An Innovative One-step Vibration-based Approach for Structural Damage Localization and Severity Estimation in Plates via Modal Strain Energy and Deep Learning

Structural health monitoring (SHM) has encountered increasing demands for advanced computational methods capable of effectively addressing damage detection challenges in structural engineering. This paper presents a novel vibration-based approach for non-destructive damage identification in plate-type structures through the integration of the modal strain energy (MSE)-based technique with convolutional neural networks (CNNs). In this framework, damage is simulated on an aluminum plate by reducing the thickness of selected finite elements. The relative MSE data, extracted from undamped free vibration analyses conducted using finite element modeling, are employed as inputs to the CNNs. The proposed approach is designed to accurately detect both the location and the severity of structural damage. Its performance is assessed across various damage scenarios using multiple evaluation indices, with predicted outcomes compared against predefined damage levels. A key novelty of this investigation lies in the design of a CNN architecture that effectively captures the distinctive patterns embedded in the MSE distribution, thereby enabling highly accurate identification of the location as well as the severity of structural damage. Additionally, the study systematically investigates the effectiveness of using individual mode shapes and their combinations. The results show that using multiple mode shapes, particularly the first three bending modes, greatly enhances the accuracy and stability of the predictions. This process of selecting mode shapes helps optimize the input configuration for the CNN, enhancing the overall efficiency and reliability of the presented approach. The findings indicate that the proposed one-step approach achieves high accuracy in detecting damage, thereby offering a reliable and data-efficient solution for SHM in plate structures.