Faster R-CNN 기반의 블레이드 결함 검출 모델

초록

This study, based on Faster R-CNN algorithm, constructed a model that can efficiently detect surface defects in internal blades. The main defects on blades were analyzed to produce specimens with dented and punched damages. Depending on the amount of data, this study applied various filters and factors to expand usable data for training and building the deep learning algorithm. The results of the study showed around 97% accuracy for punch defects and around 95% for dent defects. Further studies with additional data acquisition will be conducted to improve the proposed deep learning model.