A Smart Data-Driven Fault Diagnosis Method for Sustainable and Healthy Building System Operations

Huge amounts of building operational data are being collected by the building automation system in modern buildings, providing an ideal platform for developing data-driven methods for building energy management. Compared with traditional human-centric building management methods, data-driven methods are more efficient and have attracted significant attention from academic researchers and industry professionals. The main challenge is how to effectively extract useful insights from massive building operational data, especially when the original data are noisy and of poor quality. This study proposes a data-driven diagnosis method for the sustainable and healthy operations of building services systems. The chiller system is selected as the research target, considering it generally consumes the most energy and has the greatest energy saving potential. The method is developed based on both unsupervised and supervised machine learning techniques. Firstly, a steady-state detection method has been developed using unsupervised clustering analysis. It aims to automatically remove transient operational data to ensure the quality of the follow-up data analysis. Secondly, various supervised machine learning techniques have been used to develop classification models to identify typical faults in chiller operations. Thirdly, the method has been validated using actual chiller operational data. Different metrics, such as accuracy, fault detection rate, misdiagnosis rate and false alarm rate, have been adopted for performance evaluation. The method can be applied to enhance the efficiency for practical building management. The research outcomes are beneficial for the development of sustainable and healthy building energy management.