Análise numérica e experimental de técnicas de detecção de danos estruturais baseados em inteligência computacional

Abstract

The detection of damage or structural alteration through dynamic responses has been object of study of several researchers who, for decades, have tried to develop technologies to accurately assess the state of integrity of the structure, commonly called "structural health". Several factors can lead to changes in the mechanical properties of a structure and, often, these differences are only visually apparent in advanced deterioration conditions. A way to avoid this situation is using continuous monitoring of structural integrity, with the main objective of detecting changes that could be signs of damage. Therefore, it is possible to inform the structure´s operator that there is an abnormal behavior, still at an early stage, so that variations can be implemented to ensure the operation of the enterprise and the safety of its users. These monitoring systems rely on effective computational tools to evaluate the information that continually arrives from the structure. In order to accomplish this, techniques based on computational intelligence have been pointed out as promising. Thus, in this work, the use of three methods of Artificial Intelligence (AI) applied to damage detection problems in structures will be evaluated: Artificial Neural Networks, Support Vector Machines and K-nearest neighbors. The input parameters for the AI algorithms are statistical indicators extracted from the monitored signals. The dynamic responses of a tested double fixed frame by the author and the classic case of the Z24 bridge are used for the evaluation of the algorithms. The results obtained indicate that these algorithms have an excellent performance when applied to structural health assessment problems. Finally, the present study informs the advantages and disadvantages of each technique that has been evaluated, approaching the perspectives for possible future work