158 A Smart Sampling Strategy for One-at-a-Time Sensitivity Experiments (PSAM-0360)

In mathematical modeling, sensitivity analysis is the discipline that identifies which subset of the input parameters drives most of the uncertainty in the output under investigation. The reliability of a sensitivity analysis experiment is naturally conditioned on the good quality of the sample used to estimate the model response. One-At-a-Time (OAT) experiments constitute the simplex example of sensitivity analysis designs, where factors are moved One-At-a-Time, keeping all the other inputs fixed.

This work presents an improvement of the OAT design developed by Morris in 1991. The Morris sensitivity analysis experiment has shown its effectiveness in screening unimportant factors in complex models with a small number of model evaluations. However, the Morris sampling design might not explore efficiently the inputs' domain when the number of affordable model evaluations is extremely low.

We propose here an improved strategy that allows for a better searching of the space of the input factors, without increasing the number of model evaluations. The efficiency of the new design is tested and compared with the original strategy.