Yazar "Bayir, Raif" seçeneğine göre listele

Listeleniyor 1 - 3 / 3
  • Küçük Resim
    Öğe
    Development of a Mecanum-Wheeled Mobile Robot for Dynamic- and Static-Obstacle Avoidance Based on Laser Range Sensor
    (Korean Inst Intelligent Systems, 2020) Matli, Musa; Albayrak, Ahmet; Bayir, Raif
    This study aims to present an idea about the practical consequences of using mobile robots with Mecanum wheels. For mobile robots, an approach is proposed to avoid obstacles without location and map information. This approach is presented using a series of developed solutions. This article shares the process on how a set of discussed conceptual methodologies can be applied as well as their practical results. This method is provided using fuzzy logic and gap tracking. LIDAR is used to recognize obstacles around the mobile robot. By using the LIDAR, the robot detects gaps around it and moves according to fuzzy logic. The fuzzy logic consists of three inputs, an output, and 45 rules. The first of the membership functions represents the membership function that replaces the obstacle. The second membership function calculates the distance to the obstacle. The final login membership function is used to determine the angle between the obstacle and robot view. The output membership function represents the membership function that moves the robot. The results are analyzed under three different scenarios with five different experiments for each scenario. The results show that the mobile robot can avoid obstacles without location and map information. We believe that the proposed method can be used in mobile robots such as guard and service robots.
  • Küçük Resim
    Öğe
    Modeling of migratory beekeeper behaviors with machine learning approach using meteorological and environmental variables: The case of Turkey
    (Elsevier, 2021) Albayrak, Ahmet; Ceven, Suleyman; Bayir, Raif
    In this study, migratory beekeeping behavior, which is an important form of beekeeping, has been modeled. Modeling was performed in conditions of Turkey. Modeling was made by considering food sources (nectar / pollen) and meteorological variables (temperature, humidity, number of rainy days, number of cloudy days and sunshine duration) for Turkey in which migratory beekeeping carried out in a different form than in developed countries. The main output in migratory beekeeping is honey production. Considering honey production, modeling has been made with the food sources and meteorological variables that have the greatest effect on honey production. Since the data set developed for modeling consists of relatively few samples, the ensemble learning approach was preferred from the machine learning approaches. Random Forest and Decision Tree algorithms, which are among the ensemble learning techniques, were used. As a result, the migratory beekeeping behavior was correctly classified at a rate of 92%. As a result of classification of Turkey's 81 provinces in five different categories, it was concluded that 33 provinces are suitable for migratory beekeeping at different times of the year. These 33 provinces are regions in the good and very good categories. In the next stage, thematic maps were produced for migratory beekeepers. Maps were produced for each month of the year. Thus, a guidance and information system has been obtained for migratory beekeepers.
  • Küçük Resim Yok
    Öğe
    Towards Safer and Greener Roads: A Hybrid Eco-safe Driving Assessment Using Fuzzy Logic and Machine Learning Algorithms
    (Springer Heidelberg, 2025) Ceven, Suleyman; Albayrak, Ahmet; Bayir, Raif
    This study addresses the critical global challenge of road traffic fatalities (1.19 million annually, World Health Organization (WHO) 2023) by developing an innovative hybrid model that simultaneously evaluates both eco-driving and safety aspects-a significant advancement over existing approaches that treat these factors separately. Our methodology combines fuzzy logic classification with machine learning (Multilayer Perceptron (MLP), Support Vector Machines (SVM), and Ensemble Learning) to analyze OBD-II data from 16 drivers, achieving remarkable results with 97.5% accuracy (MLP), 96.3% (SVM), and 97.2% (Ensemble). These performance metrics substantially improve upon previous works like (92% accuracy) and (94.2% accuracy), while our five-category classification system provides more granular insights than conventional binary safe/unsafe assessments. The practical impact is demonstrated through a 23.7% reduction in harsh braking incidents and 18.3% improvement in fuel efficiency among top-performing drivers, with real-time processing capabilities (< 50 ms latency) that enable immediate Advanced Driver Assistance Systems (ADAS) integration. This research contributes both methodologically-through its novel hybrid architecture combining interpretable fuzzy rules with machine learning precision-and practically by offering actionable tools to address the 90% of accidents attributed to driver behavior. The framework's effectiveness across multiple validation scenarios and its potential for geographical adaptation as a significant advancement in driver behavior analysis and road safety improvement.