Using Design of Experiments to model the effect of uncertainties in substructure coupling

In this paper, the effect of component variability (due to dimensional tolerances) on the dynamics of an assembled structure is modeled using procedures derived from Design of Experiments (DOE). Specifically, the possibilities offered by factorial design, in order to identify a regression model of the effect of uncertainties and of their interactions, are explored. Of course, the number of numerical experiments, required to fit a regression model, is much less than the number of realisations required for the implementation of Monte Carlo simulation, presented in the companion paper [1]. The regression model can then be used instead of the physical model to evaluate the dynamic behaviour of the assembled structure. The procedure is verified by comparing the output of the regression model with results of the physical model. Furthermore, the percent contributions of different uncertainties are evaluated, allowing to select which tolerance fields should be narrowed first in order to reduce the dynamic variability of the assembled structure.