LAUSR

Abstract In recent years, uncertainty appears in different aspects of physical simulations including probabilistic boundary, stochastic loading, and multiscale modeling. Stretching across engineering domains and applied mathematics, uncertainty quantification is a multi-disciplinary field which is an inseparable part of risk analysis. However, many real-world problems deal with large number of simulations or experiments. Considering the limited budget and time to perform all these efforts (specially for practitioners), an essential task is to reduce the computational cost in an uncertain environment. This paper proposes to use a matrix completion technique for reducing the overall computational cost of engineering systems when they are subjected to the simultaneous effects of aleatory and epistemic uncertainties with high dimensions. The proposed method is further improved using hidden information in the uncertain variables based on clustering techniques. Several parametric and Monte Carlo simulations were performed to demonstrate the accuracy of our method with different compression ratios. Experimental results show a decent overall performance of our technique for high-dimensional hybrid uncertain systems.