pymooCFD - A Multi-Objective Optimization Framework for CFD

Modern computational resource have solidified the use of computer modeling as an integral part of the engineering design process. This is particularly impressive when it comes to high-dimensional models such as computational fluid dynamics (CFD) models. CFD models are now capable of producing results with a level of confidence that would previously have required physical experimentation. Simultaneously, the development of machine learning techniques and algorithms has increased exponentially in recent years. This acceleration is also due to the widespread availability of modern computational resources. Thus far, the cross-over between these fields has been mostly focused on computer models with low computational costs. However, this is slowly changing through the continued rapid development of both fields. The pymooCFD platform seeks to unite these fields of study by connecting a state-of-the-art library of optimization algorithms with industry leading CFD solvers. To begin with, this platform is important for testing the effectiveness of applying new optimization algorithms to CFD. Additionally, machine learning has been shown to help improve CFD models; this platform could serve to facilitate the development of better CFD models. In this paper, the pymooCFD platform is applied to three different optimization problems. First, for validation purposes, the platform was used to conduct a well documented optimization problem, the reduction of drag around a circular cylinder through oscillating rotation around it\u27s central axis. Second, the platform was applied to a Large Eddy Simulation (LES) to Reynolds-Average Navier-Stokes (RANS) model simplification. Lastly, the platform was applied to optimizing the direction, power and location of portable air purifiers in a room with six computer simulated persons (CSPs). The results show that the pymooCFD is a powerful tool for applying optimization algorithms to CFD. The validation of the platform was successful. A novel approach to CFD model simplification, called boundary condition calibration, is proposed. Finally, conclusions were drawn about optimal configuration of portable air purifiers within indoor spaces. These conclusions should serve to inform the experiments need to draw qualitative conclusion and create health advisories. Code Repository: https://github.com/gmclove/pymooCF