Master's thesis Weiyu Rong
Scaling method for data-driven fault detection and interpretation of the operation of building energy systems based on expert knowledgeCopyright: EBC
A significant portion of energy consumption in both residential and non-residential buildings is contributed by building services, like space heating, ventilation and air conditioning (HVAC). Nonoptimal operation and faults in HVAC systems could cause uncomfortable indoor environment, malfunction of equipment and large amounts of energy waste. An effective solution for this problem is the automated fault detection and diagnostics (FDD) method. Modern building energy monitoring systems collecting extensive data enable deep insights into the energy performance and fault detection and diagnostics process.
In this thesis, a data-driven method based on expert knowledge for fault detection and interpretation of operation in the building energy system is proposed. This method is based on scalable Key Performance Indicators (KPI) derived from time series data of various data points in the building energy system. A fuzzy logic system is utilized in this method to represent expert knowledge and mimic human reasoning considering vagueness of a linguistic expression and uncertainty in real-life systems. This method is developed on aedifion IoT cloud platform, which provides historical data of the energy monitoring system in E.ON ERC Main building and an analytics framework for the implementation of this method. As the basis of this method, an expert knowledge base for heat pump systems containing fuzzy sets for KPIs and fuzzy rules is built up. The presented method was validated not only with the historical data of the ERC’s heat pump, but also with a dataset from the National Institute of Standards on Technology (NIST), that contains experimental data with seven imposed faults of a heat pump system. The results of the validation show that this method is able to detect and diagnose faults in heat pump systems and derive corresponding recommendations on action automatically. Finally, an auto-tuning of the fuzzy logic system and a rule learning process with the NIST dataset was also tested and shown a potential combination of the fuzzy logic system and other learning algorithms for the application of FDD in building energy systems.