Damage Classification Utilizing Autoencoders and Convolutional Neural Networks

FLAVIO BARBOSA; LUCAS RESENDE; RAFAELLE FINOTTI; ALEXANDRE CURY; HERNÁN GARRIDO; MARTÍN DOMIZIO; OSCAR CURADELLI

Cartagena de Indias

Workshop; Latin American Workshop on Structural Health Monitoring (LATAM-SHM2023); 2023

CIMNE

Structural Health Monitoring (SHM) is an increasingly important field in civil engineering, as it playsa crucial role in detecting changes in the condition of structures and identifying potential damage. To achieve this, SHM strategies often employ Artificial Intelligence (AI) techniques, which involve analyzing raw dynamic data obtained from structures to perform classifications or extract relevant features. Among the various AI algorithms used in SHM, autoencoders and convolutional neural networks (CNNs) have emerged as promising solutions, and this article focuses on discussing their application. In this study, the researchers applied both autoencoders and CNNs to identify eight different damage scenarios in a beam. They utilized the time histories of the tested structure obtained through monitoring with a high-speed camera using image processing techniques.By comparing the results obtained from both algorithms, the researchers were able to highlight theirrespective strengths and weaknesses in the context of SHM. This information can assist engineers and researchers in selecting the most appropriate algorithm based on the specific requirements of their monitoring project.