Sensitivity-enhanced generalized polynomial chaos for efficient uncertainty quantification

Abstract

We present an enriched formulation of the Least Squares (LSQ) regression method for Uncertainty Quantification (UQ) using generalised polynomial chaos (gPC). More specifically, we enrich the linear system with additional equations for the gradient (or sensitivity) of the Quantity of Interest with respect to the stochastic variables. This sensitivity is computed very efficiently for all variables by solving an adjoint system of equations at each sampling point of the stochastic space. The associated computational cost is similar to one solution of the direct problem. For the selection of the sampling points, we apply a greedy algorithm which is based on the pivoted QR decomposition of the measurement matrix. We call the new approach sensitivity-enhanced generalised polynomial chaos, or se-gPC. We apply the method to several test cases to test accuracy and convergence with increasing chaos order, including an aerodynamic case with $40$ stochastic parameters. The method is found to produce accurate estimations of the statistical moments using the minimum number of sampling points. The computational cost scales as $\sim m^{p-1}$, instead of $\sim m^p$ of the standard LSQ formulation, where $m$ is the number of stochastic variables and $p$ the chaos order. The solution of the adjoint system of equations is implemented in many computational mechanics packages, thus the infrastructure exists for the application of the method to a wide variety of engineering problems.

References in corpus (10)

• Data-Driven Sparse Sensor Placement for Reconstruction
• An Introduction to Adjoint Problems
• Compressive Sampling of Polynomial Chaos Expansions: Convergence Analysis and Sampling Strategies
• Sparse Pseudospectral Approximation Method
• Enhancing ${\ell }_1$-minimization estimates of polynomial chaos expansions using basis selection
• Least Squares Polynomial Chaos Expansion: A Review of Sampling Strategies
• Coherence Motivated Sampling and Convergence Analysis of Least-Squares Polynomial Chaos Regression
• Sparse Polynomial Chaos Expansions via Compressed Sensing and D-optimal Design
• On Polynomial Chaos Expansion via Gradient-enhanced ${\ell }_1$-minimization
• A gradient enhanced ${\ell }_1$-minimization for sparse approximation of polynomial chaos expansions

Cited by in corpus (2)

• Sobolev neural network with residual weighting as a surrogate in linear and non-linear mechanics
• Variational Bayesian surrogate modelling with application to robust design optimisation