The fusion of flow field data with multiple fidelities

We propose a spatial-temporal multi-fidelity Gaussian process regression framework for the fusion of flow field data with various availabilities and fidelities but not sufficiently large to train neural networks commonly encountered in fluid mechanics studies. For example, experiments lead high fidelity sparse time space, while most numerical are generally regarded as less accurate spatially temporally continuous. The proposed aims at generating new set fused by combining merits those space. Numerical simulations [e.g., direct simulation (DNS), eddy simulation, Reynolds-averaged Navier–Stokes] around National Advisory Committee Aeronautics 0012 airfoil performed collect original raw fidelities, fraction DNS result is used mimic high-fidelity experimental data. It found that accuracy increases density points until reaching threshold, above which becomes insensitive. This limit can be overcome introducing extra dimensions, such gradients low-fidelity field. By examining error fields, it tune fields only within limited local region. firmly improved more or higher levels spatial if captures temporal development.