ABSTRACT. Digital transformation requires not only technological updates but also the restructuring of business processes through a data-driven approach. Enterprise Resource Planning (ERP) systems are central to this transformation, enabling business processes and driving digitalization strategies across organizations. The accuracy and governance of master data, which forms the foundation of ERP systems, are essential for successful transformation. In widely used ERP platforms like SAP, master data in the Plant Maintenance (PM) module—such as equipment and maintenance plans—is critical for operational efficiency, proactive maintenance, and sustainability.

In distributed enterprise environments, managing PM master data in a standardized, consistent manner presents major challenges in data integrity, process standardization, and governance. These highlight the need for auditable, centralized data management models at the corporate level to ensure sustainable digital transformation.

This study proposes a centralized data management model to improve the quality and governance of SAP PM master data in response to digital transformation requirements in maintenance operations. The model includes workflow-based approvals, rule-driven validation, and reporting tools to standardize data and ensure consistency. Results show improved data accuracy, reduced inconsistencies, and greater traceability. The study offers practical insights by addressing the intersection of digital maintenance management and data quality.

ABSTRACT. Nowadays, technological advances have led to significant increases in the processing capacity of edge devices, and it is becoming increasingly important to run AI-based object detection algorithms on such hardware. In this context, using the Jetson AGX Orin platform developed by NVIDIA, this study evaluates various optimizations to improve the performance of convolutional neural network (CNN) based object detection models. In particular, the preprocessing and postprocessing steps required for the model to make decisions during object detection were converted to ONNX and TensorRT formats and tested in different working formats. In these analyses based on the YOLOv9-c and YOLOv11-m models, these steps were run on both CPU and GPU and the performance metrics obtained are presented comparatively. As a result of the tests, an extraction speed of 88.45 FPS was obtained for the YOLOv9-c model when the preprocessing and postprocessing layers were run in TensorRT format, while this value reached 108.32 FPS for the YOLOv11-m model. When the preprocessing and postprocessing layers of the same models were run with CPU-based ONNX, 26.28 and 27.64 FPS values were achieved, respectively. Accordingly, TensorRT optimization results in a performance improvement of approximately 236.5% for the YOLOv9-c model and 291.8% for the YOLOv11-m model. These results show that optimizing the pre- and post-processing layers and converting them to TensorRT format offers a significant advantage for real-time object detection applications in embedded systems.

ABSTRACT. This study analyzes land use/land cover (LULC) dynamics in the Cappadocia region, a culturally and ecologically significant UNESCO World Heritage site, over the period of 2015–2024. The research utilizes the Google Earth Engine (GEE) platform and Landsat 8 satellite imagery combined with supervised machine learning classification. The landscape was categorized into three main classes: agricultural land, natural land, and urban/built-up areas. To ensure reliable classification, four machine learning algorithms were compared: Random Forest (RF), Classification and Regression Trees (CART), Support Vector Machines (SVM), and Naive Bayes. Among these, RF demonstrated the highest performance, achieving a Kappa coefficient of 0.6327 and an overall accuracy of 75.82%, and was therefore selected for the final spatio-temporal assessment. The analysis indicates a notable expansion of agricultural land, increasing from 25.3% to 31.8% of the total area during the study period, predominantly at the expense of natural landscapes. These findings highlight both the dynamic nature of land transformation in Cappadocia and the effectiveness of integrating GEE with machine learning techniques. Additionally, the study emphasizes the importance of model comparison in ensuring the accuracy and reliability of LULC classification in sensitive environments.

ABSTRACT. In this study, a multi-purpose autonomous agricultural robot was developed to increase agricultural productivity, reduce labor requirements, and enable precision agriculture. The developed system features a modular and expandable architecture to perform diverse agricultural tasks. Electro-optical sensors such as LIDAR and cameras, along with motion sensors such as IMU and GNSS-RTK, are integrated to allow the robot to sense its environment and determine its location with centimeter-level precision. These sensors provide 360° detection for obstacle detection, while the combination of IMU and GNSS-RTK enables field location determination. The robot is equipped with a mission computer that runs sensor fusion and real-time control algorithms. Additionally, a computer runs artificial intelligence software for plant recognition, depth perception, weed, road, and obstacle classification, and network communication software. Electric motors and actuators control the robot's movements under the control of the mission computer. The developed platform is modular, and the ISOBUS bus is used to integrate different equipment to perform various tasks, such as planting, weed removal, and spraying. In this study, an autonomous agricultural robot suitable for collective work has been developed that will minimize human labor requirements and errors for agricultural tasks and provide high productivity and sustainability.

ABSTRACT. The Battery Management System (BMS) is critical for the safe and efficient operation of electric vehicles (EVs). Insufficient cooling at elevated battery temperatures can lead to capacity loss, performance degradation, and serious safety risks. For this reason, decision makers need to select suitable battery cooling systems. This study proposes multi criteria decision making model to select a battery cooling system for BMS. Based on criteria such as technical adequacy, economic feasibility, and operational suitability battery cooling alternatives such as air cooling, direct liquid cooling, indirect liquid cooling, phase change material (PCM) cooling, and heat pipe systems are ranked using the Analytical Hierarchy Process (AHP). In this study, commonly employed battery cooling alternatives are compared with each other using AHP’s pairwise comparisons. The results indicated that direct liquid cooling emerged as the most suitable alternative with its technical superiority. Indirect liquid cooling is ranked second, primarily for its balance of applicability and performance. Heat pipe systems are identified as a promising option in terms of technical efficiency. PCM and air-cooling systems, when applied independently, are identified as low-cost alternatives suitable for applications with light cooling requirements.

ABSTRACT. This study investigates the dynamics of the moderate geomagnetic storm that occurred on April 16, 2024. Magnetic storms are formed when high-energy particles from the Sun interact with the Earth's magnetosphere. These storms were evaluated using geomagnetic indices such as Dst, Ap, Bz_GSE, Bz_GSM and F10.7. The analysis showed that during the storm, the Dst index decreased to -65 nT, the Bz component reached -84 nT, the Ap index increased to 56, and the F10.7 index remained stable at 200.2. According to the correlation analysis, there is a strong negative correlation between Ap and Dst (r = -0.72), a negative correlation between Bz and Ap (r = -0.64) and a positive correlation between Bz and Dst (r = 0.47). The results reveal that the main determinants of magnetic storms are a decrease in Bz to negative values, an increase in Ap and sudden drops in Dst. Modeling with artificial neural networks provided highly accurate predictions for the Dst and Ap indices. The study emphasizes the importance of using more advanced machine learning models in future space weather forecasts.

ABSTRACT. In this study, in order to increase the environmental awareness of autonomous agricultural robots, the map of the indoor greenhouse area, the detection of obstacles and objects, and the real-time localization on this map were performed by combining the data obtained from 360° LIDAR and IMU sensors with SLAM algorithms. In the study, distance data were obtained through the 360° LIDAR sensor, and orientation and acceleration information were integrated with the IMU sensor. In order to obtain the robot’s position information in real time, the obtained data were processed with the SLAM algorithm and a map was created. It has been shown that the proposed SLAM-based sensor fusion for the use of autonomous agricultural robots in indoor greenhouse applications can be successfully used especially in scenarios with narrow passages, irregular surfaces, and dense plant rows. In cases where traditional GNSS-based methods are insufficient due to signal attenuation in the greenhouse environment, high-accuracy localization has been achieved by integrating LIDAR and IMU data. In addition, the Extended Kalman Filter (EKF) and Graph-SLAM approach used for data fusion were compared, and it was determined that Graph-SLAM produced a consistent map. The results reveal that the SLAM-based LIDAR–IMU fusion approach provides an effective solution for navigation, task planning, and obstacle avoidance tasks in autonomous agricultural robots.

ABSTRACT. Electricity transmission lines are one of the most critical parts of the power grid. Power outages caused by high voltage power grid failures; It is important in terms of energy supply security, economic losses and social impacts. In this study; the effects of atmospheric conditions on the high voltage electricity network in coastal settlements in terms of operation depend on the changing temperature; using the data of the main insulation limit in the conductors (corona discharge), and other High Voltage (HV) power grid fault situations in terms of humidity and dew point; classification study was carried out. The importance of predictive maintenance methods and the increasing role of weather data used in these methods are explained.

ABSTRACT. This paper details an experimental implementa- tion of a cloud-based system designed for real-time food/non- food image classification and food categorization. Leveraging prominent image datasets and complementing them with user- generated content, the core objective is to develop a robust Food Model capable of distinguishing culinary items from non- culinary objects. A key innovation lies in the proposed video processing module, which converts continuous video streams into discrete frames for subsequent classification by a dedicated service. This integrated architecture facilitates a dynamic feed- back mechanism where real-time analysis of visual input drives the food identification process. The project aims to establish a foundational framework for rapid, accurate content filtering, offering significant implications for dietary assessment and food science applications by enabling immediate, generalized food referencing within live feeds.

ABSTRACT. This paper explores the innovative applications of language models within the realm of digital defense. It investigates how these models can be utilized to address complex challenges in security and safety across digital environments. The study examines the potential of language models to enhance security protocols, improve threat detection, and automate security related tasks, thereby contributing to the development of more resilient and adaptive cyber security strategies.

ABSTRACT. Health risk stratification plays a vital role in identifying individuals who are at an elevated risk of developing chronic diseases and thereby optimizing healthcare resource allocation effectively. Nonetheless, the application of machine learning (ML) in this domain is frequently constrained due to the challenges associated with accessing real Electronic Health Records (EHRs), primarily due to privacy and ethical considerations. This study introduces a privacy-preserving and lightweight ML framework that utilizes synthetically generated EHR data to achieve effective and explainable patient risk stratification. The framework employs computationally efficient models, including logistic regression, decision trees, and pruned random forests, to categorize patients into low-, moderate-, and high-risk groups based on their demographic, clinical, and medication attributes. This system is specifically designed for low-resource settings where rapid computation and minimal infrastructure are essential. SHAP (Shapley Additive Explanations) is employed to enhance model interpretability, clearly delineating the features that most significantly influence each prediction. Experiments conducted on synthetic cohorts demonstrate that the proposed approach delivers robust predictive accuracy while substantially reducing computational demands. The findings underscore the potential of combining synthetic data with lightweight AI to develop scalable, interpretable, and privacy-preserving health risk prediction solutions, particularly in healthcare environments where data sensitivity and resource constraints are prevalent.

ABSTRACT. This paper introduces a novel approach for identifying and extracting covert information concealed within digital media, particularly focusing on steganographic techniques used for malicious purposes. Addressing the escalating threat of hidden payloads in phishing attacks and malware distribution, this work leverages advanced adversarial networks to effectively unmask concealed data. By providing a robust method to crack prevalent information hiding algorithms, this paper contributes vital tools for proactive threat intelligence, digital forensics, and fortifying cybersecurity defenses against sophisticated forms of information concealment.

ABSTRACT. This work investigates a Sliding Mode Controller combined with a Iterative Learning Neural Network (SMC-ILNN) for the suppression flutter and limit cycle oscillations (LCOs) phenomena of an aeroelastic wing-section model. The proposed method enables the estimation of the model’s dynamics based on momentum gradient descent optimization algorithm for the RBFNN training. The training process utilizes online sensor data measurements of plunge and pitch motions, which characterize the aeroelastic model dynamics. This model is actuated by leading and trailing edge control surfaces (LEC and TEC), while accounting for quasi-steady aerodynamic forces and structural stiffness nonlinearities. The designed control strategy aims to achieve rapid attenuation of dynamic instabilities and an expansion of the flight speed range. Numerical simulations validate the effectiveness of the proposed approach in mitigating flutter and LCOs, demonstrating significant enhancements in both aeroelastic stability and flight performance.

ABSTRACT. The article describes the complete design, build, and testing of an inexpensive, gesture-controlled robot that is stable. The system makes use of an MPU6050 Inertial Measurement Unit (IMU) to capture hand gestures, which are then translated by an Arduino Nano microcontroller. Control signals that have been processed are sent wirelessly using nRF24L01 2.4GHz transceiver modules to a receiver module mounted on a mobile robot. The Arduino receiver converts these signals and drives DC geared motors through an L298N motor driver for contactless and intuitive robot motion control. The system provides an invaluable advantage over traditional control systems by facilitating natural and interactive human-robot interaction (HRI) experience. Possible applications include assistive technology for mobility-impaired persons, surveillance in hazardous environments, interactive learning of science, technology, engineering, and mathematics (STEM), and advanced gaming. The work delves into existing methodologies, provides a detailed account of the proposed system architecture and its deployment, presents an in-depth analysis of experimental results, and outlines a future-improvement roadmap. The future-improvement plans consist of AI-based gesture recognition for complex command sets, ultrasonic sensors for self-implemented obstacle detection, GPS for out-of-office navigation, and voice command integration to create a multimodal control system. The findings confirm that gesture control based on inertial sensors is an achievable, efficient, and affordable paradigm for the next generation of intelligent robots.

ABSTRACT. This study employs a logistic regression model to analyze the 1988 Multinational Heart Disease Dataset, which includes data from Cleveland, Hungary, Switzerland, and Long Beach V. This diverse dataset, comprising 76 attributes (including a target variable indicating heart disease presence), was meticulously preprocessed to address duplicates and outliers, ensuring the dataset's robustness. Exploratory Data Analysis (EDA) revealed a slight imbalance in the target variable and identified key relationships through various visualizations. The logit model was selected over the probit model based on AIC and BIC criteria, and a general-to-specific approach was used for variable selection. Rigorous diagnostic tests, including the Linktest, Hosmer-Lemeshow test, Osius-Rojek test, and Stukel test, confirmed the model's validity. Significant findings include a positive relationship between age and heart disease risk, with an interaction indicating that higher physical fitness may slightly mitigate this risk. Chest pain types, particularly non-anginal and asymptomatic pain, emerged as strong predictors, underscoring the need for careful symptom assessment. Unexpected results regarding exercise-induced angina and gender-specific risk suggest areas for further research. The number of major vessels affected by fluoroscopy was identified as a critical predictor, highlighting its clinical relevance. Additionally, the study proposes hypotheses on the impact of physical activity and dietary interventions on heart disease risk, both consistent with existing health knowledge. These insights contribute to heart disease prevention strategies and offer actionable implications for healthcare practices and public health policies.

ABSTRACT. Dijital dönüşüm yönetimi, yalnızca teknolojik altyapıların yenilenmesini değil, aynı zamanda kurumsal süreçlerin veri odaklı bir yaklaşımla yeniden yapılandırılmasını da gerektirmektedir. Bu dönüşümün merkezindeki Kurumsal Kaynak Planlama (ERP) sistemleri, iş süreçlerinin yürütülmesini sağlayarak organizasyonların dijitalleşme stratejilerinde kritik bir rol üstlenmektedir. ERP sistemlerinde süreçlerin temelini oluşturan ana verilerin doğruluğu ve yönetimi, dijital dönüşümün başarısında belirleyici bir faktördür. Dünyada yaygın olarak kullanılan SAP ERP sisteminin Bakım Yönetimi (PM) modülündeki ekipman ve bakım planı gibi ana veriler; operasyonel verimlilik, bakım planlaması ve iş süreçlerinin sürdürülebilirliği açısından büyük önem taşımaktadır.

Özellikle dağıtık yapıya sahip organizasyonlarda, bakım ana verilerinin standart ve tutarlı biçimde yönetilmesi; veri bütünlüğü, süreç standardizasyonu ve yönetişim açısından önemli zorluklar doğurmakta, dijital dönüşüm sürecinin sürdürülebilirliğini sağlamak için kurumsal düzeyde standartlaştırılmış, denetlenebilir ve merkezi veri yönetimi modellerine gereksinimi ortaya koymaktadır.

Bu çalışma, bakım operasyonlarında dijital dönüşüm gereksinimlerine yanıt olarak, SAP PM ana verilerinin kalite ve yönetişim süreçlerini iyileştirmeyi önermektedir. Model, iş akışı tabanlı onay mekanizmaları, kural bazlı doğrulama sistemleri ve merkezi raporlama araçları ile veri süreçlerinin standartlaştırılmasını ve veri bütünlüğünün sağlanmasını hedeflemektedir. Uygulama sonuçları, veri doğruluğunda artış, tutarsızlıklarda azalma ve süreç izlenebilirliğinde iyileşme sağlamış; çalışmayla dijital bakım yönetimi ve ana veri kalitesi etkileşimine dair literatürdeki önemli bir boşluk doldurularak uygulamaya yönelik somut katkılar sunulmuştur.