샌드위치 복합재의 결함 탐지 및 정량화를 위한 일치 추적 분해 기반 디노이징 기법 개발

In this paper, a damage detection and quantification method for sandwich composites using matching pursuit decomposition (MPD) is presented. Sandwich composites with and without delamination were fabricated using the hand lay-up and hot-press methods, and the location and size of delamination were confirmed using flash thermography. An ultrasonic wave propagation experiment using the pitch-catch method was set up to collect data from healthy and damaged samples. The acquired signals were estimated and decomposed using MPD and compared with signals denoised using fast Fourier and wavelet transforms. The denoised signals were trained by a 1-D CNN model with the same number of layers and filters. The proposed method showed improved accuracy and stability than the traditional method. In addition, more reliable mode separation in the time-frequency representation could be confirmed, extending the possibility of MPD-based signal preprocessing in deep learning training.