http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Kidong Lee(이기동),Sung Yi(이성),Soongkeun Hyun(현승균),Cheolhee Kim(김철희) 대한용접·접합학회 2021 대한용접·접합학회지 Vol.39 No.1
With the development of deep learning technology, research on classification and regression models on welding phenomena using convolution neural networks (CNNs) are gradually increasing. Part 1 of this study introduced the characteristics of deep learning models using CNNs and their application to welding studies. In this paper, we reviewed recent welding research papers to analyze how to evaluate CNN models and visualize the modeling output, and details of evaluation index, comparison models, and visiualization methods were explained.
Kidong Lee(이기동),Sung Yi(이성),Soongkeun Hyun(현승균),Cheolhee Kim(김철희) 대한용접·접합학회 2021 대한용접·접합학회지 Vol.39 No.1
During machine learning algorithms, deep learning refers to a neural network containing multiple hidden layers. Welding research based upon deep learning has been increasing due to advances in algorithms and computer hardwares. Among the deep learning algorithms, the convolutional neural network (CNN) has recently received the spotlight for performing classification or regression based on image input. CNNs enables end-to-end learning without feature extraction and in-situ estimation of the process outputs. In this paper, 18 recent papers were reviewed to investigate how to apply CNN models to welding. The papers was classified into 5 groups: four for supervised learning models and one for unsupervised learning models. The classification of supervised learning groups was based on the application of transfer learning and data augmentation. For each paper, the structure and performance of its CNN model were described, and also its application in welding was explained.
Al/Fe 이종재의 마이크로 마찰교반 맞대기용접 적용성 평가
유현정(Hyeonjeong You),안영남(Youngnam Ahn),이성(Sung Yi),현승균(Soongkeun Hyun),김철희(Cheolhee Kim) 대한용접·접합학회 2021 대한용접·접합학회지 Vol.39 No.3
In the automobile industry, there is an increasing demand for Al/Fe dissimilar metal joining. Friction stir welding (FSW) is an efficient solid-state welding method to achieve high-quality Al/Fe dissimilar metal welding. Here, we reviewed the previous studies on butt FSW of thin Al/Fe sheets and conducted feasibility tests to investigate the applicability of micro FSW with a base material thickness of 1 mm or less. Most of the past literature, except for one study that adopted 1.12 mm-thick specimens, has worked with a base metal thickness of 1.5 mm or more. Selecting appropriate parameters can lead to a weld strength that is more than 90% of the base metal strength. Through feasibility tests on 2 mm-thick specimens, we could derive the welding conditions to obtain sound welds and the required joint strength. An adequate range (0.5-0.75 mm/rev) of advance per revolution was recommended to ensure the weld strength. A feasibility test on 1 mm-thick specimens revealed the possibility of melting of Al base metal during FSW of 1 mm-thin sheets; moreover, a low tool rotation speed was found to be crucial in ensuring the weld joint strength. The maximum weld strength for 1 mm-thick specimens was 200 MPa, which is 117% of the required weld strength.
머신러닝을 이용한 레이저 용접부의 모델링 Part I: Al/Cu 이종재료 겹치기 레이저용접부의 용입깊이
이기동(Kidong Lee),강상훈(Sanghoon Kang),강민정(Minjung Kang),이성(Sung Yi),현승균(Soongkeun Hyun),김철희(Cheolhee Kim) 대한용접·접합학회 2021 대한용접·접합학회지 Vol.39 No.1
Thin sheets of Al/Cu dissimilar materials are overlap welded for the electrical connection of secondary battery electrodes by laser welding. The weld penetration depth is an important joint quality to ensure strength and electrical conductance. It is difficult to predict the penetration depth using analytical methods because of the high laser reflection and small thickness of the base materials. Several machine learning algorithms were investigated to develop regression models for the penetration depth. The models included linear regression, decision tree, supported vector regression, Gaussian process regression, and decision tree ensemble model groups. The regression models with high degrees of freedom showed excellent mean absolute percentage errors (MAPE) and coefficients of determination (R²). In particular, the Gaussian process regression model with exponential kernels had an MAPE of 0.2% and an R² of unity.
머신러닝을 이용한 레이저 용접부의 모델링 Part Ⅱ: 고강도강 겹치기 레이저용접부의 형상 및 기계적 거동
유현정(Hyeonjeong You),강민정(Minjung Kang),이성(Sung Yi),현승균(Soongkeun Hyun),김철희(Cheolhee Kim) 대한용접·접합학회 2021 대한용접·접합학회지 Vol.39 No.1
In accordance with the requirements of lightweight automobiles, the application of high-strength steel sheets to car bodies is continuously increasing. The strength of the laser overlap welds is determined by the strength distribution of weldments and the bead width at the faying surface. In the case of high-strength steel sheets, it is difficult to predict the fracture load and fracture mode during the tensile shear test of the weldment owing to the high strength of the base material, softening of the heat affected zone (HAZ), and small bead width. In this study, we investigated machine learning algorithms, including artificial neural networks, to develop a fracture mode classification model and regression models for joint strength and bead width. Machine learning algorithms have shown excellent performance in predicting mechanical behaviors during tensile shear tests. Among the machine learning regression algorithms, Gaussian process regression showed the best regression ability. The R² values for the bead width and fracture load models were 0.98 and 0.99, respectively. Several machine learning models, including shallow neural networks, have shown perfect estimates for fracture locations.