Fault Detection and Diagnosis as a Predictor of Energy Consumption in Special-Purpose Buildings

Faults in building equipment generally result from inefficient operation and poor preventative maintenance, resulting in higher energy consumption, shorter equipment lifetime, and an uncomfortable indoor environment. Thus, improving maintenance procedures is one of the essential aspects of reducing the building energy consumption gap and enhancing occupant satisfaction. In this context, the applications of fault detection and diagnosis (FDD) approaches have drawn the attention of scientists through utilising operational data and applying artificial intelligence (AI) to develop sophisticated algorithms to implement the FDD work. The data-driven FDD techniques depend primarily on operational data and do not demand a thorough knowledge of the system’s background; however, a significant amount of data is needed. Thus, this study proposes a data-driven FDD strategy for recognising various sorts of building system faults by developing a baseline model that can be utilised to monitor building energy performance. As a result, an ensemble forecasting approach (XGBoost) was used at the whole-building scale in a public building in a hot and humid region. The XGBoost exhibited an outstanding ability to forecast energy usage with a relatively small RMSE (around 0.33) as well as excellent fault detection accuracy for most fault scenarios even with low fault severity, with an average accuracy of 85%, with one notable exception of the dirty filter fault, which was identified with an average detection accuracy of 73%.