Abstract : The main objective of the present research works is to propose a contribution in the elastodynamic domain by addressing the updating methods. The innovations proposed here lie in the formulation of strategies for the reanalysis and parametric identification problems, practicable for large industrial models.
First part : Approximate reanalysis algorithms
When design parameter values are modified, it is necessary to recalculate the output behavior (eigenvalues and modes) of the new model, but large industrial models preclude the exact reanalysis. Our strategy is based on the Rayleigh-Ritz method and includes the contribution of the static residual vectors. These terms improve the precision of the predicted eigensolutions.
Second part : Implementation of a forward estimation procedure for model updating based on genetic algorithms
We propose to adapt an evolutionary computation method to the parametric identification of large models. Given a cost function (eg model-structure distances), the procedure is able to search throughout the parameter space. Additional heuristic mechanisms are added to localize minima.
Third part : Proto–Dynamique software
This part has for objective to present the environment which allowed to develop the techniques and to perform the numerical tests. Proto is a Matlab application and is organized as an opened platform containing a various number of analysis tools and updating methods.