Parallel Multispecies Genetic Algorithm for Physics and Parameter Estimation in Structural Dynamics

Abstract

The distinction between model structure verification and model updating is logically clear. In the former, one adjusts the physics to better match experimental data; in the latter the physical parameters of the model are adjusted. The current state of the practice is to iterate back and forth between both problems until reasonable agreement with experimental data is obtained. In this work, the selection of model physics from a predefined set and the simultaneous updating of the corresponding physical parameters are investigated in an unified computational framework. A newly developed parallel (computational efficiency) multispecies (physics models) genetic algorithm is proposed and evaluated. The genetic algorithm “evolves” both model physics and parameters to improve the correlation of analytical models and corresponding test data. Two new genetic operators are developed. The first is a crossover function, which enables the sharing of the genes between two species. The second is a transmutation operator, which allows a species member to transform into a different species. The proposed algorithms are demonstrated and evaluated using three different “numerical” experiments.