Efficient sensitivity analysis of EM structures using NLPLS-based PCE

Abstract

This paper presents a Non-Linear Partial-Least-Squares-based Polynomial Chaos Expansion (NLPLS-based PCE) approach for high dimensional global sensitivity analysis. NLPLS-based PCE effectively reduces the system dimensionality using NLPLS and simultaneously extracts the statistical information on the same sample set, i.e. variance, using PCE. A post-processing step is applied to transform the NLPLS-based PCE surrogate to a standard PCE surrogate, described by the full set of system parameters. A variance-based global sensitivity analysis is then applied, quantifying each system parameter’s sensitivity as the partial influence on the total variance of the performance variable. This method is illustrated using a high-dimensional problem, namely, a 37-variable passive diplexer structure, requiring 30 analysis points to obtain a converged global sensitivity analysis. The diplexer is optimized for manufacturing and the fabricated diplexer performs as expected. Another 37-variable global sensitivity analysis is performed and compared to the original diplexer.