Aplicação de metodologias de reconhecimento de padrões para detecção de dano em vigas de concreto

Abstract

In the last decades, permanent Structural Integrity Monitoring (SHM) systems have been proposed to increase the reliability of the structural condition assessment and support the bridge maintenance process. Working with the intention of complementing the structural analysis of engineering structures in service, this project aims to apply methodologies of pattern recognition to identify damage in concrete beams, in order to analyze the information coming from monitoring systems, Permanent or temporary, including non-destructive technology, in order to help in the future the process of maintaining bridges and special structures. For this purpose, we have applied pattern recognition algorithms to identify concrete beam damage: wavelets, support vector machines and the gaussian mixtures model. The beams had different damage levels: intact, 30% damage and 60% beam height. For the analysis of the wavelets, the functions of Haar and Daubechies were used. For the support vector machines, the polynomial and RBF (Radial Basis Function) functions kernel were tested and three clusters were used in the gaussian mixtures model. The objectives proposed by the work were achieved, that is, the three methods were able to distinguish the different damage patterns. Using the wavelets, the Daubechies family presented better quality. For the support vector machines, the best results were obtained with the RBF kernel function. In relation to the Gaussian mixture models, the method also recognizes the existence of different levels of damage, although this recognition is not complete. The evaluated parameters were flexural stiffness as a function of displacement. The conclusions were obtained from statistical methods, the LIBSVM and the EM algorithm (Expectation Maximization).