Modelling Uncertainty in Computational Electromagnetics

Andrew Austin

Sarris Group, Electromagnetics

Abstract:

The numerical simulation of Maxwell's equations is an integral part of the design process for many electromagnetic structures and devices. However, uncertainties---inherent in the description of the problem or introduced by manufacturing processes---are difficult to capture and characterize using existing computational methods. For example, manufacturing tolerances introduce uncertainty in the physical dimensions, which “propagates” through the structure or device to induce uncertainty in the response and outputs. Characterizing the randomness in the simulated response is an essential step in the design and validation process to estimate the sensitivity of the predictions and to set realistic design margins. Statistics computed via traditional approaches, such as the Monte Carlo method, generally converge slowly, which tends to limit their application for computationally large or complex problems.

This talk will focus on modelling uncertainty in computational electromagnetics (in particular, the FDTD method and ray-tracing) using generalized polynomial chaos (gPC). gPC approximates quantities in a stochastic process as the finite expansion of orthogonal basis polynomials in the random input parameter space, and can converge significantly faster than the Monte Carlo method.

Biography:

Andrew Austin received the B.E. (Hons.) and Ph.D. degrees in electrical and electronic engineering from the University of Auckland, in 2007 and 2012 respectively. He is currently a postdoctoral fellow in the electromagnetics group at the University of Toronto.