Global Optimization Techniques for Dynamic Model Updating of Civil Engineering Structures

Abstract

Due to uncertainties associated with material properties, structural geometry and boundary conditions and connectivity as well as inherent simplifying assumptions in the development of structural analysis (SA) models, the actual behaviour of structures often differs from model‟s predictions. On-site measurements may reveal those differences. In model updating, dynamic measurements such as natural frequencies, mode shapes and damping ratios are correlated with their SA model counterparts to calibrate the SA model. The number of measurements available is usually much smaller than the number of uncertain parameters in the SA model, and, consequently, not all uncertain parameters are selected for model updating. In this research, traditional sensitivity-based model updating method and global optimization algorithms (GOAs) are explored on two experimental structures. Three GOAs, Particle swarm optimization (PSO), Genetic algorithms (GA) and Simulated annealing (SiA), have been investigated for their efficiency and accuracy in model updating problems. Initially, the three GOAs were applied to a bookshelf-type laboratory structure and it was found that PSO proved to be computationally more efficient and accurate than the other two algorithms. The PSO was then applied to a dynamically tested full-scale cable-stayed pedestrian bridge. The limitations of sensitivity based method in dynamic model updating of full scale structure are explored. A combination of Sequential niche technique (SNT) with PSO was also proposed to systematically search the full domain and gave more confidence to the analyst by systematically exploring the full search space. Finally, a damage estimation method was also proposed, using a multi-objective optimization (MOO) technique which simultaneously updated the damaged as well as the undamaged structural model. The technique was applied to a numerical beam model with some noise level to account for the experimental errors. It was found that the proposed method gives relatively better damage estimation than single-objective optimization (SOO). False detections in undamaged elements were found and a regularization technique was adopted to mitigate false detections.