Shiga University Repository
Not a member yet
15011 research outputs found
Sort by
The effectiveness of nudges in reducing methane emissions from agriculture
Get PDFdepartmental bulletin pape
Wildlife Conflict Management Initiatives in Bavaria, Germany: The Case of Beaver Management
Get PDFWildlife conflict management for human communities is a critical challenge for biodiversity conservation across Europe. The Eurasian beaver （Castor fiber）, reintroduced to many regions including Germany, exemplifies these conflicts due to its ecological benefits as an ecosystem engineer as well as its negative impacts on agriculture, forestry, infrastructure, and water management. This study examines the comprehensive approach to beaver management in Bavaria, Germany, highlighting how stakeholder engagement, flexible policy frameworks, and economic support mechanisms collectively contribute to conflict mitigation. Bavaria’s distinctive management structure, particularly the deployment of specialized beaver managers and advisors, bridges administrative agencies and local communities, facilitating targeted interventions and effective communication. Financial support schemes vary significantly by sector, reflecting differing societal values: agriculture prioritizes immediate economic compensation for crop damage; forestry focuses on sustainable forest management incentives; environmental conservation sectors emphasize biodiversity and habitat protection through long-term ecosystem management; and infrastructure and water management stress safety and preventive measures. These differences underscore the importance of adaptable and context-specific policies. Insights from the Bavarian experience highlight three critical elements: establishing localized expert intermediaries, offering diverse economic support aligned with stakeholder values, and balancing ecological protection with pragmatic population management. This multifaceted model demonstrates significant applicability to other regions confronting similar wildlife-human conflicts, suggesting a practical pathway towards sustainable coexistence between human societies and wildlife not only in Europe but also around the world.departmental bulletin pape
An Analysis of Factors on the Retention of Memories of University Students about Nature Experience Programs in their Elementary School Ages
Get PDFThe purpose of this study is to clarify factors in the retention of memories garnered from nature
experience programs during elementary school ages by analyzing the memories of students at Shiga
University. For this purpose, 91 descriptions of memories in the class assignment were interpreted and
analyzed through a co-occurrence network analysis using KH Coder.
A remarkable nature experience program included among their memories was a “fish catching program”,
i.e. catching and skewering a fish by themselves and eating it after grilled by adult leaders. This study
suggests that doing the activity by oneself was a significant factor in the retention of memories from that
program. It also suggests that retention is stronger when palm feeling is involved.departmental bulletin pape
フシメ ニ チグハグ マチマチ アラアラ 2025ネンド コクリツ ハンセン ビョウ シリョウ カン ギャラリー テン センゴ 80ネン センソウ ト ハンセン ビョウ セイゴ ヒョウ イン マイ ライフ 2
Get PDFtechnical repor
ケイジョウ トクチョウ リョウ オ カツヨウ シタ ダイカスト セイヒン ノ フリョウ ヨソク オヨビ ケイジョウ サイテキカ
Get PDF滋賀大学博士（データサイエンス）The die casting process is a pivotal manufacturing technology, particularly in the automotive industry, enabling the mass production of complex metal components by injecting molten metal into dies under high pressure. While this process allows for efficient production and weight reduction in components, it presents challenges such as defect prediction and reduction of development lead times due to complex shape designs. Traditional simulation methods for predicting defects like soldering and die cracking require detailed model information and are computationally intensive, making them less practical for early-stage design considerations.
This research focuses on developing surrogate models using Variational Autoencoders (VAEs) and Multilayer perceptrons (MLPs) to predict defects in die-cast products based solely on geometrical features. By leveraging machine learning techniques, the proposed method aims to provide accurate predictions with reduced computational effort, facilitating more efficient product development cycles. The surrogate model transforms high-dimensional shape data into a low-dimensional latent space using VAEs, capturing essential geometrical characteristics. A MLP is then trained using these feature vectors to predict defect occurrence, such as soldering and die cracking.
The study analyzed 113 types of transaxle parts and 130 engine block parts produced by Toyota Motor Corporation. For soldering defect prediction, the surrogate model demonstrated a higher F1 score compared to the conventional rule-based model, significantly reducing the false positive rate. This improvement is attributed to the model's ability to learn from data where design interventions, like additional cooling, were already in place, effectively capturing both physical phenomena and practical manufacturing constraints.
For die cracking prediction, the surrogate model achieved a certain level of accuracy by evaluating the risk based on structural features and manufacturing conditions. However, limitations were noted due to the model's reliance on surface shape alone, without considering internal structures like cooling circuits that can significantly impact stress distribution. This highlights the need for incorporating more detailed information, such as die design features and thermal properties, to enhance prediction accuracy.
The research also explored optimization techniques for shape design in cast components. Traditional optimization methods are computationally intensive when combined with Computer-Aided Engineering (CAE) simulations. By utilizing the surrogate model, the study enabled efficient shape optimization to suppress defects, significantly reducing computational costs. Constraints were applied in the latent space to ensure shape modifications not deviating significantly from the original design, maintaining manufacturing precision and meeting design requirements.
Further, the study evaluated advanced generative models such as VAE-GAN, SAGAN, and Generative Voxel Net to improve shape generation accuracy. These models aim to incorporate higher-resolution shape features into the surrogate model, enhancing defect prediction accuracy. The evaluation focused on how each method affects the generation accuracy of voxel models and the predictive performance of the surrogate model, providing insights for selecting optimal methods in future shape optimization endeavors. In conclusion, the surrogate modeling approach using VAEs and NNs presents a promising solution for early-stage defect prediction in die casting, enabling efficient product development and optimization. While the models show significant improvements over traditional methods, incorporating more comprehensive data and advanced generative techniques can further enhance accuracy and applicability across various defect types and manufacturing processes.doctoral thesi
第3章 総合的な学習の時間の研究 2情報の時間 情報社会を多面的・多角的にとらえさせる実践的研究―第三学年「著作権とAI」の単元構想を通して―
Get PDFdepartmental bulletin pape
GNN ト タイショウ ガクシュウ ニ モトヅイタ マルチ モーダル ユウゴウ ニ ヨル ショクヒン スイセン システム ノ コウチク
Get PDF滋賀大学修士（データサイエンス）master thesi