Yıldız Technical University Research Information System
Not a member yet
91324 research outputs found
Sort by
Fed3C: federated clustering-based centralized classification
No full textThe success of artificial intelligence (AI) and machine learning (ML) applications depends on the analysis of large and diverse datasets. However, concerns regarding personal data privacy and data sharing, especially in datasets containing sensitive information, restrict the widespread use and sharing of such data. Federated learning (FL) offers a solution to these issues by enabling multiple users to collaboratively train a global model without the need to share their data. In this study, Federated Clustering-Based Centralized Classification (Fed3C) is proposed, where data belonging to the same class is divided into subsets, centroids are generated for each subset, and these centroids are shared with the server. The optimal number of centroids is determined using Bayesian optimization, and the centroids are generated using the K-means method. These centroids are then sent to the server, where classification is performed using K-nearest neighbors (KNN). The success of the proposed method has been tested on five different datasets, and the effects of changing the maximum number of centroids, the number of clients, the number of iterations, and the optimization algorithm on the method’s performance have been examined. The results demonstrate that the proposed approach, which does not require direct original data sharing, is effective in improving model performance. In particular, in real-world scenarios where the amount of data belonging to a single client is insufficient, the proposed method achieves a notable increase in success by involving multiple clients in the training process
Zaman-Kesirli Genelleştirilmiş Burgers-Huxley Denklem Ailesinin Analitik Çözümleri Üzerine
No full textCan electric trucks be a viable green logistics and transportation solution? Modeling a joint logistics-and-charging-infrastructure network design problem
No full textAs one of the most substantial contributors to the rapidly increasing carbon emissions, the greening and decarbonization of the transportation and logistics sectors are essential. In this context, electric trucks (E-trucks) have garnered global interest due to their potential to significantly reduce tailpipe emissions in the freight transport sector. However, inadequate charging infrastructure is a substantial barrier to the widespread adoption of E-trucks. This paper investigates a charging infrastructure development model led by an industry cluster in order to better meet the emission target. To this end, a new optimization model is formulated for a joint logistics-and-charging-infrastructure network design problem. The model aims to minimize the total cost of operating the logistics system and charging infrastructure network while simultaneously ensuring accessibility to charging stations. Numerical experiments based on a case study in Nepal were conducted to validate the proposed optimization model. The results demonstrate potential reductions of up to 33.3% in total logistics costs and 55.9% in emissions related to transportation through the transition to electric power. This analysis highlights the economic viability and environmental benefits of adopting E-trucks in green logistics and transportation, supported by an industry-spearheaded business model for developing charging infrastructure
Integrating stable diffusion via remote server APIs for enhanced parametric design workflows
No full textThe current advancements of deep learning models offer potential applications for computational design through sets of generated images controlled by parametric inputs, yet they remain disconnected from geometry-driven parametric tools. For this reason, we study the implications of text and image-based generation methods to be used in traditional parametric design procedures. We implement this study by integrating Stable Diffusion and ControlNet to Rhino Grasshopper through a Python-based remote-API plug-in. This API allows a direct connection to the diffusion-based image generation methods without any middleware. Our main contribution is to enable architects and designers to interactively generate and investigate new design ideas in their native parametric design environment. We evaluate potential impact on parametric design education with 15 architecture students using a single GPU server running Stable Diffusion v1.5 across three exercises: Text-to-Image, Image-to-Image using Rhinoceros view captures, and Parametric-Model-to-Image with ControlNet. Quantitative results showed that the API-enabled image generation averaged 4–15 seconds per image, allowing seamless integration with parametric workflows for all 15 students in a classroom setting. Performance evaluations show that our approach offers significantly improved efficiency and responsiveness compared to existing diffusion-based tools, highlighting its suitability for seamless integration within parametric design environments. Qualitative feedback indicated improved design ideation, greater fluency in prompt engineering, and enhanced understanding of parametric logic through iterative visual experimentation. These findings demonstrate the potential of real-time AI integration to augment both conceptual design and parametric design education
Thin and flexible Mg₀.₇Ni₀.₃Fe₂O₄ nanoparticle-reinforced chitosan/polyvinyl alcohol bionanocomposite films with tunable dielectric properties for energy storage devices
No full textMechanical Performance and Structural Integrity of 3D-Printed Polylactic Acid in Tensile Testing: Influence of Hole Fabrication Technique and Process Parameters
No full textThis study presents a systematic investigation of the tensile behavior of FFF-printed PLA specimens, with a specific emphasis on the role of hole fabrication methods—post-drilled versus integrated printed holes—on structural integrity. Unlike prior works that primarily addressed raster orientation and infill effects, this research isolates the influence of hole manufacturing techniques under standardized ASTM D638 and D5766 testing. Stress concentration factors (Kt) were calculated using classical analytical expressions, and their limitations for anisotropic FFF parts are acknowledged and further discussed in the Results and Discussion section. The results revealed that, although raster angle and infill density affected overall strength, the decisive factor was the method of hole generation: post-drilled holes consistently outperformed printed-hole counterparts in tensile resistance and failure behavior. Microscopic analysis confirmed that printed holes introduced interlayer misalignment and shell–infill discontinuities, accelerating crack initiation. These findings demonstrate that hole geometry alone is insufficient to guarantee mechanical reliability, and that the fabrication method of stress concentrators must be considered a critical design parameter in FFF applications
Integrating Machine Learning with Optimization to Solve Time-Dependent Cash in Transit Vehicle Routing Problem with Time Windows
No full textCash in Transit (CIT) involves the transportation of banknotes, coins, and other valuables. The transportation of these items inherently presents certain risks, necessitating security measures to protect the process. This study considers the time-dependent vehicle routing problem, where a fixed-capacity vehicle fleet collects predetermined cash from customers. The study consists of two stages. First, we apply a comparison of the most common machine learning methods, such as multiple linear regression, polynomial regression, decision trees, random forests, support vector machines, multilayer perceptron, and generalized regression neural networks, to predict traffic conditions that directly affect the speed of the vehicles. Second, the estimated speed values are utilized in the mathematical model to minimize travel time between customers during the cash collection operation. Subsequently, a real-life application is conducted in Istanbul to evaluate the effect of the proposed model, and it is solved using GAMS . This study is a starting point for future research focusing on using machine learning to predict dynamic parameters such as traffic density
The emergence of collective moral elevation through multilevel individual and organizational processes
No full textPurpose: This paper aims to extend the concept of moral elevation from the individual level to the team/group level by introducing the notion of collective moral elevation (CME) and elucidating its emergence. Design/methodology/approach: Drawing on social exchange theory and social capital arguments, this paper presents a conceptual framework that outlines the mechanisms through which moral norms, organizational symbols, social cohesion and repeated interactions facilitate the transition from individual internalization of norms to collective aggregation of moral elevation. Findings: The proposed framework emphasizes the processual nature of CME, highlighting the importance of understanding sequences of events rather than merely examining static relationships between variables. Originality/value: This paper is one of the pioneering academic works to offer a novel perspective on moral elevation, exploring its manifestation at the collective level and elucidating the dynamics of its emergence and evolution within teams and groups in organizational settings. Our proposed framework explicates how moral norms, organizational symbols, social cohesion, emulation and their iterations allow individuals to transition from individual internalization of norms to collective aggregation of moral elevation