Uncertainty Quantification of Structural Systems with Subset of Data

TL;DR

This paper introduces a hybrid uncertainty quantification method for structural responses under variable ground motions, using matrix completion, machine learning, and regression to reduce computational effort while maintaining accuracy.

Contribution

It proposes a novel combination of sampling, machine learning, and regression techniques for efficient uncertainty quantification in structural response analysis.

Findings

Effective estimation of structural responses with limited simulations

Improved accuracy through unsupervised sampling technique

Regression model enhances response prediction

Abstract

Quantification of the impact of uncertainty in material properties as well as the input ground motion on structural responses is an important step in implementing a performance-based earthquake engineering (PBEE) framework. Among various sources of uncertainty, the variability in the input ground motions, a.k.a. record-to-record, greatly affects the assessment results. The objective of this paper is to quantify the uncertainty in structural response with hybrid uncertainty sources. In this paper, multiple matrix completion methods are proposed and applied on a case study structure. The matrix completion method is a means to estimate the analyses results for the entire set of input parameters by conducting analysis for only a small subset of analyses. The main algorithmic contributions of our proposed method are twofold. First, we develop a sampling technique for choosing a subset of representative simulations, which allows improving the accuracy of the estimated response. An unsupervised machine learning technique is used for this purpose. Next, the proposed matrix completion method for uncertainty quantification is further refined by incorporating a regression model that is trained on the available partial simulations. The regression model improves the initial sampling as it provides a rough estimation of the structural responses. Finally, the proposed algorithm is applied to a multi-degree-of-freedom system, and the structural responses (i.e., displacements and base shear) are estimated. Results show that the proposed algorithm can effectively estimate the response from a full set of nonlinear simulations by conducting analyses only on a small portion of the set.