Using deep learning algorithm to predict flow boiling pressure drop under ultrasound fields in vertical mini-channels

Abstract

Two-phase flow in mini/micro-channels can meet the high heat dissipation requirements of many state-of-the-art cooling solutions. However, there is lack of accurate universal methods for predicting parameters like pressure drop in these configurations. In recent years, deep learning has been brought to wide attention that could be employed for data-driven estimation gas-liquid two-phase pressure drop in mini-channels. This paper put forward a new multi-scale convolution neural network (MSCNN) and time-frequency representation (TFR) model based on deep learning was proposed to predict pressure drop of gas-liquid two-phase flow in mini-channels under ultrasound. Time-series pressure drop fluctuation data of gas-liquid two-phase were transformed into useful TFRs data by wavelet transform. TFR could effectively reveal the pressure drop fluctuation of gas-liquid two-phase in mini-channel, but The TFRs data scale was the high dimensionality and consumes computer resources. Therefore, bilinear interpolation was used to reduce TFRs data scale and as input of deep learning model. Compared with the traditional convolutional neural network (CNN) model structure, MSCNN model structure has global and local information synchronization. The prominent features are helpful to predict flow boiling pressure drop in mini-channel and to the automatic learning of MSCNN. Experiments showed that the prediction performance of MSCNN model has been greatly improved than the traditional data-driven.