Estimation of Acoustic Power in Work-Recovery Pulse Tube Cooler Based on Machine Learning

Abstract

The work-recovery pulse tube cooler (WRPTC) has an intrinsic higher efficiency than the orifice or inertance pulse tube coolers at high operating temperature. A novel approach to estimate the acoustic power of a given WRPTC using deep learning (DL) algorithm is developed in this paper. Four typical operating parameters, including working frequency, driving voltage, refrigerating temperature and cooling capacity, are selected as the inputs for the DL model. This model is trained by existing experimental data and to predict the reasonable acoustic power in the future as the operating conditions changes. More than 80% of the total experimental data are adopted to train the DL model, while the rest of those are adopted as testing set for validation of the predicting results. In the end, a mean relative error of 3.5% between the prediction and the experiments is observed for the estimation of the acoustic power in the WRPTC. It is worth mentioning that the DL model could be applicable to estimate acoustic power in different WRPTCs with distinct geometries and varying operating conditions, due to the adaptive ability of machine learning with specific training sets and the internal automatic optimization strategy.