Surrogate based uncertainty quantification in pyrolysis modeling: A comparative analysis

In this work we analyze and compare the uncertainty propagation in a comprehensive pyrolysis model that captures an experimental setup using different approaches: namely polynomial chaos expansion (PCE), proper orthogonal decomposition with interpolation (PODI), Gaussian process regression (GPR) and quasi-Monte Carlo simulations. We start with inverse modeling to obtain the nominal parameter values for a set of kinetic and thermal parameters that devise the essential equations of the model. This is followed by a global sensitivity analysis to choose the most critical parameters for which the stochastic uncertainty is studied in the output over time. For the uncertainty, we observe the convergence behavior for mean, standard deviation, skewness and kurtosis. The statistical moments convergence analysis shows PCE to exhibit comparatively poor convergence behavior for standard deviation, and divergence for skewness and kurtosis. On the other hand, the computational effort required for GPR does not scale well in terms of accuracy with increased number of samples. Whereas, PODI is observed to converge faster while being comparatively less computationally intensive than other surrogate approaches. In conclusion, PODI poses as the most prominent approach among the aforementioned for uncertainty quantification in highly resolved pyrolysis and flame spread models.