management: Processing, visualization, and verification of the available datasets to ensure sufficient quality for AI algorithms; data labeling where necessary for supervised learning.
- Literature Review: Evaluation of AI methods based on deep learning, and selection of the most promising ones for preliminary testing and further implementation.
- Training of Algorithms: Training of fault detection algorithms using time series data from FWAG. Evaluation of the transferability of AI methods to similar systems.
- Development of Simulation Models: Creation of simulation models for the most common HVAC systems in FWAG buildings. These will be tested by FWAG to refine and improve existing rule-based fault detection algorithms.
- Synthetic Data Generation: Use of simulation models to generate synthetic data with simulated typical component faults, which can then be used to train additional AI models.
AIT's Role in the Project
AIT, as the scientific partner, plays a central role in the project, especially in terms of data preparation, development and validation of simulation models, as well as the testing, training, and evaluation of AI-based fault detection algorithms.
Innovative Integration of Simulation and AI
For a large number of buildings, manual fault detection in ventilation systems by operators is difficult to manage. Simple rule-based algorithms can help detect certain faults. The simulation of fault scenarios allows for the fine-tuning of detection rules as well as controller parameter settings. On the other hand, simulation models can be used to generate synthetic data for further machine learning models, which require large datasets for supervised learning. This combination of system simulation and data-driven algorithms offers innovative potential. If synthetic time series with various typical component faults can be generated, classification algorithms can be trained using these synthetic time series to differentiate between types of faults.
