A fault diagnosis system for heat pumps

Abstract

During the operation of heat pumps, faults like heat exchanger fouling, component failure, or refrigerant leakage reduce the system performance. In order to recognize these faults early, a fault diagnosis system has been developed and verified on a test bench. The parameters of a heat pump model are identified sequentially and classified during operation. For this classification, several 'hard' and 'soft' clustering methods have been investigated, while fuzzy inference systems or neural networks are created automatically by newly developed software. Choosing a simple black-box model structure, the number of sensors can be minimized, whereas a more advanced grey-box model yields better classification results.