An early prototype for fault detection and diagnosis of Air-Handling Units

Abstract

The built environment is responsible for nearly 35% of energy consumption and is undergoing a digital transformation. Up to 30% of energy is consumed inefficiently due to inadequate setup and/or incomplete utilization of available data. An efficient fault detection and diagnosis (FDD) strategy for air handling units is key to addressing this gap. Even though numerous FDD approaches have been published, real-world applications are far more complex and rarely discussed. This paper deals with FDD tool prototyping and integration aspects and discusses its development for air handling units deployed at 2 case-study buildings located in the Netherlands. The design and development of the FDD tool follows a structured 4 step process. Firstly, literature research is utilized to narrow the design space and establish a complete use case for developing the FDD tool. Secondly, the developed use case is handled utilizing a data-driven strategy to generate fault symptoms using a state-of-the-art extreme gradient boosting algorithm (XGBoost). Thirdly, the detected faults are isolated with a diagnostic Bayesian network. This way the fault detection and diagnosis aspects are separately handled. Lastly, integration of the prototyped tool with a commercially operated continuous monitoring system, currently being utilized to monitor 400 buildings, is discussed. Upon experimental validation, diagnosis specificity exceeding 90% is realized. It is further observed that the prototyped FDD tool could prevent up to 33% of chiller consumed energy. Moreover, the results presented will contribute to the adoption and deployment of AI-based FDD strategies in commercial applications.