Experimental investigation and validation on an air-source heat pump frosting state recognition method based on fan current fluctuation signal and machine learning

Abstract

Thick frost on the evaporators of air-source heat pumps and refrigeration systems can degrade system performance. Defrosting in time based on precise frosting state recognition method is of great importance. Indirect measuring recognition method with low accuracy is widely used in commercial systems, for cheapness and simpleness, needing improvement urgently. While accurate direct measuring methods are usually expensive and complicated. Therefore, a novel frosting state recognition method with high accuracy is proposed, which can be realized with a simple current sensor. The method is based on the micro fluctuation in the evaporator fan current (rather than current amplitude) caused by the perturbed air due to frost. An experimental setup is built to obtain fan current samples. Combining three feature extraction approaches and three classifiers, four frost state recognition methods using fan current fluctuation are merged. They are compared and studied based on the experimental samples. Results show original signal + 1D-CNN method has the best identification performance, reaching 95.74 ± 1.73 % accuracy at −10 °C evaporator air temperature. It reveals 94.53 ± 1.06 % accuracy in a temperature range of −5∼-20 °C, and 94.73 ± 1.00 % accuracy for another fan with the same model. The method has significant improvement in performance and good potential for further study and application.