양이온성 η3-알릴 팔라듐 촉매를 사용한 폴리(5-헥실-2-노보넨-co-5-메틸에스터-2-노보넨)의 비닐 부가 중합 거동

황동준 인하대학교 일반 대학원 2006 국내석사

국립대학 도서관의 지역사회 연계서비스 개선방안

황동준 경북대학교 대학원 2025 국내석사

Over the past few decades, community engagement in national university libraries has been extensively discussed in library literature. These collaborations aim to provide educational opportunities, fulfill responsibilities to local communities, and share research resources. This dissertation investigates comprehensive strategies to strengthen community engagement services offered by national university libraries in South Korea. To achieve this, a study was conducted involving the websites and literature of 10 national universities in South Korea and state universities in North America. Additionally, semi-structured, in-depth interviews were conducted with eight national university librarians from March to May 2024. Interview data were analyzed through the constant comparative method. Through the case studies, it is evident that the range of services provided to local communities has significantly expanded compared to their early stages. This expansion has been classified into four primary areas for analysis: the utilization of facilities and data, support for education and learning, cultural programs, and institutional collaborations, and partnerships. Although the scope of these services has broadened, several challenges have also emerged. University library staff reported an increased workload and negative perceptions from university members. Despite the growing demand for services, participants expressed concerns regarding reduced budgets and staffing levels, leading to operational challenges. To address these issues, participants recommended that the government develop a systematic plan for community engagement services and implement it systematically. Additionally, an integrated local website with expanded functionalities for resource sharing and information exchange among librarians and stakeholders was suggested to enhance utilization. These findings highlight the potential of national university libraries to strengthen their ties with the communities they serve. The significance of this research lies in its provision of guidelines based on case studies of community engagement services in university libraries. By drawing on these studies, the research offers practical insights into how libraries can collaborate with local communities to effectively deliver and manage a variety of services. Through a careful examination of the current programs, this research serves as a valuable reference for developing future services that strengthen community engagement. Consequently, libraries can refer to this research when formulating plans for community engagement services at universities.

반도체 산업의 테스트베드 구축에 대한 고찰

황동준 경북대학교 대학원 2021 국내석사

The semiconductor industry can be said to be a key industry that has led the domestic industry. The domestic semiconductor industry is facing a growth limit due to quantitative growth centered on memory semiconductor. This is the system semiconductor market with the highest growth potential in the semiconductor industry and the technology to improve precision and refine the manufacturing process for highly integrated semiconductors according to the limitations of Moore's Law are at the level of latecomers. Materials, parts, and equipment in the semiconductor industry are the basic raw materials, but the localization rate of equipment and materials necessary for production is low due to the lack of technology of domestic suppliers, so raw materials are dependent on imports from abroad. Accordingly, there is a need for a case study on a test bed that can support the R&D of domestic material, parts and equipment companies and evaluate the performance of the developed product. As a way to solve such structural problems in the semiconductor industry, this study intends to consider how to build a Korean Test-bed suitable for the current state of the semiconductor industry in Korea through domestic and overseas Test-bed construction case studies and industry survey data analysis. These considerations will be helpful in strengthening the competitiveness of materials and parts technology in the semiconductor industry and creating a robust industrial ecosystem in the actual semiconductor industry by suggesting a support plan for basic performance evaluation prior to application to the semiconductor manufacturing process.

연합학습 성능 모니터링 및 관리 도구

황동준 忠南大學校 大學院 2022 국내석사

Federated learning is a technology in which a large number of clients and one central server work together to learn a global model while protecting personal information while decentralizing data. This technology can improve the learning model and improve the quality of service without leaking personal data to the user. However, the importance of client personalization due to the characteristics of protecting personal information has increased, and there is a problem that must be managed in order to improve the performance of the model. The communication environment, data distribution, and class / label heterogeneity of the clients participating in the associative learning rather hinder the clients participating in the learning. In previous studies, there is a way to select the best client set for each learning round through the performance of the model, but again, the participation of heterogeneous clients in each round makes it difficult to determine the performance management policy that is appropriate for the situation. To solve this problem, this paper implements a performance management tool that uses client selection methods to select the best set of clients, based on a dashboard that can monitor federated learning performance data. Federated learning performance monitoring and management tool supports interpretations that allow administrators to identify potential anomalous phenomena (overfitting, atypical data, long-term learning time, anomalous data) at the overview, learning rounds, and model levels. It also helps to directly manage performance indicators (loss, entropy, learning elapsed time, falsely predicted probability) if the user experience determines that the situation is not appropriate. In addition, the tool can investigate the entire process including all clients who participated in the learning, analyze by visualizing the performance index of the learning model for each round, and evaluate the contribution ranking of the clients by the performance index. In a case study, we'll see how to monitor performance metrics as a result of performing federated learning while collecting eight days of user behavior data. The test accuracy was shown to reach a maximum of 93%, and the test loss was 0.23, indicating that the learning loss gradually converged. It also shows that the Fashion-MNIST dataset can be used to apply client selection methods through performance criteria to select the best client. First, performance management by loss index identified the overfitting phenomenon, selected the client, and reduced the test loss variation range from 0.21 up to 0.06. Second, performance management with entropy metrics eliminated clients with heterogeneous data and improved accuracy by up to 5%. Finally, since a long learning time is inefficient for overall learning, specific clients are excluded from learning, and the effectiveness of the tool is checked by removing clients with abnormal data as an erroneously predicted probability indicator. 연합학습은 다수의 클라이언트와 하나의 중앙 서버가 협력하여 데이터를 탈중앙화한 상태로 개인 정보를 보호하면서 글로벌 모델을 학습하는 기술이다. 이 기술로 사용자에게 개인 데이터를 유출하지 않고 학습 모델을 개선하고 서비스 품질을 높일 수 있다. 그러나 개인 정보를 보호하려는 특성으로 인한 클라이언트 개인화의 중요성이 커지면서 모델의 성능 개선을 위해 관리해야 하는 문제들이 생겨나고 있다. 연합학습에 참여하는 클라이언트의 통신 환경이나 데이터 분포, 클래스/레이블의 이질성으로 인해 학습에 참여하는 클라이언트가 오히려 장애가 된다. 기존 연구에서 모델의 성능을 통해서 학습 라운드마다 최적의 클라이언트 집합을 선택하는 방법이 있지만, 이 또한 라운드마다 이기종 클라이언트의 참여는 상황에 알맞은 성능 관리 정책을 결정하는 어려움이 있다. 본 논문은 이 문제를 해결하기 위해 연합학습 성능 데이터 모니터링이 가능한 대시보드를 기반으로 클라이언트 선택 방법을 사용하여 최적의 클라이언트 집합을 선택하는 성능 관리 도구를 구현한다. 연합학습 성능 모니터링 및 관리 도구는 개요, 학습 라운드 및 모델의 수준에서 관리자가 잠재적인 이상 현상(과적합, 이기종 데이터, 오랜 학습 시간, 이상 데이터)을 식별할 수 있는 해석을 지원하고 더 나아가 사용자 경험을 통해 적절하지 못한 상황이라 판단했을 경우 성능지표(손실, 엔트로피, 학습 경과 시간, 잘못 예측한 확률)를 직접 관리하도록 돕는다. 또한, 도구는 학습에 참여한 모든 클라이언트를 포함하는 전체 과정의 조사, 라운드별 학습 모델의 성능지표 시각화를 통한 분석, 그리고 성능지표를 통한 클라이언트의 기여도 순위 평가가 가능하다. 사례 연구로는 8일 동안 사용자 행동 데이터를 수집하면서 연합학습 수행한 결과로 성능 지표들을 모니터링하는 방법을 확인한다. 테스트 정확도는 최대 93%까지 달성함을 보였고 테스트 손실은 0.23으로 학습 손실과는 점차 수렴된다는 것을 보였다. 그리고 Fashion-MNIST 데이터셋을 사용하여 성능 기준을 통해 클라이언트 선택 방법을 적용하여 최적의 클라이언트를 선택할 수 있음을 보인다. 첫 번째로 손실 지표를 통한 성능 관리는 과적합 현상을 식별하고 클라이언트를 선택하여 테스트 손실 변화폭을 0.21에서 최대 0.06까지 줄였다. 두 번째는 엔트로피 지표를 통한 성능 관리는 이기종 데이터를 가진 클라이언트를 제거하여 정확도를 최대 5%를 개선했다. 마지막으로 오랜 학습 시간은 전체적인 학습에 비효율적이기에 클라이언트를 특정해 학습에서 제외하며, 잘못 예측한 확률 지표로 이상 데이터를 가진 클라이언트를 제거를 통해 도구의 효율성을 확인한다.

주요 범용 고분자 제조공정의 재해분포 특성 및 공정의 안전성 확보방안

황동준 부산대학교 대학원 2017 국내석사

Cu-10wt%Sn 소결베어링 제조시 공정변수가 소결특성에 미치는 영향

황동준 아주대학교 2002 국내석사

90wt%Cu-10wt%Sn, oil-less bearing 재료의 소결실험에서 공정변수들 중에서 분말입도, 분말종류, 성형밀도 및 소결온도가 치수변화, rupture strength, 기공률 및 함유량에 미치는 영향을 연구하였다. 구리분말은 전해동분과 분무환원동분 두 가지를 사용하였으며 분말 입도는 전해동분의 경우 상용분말 ECu-200. ECu-325, 분급된 분말인 -200~+325mesh 및 -325mesh를 사용하였고 분무환원동분은 ACu-140, ACu-325, 분급된 분말 -200~+325mesh를 사용하였다. 시편은 이론밀도의 70%, 75%. 80%로 성형되었고 암모니아 분위기에서 740도와 800도 온도에서 소결되었다. 성형체의 소결 전후 치수를 비교하여 보면 Herring의 이론에서 밝혀진 바와 같이 분말입도가 작을수록 수축이 일어났으며 성형밀도가 낮을수록, 소결온도가 높을수록, 소결후 수축이 일어났으며 분무환원동분 보다 전해동분의 치수변화율이 더 크다. Rupture strength는 미세한 분말의 분포가 더 높은 전해동분이 분무환원동분 보다 높았으며 성형밀도가 높고 소결온도가 높을수폭 강도가 높다. 소결온도가 높을수록 합금화가 더 진행되기 때문에 강도는 소결온도 증가에 영향을 크게 받는다. 기공률 및 함유량은 분무환원동분을 사용한 시편에 1차기공인 공극이 더 많이 존재하기 때문에 전해동분 보다 분무환원동분에서 더 높게 나타났으며 주석 성형밀도가 낮고 소결온도가 높을수록 용융량이 많아지면서 용융액의 이동이 용이하기 때문에 2차기공인 주석분말 자리가 더 많이 존재하기 때문에 높게 나타났다. The dimensional change along with the microstructure, particularly the pore structures, and the sintered strength after the sintering of Cu-10wt%Sn bronze were investigated with regards to the size and shape of copper powder, compaction density and sintering temperature. Increasing the compaction density generally decreased shrinkage rate at all conditions. Electrolytic copper powder having high surface area produced more dimensional change and distortion than atomized copper powder, but the effect is much more pronounced at the higher sintering temperature of 800℃. Little difference in dimensional change was observed between the two powders when sintered at 740℃ at all compaction densities. The lower sintering temperature produced a better dimensional change stability but a much lower sintered strength. At all processing conditions, the ratio of open porosity to total porosity is higher and the pores are more uniformly distributed when atomized powder was used. The differences of the dimensional change, pore structure and the strength can be explained by comparing the relative amount of the transient liquid phase formed and the duration of it's existence in regards to the shape and the extent of the boundaries between the copper powders during the sintering.

Three-dimensional nanostructured multilayers from cross-linkable block copolymer

황동준 高麗大學校 大學院 2010 국내석사

In this work, we fabricated three dimensional nanotemplates using crosslinkable block copolymers (BCPs). We synthesized crosslinkable BCPs, P(S-r-(S-N3))-b-PMMA, in which 1.5 mol % of crosslinkable azide (N3) groups were incorporated. The block composition was controlled to have cylindrical or lamellar microdomains. First, thin films of P(S-r-(S-N3))-b-PMMA block copolymers, exhibiting either perpendicularly oriented cylinders or lamellae, were prepared on the silicon substrates that were neutralized with PS-r-PMMA random copolymers. To construct the BCP multilayers, the films were thermally crosslinked, and then cylinder- or lamellar-forming BCPs were prepared on top without disturbing the underlying layers. Interestingly, it was observed that lamellae on cylinder layers and cylinders on lamellae layers exhibit the perpendicular orientations. This may suggest that the underlying layers could neutralize the interfacial interactions as PS-r-PMMA random copolymers. We attribute such registration to the minimization of wetting energy at the interface, while the perpendicular orientation of lamellae can be also affected by other factors, such as the nematic interaction, quenching of polymer chains at the surface, the substrate roughness, etc. These results demonstrate a promising strategy for designing well-defined 3D nanotemplates for advanced BCP nanolithography.

이웃한 두 개의 원형 실린더 주위의 유동 및 입자 거동 해석

황동준 금오공과대학교 2016 국내석사

Numerical simulations of particle-laden flow around two nearby circular cylinders are carried out for side-by-side and tandem arrangements. The present study focuses on the effects of particle Stokes number and the spacing between two cylinders on the particle dispersion and deposition characteristics as well as flow structure. The motion of particles is calculated by a Lagrangian approach based on one-way coupling. Results show that the particle dispersion pattern changes significantly due to the modification of wake structure for both cylinder arrangements. However, only the tandem arrangement affects the particle deposition strongly, especially on the downstream cylinder. The deposition efficiency of downstream cylinder is much smaller than that of single cylinder and it becomes zero for small cylinder spacings.

디지털 유료교육컨텐츠 이용고객의 만족도 및 재구매의도에 관한 연구

황동준 숭실대학교 대학원 2007 국내석사

Al2O3 박막 식각도에 영향을 주는 요소 연구

황동준 한양대학교 대학원 2023 국내석사