Study on data universality and machine learning for methane transcritical flow and heat transfer in PCHE channel with different boundary conditions

In order to integrate data under various conditions to carry out machine learning, it is necessary to establish the principle of data universality. In this work, the methane transcritical flow and heat transfer in PCHE was numerically studied with three boundary conditions to propose the data universality principle. That is, when the local boundary parameters, flow state and thermal physical properties are similar, the local heat transfer coefficient and pressure drop will be equal. The machine learning methods of ANN and LightGBM models were employed to predict the local flow and heat transfer parameters in the channel. The results indicate that, under multiple conditions, the ANN model achieves a mean absolute percentage error (MAPE) less than 3 % and the R² over 0.999. The prediction on new cases are also highly consistent with the simulation results. The data universality principle laid the key foundation of machine learning prediction on various conditions.