NON-DESTRUCTIVE DETECTION OF MICRO DELAMINATION IN GLASS FIBER REINFORCED POLYMER COMPOSITES USING TERAHERTZ WAVE WITH CONVOLUTION NEURAL NETWORK

Abstract

The algorithm for detecting micro-delamination inside the glass fiber reinforced polymer (GFRP) was studied by training the terahertz (THz) signal based on the convolutional neural network (CNN). THz signals with respect to the thickness of delamination in GFRP specimens were obtained through the reflection mode of the Terahertz Time-Domain Spectroscopy (THz-TDS) system. Peaks of the THz signal reflected from the top surface, micro-delamination, and the bottom surface of the GFRP specimens were classified, respectively. Then, after transforming 1D-THz signals to 2D-spectrograms through Short-Term Fourier Transform (STFT), the THz signals were trained through a CNN. Based on this, the probability map that can predict the thickness of micro-delamination from the THz signal was derived. As a result, the thickness of micro-delamination could be successfully predicted.