기계학습 모델 복잡도에 따른 템퍼드 마르텐사이트 경도 예측 정확도 비교 연구

전준협,김동응,홍준호,김휘준,이석재 대한금속·재료학회 2022 대한금속·재료학회지 Vol.60 No.9

We investigated various numerical methods including a physical-based empirical equation, linear regression, shallow neural network, and deep learning approaches, to compare their accuracy for predicting the hardness of tempered martensite in low alloy steels. The physical-based empirical equation, which had been previously proposed with experimental data, was labelled and used in the present study. While it had a smaller number of coefficients, the prediction accuracy of the physical-based empirical equation was almost similar to that of the regression model based on the response surface method. The prediction accuracy of the machine learning models clearly improved as the number of layers increased and became more complicated in structure before the model began to overfit. The key point we found was that a single layered neural network model with optimized hyperparameters resulted in similar or better hardness prediction performance compared to deep learning models with a more complex architecture. We also analyzed 18 research papers from the literature which used neural network models to predict the hardness of steels. Only two recent papers adopted a convolutional neural network, as a kind of deep learning model, in a new attempt to predict hardness. The other 16 papers from 1998 to 2021 commonly chose shallow neural network models because a more complicated model is less effective than a simple model for regression problems with well-labeled experimental data in materials science and engineering.

침탄 공정 대체를 위한 방전 플라즈마 소결 방법

전준협,이준호,서남혁,손승배,정재길,이석재 한국분말재료학회 2022 한국분말재료학회지 (KPMI) Vol.29 No.3

An alternative fabrication method for carburizing steel using spark plasma sintering (SPS) is investigated. The sintered carburized sample, which exhibits surface modification effects such as carburizing, sintered Fe, and sintered Fe–0.8 wt.%C alloys, is fabricated using SPS. X-ray diffraction and micro Vickers tests are employed to confirm the phase and properties. Finite element analysis is performed to evaluate the change in hardness and analyze the carbon content and residual stress of the carburized sample. The change in the hardness of the carburized sample has the same tendency to predict hardness. The difference in hardness between the carburized sample and the predicted value is also discussed. The carburized sample exhibits a compressive residual stress at the surface. These results indicate that the carburized sample experiences a surface modification effect without carburization. Field emission scanning electron microscopy is employed to verify the change in phase. A novel fabrication method for altering the carburization is successfully proposed. We expect this fabrication method to solve the problems associated with carburization.

레이저 분말 베드 용융법으로 제조된 AlSi10Mg 합금의 경도 예측을 위한 설명 가능한 인공지능 활용

전준협,서남혁,김민수,손승배,정재길,이석재 한국분말재료학회 2023 한국분말재료학회지 (KPMI) Vol.30 No.3

In this study, machine learning models are proposed to predict the Vickers hardness of AlSi10Mg alloys fabricated by laser powder bed fusion (LPBF). A total of 113 utilizable datasets were collected from the literature. The hyperparameters of the machine-learning models were adjusted to select an accurate predictive model. The random forest regression (RFR) model showed the best performance compared to support vector regression, artificial neural networks, and k-nearest neighbors. The variable importance and prediction mechanisms of the RFR were discussed by Shapley additive explanation (SHAP). Aging time had the greatest influence on the Vickers hardness, followed by solution time, solution temperature, layer thickness, scan speed, power, aging temperature, average particle size, and hatching distance. Detailed prediction mechanisms for RFR are analyzed using SHAP dependence plots.

IN 939 W 합금의 소결 승온 속도에 따른 물리적 특성과 미세조직 분석

전준협,이준호,서남혁,손승배,정재길,이석재,Jeon, Junhyub,Lee, Junho,Seo, Namhyuk,Son, Seung Bae,Jung, Jae-Gil,Lee, Seok-Jae 한국분말재료학회 (*구 분말야금학회) 2022 한국분말재료학회지 (KPMI) Vol.29 No.5

Changes in the mechanical properties and microstructure of an IN 939 W alloy according to the sintering heating rate were evaluated. IN 939 W alloy samples were fabricated by spark plasma sintering. The phase fraction, number density, and mean radius of the IN 939 W alloy were calculated using a thermodynamic calculation. A universal testing machine and micro-Vickers hardness tester were employed to confirm the mechanical properties of the IN 939 W alloy. X-ray diffraction, optical microscopy, field-emission scanning electron microscopy, Cs-corrected-field emission transmission electron microscopy, and energy dispersive X-ray spectrometry were used to evaluate the microstructure of the alloy. The rapid sintering heating rate resulted in a slightly dispersed γ' phase and chromium oxide. It also suppressed the precipitation of the η phase. These helped to reinforce the mechanical properties.

설명가능한 인공지능을 통한 마르텐사이트 변태 온도 예측 모델 및 거동 분석 연구

전준협,손승배,정재길,이석재 한국열처리공학회 2024 熱處理工學會誌 Vol.37 No.3

Martensite volume fraction significantly affects the mechanical properties of alloy steels. Martensite start temperature (Ms), transformation temperature for martensite 50 vol.% (M50), and transformation temperature for martensite 90 vol.% (M90) are important transformation temperatures to control the martensite phase fraction. Several researchers proposed empirical equations and machine learning models to predict the Ms temperature. These numerical approaches can easily predict the Ms temperature without additional experiment and cost. However, to control martensite phase fraction more precisely, we need to reduce prediction error of the Ms model and propose prediction models for other martensite transformation temperatures (M50, M90). In the present study, machine learning model was applied to suggest the predictive model for the Ms, M50, M90 temperatures. To explain prediction mechanisms and suggest feature importance on martensite transformation temperature of machine learning models, the explainable artificial intelligence (XAI) is employed. Random forest regression (RFR) showed the best performance for predicting the Ms, M50, M90 temperatures using different machine learning models. The feature importance was proposed and the prediction mechanisms were discussed by XAI.

태안 근소만 갯벌에서 모래살포와 경운이 바지락 자연 종패 발생에 미치는 영향

전준협,정희도,박광재,이규현,안현미,이희중,최윤석,송재희,강희웅 한국패류학회 2019 The Korean Journal of Malacology Vol.35 No.2

This study was carried out to investigate the effects on the occurrence of clam spats resources by sand supplement and plowing at the natural seedling fields of manila clam (Ruditapes philippinarum) in Geunso bay, Taean-gun, of the west coast of Korea. The sand addition and plowing were conducted on the tidal flats of Pado-ri and Beopsan-ri in Geunso bay in August 2013. We monitored the variation of newly recruited clam spats (shell length (SL) ≤ 5 mm), adult clams (SL ≥ 10 mm) and the change of sediment from January 2014 to December 2015. In the Pado-ri tidal flats (with 2-4 hours mean air-exposure time), the density of newly recruited clam spats in the experimental site (sand added) was higher 1.5-4.0 times than that of control (sand not added) from October 2014 to April 2015. And the highest clam density was 25,750 ± 1,708 clams per square meter in January 2015. The density of adult clams in Pado-ri was also higher in the experimental site than control until June 2015. However, despite lots of juvenile clams (SL ≤ 20 mm) were observed in control site, adult clam density in Pado-ri has changed from July 2015 due to decrease of sand proportion in sediments. In Beopsan-ri experimental site (with 4-6 hours mean air-exposure time), the newly recruited clam spats was less than 3,000 clams per square meter, which was lower than control. As the results of the study, it seems that clam spats increased by sand addition and plowing are likely to be decreased in a short period of time due to deposition of fine sediment and other adverse environments (such as long air-exposure time). Therefore it seems to be necessary to examine carefully the various environmental factors before trying to improve the environment of natural clam seedling beds by sand supplement and plowing.

고진공 고압 다이캐스팅으로 제조된 AA365 합금의 미세조직과 기계적 특성 에 미치는 T6 열처리의 영향

전준협,손승배,이석재,정재길 한국열처리공학회 2024 熱處理工學會誌 Vol.37 No.3

We investigate the effect of T6 heat treatment on the microstructure and mechanical properties of AA365 (Al-10.3Si-0.37Mg-0.6Mn-0.11Fe, wt.%) alloy fabricated by vacuum-assisted high pressure die casting by means of thermodynamic calculation, X-ray diffraction, scanning and transmission electron microscopy, and tensile tests. The as-cast alloy consists of primary Al (with dendrite arm spacing of 10~15 μm), needle- like eutectic Si, and blocky α-AlFeMnSi phases. The solution treatment at 490 °C induces the spheroidization of eutectic Si and increase in the fraction of eutectic Si and α-AlFeMnSi phases. While as-cast alloy does not contain nano-sized precipitates, the T6-treated alloy contains fine β' and β' precipitates less than 20 nm that formed during aging at 190℃. T6 heat treatment improves the yield strength from 165 to 186 MPa due to the strengthening effect of β' and β' precipitates. However, the β' and β' precipitates reduce the strain hardening rate and accelerate the necking phenomenon, degrading the tensile strength (from 290 to 244 MPa) and fracture elongation (from 6.6 to 5.0%). Fractography reveals that the coarse α-AlFeMnSi and eutectic Si phases act as crack sites in both the as-cast and T6 treated alloys.

HVOF 용사 코팅 공정 조건에 따른 코팅층의 기공도 예측

전준협,서남혁,이종재,손승배,이석재 한국분말재료학회(구 한국분말야금학회) 2021 한국분말재료학회지 (KPMI) Vol.28 No.6

The effect of the process conditions of high-velocity oxygen fuel (HVOF) thermal spray coating on the porosity of the coating layer is investigated. HVOF coating layers are formed by depositing amorphous FeMoCrBC powder. Oxygen pressure varies from 126 to 146 psi and kerosene pressure from 110 to 130 psi. The Microstructural analysis confirms its porosity. Data analysis is performed using experimental data. The oxygen pressure-kerosene pressure ratio is found to be a key contributor to the porosity. An empirical model is proposed using linear regression analysis. The proposed model is then validated using additional test data. We confirm that the oxygen pressure-kerosene pressure ratio exponentially increases porosity. We present a porosity prediction model relationship for the oxygen pressure-kerosene pressure ratio.

반복 템퍼링이 AISI 4340 강의 미세조직과 기계적 특성에 미치는 영향

박정빈,전준협,이주헌,손승배,이석재,정재길 한국열처리공학회 2023 熱處理工學會誌 Vol.36 No.1

We investigated the effect of multiple tempering on the microstructure and mechanical properties of AISI 4340 steel. The austenitized and quenched AISI 4340 steels were tempered at 550, 600, and 650oC for 1, 2, and 4 h by single-tempering (ST). The multiple tempering was conducted for 4 h by double-tempering (DT, 2 h + 2 h), and quadruple-tempering (QT, 1 h + 1 h + 1 h + 1 h). As tempering temperature increases, yield strength and ultimate tensile strength decrease and elongation increases due to recovery and recrystallization of martensite and coarsening of carbides. At 550oC, as the number of tempering cycles increases, the yield strength and tensile strength decrease at the expense of fracture elongation. At 600 and 650oC, the yield strength and tensile strength increase with increasing the number of tempering cycles while fracture elongation maintains similar values. The multiple tempering at the same tempering time of 4 h improves the modulus of toughness at all tempering temperatures, which is presumed to be due to the change in carbide precipitation behavior by multiple tempering.

임계간 온도에서 열처리한 구상흑연주철의 미세조직 및 경도 예측

서남혁,전준협,송수영,김종수,김민수 한국주조공학회 2023 한국주조공학회지 Vol.43 No.6

본 연구에서는 임계간 온도 범위에서 열처리한 구상흑연주철의 열처리 온도에 따른 물성 예측을 위해 , 인장강도 450 MPa 급구상흑연주철을 다양한 온도에서 열처리한 후 공냉하여 물성 예측에 필요한 미세조직을 분석하고 브리넬 경도를 측정하였다 . 임계간 온도 구간에서 열처리 온도가 증가할수록 구상흑연주철 내 페라이트 분율은 감소함과 동시에 펄라이트 분율은 증가하였으나 , 흑연 구상화율 및 구상흑연입수는 주방상태에서 측정된 값과 유사하였다 . 열처리한 구상흑연주철의 브리넬 경도는 열처리 온도가 증가할수록 점점 증가하였다 . 측정된 합금 조성 및 각 안정상의 분율 , 그리고 문헌에 알려진 구상흑연주철의 브리넬 경도 예측 모델을 활용하여 열처리 온도 별 구상흑연주철의 경도 값을 계산해 본 결과 , 측정값과 매우 유사한 값을 얻을 수 있었다 . 또한 열역학 계산을 통해 예측된 상분율을 활용하여 정확한 경도 예측이 가능할지 확인해보기 위해 , 열처리 온도 별로 구상흑연주철 내 흑연, 페라이트 및 오스테나이트의 부피를 계산한 후, 이를 면적으로 변환하여 동일한 구상흑연주철의 경도 예측 모델에 적용하였다 . 이렇게 열역학 계산과 경도 예측 모델을 동시에 활용하여 계산된 구상흑연주철의 경도 값은 실제 측정한 브리넬 경도 대비 최대27의 오차 범위 내에서 유사한 값을 나타내었다. In order to predict the mechanical properties of ductile cast iron heat treated in an intercritical temperature range, samples machined fromcast iron with a tensile strength of 450 MPa were heat-treated at various intercritical temperatures and air-cooled, after which a micro-structural analysis and Brinell hardness test were conducted. As the heat treatment temperature was increased in the intercritical temperaturerange, the ferrite fraction in the ductile cast iron decreased and the pearlite fraction increased, whereas the nodularity and nodule count didnot change considerably from the corresponding values in the as-cast condition. The Brinell hardness values of the heat-treated ductile castiron increased gradually as the heat treatment temperature was increased. Based on the measured alloy composition, the fraction of each sta-ble phase and the hardness model from the literature, the hardness of the ductile cast iron heat treated in the intercritical temperature rangewas calculated, showing values very similar to the measured hardness data. In order to check whether it is possible to predict the hardness ofheat-treated ductile cast iron by using the phase fraction obtained from thermodynamic calculations, the volumes of graphite, ferrite, and aus-tenite in the alloy were calculated for each temperature condition. Those volume fractions were then converted into areas of each phase forhardness prediction of the heat-treated ductile cast iron. The hardness values of the cast iron samples based on thermodynamic calculationsand on the hardness prediction model were similar within an error range up to 27 compared to the measured hardness data.