Structural model updating and prediction variability using Pareto optimal models

• Published in: Computer methods in applied mechanics and engineering 2008-11, Vol.198 (1), p.138-149
• Main Authors: Christodoulou, Konstantinos, Ntotsios, Evaggelos, Papadimitriou, Costas, Panetsos, Panagiotis
• Format: Article
• Language: English
• Subjects: Model updating; Multi-objective optimization; Pareto optimality; Reliability; Structural identification
• ISSN: 0045-7825; 1879-2138
• DOI: 10.1016/j.cma.2008.04.010

A multi-objective identification method for structural model updating based on modal residuals is presented. The method results in multiple Pareto optimal structural models that are consistent with the experimentally measured modal data and the modal residuals used to measure the discrepancies between the measured and model predicted modal characteristics. These Pareto optimal models are due to uncertainties arising from model and measurement errors. The relation between the multi-objective identification method and the conventional single-objective weighted modal residuals method for model updating is investigated. Using this relation, an optimally weighted modal residuals method is also proposed to rationally select the most preferred model among the alternative multiple Pareto optimal models for further use in structural model prediction studies. Computational issues related to the reliable solution of the resulting multi-objective and single optimization problems are addressed. The model updating methods are compared and their effectiveness is demonstrated using experimental results obtained from a three-story laboratory structure tested at a reference and a mass modified configuration. The variability of the Pareto optimal models and their associated response prediction variability are explored using two structural model classes, a simple 3-DOF model class and a higher fidelity 546-DOF finite element model class. It is demonstrated that the Pareto optimal structural models and the corresponding response and reliability predictions may vary considerably, depending on the fidelity of the model class and the size of measurement errors.