Detecting Soft Faults in Heat Pumps

Heat pumps are critical for energy-efficient heating and cooling, but their performance can be compromised by soft faults like condenser silting. It is vital to detect and fix such faults early in order to ensure optimal performance and longevity of heat pump systems, and consequently optimize the positive effect of heat pumps to our environment. In this paper, we tackle the problem of early fault detection and propose a supervised machine learning approach that detects soft faults. In particular, we used a random forest approach to learn the regular behavior of heat pumps. We detect faults via comparing the expected behavior obtained from the learned model with the current behavior. In addition to the description of the used methodology, we provide and discuss the results obtained from an experimental study that is based on synthetic data of two different heat pumps.