Structural identification of a full-scale tied arch bridge span using genetic algorithms

Fostered by advancements in computational capabilities and the development of low-cost structural health monitoring systems, structural identification has emerged as a promising experimental technique offering contributions to several applications in performance-based civil engineering. Fundamentally, the methodology provides a framework for determining the mechanical properties of in-service civil structures by leveraging experimental measurements to update a physics-based model of the structure. Use of these calibrated high-fidelity finite elements models, and the parameters identified in the analysis, has been proposed for numerous applications in condition assessment, structural health monitoring, vibration-based damage detection, and other areas supporting decision-making support in infrastructure management. In this thesis, structural identification of a full-scale tied arch bridge span is performed using genetic algorithms to solve the optimization problem. The study leverages ambient vibration monitoring test data acquired by a wireless sensor network consisting of 48 accelerometers distributed across the tie girders of the span. Stochastic subspace state-space system identification is used to experimentally estimate a set of twenty mode shapes of the bridge with their undamped natural frequencies and damping ratios. An idealized finite element model of the span is developed to analytically predict these dynamic properties, and this idealized model is correlated to the measured response to indicate discrepancies. A parametric sensitivity analysis is then performed to identify the most meaningful uncertain parameters within the finite element model for subsequent model updating. Optimization of the model correlation through tuning uncertain parameters is achieved by minimizing an objective function using a parallel implementation of the genetic algorithm capable of exploring large population sizes. In total, sixteen different scenarios of model updating using the genetic algorithm are explored to identify the effects of varying the number of modes included in the objective function as well as the number of uncertain parameters included in the model updating routine. The results indicate that the identified parameter assignments may be highly sensitive to these factors, especially the number of modes included in the objective function. The strongest model correlation is achieved using all 20 modes in the objective function and the largest number of uncertain parameters in the model. The improvement in model correlation relative to the idealized finite element model is presented to contribute a real-world case study to the field of structural identification.