합성곱 신경망 기반 복합재료의 손상 위치 탐지 방법

초록

This study proposed a method to detect the location of damage in a composite material based on the Convolutional Neural Network(CNN). Existing research targets isotropic materials such as metals and ceramics. These studies have numerically calculated the location of the damage using the velocity of the elastic waves is constant which is due the damage. However, this method is difficult to apply to anisotropic materials, such as fiber reinforced plastic(FRP), in which the speed of elastic waves varies irregularly depending on the fiber orientation and the layered structure. In this research, a 300mm x300mmx3mm plate specimen was produced for Carbon Fiber Reinforced Plastic(CFRP) to detect damage location of composite materials. After attaching a PVDF sensors to each corner of the specimen, a pencil lead breakage test(PLB) was performed to simulate elastic waves by damage. The signals obtained from each sensor were preprocessed with high-pass filtering and denoising to extract elastic waves. Scalogram was extracted by Continuous Wavelet Transform(CWT) from extracted elastic waves. The training dataset was created with scalograms and damage locations as input data and output data respectively. CNN model was trained by training dataset. For verify the methodology presented in this study, three data were used as test data set. CNN model detected accurate damage locations with errors of 4.48mm, 3.65mm, 8.57mm, respectively. The model was verified to be correct as the model detected within a maximum error of 3% based on specimen size. In conclusion, the methodology for damage location detection of composite materials based on a CNN presented in this study is thought to be able to utilize not only detecting damage location of composite materials but also complex structures. Therefore, it is thought to be able to utilize monitor damage of various composite material structures.

키워드