Deep learning approach for predicting the flow field and heat transfer of supercritical hydrocarbon fuels

Flow dynamics and heat transfer of supercritical hydrocarbon fuels have significant influences on the regeneratively cooled advanced aero-engines. In this paper, based on the principle of deep learning (DL), a deep neural network with a Multilayer Perceptron (MLP) model was developed to predict the flow field and heat transfer characteristics of supercritical hydrocarbon fuels. The network was trained by the calculation of Computation Fluid Dynamics (CFD). The analysis shows that the predicted values of bulk fuel temperature, fuel velocity and heat transfer coefficient by the trained DL model match well with the CFD simulations. The linear correlation coefficients (R) between the DL model predicted values and the CFD calculated results are all bigger than 0.99. Most of the relative errors are lower than 2 %. The consumed time of the DL model is of several orders (10⁶) less than the CFD simulation. The effects of inlet Reynolds numbers, pressure and heating rate on the flow field and heat transfer of supercritical hydrocarbon fuels inside the regenerative cooling channel have been successfully learned by the DL model. Therefore, the proposed deep learning model provides an efficient tool for the sensitivity analysis and design optimization of the regenerative cooling system.