Aplicação de métodos computacionais a dados vibracionais para detecção de alterações estruturais

Abstract

Structural damage detection using dynamic measurements has led to the development of several techniques in the last decades. Most of these methods associate variations of natural frequencies, mode shapes and damping ratios to damage, like the Modal Assurance Criterion (MAC), methods based on strain energy deviation, methods based on curvature mode shapes, flexibility matrix analysis, among others. Although these aforementioned techniques are mostly efficient to identify structural alterations in numerical models, they have difficulties in practical applications with experimental data. Thus, approaches involving computational intelligence to identify structural damage can be a promising field of research. This work evaluates the Artificial Neural Networks (ANN) and Support Vector Machine (SVM) to detect structural changes based on evolution of dynamic responses. The modal characteristics and statistical indicators extracted directly from raw dynamic measurements are used as inputs to computational intelligence models. Furthermore, a new methodology based on the time history of natural frequencies and temperature records is presented. The technique consists in detecting structural changes and when they occur by using a SVM algorithm. The efficiency of the proposed methodology is analyzed through data from a numerical model of a supported beam and from a continuous monitoring of the Gabbia Tower, in Italy.