Electrical Engineering and Computer Science

Defense Event

Uncertainty Quantification for Electromagnetic Analysis via Efficient Collocation Methods

Abdulkadir C. Yucel

 
Tuesday, April 16, 2013
2:00pm - 4:00pm
3316 EECS

Add to Google Calendar

About the Event

Electromagnetic (EM) devices and systems often are fraught by uncertainty in their geometry, configuration, and excitation. These uncertainties (often termed “random variables”) strongly and nonlinearly impact voltages and currents on mission-critical circuits or receivers (often termed “observables”). To ensure the functionality of such circuits or receivers, this dependency should be statistically characterized. In this thesis, efficient collocation methods for uncertainty quantification in EM analysis are presented. First, a Stroud-based stochastic collocation method is introduced to statistically characterize electromagnetic compatibility and interference (EMC/EMI) phenomena on electrically large and complex platforms. Second, a multi-element probabilistic collocation (ME-PC) method suitable for characterizing rapidly varying and/or discontinuous observables is presented. Its applications to the statistical characterization of EMC/EMI phenomena on electrically and complex platforms and transverse magnetic wave propagation in complex mine environments are demonstrated. In addition, the ME-PC method is applied to the statistical characterization of EM wave propagation in complex mine environments with the aid of a novel fast multipole method and fast Fourier transform-accelerated surface integral equation solver -- the first-ever full-wave solver capable of characterizing EM wave propagation in hundreds of wavelengths long mine tunnels. Finally, an iterative high-dimensional model representation technique is proposed to statistically characterize EMC/EMI observables that involve a large number of random variables. The application of this technique to the genetic algorithm based optimization of EM devices is presented as well.

Additional Information

Sponsor: Eric Michielssen

Open to: Limited