Identificação de danos em painéis metálicos baseada em vibrações em conjunto com algoritmos de machine learning

Abstract

Structural damage monitoring in civil and mechanical structures is essential to ensure safety and functionality throughout their service life, a process known as Structural Health Monitoring (SHM). SHM employs sensors, monitoring techniques, and analysis algorithms to detect damage, fatigue, and degradation. Among experimental approaches, vibration-based SHM stands out for correlating modal and physical parameters to monitor structures using sensors such as piezoelectric patches. In recent years, Machine Learning (ML) has increasingly been applied to enhance damage detection and prognosis in SHM, demonstrating strong potential to improve efficiency and accuracy. This work aims to employ vibration-based SHM and supervised ML algorithms to identify damage in metallic panels, using an experimental setup with a metallic plate. In its initial undamaged configuration, the plate was tested, and its Frequency Response Function (FRF) was experimentally obtained through impact hammer testing and piezoelectric sensors. A finite element model of the plate was then developed and calibrated using the experimental data, enabling simulations to generate new numerical FRFs under different damage scenarios. The results from this model were used to build databases for training supervised ML classification algorithms. In the validation stage, the models demonstrated the ability to detect and locate damage, with accuracies ranging from 30% to 60% for identifying the damaged zone and from 40% to 80% for classifying damage intensity, with the best performance observed for the SVM and for datasets constructed using the RMSD index. Thus, the project enabled the development of a damage detection methodology that integrates experimental data, computational modeling, and ML algorithms, highlighting the potential of this approach for SHM applications.