Black-box calibration for complex-system simulation.

Area of Science:

• Engineering
• Computational Science
• Aerodynamics

Background:

• Predicting complex system outputs is crucial but challenging in science.
• Computationally intensive models and experiments limit parameter studies and optimization.

Purpose of the Study:

• To present Gaussian-process-based correlations for correcting simple computer models.
• To demonstrate efficient aerodynamic design using these correlations with sparse data.

Main Methods:

• Utilizing Gaussian-process correlations to calibrate simple computer models.
• Replacing computationally expensive physics-based codes with fast, statistics-based codes.
• Applying the method to two aerodynamic design examples.

Main Results:

• Successfully calibrated a 2D potential-flow solver for unmanned air vehicle wing flow.
• Optimized a racing car's rear wing using calibrated simulations accounting for overall car flow.

Conclusions:

• Gaussian-process correlations offer an efficient alternative to complex simulations and experiments.
• This approach facilitates effective aerodynamic design and optimization with limited data.