An Information-theoretical Framework for Data-driven Building Automatic Fault Detection and Diagnosis Support

Buildings are complex and integrated systems consisting of multiple sensors, sub-systems, and automatically controlled components, among which the heating, ventilation, and air conditioning (HVAC) systems are critical for building energy consumption and indoor environment quality. HVAC systems usually suffer from faults, such as malfunctioning sensors, equipment, and control systems, which significantly affect a building’s performance. Automatic fault detection and diagnosis (AFDD) tools have shown great potential in improving building energy efficiency and indoor environment quality. With the rapid advancement of internet-of-things (IoT) and sensor techniques, data-driven AFDD approach has drawn increased attentions due to its ease of implementation and low costs while achieving high fault detection and diagnosis performances. While promising, there are certain challenges, including reliable baseline constructions, issues of the use of building simulation for fault detection, and efficient fault-to-symptom root cause analyses, that hinder the performances of data-driven AFDD in a real, whole building system. The goal of this dissertation is to develop an information-theoretical framework for data-driven building automatic fault detection and diagnosis support. The first phase is to propose a decision-making metric, termed Eigen-Entropy (EE), to support the baseline construction for building AFDD by sampling from historical normal datasets. The second phase is to extend the use of EE to extract features from graph-structured data obtained from simulation building data to realize cross-datasets (simulation-to-real) fault detection improvements. The third phase is to utilize EE with causal inference to construct Bayesian Networks by determining the fault-to-symptoms relationships in the building systems. Experiments have shown the proposed information-theoretical framework has satisfactory capabilities, efficiency and efficacy for building AFDD.