A semi-labelled dataset for fault detection in air handling units from a large-scale office

Abstract

Fault detection and diagnosis (FDD) in Air Handling Units (AHUs) ensure building functions such as energy efficiency and occupant comfort by quickly identifying and diagnosing faults. Combining deep learning with FDD has demonstrated high generalization ability in this field. To develop deep learning models, this research constructed a dataset sourced from real data collected from a large-scale office in South Korea. The raw AHU data were extracted from the Building Management System (BMS) at 1-h intervals, spanning from November 2023 to May 2024. The dataset was partially labeled by annotation experts, categorizing the data into six types: normal condition, supply fan fault, total heating pump fault, return air temperature sensor fault, supply air Temperature sensor fault, and valve position fault. Additionally, semi-supervised learning methods were applied as an application example using this constructed dataset. The main contributions of this dataset to the field are twofold. First, it represents a unique dataset sourced from the real operational data of a large-scale office, which is currently non-existent in this domain. Second, the dataset's expert labeling adds significant value by ensuring accurate fault classification. Therefore, we hope that this dataset will encourage the development of robust FDD techniques that are more suitable for real-world applications.