LEE- Uçak ve Uzay Mühendisliği-Doktora
Bu koleksiyon için kalıcı URI
Gözat
Son Başvurular
1 - 5 / 26
-
ÖgeMulti agent planning under uncertainty using deep Q-networks(Graduate School, 2024-04-29)The extensive popularity of commercial unmanned aerial vehicles has drawn great attention from the e-commerce industry due to their suitability for last-mile delivery. However, the organization of multiple aerial vehicles efficiently to deliver the given set of goods within the existence of no-fly zones, numerous warehouses, limited fuel, and uncertainties are still a problem for traditional algorithms. The main challenge of planning is scalability, since the planning space grows exponentially with respect to the number of agents, and it is not efficient to let human-level supervisors structure the problem for such large-scale settings. With the recent advancements in deep reinforcement learning, algorithms such as Deep Q Networks (DQN), had unprecedented success in solving single-agent decision-making problems. Extension of these algorithms to multi-agent problems such as multi-drone delivery is very limited due to scalability issues. This work proposes an approach that improves the performance of DQN on multi-agent drone delivery problems by utilizing state decompositions for lowering the problem complexity, curriculum learning for handling the exploration complexity of delivery environments, and genetic algorithms (GA) for searching efficient packet-drone matching across the combinatorial solution space. The performance of the proposed method is shown in a multi-agent delivery by drone problem that has $10$ agents and $\approx10^{77}$ state-action pairs. Comparative simulation results are provided to demonstrate the merit of the proposed method. Compared with the conventional DQN schemes, and recently developed utility decomposition techniques, the proposed genetic algorithm-aided multi-agent DRL outperformed the rest in terms of scalability and convergent behavior. The prior techniques become intractable quickly at a large number of agents within the context of delivery by drone problem. The basic DQN algorithm fails to find a solution for three agents in a 10x10 drone delivery scenario within a reasonable number of steps, but the deep correction method successfully converges after approximately 1 million Bellman updates. Furthermore, applying the deep correction method increases the learning capacity to five agents and converges around 35 million Bellman updates. However, using this method does not lead to convergence with ten agents in a manageable way. With powerful computing resources, it becomes clear that while single-agent models set an initial computational standard, increasing the number of agents introduces complexity, as seen through immediate convergence difficulties in a three-agent DQN setup. Although there is promise with three- and five-agent configurations using Deep Correction, the model with ten agents exceeds the threshold for convergence within 24 hours, emphasizing the delicate balance between agent quantity and computational feasibility. The utilization of drone delivery simulation presents intricate challenges, including restricted airspace, fuel limitations, and the pick-and-place scenario. The study demonstrates that employing a method involving packet distribution through genetic algorithms effectively minimizes the complexity in resolving tasks for 10 agents within 5.74 minutes. Subsequently, the reduced problem is handled by deep Q-network inference models with Curriculum Learning and Prioritized Experience Replay, achieving execution times measured in milliseconds. This two-fold approach skillfully learns the dynamic nature of delivery problems without requiring prior domain knowledge input amid uncertain environmental conditions prone to altering actions. Furthermore, visual evidence at various time steps during execution illustrates how integrating GA-based packet distribution empowers the proposed base DQN model with Curriculum Learning and PER framework to tackle scenarios involving 10 agents – an accomplishment deemed unattainable by other explored solutions within reasonable time frames and computational resources. In conclusion, the combination of deep reinforcement learning and genetic algorithms provides a promising approach for efficient and effective delivery with multi-agent drones under uncertainty.
-
ÖgeAdvanced visual odometry and depth estimation techniques for unmanned aerial systems (UAS) in U-Space environments(Graduate School, 2024-12-11)This thesis explores advanced techniques in visual odometry (VO) and depth estimation for Unmanned Aerial Systems (UAS), specifically within the context of U-Space environments. U-Space, as a European initiative, aims to ensure the safe, efficient, and secure integration of UAS into airspace. This work contributes to this goal by addressing two critical aspects of UAS navigation: precise visual odometry and reliable depth estimation. Chapter 1 - Introduction: The introduction presents the context of U-Space, outlining its evolution and the services it offers, with a focus on emergency management. The challenges of autonomous contingency planning in UAS operations are highlighted, particularly in relation to visual odometry and depth estimation. Chapter 2 - Canonical Trinocular Feature-Based Visual Odometry: This chapter proposes a novel trinocular camera configuration to enhance VO for UAS. The research compares two trinocular setups—inline and 45-degree—with traditional binocular setups, testing them in various scenarios (horizontal, vertical, hybrid, and long). The results demonstrate that the 45-degree trinocular configuration with a standard lens offers significant improvements in both accuracy and computational efficiency, reducing the computational effort to 40\% of that required by binocular systems while delivering more accurate results. However, when a fisheye lens is used, the benefits are less pronounced, particularly in vertical and long scenarios. Chapter 3 - Fine-Tuning Monocular Depth-Estimator Artificial Neural Networks Trained on Synthetic RGB-D Datasets for Real Scenes: This chapter addresses the challenge of depth estimation for UAS using monocular cameras, which are cost-effective but typically less reliable than stereo cameras. The research investigates the effectiveness of fine-tuning deep-learning models trained on synthetic data with small real-world datasets. The results show that complete fine-tuning of all model parameters, as opposed to just the decoder, yields the best performance, especially when the available real data is limited to less than 12.5\% of the data required for optimal model performance. This finding is crucial for applications where only limited real-world data is available. Conclusion: The thesis concludes that the proposed trinocular VO configuration significantly enhances the accuracy and efficiency of UAS navigation, particularly in complex U-Space environments. Additionally, it establishes the importance of fine-tuning depth estimation models with real-world data, even when such data is scarce, to improve the reliability of UAS in operational scenarios. These advancements contribute to the broader goal of integrating UAS into airspace, ensuring they can operate safely and effectively under various conditions.
-
ÖgeInvestigation of impact behavior of polycarbonate panels under projectile impact loading(Graduate School, 2024-07-24)This study embarks on a comprehensive exploration of the damage inflicted upon polycarbonate panels by a spherical steel projectile, employing a combination of numerical simulations and experimental investigations. Within the scope of this study, projectile impact tests are launched to the panel center and 160 mm away from the panel. All panel thickness is determined as 2.1 mm. In addition, all polycarbonate panels are measured as 500 x 500 mm2. Single curved panels were designed to have a radius of 1000 mm and 2000 mm and were produced in accordance with these determined dimensions. The trade name of the polycarbonate material used is Lexan 9030. Furthermore, all panel designs were made using Catia. The material properties crucial for the analysis are meticulously acquired through a series of tests, including the split Hopkinson pressure bar test (SHBT), static tensile test (STT), and static compression test (SCT). Static tensile and compression tests were carried out at ITU Aerospace Engineering Composite Structure Laboratory. Tensile and compression tests were performed according to ASTM D638 and ASTM D695 standards, respectively. According to these standards, static tensile tests were performed at 5 mm/min. Static compression tests were carried out at a speed of 1.3 mm/min. Split Hopkinson bar tests were performed to determine the material properties at high strain rates.
-
ÖgeA feedback star identification algorithm via regularized pattern recognition using a unique feature extraction(Graduate School, 2024-07-11)This thesis presents a star identification algorithm integrated with preprocessing. Star sensors, which are highly reliable for attitude determination use of spacecrafts and satellites, relies on star identification algorithms. The star identification algorithm proposed in this study is capable of functioning either in lost-in-space method or recursive method. Both methods utilize a unique feature extraction scheme. This novel approach of feature extraction method extracts a single vector from each captured image instead of treating each star as a separate object. This cumulative approach aims to save a significant amount of memory space while taking the entire catalog into account for elevated accuracy. A database containing stars from the catalog is constructed using the unconventional features extracted from each corresponding field-of-view. The databases may differ in size and detail dependent on the parameters of overlapping ratio and brightness threshold. These parameters have a significant effect on accuracy and complexity of the method. The method aims to estimate the inertial boresight vector and the rotation angle about it. This is a novel approach that is carried out by matching frames but not matching individual stars, star pairs, star triangles or star polygons. Both star identification methods rely on pattern recognition and regularization successively. First, a 1NN classifier is used to perform a coarse estimation with limited accuracy specified by the characteristics of the database with predetermined parameters. The coarse estimation is exactly the database vector that is most similar to the observation vector. Subsequently, a dictionary is generated using the neighbor database vectors of the most similar database vector. The final estimation is obtained by conducting a regularization method for fine estimation. A solution coefficient vector is yielded through regularization. The estimates of boresight vector and rotation angle are retrieved using the solution coefficient vector. This is the output of the lost-in-space star identification method. Since the lost-in-space algorithm is very sensitive to false stars, an additional false star filtering algorithm is developed. This algorithm is based on density-based clustering. A disparity list is created using two successive image frames. After estimating true stars by implementing density-based clustering on the disparity list, false stars are removed. Using the successive frames containing only estimated true stars, an affine transformation matrix is obtained by a regression analysis procedure. In order to overcome the issues tackling the lost-in-space star method, the recursive star identification method is developed. Apart from the algorithmic structure taken from the lost-in-space method, it possesses an update mechanism that ensures usage a much smaller portion of the database to reduce computational complexity and average run time. Also, the false star filtering avoids sensitivity to false stars.
-
ÖgeHelikopter ana rotor uç geometrisinin aeroakustik açıdan optimizasyonu ve rotor hareketlerinin aeroakustik etkilerinin incelenmesi(Lisansüstü Eğitim Enstitüsü, 2024-05-31)Helikopter hem mekanik hem de aerodinamik açıdan karmaşık bir uçan makinedir. İtki/taşıma üretiminin dönel doğası, helikopterleri daha yüksek güç ihtiyacıyla birlikte çok titreşimli ve gürültülü hale getirir. Bir helikopter tasarlanırken genel maliyetin yanı sıra güç talebi, titreşim ve gürültünün de en aza indirilmesi gerekir. Dolayısıyla bu dört parametre, helikopter tasarımının en temel Araştırma ve Geliştirme etkenleri olarak değerlendirilebilir. Helikopterlerin yük ve yolcu taşımacılığında sivil amaçlı kullanımı, gelişen sosyal ve ticari ihtiyaçlar çerçevesinde özellikle büyüyen metropol şehirlerde hızlı bir artış göstermiş, buna bağlı olarak çevre için önemsenebilir bir gürültü kaynağı olmaya başlamıştır. Helikopterlerin şehir kullanımındaki artışı , yönetmeliklerde tanımlanan kabul edilebilir gürültü seviyelerini sağlamasını gerektirmektedir . Ayrıca helikopterler, hareket kabiliyetleri dolayısı ile nokta hedefler için kritik öneme sahip olduklarından, askeri amaçlı kullanım için de vazgeçilmezdir. Askeri amaçlı kullanımda, gerek intikal gerek boşaltmada helikopterin genel gürültü seviyesinin gizlilik unsuru için bozucu bir etkiye sahip olduğu da temel bir gerçektir. Bu iki durumdan hareket ile helikopterin asıl gürültü kaynaklarından biri olan ana rotorun, gerek askı durumunda gerekse ileri uçuşta rotor uç geometrisinin gürültüye katkısının incelenmesi ve düşük gürültü yaratacak geometrinin tahmini, gürültü kirliliğini azaltma anlamında büyük önem taşımaktadır. Bu gerçeklik, ABD ve Avrupa Devletlerini yeni gürültü yönetmelikleri oluşturmaya, DNW (Alman Nederland Rüzgar Tüneli) ve Onera (Fransa) gibi Deneysel Araştırma Kurumlarına yeni araştırma projeleri için yatırımlar yapmaya teşvik etmiştir. Rotor tarafından üretilen gürültünün kaynakları; kalınlık gürültüsü, yükleme gürültüsü, yüksek hızda atım gürültüsü, kanat-girdap etkileşimi gürültüsü, geniş bant gürültüsü olmak üzere çeşitlilik arz eder. Bu sebeple helikopter tasarım çalışmalarında, rotor araç akış alanının ve buna bağlı gürültünün doğru sayısal analizi hala çok zorlu bir iştir. Aeroakustik tahminler, deneysel ölçümlerle eşleşen kabul edilebilir gürültü rakamlarına ulaşmak için on milyonlarca veya daha fazla sayıda akışkan ağ örgüsü ve yüksek kapasiteli bilgisayar altyapısı gerektirir ki bu durum tasarım çalışmalarının süresini uzatmakta ve sonuca ulaşmayı zorlaştırmaktadır. Yapılan bu çalışmalara ait detaylar 1.2. Literatür Araştırması bölümünde ayrıntılı olarak ele alınmıştır. Tez kapsamında, bir helikopterin ana rotor uç geometrisinin rotor kaynaklı gürültüye etkisinin incelenmesi ve rotor ucu şeklinin en az gürültüyü yaratacak biçimde optimize edilmesi ve rotor hareketlerinin yaratılan gürültüye etkisini tahmin etmek üzere matematik model ve sayısal çözümün oluşturulması hedeflenmiştir.