딥러닝을 이용한 미소 파단면 해석에 관한 연구(1) : 신경망 CNN 기법에 의한 파면 분류Study on Microscopic Fracture Surface Analysis based on Deep Learning (1) : Fracture Surface Classification by Convolutional Neural Network
- Other Titles
- Study on Microscopic Fracture Surface Analysis based on Deep Learning (1) : Fracture Surface Classification by Convolutional Neural Network
- Authors
- 이승진; 최성대
- Issue Date
- Dec-2022
- Publisher
- 한국기계가공학회
- Keywords
- Deep Learning(딥러닝); Convolutional Neural Network(합성곱 신경망); Fractography(파면분석학); Microscopic Fracture Surface(미소 파면); Fracture Surface Classification(파면 분류)
- Citation
- 한국기계가공학회지, v.21, no.12, pp 1 - 8
- Pages
- 8
- Journal Title
- 한국기계가공학회지
- Volume
- 21
- Number
- 12
- Start Page
- 1
- End Page
- 8
- URI
- https://scholarworks.bwise.kr/kumoh/handle/2020.sw.kumoh/26147
- DOI
- 10.14775/ksmpe.2022.21.12.001
- ISSN
- 1598-6721
2288-0771
- Abstract
- Fracture surface contains many important information of machine fracture such as crack origin, direction,number of cycle, types and defects. By observing microscopic fracture surface, cause of fracture can befigured out. However it is difficult to pass on expertise and the number of those engineers is lately dicreased.
Recently, deep learning is popular because of 4th industrial revolution. Deep learning is widely used in thevision recognition. Moreover, Convolutional Neural Network(CNN) is a kind of deep learning and optimizedto analyze visual imagery and it used to convergence with vision recognition and classification.
In this research, three kind of fracture surfaces are classified by using CNN. In order to improveclassification accuracy, different number of filters are set and tested. And several times of imageaugmentation method is applied. Finally, 95% of classification accuracy is printed. Developed program canclassify random microscopic fracture surface images.
- Files in This Item
-
- Appears in
Collections - School of Mechanical System Engineering > 1. Journal Articles
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.