Deep learning architecture to predict natural vibration frequencies of damaged structures

Abstract

Structural damage can arise from various unforeseen causes. Such damage exerts a substantial impact on the load-carrying capacity of the structure. In this study, we propose a Deep Neural Network (DNN) serving as a surrogate model to determine the severity of damage in beam structures Initially, a finite element model (FEM) was constructed in MATLAB to generate the training and testing datasets. Subsequently, a multi-layer deep learning architecture utilizing an artificial neural network is constructed. The Deep Neural Network is trained on this dataset, which encompasses numerous damage scenarios, to predict the output parameters (specifically, the first three natural frequencies of the structure). The reliability of the Deep Neural Network was subsequently verified on the test dataset, achieving an R^2value greater than 0.99. Consequently, this Deep Neural Network can serve as a substitute for the finite element method, thereby significantly accelerating the model updating process within the damage prediction framework