Robust Design Optimization


Real-world optimization problems often involve uncertainties. Not accounting for such uncertainties can lead to theoretical optima which are not at all optimal or practical in real-life. Robust optimization aims to find solutions which are optimal with respect to performance in the theoretical optimization model, but also good/stable with respect to variations caused by uncertainties or noise. Robust optimization is therefore aimed at finding optimal solutions that are also meaningful in practice.

The goal of this project is to find optimization methods that can deal with uncertainties and noise within optimization problems. Besides focussing on the theoretical backgrounds of the effects of noise and uncertainty in optimization problems, the aim is also to develop methods that can be applied in practice.

Two practical optimization problems are studied as test-cases within this research are 1) automated design of drug molecules 2) robus design of a drawbead process. In the automated design of drug molecules the assessment of the quality of candidate solutions is vaguely defined and methods are needed to still provide promising solutions, despite the uncertainty in the quality assessment. In the robust design of a drawbead process the aim is to design a process that yields proper parts, but which requires minimal material. Here it is very important to find solutions that are also stable variations of the material properties.

Figure 1: Four stages of the forming of the blank during the deep drawing process.