ABSTRACT. Reducing energy consumption in embedded systems has become a critical requirement, particularly for battery-powered portable devices, Internet of Things (IoT)-based sensor networks, and industrial applications. Energy consumption is a fundamental design parameter for IoT-based sensor networks, medical devices, and portable industrial solutions. Extending battery lifetime and ensuring sustainable energy usage have emerged as global priorities in recent years, not only from an engineering perspective but also due to their environmental impact.

To address these needs, the study to be conducted by Demsay Electronics R&D Center adopts a hardware- based power management approach and aims to present an innovative solution particularly focused on

extending battery and cell lifetime. In this work, the nPzero IC (Power Saving chip) developed by NanoPower will be utilized to experimentally evaluate the low-power mode transition performance of embedded systems.

The current consumption in active and standby modes will be observed using precision multimeters and low- current oscilloscopes, in accordance with IEEE standards, in order to technically demonstrate the efficiency

provided by the chip. The nPzero IC, with its ultra-low standby current and fast wake-up capability, offers a more fundamental and hardware-based solution compared to software-driven optimization techniques. This feature is expected to extend the operational lifetime of a typical IoT device from weeks to months, or even years. The anticipated results are not limited to reducing power consumption but also include multi-dimensional benefits such as prolonged operational lifetime in IoT devices, enhanced battery durability in medical devices to ensure patient safety, improved reliability of portable communication systems in defense electronics, and reduced maintenance frequency in industrial automation. Within the scope of this study, it is emphasized that fewer battery replacements and reduced electronic waste will provide a direct and meaningful contribution to environmental sustainability.

ABSTRACT. The transition to Industry 5.0 requires production systems to ensure not only efficiency but also compliance, traceability, and reliability. This study proposes an IoT-enabled smart manufacturing framework integrated with ERP systems to directly collect and analyze data from production devices. Continuous collection of machine and sensor data provides critical insights for process optimization while also ensuring compliance with international standards. The proposed framework enables real-time tracking of production stages, recording the duration of each phase and identifying inefficiencies. A key feature is the integration of quality control mechanisms: production is automatically stopped when quality approvals are not verified, preventing defective products from passing to subsequent stages. This approach reduces errors, minimizes waste, and ensures consistent product quality. The system also utilizes QR code-based tracking to ensure end-to-end product traceability. Each product is assigned a unique QR code linked to ERP records, precisely tracking production history, quality controls, and compliance documentation. This approach increases transparency across the supply chain and supports regulatory oversight. Combining device-level data collection, ERP integration, QR code traceability, and automated quality assurance, this framework offers a scalable way for businesses to modernize their operations, aligning with Industry 5.0 principles that emphasize resilience, human-centricity, and sustainability.

ABSTRACT. This study presents a functional IoT design that combines wired and wireless communication infrastructures on the same board. The designed board serves as a flexible platform for data collection, processing, and transfer to different network environments from field devices. The architectural structure is built on reliability, multi-protocol support, and the ability to power various power sources for industrial applications. On the wired communication side, the board is equipped with Ethernet, serial-based industrial protocols, and isolated data paths. On the wireless communication side, it supports Wi-Fi, Bluetooth, cellular network connectivity, and low-power wide-area networks. This makes the system adaptable to both short-range applications and long-range field deployments. The input/output layer provides broad compatibility for analog and digital signals, enabling direct connection to sensors commonly used in the industry. The power architecture can operate with both external power supply and alternative sources, allowing the system to easily adapt to different usage scenarios. As a result, the developed IoT board: By combining multiple communication standards on a single hardware, it offers flexible, reliable and fast integration opportunities in different sectors such as production, agriculture, energy and logistics.

ABSTRACT. With the miniaturization of electronic devices, the fast and error-free placement of micro and miniature-scale semiconductor components has become a critical requirement. In particular, the thermoelectric modules, which use semiconductor legs as small as 1x1x1.2 mm³, are still being placed manually, resulting in a time-consuming, costly, and error-prone production process. This study presents the “Next-Generation Control and Placement System,” developed under the TÜBİTAK 1507-SME R&D Startup Support Program. The system is designed as a desktop-type intelligent intermediate control and placement device, compatible with standard pick-and-place machines. The device integrates an STM32-based control board, Raspberry Pi-supported image processing infrastructure, and precise sensor systems to automate material positioning, sorting, and verification. The developed prototype is equipped with camera-assisted image processing algorithms, magnetic sensors, and a vibration mechanism. In this way, empty slots are automatically detected, excess materials are sorted, and critical safety checks are ensured. In addition, IoT-enabled communication infrastructure has been tested for data transfer via GSM, Wi-Fi, and Bluetooth. The results demonstrate that the device reduces manual production time by more than 80%, enabling the production of hundreds of thermoelectric modules within one hour. This approach not only improves manufacturing efficiency but also reduces costs, minimizes error rates, and contributes to the digitalization processes of the sector. This study provides a unique solution for semiconductor manufacturing in Türkiye, enhancing national know-how and contributing to industrial competitiveness.

ABSTRACT. Industrial refrigeration cabinets are used as cold chain display cabinets in various areas such as hotels, supermarkets and gas stations. These cabinets are produced with doors, without doors or as freezers. Icing occurs due to the humidity of the air passing over the evaporator in the industrial refrigeration cabinets and the low evaporator operating temperature. This icing causes the air flow to slow down and the products cannot be stored at the desired temperatures. Various methods have been developed to prevent icing or to melt the ice. Some of these have become widespread as closed cycle method, closed cycle method with hot gas and resistance defrosting. The application of these methods is determined according to the type of cabinet. The ambient conditions during the tests were set at 25°C and 60% humidity as per ISO 23953 standard. However, these ambient conditions cannot be achieved in every region where the cabinet operates, and the cabinet can be used in environments with higher humidity and temperature. In this study, image processing techniques were used to investigate the feasibility of detecting icing on the evaporator for cabinets operating in different regions.

ABSTRACT. Object detection is one of the most important tasks in computer vision and remote sensing and is often accomplished using deep learning algorithms. It has many application areas such as object detection, autonomous vehicles, security system, medical imaging, industrial automation and defense industry. As a result of technological developments, it has increasingly more areas of use. With the wide range of these usage areas and the recent developments in deep learning algorithms, object detection studies are becoming increasingly important. Performance evaluation of rapidly developing algorithms on large data sets is also becoming an important problem. Convolutional Neural Networks (CNN) based algorithms are generally used to solve this problem. In this study, the performances of YOLOv5n, YOLOv5s, YOLOv6n, YOLOv6s, YOLOv8n and YOLOv8s models, which are versions of YOLO, a CNN-based algorithm, were analyzed using the military aircraft detection data set. According to the mAP50, mAP50-95, precision and recall values determined as performance metrics, the model that gave the best results was the YOLOv8s model. The fastest working models are YOLOv5 models.

ABSTRACT. In recent years, significant advances in technology is led to significant research in the fields of computer vision and artificial intelligence. In the critically important field of security, the development of artificial intelligence technology has accelerated research and led to significant advances. Security protection to counter potential threats to critical infrastructures is taken by their own Institutions. X-ray machines play a crucial role in these security measures, allowing users to check the contents of their luggage upon entering a building. The X-ray machine allows high-density items in luggage to be monitorized using X-rays, and these images are used to identify potentially threatening objects by the experts. Therefore, the security of an institution is directly related to the expertise, physical, and mental state of the relevant personnel. In this study, we analyzed the classification results of dangerous objects by examining the results, which are obtained from the Yolov11, Yolov12, Faster R-CNN Resnet50, and Faster R-CNN MobileNetV3-Large models to eliminate personnel dependency and minimize errors in detecting hazardous objects. According to the Precision, Recall, F1 Score, mAP@50 metric values, the best result is obtained by using YOLOv12 model.

ABSTRACT. Today, the need for instant access to information, traceability, and operational flexibility is increasing in the production and logistics sectors. To address these needs, the digital labeling system, powered by a wireless, Wi-Fi Mesh architecture and developed by Demsay Electronics R&D Center, is designed for use on production lines and warehouse shelves. The system consists of digital labels with OLED displays, dynamic QR code support, a master controller, and central management software. Wireless communication is established between labels via MAC addresses, eliminating the need for any internet infrastructure. This system offers high flexibility in all areas, such as logistics warehouses, production lines, and retail markets. Each label receives data from the central control software in real time and automatically updates the displayed information. Additionally, dual LED indicators on the labels provide visual notifications to the user both in case of data updates and when the relevant label needs to be located in the field. The integration of the developed system minimizes errors related to manual data entry in production and inventory management processes, reduces operational burdens related to labor, and minimizes paper consumption, contributing to environmental sustainability. The project has already been implemented and is being successfully implemented in Demsay Elektronik's own production and warehouse areas. This report will detail the system's architectural structure, wireless communication protocol, ERP integration, dynamic content management, and field application experiences. It will also assess the system's place within the IoT ecosystem and its potential for widespread deployment across sectors.

ABSTRACT. In this study, a novel machine learning-based framework for intrusion detection is proposed by merging two real-world network datasets to simulate diverse cybersecurity threat scenarios. The methodology combines supervised, unsupervised, and deep learning approaches to classify and detect network anomalies effectively. Supervised models such as LightGBM, Random Forest, and Gradient Boosting were trained and evaluated using standard classification metrics, while unsupervised methods including Isolation Forest, K-Means, and DBSCAN were applied for anomaly detection without labeled data. Additionally, deep learning models like SVM, CNN, and LSTM were used to further enhance detection capabilities. The results show that LightGBM achieved the highest overall accuracy (92.74%), while Isolation Forest performed best in detecting attack patterns (50.8% recall). A visual analytics dashboard was developed using Power BI to provide deeper insights into user behaviour, protocol usage, and model predictions. The study highlights the trade-offs between model sensitivity and false positive rates and provides a comprehensive comparison of techniques suitable for real-time intrusion detection applications

ABSTRACT. As project-based work continues to gain popularity in organizations for its ability to enhance teamwork, flexibility, and resource optimization, assigning individuals and tasks becomes more critical which is widely addressed in literature. This paper tackles a problem of assigning individuals—each characterized by multiple attributes such as skills, experience etc.— and tasks —each characterized by required skills and durations— to teams, while ensuring balanced teams and workload. We propose a Mixed-Integer Linear Programming (MILP) model, developed using the goal programming technique which includes three phases and, in each phase, one of the deviations (related to balancing) is minimized. In this way, the model can respond to varying priorities of organizations. To validate the effectiveness of our proposed model, computational experiments on problem instances of varying sizes, including up to 24-36 individuals and tasks are conducted—a scale commonly encountered in practice. Results demonstrate that our model can provide optimal or near-optimal solutions within a reasonable computational time. We also conduct experiments using single objective function (no priority defined). Results indicate that, compared to model with single objective function, our approach achieved better solutions in more than one-third of the experiments with respect to the prioritized components.

ABSTRACT. This study aims to predict future usage trends by analyzing e-bike usage in a province using time-series models. E-bikes, as an important component of sustainable transportation, increase urban mobility and reduce traffic congestion. In this context, accurate demand forecasting plays a critical role in the effective management and operational planning of bike-sharing systems. In this study, forecasts were first made using the univariate time-series model ARIMA using only bicycle usage data. Then, the multivariate model SARIMAX was applied by incorporating exogenous factors such as weather. The dataset included daily bicycle usage information and meteorological variables such as temperature, precipitation, and wind speed. Model performance was evaluated using RMSE and MAE metrics. The results revealed that model performance varied across regions. While the univariate model ARIMA performed best in four regions, the multivariate model SARIMAX yielded superior results in most regions. It was observed that incorporating exogenous factors such as weather into the model increased forecast accuracy. The findings contribute to optimizing field operations of bike-sharing systems and improving urban transportation planning. Thus, it becomes possible to use resources more efficiently and increase user satisfaction.