이중 인공지능을 이용한 Al 7075 합금에서의 압광 균열 진단 연구 (Diagnosis of Mechanoluminescent Crack Based on Double Deep Learning in Al 7075)

서지정보

· 발행기관 : 대한금속·재료학회
· 수록지 정보 : 대한금속·재료학회지 / 61권 / 12호 / 958 ~ 964페이지
· 저자명 : 박태오, 신윤우, 이승환, 좌비오, 권용남, Suman Timilsina, 장성민, 조철우, 김지식

초록

The phenomenon of mechanoluminescence (ML) refers to the emission of light induced bymechanical stimulation applied to mechano-optical materials for example SrAl2O3:Eu,Dy (SAO). Numeroustechnologies on the basis of ML have been presented to visualize the stress or strain in various structures forthe applications including structural health monitoring. As a result, extensive attention has been devoted tothe design, synthesis, characteristics, optimizations, and applications of ML materials. However, challengesstill remain in the standardization of ML measurement and evaluation, thereby commercially viable MLapplications are currently unavailable. To overcome these difficulties, present study proposes ML measurementand evaluation techniques employing the ML fracture mechanics, finite element method, and dual deeplearnings. For the effective normalization of visualized ML images under fixed initial ML intensity condition,continuous UV irradiation above the critical ML power density has been subjected to tensile and compacttension (CT) specimens. Therefore, Plastic Stress Intensity Factor (SIF) as well as crack tip stress field havebeen extracted successfully from normalized ML images based on ML fracture mechanics. To complement andverify the ML analysis, numerical FEM simulation and analytical ASTM calculation have been also provided.
Finally, a double deep learning consists of Generative Adversarial Networks (GAN) and Convolutional NeuralNetworks (CNN) has been trained and tested for the standard evaluation of in-situ ML images.

영어초록

The phenomenon of mechanoluminescence (ML) refers to the emission of light induced bymechanical stimulation applied to mechano-optical materials for example SrAl2O3:Eu,Dy (SAO). Numeroustechnologies on the basis of ML have been presented to visualize the stress or strain in various structures forthe applications including structural health monitoring. As a result, extensive attention has been devoted tothe design, synthesis, characteristics, optimizations, and applications of ML materials. However, challengesstill remain in the standardization of ML measurement and evaluation, thereby commercially viable MLapplications are currently unavailable. To overcome these difficulties, present study proposes ML measurementand evaluation techniques employing the ML fracture mechanics, finite element method, and dual deeplearnings. For the effective normalization of visualized ML images under fixed initial ML intensity condition,continuous UV irradiation above the critical ML power density has been subjected to tensile and compacttension (CT) specimens. Therefore, Plastic Stress Intensity Factor (SIF) as well as crack tip stress field havebeen extracted successfully from normalized ML images based on ML fracture mechanics. To complement andverify the ML analysis, numerical FEM simulation and analytical ASTM calculation have been also provided.
Finally, a double deep learning consists of Generative Adversarial Networks (GAN) and Convolutional NeuralNetworks (CNN) has been trained and tested for the standard evaluation of in-situ ML images.

참고자료

· 없음