Name: THIAGO ARTUR MENDES DE SOUZA
Publication date: 06/04/2026
Examining board:
Name |
Role |
|---|---|
| DIOGO RODRIGO FERREIRA RIBEIRO | Examinador Externo |
| JOAO VICTOR FRAGOSO DIAS | Presidente |
| MARCOS ANTONIO CAMPOS RODRIGUES | Examinador Interno |
| TULIO NOGUEIRA BITTENCOURT | Coorientador |
Summary: The Brazilian transportation infrastructure comprises thousands of bridges, many of which are
aging and subject to degradation processes, highlighting the need for effective monitoring to
ensure safety and extend their service life. In this context, the finite element method (FEM) has
consolidated itself as a fundamental tool for the analysis and monitoring of these structures.
However, the reliability of these models depends on a calibration process, which seeks to reduce
discrepancies between numerical responses and experimental data. This study aimed to develop
and evaluate computational strategies for model calibration. To this end, three algorithms
widely used in the literature were investigated: Genetic Algorithm (GA), Particle Swarm
Optimization (PSO), and Bayesian Optimization (BO). The study also analyzed relevant aspects
of the calibration process, such as the influence of the algorithms' hyperparameters and the
application of sensitivity analysis for variable selection. The methodology was applied to four
structural problems with varying levels of complexity: three synthetic structures – a simply
supported beam, a slab on girders, and a truss bridge – and one real structure – a pedestrian
bridge located at Poli-USP in São Paulo. The results indicate that the analyzed techniques can
successfully drive the structural model calibration process, significantly reducing the errors
between numerical and reference modal responses. It was also verified that the sensitivity
analysis step contributes to reducing the problem's dimensionality and increasing the efficiency
of the optimization process. BO demonstrated particular efficiency regarding the required
number of objective function evaluations, while the population-based algorithms exhibited
greater robustness in exploring the search space. As the main contribution of this work, a Python
computational package was developed to integrate optimization routines, sensitivity analysis,
and an interface with finite element models, allowing the systematization and automation of the
structural model calibration process. The tool provides a foundation that facilitates the
application and expansion of calibration methodologies in engineering problems.
Keywords: numerical modeling, model updating, modal analysis, optimization algorithms,
bridges
