Samenvatting

Anomalies in Heating, Ventilation, and Air Conditioning (HVAC) systems of buildings can lead to performance degradation, energy inefficiencies, and occupant discomfort. Automated anomaly detection is essential to prevent these issues. Machine learning based anomaly detection techniques are effective in identifying such faults, particularly when leveraging historical sensor data from HVAC systems. In this study, supervised machine learning models, including Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) networks, were developed to detect anomalies in air pressure within Air Handling Units (AHUs). These models were trained on a 14-month dataset comprising operational measurements such as fan speed, damper positions, and room set points, which significantly impact ventilation in HVAC systems and the air pressure operation. Given the absence of labeled fault data, synthetic fault injection was employed to simulate operational failures, enabling effective model training and evaluation. The LSTM model successfully detected 22 out of 23 fault events in the test set with zero false positives. The approach emphasizes energy savings, system reliability, and proactive fault management in HVAC systems, addressing the growing demand for efficient and sustainable building operations.