Uncertainty Quantification and Global Sensitivity Analysis of Radiated Susceptibility in Multiconductor Transmission Lines using Adaptive Sparse Polynomial Chaos Expansions

Yu Zhu, Ke Wang, Quanyi Yu, Dayong Wu, Yang Zhang, Tong ZhangThe Applied Computational Electromagnetics Society Journal (ACES) 2021 · DOI

Abstract

This study analyzes the uncertainties of the radiated susceptibility in multiconductor transmission lines (MTLs), and introduces an adaptive sparse polynomial chaos expansion combining hyperbolic truncation scheme with orthogonal matching pursuit method (AS-PCE (OMP)). This method is used as the basis to realize the uncertainty quantification (UQ) of radiated susceptibility and global sensitivity analysis (GSA) of input variables to output variables. GSA considers the influencing factors of the incident field and transmission-line geometric parameters. The global sensitivity indices of each input variable are calculated for varying impedance loads. The accuracy and efficiency of the proposed method are verified compared with the results of the polynomial chaos expansion based least angle regression method and Monte Carlo methods.

Introduction

Radiated susceptibility in multiconductor transmission lines (MTLs) is a critical concern for electromagnetic compatibility (EMC) in complex electronic systems, ranging from electric vehicles to aircraft. Real-world conditions introduce significant uncertainties in parameters such as incident field angles and geometric tolerances. Traditional Monte Carlo (MC) methods for quantifying these uncertainties are computationally prohibitive for rapid industrial design. This paper presents an Adaptive Sparse Polynomial Chaos Expansion using Orthogonal Matching Pursuit (AS-PCE (OMP)) to provide a highly efficient and accurate surrogate model for radiated susceptibility analysis, achieving a nearly 300x speedup over traditional simulations.

Methodology

The proposed framework utilizes Polynomial Chaos Expansion (PCE) to represent system outputs as a series of orthogonal polynomials, essentially creating a "digital twin" of the physics model. To manage the "curse of dimensionality," a hyperbolic truncation scheme is employed, focusing on low-order interactions between variables based on the sparsity-of-effects principle.

The Orthogonal Matching Pursuit (OMP) algorithm—a "greedy" search method—is then used to iteratively select only the most influential polynomial terms. The optimal polynomial degree is determined adaptively using the Leave-One-Out (LOO) error ($\epsilon_{LOO}$), ensuring high precision without over-engineering the model.

Simulation Setup

The study analyzes a (2+1) transmission-line system consisting of two wires over a ground plane. Nine input variables are assigned specific probability distributions, including incident field angles ($\theta, \psi, \eta$), field amplitude ($E_0$), and geometric parameters such as wire radius ($r$), heights ($h_1, h_2$), distance ($d$), and length ($L$). Training data is generated using Latin Hypercube Sampling (LHS), with 600 samples found to be optimal for balancing speed and accuracy.

Configuration of the (2+1) multiconductor transmission line system and incident field parameters.

Performance and Accuracy

The AS-PCE (OMP) method demonstrates exceptional efficiency, completing the analysis in 3.73 seconds, compared to 1037.77 seconds for a standard 10,000-run Monte Carlo simulation. Furthermore, the method achieves significantly higher numerical precision than the competing Least Angle Regression (LARS) approach, with error floors reaching $10^{-18}$ compared to $10^{-1}$ for LARS in certain tests. This accuracy is maintained across different load conditions, as shown in the probability density mappings of induced currents.

Probability density functions of the induced current for low (50 Ω) and high (10 kΩ) impedance loads, comparing AS-PCE(OMP), MC, and AS-PCE(LARS).

Global Sensitivity Analysis (GSA)

GSA reveals that the dominant factors influencing susceptibility vary significantly with load impedance:

  • Low-impedance systems (50 Ω): The orientation of the incoming electromagnetic field (angles $\psi$ and $\eta$) is the dominant factor in noise pickup.
  • High-impedance systems (10 kΩ): The physical length ($L$) of the wire becomes the most critical variable, especially as frequencies increase.

These insights allow engineers to identify which physical tolerances (like wire height or incident angle) must be tightened to ensure electromagnetic compatibility.

Conclusion

The AS-PCE (OMP) method provides a robust and efficient tool for UQ and GSA in MTL systems. While the current study focuses on a (2+1) configuration up to 200 MHz, the framework establishes a foundation for analyzing larger cable harnesses and higher frequency ranges. By drastically reducing computational costs, this approach enables rapid design optimization under real-world uncertainty.

Summary updated

Abstract

This study analyzes the uncertainties of the radiated susceptibility in multiconductor transmission lines (MTLs), and introduces an adaptive sparse polynomial chaos expansion combining hyperbolic truncation scheme with orthogonal matching pursuit method (AS-PCE (OMP)). This method is used as the basis to realize the uncertainty quantification (UQ) of radiated susceptibility and global sensitivity analysis (GSA) of input variables to output variables. GSA considers the influencing factors of the incident field and transmission-line geometric parameters. The global sensitivity indices of each input variable are calculated for varying impedance loads. The accuracy and efficiency of the proposed method are verified compared with the results of the polynomial chaos expansion based least angle regression method and Monte Carlo methods.
BibTeX
@article{zhu2021uncertainty,
  author = {Yu Zhu and Ke Wang and Quanyi Yu and Dayong Wu and Yang Zhang and Tong Zhang},
  title = {{Uncertainty Quantification and Global Sensitivity Analysis of Radiated Susceptibility in Multiconductor Transmission Lines using Adaptive Sparse Polynomial Chaos Expansions}},
  journal = {The Applied Computational Electromagnetics Society Journal (ACES)},
  year = {2021},
  doi = {10.13052/2021.aces.j.361004},
}
Figure