Zeger's Ph.D. thesis is online available

2019/02/05

Zeger's Ph.D. thesis is online available

Simulation and robust optimization for electric devices with uncertainties

The dissertation deals with modeling, simulation and optimization of low-frequency electromagnetic devices and quantification of the impact of uncertainties on these devices. The emphasis of these methods is on their application for electric machines.

A Permanent Magnet Synchronous Machine (PMSM) is simulated using Iso-Geometric Analysis (IGA). An efficient modeling procedure has been established by incorporating a harmonic stator-rotor coupling. The procedure is found to be stable. Furthermore, it is found that there is strong reduction in computational time with respect to a classical monolithic finite element method. The properties of the ingredients of IGA, i.e. B-splines and Non-Uniform B-Splines, are exploited to conduct a shape optimization for the example of a Stern-Gerlach magnet. It is shown that the IGA framework is a reliable and promising tool for simulating and optimizing electric devices.

Different formulations for robust optimization are recalled. The formulations are tested for the optimization of the size of the permanent magnet in a PMSM. It is shown that under the application of linearization the deterministic and the stochastic formulation are equivalent. An efficient deterministic optimization algorithm is constructed by the implementation of an affine decomposition. It is shown that the deterministic algorithm outperforms the widely used stochastic algorithms for this application.

Finally, different models to incorporate uncertainties in the simulation of PMSMs are developed. They incorporate different types of rotor eccentricity, uncertainties in the permanent magnets (geometric and material related) and uncertainties that are introduced by the welding processes during the manufacturing. Their influences are studied using stochastic collocation and using the classical Monte Carlo method. Furthermore, the Multilevel Monte Carlo approach is combined with error estimation and applied to determine high dimensional uncertainties in a PMSM.

go to list