A methodology for exploiting the tolerance for imprecision in genetic fuzzy systems and its application to characterization of rotor blade leading edge materials

A methodology for obtaining fuzzy rule-based models from uncertain data is proposed. The granularity of the linguistic discretization is decided with the help of a new estimation of the mutual information between ill-known random variables, and a combination of boosting and genetic algorithms is used for discovering new rules. This methodology has been applied to predict whether the coating of an helicopter rotor blade is adequate, considering the shear adhesion strength of ice to different materials. The discovered knowledge is intended to increase the level of post-processing interpretation accuracy of experimental data obtained during the evaluation of ice-phobic materials for rotorcraft applications.

► Imprecise probability-based bounds of the mutual information.
► Numerical algorithm for estimating the mutual information with ill-known data.
► Methodology for discovering expert knowledge from imprecise data.
► Rule-based model of shear adhesion strength of ice for rotorcraft applications.