The aim of this exercise is to introduce participants to the aerodynamic shape optimization that they can later use together with both steady PANS simulations but also with LES, PANS, DES or URANS simulations of vehicle flows. During the last day of the course we will discuss different approaches for shape optimization including automatic shape optimization and surrogate model based optimization and how these can be used together with steady and unsteady CFD simulations.
One way of making the shape optimization of the vehicle feasible is to use the surrogate model based optimization to estimate the response of CFD simulation. In this exercise we are going to use response surface approximation (RSA) for multi-objective optimization of vehicle aerodynamics. The aim of the task is to find an optimal aerodynamic shape of a generic three-dimensional train and a vortex generator used for passive flow control on vehicles. The two cases are taken from our publication . Although there are commercial packages available for surrogate model based optimization, their application for shape optimization of vehicles is not straightforward and there are many different steps in the process where the user must have deep understanding of what is happening behind the buttons of the software. This is in particular important when unsteady simulations such as LES or DES are used. The results from such unsteady simulations of vehicle flows can easily result in noisy data from the exploration of design domain which requires good skills of the user in construction of surrogate models, statistical analysis and the optimization procedure.
During this exercise the participants will learn different steps in the aerodynamic shape optimization:
- How to make geometric parameterisation of the model.
- How to make design of experiments (DOE).
- How to perform the CFD simulations in design points.
- How to choose among different surrogate models and how to construct them.
- How to perform search for the optimal design (the optimization process).
MATLAB codes are prepared for the participants for every step of the optimization process. For example, we will use the results of our CFD simulations to construct different response surface models such as polynomial models or neural networks. The participants will learn how to improve the surrogate models and how to obtain the best fit with the CFD data but also the best model for the optimization purpose.
An aerodynamic shape optimization is always a multi-objective optimization procedure. Participants will learn how to perform optimization using genetic algorithms and find so called Pareto optimal solutions. Using Pareto optimal solutions they will be able to make trade-offs between different objective functions such as drag, cross-wind stability, aero-acoustic properties of vehicles etc.
 Krajnovic, S.: Shape optimization of high-speed trains for improved aerodynamic performance. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, 223 (5) pp. 439-452.