Sparse Sampling and Fusion of Multi-Fidelity Aerodynamic Heating Data

To reduce the acquisition cost of aerodynamic heating data for high-speed aircraft by leveraging diverse data sources, this paper proposes a multi-source data fusion method based on sparse sampling and multi-fidelity neural network. First, a large volume of low-fidelity data is generated using low-cost engineering calculation methods along the flight trajectory, and sparse sampling is performed according to the data variation trends. Then, CFD numerical simulation is conducted on the sampled flight parameters to obtain a limited set of high-fidelity data. Finally, high-fidelity data across the flight trajectory is estimated by combining high-fidelity and low-fidelity data through multi-fidelity neural network. The effectiveness of the fusion is verified using heat flux data at the nose tip of a certain aircraft under the simulated flight trajectory. Results indicate that the proposed fusion approach can achieve a prediction error of less than 4% while using only 5% of the CFD data points from the original trajectory, significantly reducing the CFD simulation workload and demonstrating the method's potential for engineering applications.