Peer-benchmarking-based fault detection in a residential geothermal heat-pump fleet
The growing adoption of heat pumps for heating and cooling in the residential sector raises the critical issue of ensuring their operational reliability over time. Fault Detection and Diagnosis (FDD) methods can identify soft faults at an early stage, thereby preventing their evolution into critical failures or long-term performance losses. Among the various FDD approaches, process history based techniques are becoming increasingly popular due to their minimal requirement for prior system knowledge. However, these methods typically require large volumes of labelled operational data, which may be unavailable — especially in cases of limited historical records (e.g., new machines or restricted data storage) or when no preliminary laboratory testing has been conducted. This study proposes an unsupervised methodology based on peer-benchmarking to compare the performance of similar, though not strictly identical, heat pumps with the aim of identifying potential outliers. These outliers are then analysed to highlight true anomalies, indicative of soft faults. This methodology is tested on a real-world dataset of 199 residential geothermal heat pumps, with thermal capacities ranging from 8 to 14 kW, monitored over a one-year period. This study assesses the suitability of peer benchmarking via outlier detection for fault diagnosis. In the case study, the methodology detected five outliers, three of which were associated with confirmed heat pump faults. These findings demonstrate the capability of this approach to identify malfunctioning units in a heterogeneous, label-free heat pump fleets.