Reconstruction of temperature field in nanofluid-filled annular receiver with fins using deep hybrid transformer-convolutional neural network

Abstract

This study proposes a deep Hybrid Transformer-Convolutional Neural Network (HTCNN) to reconstruct the temperature field in the nanofluid-filled Parabolic Trough Collector receivers. The proposed HTCNN is improved from U-net architecture by replacing all the skip-connections with residual blocks and embedding a Transformer module. The novel method tackles the challenges of high computational cost inherent in numerical simulation, and the constraints posed by limited data availability in data-driven approaches. The results observed that the HTCNN enables estimating the temperature field with high accuracy and remarkably fast speed; furthermore, it doesn't depend on the dataset scale: trained with only 18 data, the HTCNN reaches an average prediction accuracy higher than 99.92% on the test set and a prediction speed of more than 1000 times faster than numerical simulation. Our results show the impressive potential of applying the HTCNN for rapid and efficient 3D fin configuration optimization of nanofluid-filled receivers with limited data.