Large Scale Data Analytics for Fault Detection and Diagnosis of Residential HVAC Systems
Abstract
Residential heating, ventilation, and air conditioning (HVAC) equipment maintains the indoor environment with appropriate temperature and humidity levels. Meanwhile, it accounts for 51.3% of annual energy use and 40.1% of annual energy expenditures in the residential buildings in the U.S. However, residential HVAC systems often suffer from installation faults and operational faults leading to degradation in system capacity or even complete breakdowns, causing extra energy consumption and occupant discomfort. Fault detection and diagnosis (FDD) methods assist in identifying specific system faults, predicting gradual degradation and prompting necessary maintenance. Though plenty of researches have been conducted to develop FDD methods for commercial HVAC systems, relatively few researches focus on residential systems, mainly because FDD requires installation of additional sensors on each HVAC equipment, which is not cost-effective for the mass-produced residential systems.
This research fills this gap by developing statistics-based FDD methods to identify faults and monitor behavior changes simultaneously from a large number of residential HVAC systems using smart thermostat data. Two main approaches for fault detection and preliminary diagnosis are proposed in this research, namely: comparing operational features between multiple systems and monitoring the changes of operational features within each system. Following the idea of each approach, a few useful FDD algorithms are developed, including the setpoint tracking failure detector, inadequate capacity detector, control problem detector, and degradation trend detector. Additionally, the research provides general preprocessing procedures for the smart thermostat data, which could be applied to all fault detectors. The preprocessing procedures ensure the data is clean and critical features are extracted that representative of the operational conditions of each system.
The main body of this thesis presents each of the proposed detector. The setpoint tracking failure detector identifies degraded systems that cannot effectively regulate the indoor temperature around the desired setpoints. The inadequate capacity detector identifies systems with much lower cooling/heating capacity compared to other systems in the similar climate region, and in majority of the time the degradation of system capacity is imperceptible for home occupants. The control problem detector identifies systems with abnormally high cycle frequency and setpoint error in a large population, which is usually caused by control faults. Lastly, the degradation trend detector is able to detect slow system capacity degradation over time and quantify the magnitude of the degradation.
Finally, the author proposes a few future research directions of FDD for residential HVAC systems. Possible research directions include (1) improving the performance of fault detectors through verified in situ faulty systems, (2) developing deep learning models such as the recurrent neural network and the Siamese network, and (3) incorporating additional features from limited numbers of low-cost sensors.
Subject
Fault Detection and DiagnosisSmart Thermostat
Residential HVAC System
Performance Degradation
Trend Detection
Citation
Guo, Fangzhou (2021). Large Scale Data Analytics for Fault Detection and Diagnosis of Residential HVAC Systems. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /195267.