Unsupervised identification of zone-level anomalies in VAV terminal units utilizing autoencoders and PCA

• Published in: HVAC&R research Vol. 30; no. 10; pp. 1196 - 1216
• Main Authors: Parsaei, Arya, Gunay, Burak, O'Brien, William, Moromisato, Ricardo
• Format: Journal Article
• Language: English
• Published: Philadelphia Taylor & Francis 25-11-2024; Taylor & Francis Ltd
• Subjects: Air flow; Anomalies; Effectiveness; Fault detection; HVAC equipment; Position measurement; Principal components analysis; Synthetic data
• ISSN: 2374-4731; 2374-474X
• DOI: 10.1080/23744731.2024.2411161

The effective operation of HVAC systems is crucial to minimize energy inefficiencies and occupant discomfort. However, these systems can experience various problems, including hardware and software-related anomalies. In contrast to most existing fault detection and diagnostic approaches, which rely on simple rules and alarms, this study introduces novel unsupervised approaches for detecting zone anomalies in variable air volume (VAV) air handling units (AHUs). The methods utilize autoencoders (AE) and principal component analysis (PCA). To evaluate the effectiveness of the proposed methods, both a synthetic dataset and measured data from a 28-zone VAV AHU system were investigated. The proposed method successfully detected several zone temperature and airflow anomalies using the AE-based method, and several zone anomalies were also identified using the PCA-AE approach by considering four commonly available zone-level trend logs in VAV AHUs namely temperature, airflow, airflow set-point, and VAV terminal damper position. The findings demonstrated the great adaptability of the proposed methods in detecting a wide range of zone anomalies in any modern building equipped with VAV AHUs, giving operators valuable insights about the system and notifying them of potential faults at an early stage.