Smart thermostat, Residential HVAC system, Fault detection and diagnosis
Fault detection and diagnosis (FDD) using aggregated smart thermostat data is a relatively new research field, but one with immediate practical application to residential indoor climate control. This paper analyzes a cloud-based dataset which contains thermostat history records of nearly 370,000 distinct residential HVAC systems in the U.S. The large, diverse, and growing dataset enables novel methods for detecting and diagnosing faults on systems with limited sensor data. This paper proposes a statistics-based FDD method for non-variable speed heat pump and air conditioning units, and demonstrates the effectiveness with several case studies. The proposed method identifies systems within a similar climate region, and then segments and classifies the time series data based on operational mode and behavior. Various data features are then extracted from the time series segments to identify systems that exhibit poor transient behavior. Additional features are used to refine and classify the problem severity. Statistical methods are then used to compare system performance to the entire population and identify outlier behavior due to operational faults that affect system efficiency and occupancy comfort. The resulting algorithm demonstrates the potential of big data fault detection for air conditioning systems using limited cloud-based sensor information.