Conceptual implementation of the variance-based sensitivity analysis for the calculation of the first-order effects

Studies focused on models and their associated parameters are of crucial importance to engineers since models are the essential tools for engineering analysis and design. Uncertainty and successively sensitivity analysis are integral parts of such studies. Among the several methods available for performing sensitivity analysis, approaches relying on the variance-decomposition concept are well established and widely used, though they are mostly expensive in terms of the computational costs. In this study an efficient conceptual implementation for the estimation of Sobol's first-order effects is proposed and controlled through analytical and numerical benchmark problems. The competency of the implementation was tested against other approaches including EFAST. The conceptual implementation proved to provide a reliable approach for the calculation of the first-order sensitivity indices. It particularly performed stably and accurately in the case of large models with limited available data, which account for a large portion of the engineering problems. Moreover, it does not require any specifically generated sample sets or on-demand function calls and therefore is applicable to experimental data.