Prediction of Evaporating Temperature Based On Compressor Vibration Data Using Cnns and Autoencoders

Household and light-commercial refrigeration systems usually do not have embedded pressure transducers, so it is not possible to directly measure the suction pressure of the compressor, which holds relevant information for fault detection and diagnosis. This paper proposes a virtual sensor for the suction pressure, which is then used to estimate the evaporating temperature. The proposed method uses data on compressor vibration represented as spectrograms for time-frequency analysis done by convolutional neural networks and autoencoders, used for dimensionality reduction. Using a few seconds of acquisition, it was possible to infer the evaporating temperature up to a root mean square error of 2.7 ºC in an environment with unknown angular speed in an unknown compressor of the same model as the ones used in training.