측정 위치에 대한 강건성을 가지는 구조 진동 신호 기반의 결함 있는 복합재 구조물의 분류

Abstract

In the present work, a new method to classify healthy and damaged composite structures using experimentally obtained structural vibration data is proposed and evaluated. After fabricating healthy and damaged laminated composite beam specimens, structural vibration data for fixed-free boundary conditions is experimentally obtained via random excitation. The measured vibration signals are converted into images using a Short-Time Fourier Transform and used as input data for learning and testing. First, an autoencoder is used to detect the presence of damage. The autoencoder model is trained using the vibration data of the healthy composite structure. The vibration data of a healthy composite structure is input to the trained autoencoder model with the data of a damaged composite structure, and errors between the input and output data are compared to detect the presence of damage. Second, a convolutional neural network model is used to classify the healthy and damaged composite structures with two different damage locations. This study confirms that the proposed technique can effectively detect and locate damage in composite structures.