Global Robust Optimization of Computationally Expensive Systems

Global Robust Optimization of Computationally Expensive Systems: A Lavel Rotor Suspended by Fluid Film Bearings

Immerzeel, F.C.

van Ostayen, R.A.J. (mentor)
Langelaar, M. (mentor)
Eling, R.P.T. (mentor)

Mechanical, Maritime and Materials Engineering

Precision and Microsystems Engineering (PME)

Engineering Mechanics

2016-08-16

Kriging based methods enable the deterministic and robust optimization of computationally expensive systems. With a limited amount of function evaluations the optima are found in an iterative process with expected improvement as infill sampling criteria. Outputs of computer models can be stochastic and/or the models do not always succeed to perform the analysis. For the latter case, a problem is said to be affect by a hidden constraint. Regression Kriging can be included in the methods to deal with the stochastic model outputs. A new method is introduced to handle the hidden constraint in both deterministic and robust optimization. The methods are used to optimize a validated model of a Laval rotor suspended by plain journal bearings. To capture the non-linear behaviour of the self-excited vibrations, a computationally expensive time-transient run-up analysis needs to be performed. The output of this model is stochastic and the model fails to perform a run-up for some combinations of model inputs. The most influential control variables and uncertainties are indicated with an efficient global sensitivity study and used to optimize the system. With the extension of the deterministic and robust optimization, the optima are successfully obtained and can be compared.

robust optimization
efficient global optimization
metamodel
Kriging
hidden constraint
Laval rotor
hydrodynamic bearing
whirl
whip

http://resolver.tudelft.nl/uuid:e2c73349-91f6-4db7-9f24-cd90f4cbf167

Student theses

master thesis

(c) 2016 Immerzeel, F.C.