Assessment of Surrogate Models for Quantification of Uncertainty in Laminar Flame Speed

Chemically reacting flows suffer from uncertainties associated with operating conditions, chemistry, turbulence, and turbulence-chemistry interactions. Such uncertainties pose challenges to the numerical investigation of such flows in terms of reliable predictive capabilities. Uncertainty quantification (UQ) is an effective computational strategy to quantify such uncertainties. In this study, a non-intrusive forward UQ strategy is considered to examine the accuracy and efficiency of different modeling techniques to quantify uncertainties associated with laminar premixed flame when reliable data is available. The UQ study can be performed either using direct or surrogate modeling techniques. While direct modeling technique such as Monte Carlo sampling is popular and accurate, it tends to be computationally prohibitive for the investigation of chemically reacting flows. To this end, surrogate modeling techniques offer a computationally tractable approach. In this study, three popular surrogate modeling techniques, namely, polynomial chaos expansion (PCE), stochastic collocation (SC), and Gaussian process (GP) are assessed for their capabilities for the investigation of freely propagating laminar premixed flame. A globally reduced one-step and five-species chemical mechanism is considered where the pre-exponential factor in the Arrhenius kinetics is the uncertain parameter and the laminar flame speed is the output of interest.