Multi-fidelity surrogate modeling based on data extension using POD and ANN

Surrogate model of flow-field can fast provide flow-field information for various geometries in aircraft conceptual design. Due to the limited sample size of high-fidelity data, it is hard to achieve high prediction accuracy of the flow-field using single-fidelity surrogate models. In this paper, we develop a multi-fidelity modeling method of flow-field, of which the large amount of low-fidelity data can offer trends of the surrogate model, and the small size of high-fidelity data can calibrate model precision. However, flow-field data in various fidelity accumulated in historical designs could be heterogeneous. To handle this problem, we propose a data extension strategy using proper orthogonal decomposition (POD) and artificial neural networks (ANN) to transform the data into the same size. During the modeling, POD extracts low-dimensional features from high-dimensional flow-field data, while ANN maps low-fidelity data to corresponding high-fidelity data. The flow-field prediction of the NACA 4-digit airfoils is carried out to validate the proposed modeling method, and the results indicate that the modeling cost can be obviously reduced in comparison with the modeling purely by small size of high-fidelity data.