Numerical optimization methods comparison based on the CFD conduction-convection heat transfer case

In some types of industries, the best possible design, in terms of thermal and flow performance, determines the success or failure of the company. This applies, among others, to the sectional elements that form doors, windows or prefabricated building walls. The purpose of this paper is to show the possibilities and limitations of the different response surface methods for optimization in the case where natural convection phenomenon appears inside the sectional structures.

A three-layered wall with air gap is used as a cross-section heat flow model. Response surface algorithms for optimization, which can be found in commercial software, e.g. ANSYS/WORKBENCH, can help to optimize geometrical structure of components to achieve bigger or smaller heat flux value. In this paper, the optimization methodology of the design of experiments (DOE) and different response surface (RS) methods are used.

Optimal results obtained with the use of genetic aggregation, standard RS, Kriging, non-parametric regression and neural network methods are compared with direct CFD and analytical calculations. Different limitations and advantages of the RS methods make individual methods more appropriate for different issues. For a properly defined optimization problem, the heat flux value approximated for the optimal geometry agrees with the direct CFD simulations.

The presented investigations show how to use response surface optimization methods for thermal optimization of the sectional elements and their applications to obtain reliable results.

This paper presents the value of the use of RS methods in CFD-based geometry optimization. The study also shows that the RS optimization methods can approximate thermal properties under natural convection development conditions.