Experimental identification of a stochastic computational model using modal data - Application to an industrial built-up structure

We are interested in the construction of a stochastic computational model (SCM) of a dynamical structure representing a family of nominally-identical structures for which experimental modal data are available. The objective of this research consists in identifying this hyperparameters using experimental modal data (natural frequencies and mass-normalized mode shapes) and realizations of the SCM. If the experimental variability and the randomness in the SCM are high, mode crossing phenomena or mode veering phenomena may occur for the experimental modes and for the computed stochastic modes, yielding a possible mismatch between the computational modal quantities calculated using the SCM and the experimental modal data. The methodology proposed here introduces a random transformation of the computational observations (computational eigenfrequencies and computational mode shapes) in order to match them to the experimental observation of each measured structure. The hyperparameters of the SCM are identified using the maximum likelihood method.