A HYBRID LEARNING MODEL FOR ASSESSMENT BEAM DAMAGE DETECTION

Resumo

Structural damage induces local flexibility into the structure generating undesirable displacements and
vibrations. Such changes in the dynamic response can be used as a resource allowing us to discriminate the
current structural condition and to predict its useful life for short or long periods. Early damage detection and
periodic structural integrity assessment are the keys for the system to operate correctly and prolong its lifespan.
Many structural health monitoring techniques have been used in technologies that combine modern sensors and
intelligent computational algorithms. This study focuses on applying machine learning (ML) algorithms within a
multiclass framework to monitor structural integrity, enabling the identification and quantification of damage. In
this context, this paper proposes a strategy to damage detection in a beam structure based on an artificial neural
network machine learning algorithm. A damage index calculated from the natural frequency builds the input
dataset for the ML algorithms. The methodology combines supervised learning classification (artificial neural
networks) and unsupervised (cluster k-means) methods for constructing a hybrid classifier. The results show that
the hybrid classifier can correctly classify the integrity condition of the structure compared to the artificial neural
network algorithm.