Initial structural damage detection approach via FE-based data augmentation and class activation map

Abstract

Damage is generally initiated locally and spread to the entire structure. To avoid the destruction of the entire structure, it is crucial to detect and act on damage at an early stage through the real-time monitoring of the entire structure. However, the attachment of the many sensors to obtain sufficient detection resolution could change the structural dynamic characteristics of the structure. To compensate for these shortcomings, research has been conducted on digital image correlation (DIC) as a non-contact method of displacement/strain measurement. In addition, the real-time monitoring of a structure using DIC equipment is relatively straightforward. The final goal of this study is to predict the location of damage using the displacement of the structure surface which can be via DIC. This paper introduces a method for monitoring damage locations using a class activation map (CAM) network. The feasibility of the proposed process using the finite element method for an example considering the experimental situation was confirmed. To generate training data, the finite element method was used to obtain the displacement and strain of a target structure. Herein, sub-structuring approach with data augmentation using Gaussian integration point interpolation were employed to obtain the benign performance of the proposed approach. Thus, the proposed CAM network could classify the presence or absence of damage by considering strain fields. Moreover, the relevant result of the CAM network is a CAM image, which indicates the location of the damage. Finally, this CAM network was applied to a tensile specimen example and show good performance in the classification and detection of damage locations.

Identification of Damage in a Wind Turbine Blade Using Mechanical Measurements and Artificial Neural Networks

“…In this context, one has pretrained Artificial Intelligence (AI) systems based on artificial data that are further trained with real-life data. Representative works combining computational modeling by finite element and boundary element methods for the creation of damage scenarios and inverse analysis include these papers [21][22][23][24][25][26][27][28][29][30][31][32].…”

Image-to-Image Translation-Based Structural Damage Data Augmentation for Infrastructure Inspection Using Unmanned Aerial Vehicle

“…While common data augmentation methods like random flipping, rotating, and cropping have demonstrated their ability to bolster the resilience of trained DNNs, they possess limitations in encapsulating the authentic diversity seen in real-world structures. This includes the variations in building attributes such as size, shape, and color, consequently curbing these methods' efficacy [18][19][20]. In response to this issue, researchers have introduced innovative data augmentation strategies that leverage synthetic images-either generated by computers or data-driven [21,22].…”