A high-fidelity numerical modelling of standing-wave thermoacoustic engine through Computational fluid dynamics (CFD) simulations offer accurate real-life phenomena representation, at the cost of high computational burden. Training an accurate surrogate model using CFD simulation results becomes challenging due to the time-consuming aspect in gathering sufficient training data. A computationally faster lower fidelity modeling can be achieved using Thermoviscous acoustic (TVA) simulations. However, TVA analysis does not account for the nonlinear effects induced by high pressure fields. This study aims to train a multifidelity surrogate model for a thermoacoustic engine by combining scarce high-fidelity CFD simulation results with abundant low-fidelity TVA simulation results using gaussian process model, and the improvement in prediction accuracy and total simulation computational cost are compared to conventional surrogate model trained with only CFD results.
RAHIM et al. (Wed,) studied this question.