based on a global optimization method Antonella Frigerio*, Elia Bon*, Giuseppe Bruno* 8th World Congress on Structural and Multidisciplinary Optimization Lisbona, Portogallo 1-5 Giugno 2009 * ERSE SpA Modal identification methodologies, based on the Finite Element Method (FEM), are useful to set up reliable mathematical models of civil structures, which are essential to conjugate structural safety, rehabilitation design and life extension. Even though great improvements in structural modelling have been obtained so far, it is well-known that a preliminary FE model is often a poor representation of the actual behaviour of a structure owing to several simplifying assumptions depending on the incompleteness of information generally at disposal of engineers. Considering that experimental vibration data, as well as modal shapes, could be well representative of how a structure behaves, the preliminary numerical model could be corrected by means of a FE model updating procedure, known as “identification process”: an objective function, containing discrepancies between the numerical and the experimental modal data, is minimized using optimization methods which modify and adjust a preselected set of physical parameters, such as structural stiffness. Given that the relation between the experimental modal data and the physical parameters is nonlinear, an iterative optimization process is performed. The major issue related to this type of identification processes is the choice of a robust and efficient optimization algorithm to solve the least squared problem of the optimization process, which can be ill- posed. In the last few years, the Coupled Local Minimizers method (CLM) has been proposed in literature; this is a global optimization method which sets up a cooperative search mechanism, based on several local optimization analyses, performed simultaneously, which are coupled by information exchange instead of running independently. This method combines the advantage of a relative fast convergence, peculiar of local gradient-based algorithms, with a parallel strategy and information exchange typical of global approaches such as the genetic algorithms. Recently, the CLM method has been implemented in the Finite Element code MIDA (Modal IDentification Analysis) as an alternative of the Levenberg-Marquardt algorithm, used so far. At each iteration MIDA performs a modal analysis for each research point located in the domain defined by the physical parameters which have to be indentified (mainly structural stiffness), and afterwards it applies the CLM method to reduce the error between the so computed modal data and the experimental target ones. In this paper details related to the implementation of the CLM method within the MIDA identification process will be given, and the obtained benefits will be presented with reference to a full scale validation test (i.e. a large dam). This test has been chosen on the basis of the quality of the available experimental modal data, derived from a dynamic campaign based on non-destructive vibration testing. The identified numerical model has been successfully used for diagnostic purposes to investigate possible structural damages.