RECURSIVE MULTI-MODEL UPDATING OF BUILDING STRUCTURE: A NEW SENSITIVITY BASED FINITE ELEMENT APPROACH

An invaluable tool for structural health monitoring and damage detection, parametric system identification through model-updating is an inverse problem, affected by several kinds of modelling assumptions and measurement errors. By minimizing the discrepancy between the measured data and the simulated response, traditional model-updating techniques identify one single optimal model that behaves similarly to the real structure. Due to several sources of errors, this mathematical optimum may be far from the true solution and lead to misleading conclusions about the structural state. Instead of the mere location of the global minimum, it should be therefore preferred the generation of several alternatives, capable to express near-optimal solutions while being as different as possible from each other in physical terms. The present paper accomplishes this goal through a new recursive, direct-search, multi-model updating technique, where multiple models are first created and separately solved for the respective minimum, and then a selection of quasi-optimal alternatives is retained and classified through data mining and clustering algorithm. The main novelty of the approach consists in the recursive strategy adopted for minimizing the objective function, where convergence towards optimality is sped up by sequentially changing only selected subsets of parameters, depending on their respective influence on the error function. Namely, this approach consists of two steps. First, a sensitivity analysis is performed. The input parameters are allowed to vary within a small interval of fractional variation around a nominal value to compute the partial derivatives numerically. After that, for each parameter the sensitivities to all the responses are summed up, and used as an indicator of sensitivity of the parameter. According to the sensitivity indicators the parameters are divided into an indicated number of subsets given by the user. Then every subset is updated recursively with a specified order according to the sensitivity indicator.