ENBIS-16 in Sheffield
11 – 15 September 2016; Sheffield
Abstract submission: 20 March – 4 July 2016
Multicriterial Optimization of Real-World Applications and Computer Experiments
13 September 2016, 09:40 – 10:00
- Submitted by
- Nikolaus Rudak
- Nikolaus Rudak (Dortmund University of Applied Sciences), Sonja Kuhnt (Dortmund University of Applied Sciences)
- In real-world technical applications, as for example in a thermal spraying process, the quality of a product, represented by multiple properties, is usually affected by several machine parameters und some unaccounted noise. In order to understand the relationship between machine parameters and the properties of a product, the experimenter can conduct an appropriate experimental design and then fit a regression model for the unknown mean and variance of each property. Afterwards, one can find machine parameters, where the fitted mean of each product property is near a desired target value with minimal variance. We review some methods to tackle this problem, as for example by means of the JOP method (Rudak et al., 2015).
If available, computer simulations can emulate costly and complex real-world experiments. In turn, these computer experiments can be very time consuming, e.g. for the finite element method. Therefore, usually an easy to evaluate surrogate meta-model is build based on computer experiments. This meta-model might then be used for optimization with the well-known Efficient Global Optimization (EGO) algorithm. However, if further constraints have to be respected, where the constraint function is also an output of the computer simulation, one needs a modification of the EGO algorithm. We review some methods for constrained optimization of computer experiments, e.g. Durantin et al., 2016, and present results from an application in the field of radial impeller optimization.
Return to programme