Machine Learning-Driven Aerodynamic Optimization of Bluff Body Vehicle Geometry

A methodology to optimize bluff body geometries and airfoils will be performed by coupling a Bayesian surrogate model trained with high-fidelity Computational Fluid Dynamics (CFD) data and low fidelity panel-based solver XFLR5, and an optimization framework known as OpenMDAO. The bluff body has a base geometric shape representing a rectangular prism with parameterized fillets on every edge. The task at hand is delivering a geometry that can minimize drag and an airfoil with maximized lift. The commercial code, STAR-CCM+, to perform the CFD analysis and generate the initial database, which will then be used to train a data driven surrogate model. Gaussian Process Regression (GPR) will act as our surrogate model and generate an objective function to be optimized using OpenMDAO and PySwarms. Paper is meant as a demonstration of a surrogate based optimization framework, with adaptive sampling to intelligently search our design space.