Efficient global sensitivity analysis method for models of systems with functional outputs

A Global Sensitivity Analysis procedure is presented to efficiently rank influence strength of model parameters upon the model functional outputs. A modified Morris one-at-a-time algorithm is used for parameter sampling. Simulated model responses are transformed into global sensitivity functions using response discretization and a scalar Global Sensitivity Index. Two indices are tried - the reference mean of absolute value of elementary effects and an alternative index proposed here. Finally, aggregated sensitivity measures are formed, suitable for global ranking of model parameters. The two variants of the proposed procedure and a reference total sensitivity index based method are compared for a circuit model of an infra-red photo-detector front end. Accuracy advantage of the proposed index and performance advantage of the proposed procedure are shown.