Estratégia baseada em seleção de características para localização de deterioração estrutural

Abstract

Recently, structural damage detection techniques have been boosted by advances in data science technologies. In this context, the present study presents an automatic damage localization methodology based on the extraction of features from dynamic data in multi-domains associated with a filtering process. The extraction step is performed simultaneously in time, frequency, and quefrency, to diversify the acquisition of information. In machine learning, this filtering procedure is called “feature selection” and is applied here with the aim of decreasing redundancy and increasing the relevance of the feature set. The main concept is that the proposed method can adapt to the structure, providing generality about the type of geometry, material, and excitation it encounters. The damage-sensitive index is calculated from a proposed outlier analysis. The method showed promise in locating anomalies on the Z24 bridge, a full-scale construction.