LEE- Kontrol ve Otomasyon Mühendisliği-Doktora

Bu koleksiyon için kalıcı URI


Son Başvurular

Şimdi gösteriliyor 1 - 5 / 10
  • Öge
    Differential flatness-based fuzzy controller design for aggressive maneuvering of quadcopters
    (Graduate School, 2023-05-04) Güzay, Çağrı ; Kumbasar, Tufan ; 504142105 ; Control and Automation Engineering
    This study presents a new differential flatness-based single input fuzzy logic controller structure for aggressive maneuvering control alongside its real-world application on a nano quadcopter. We propose both type-1 and interval type-2 single input fuzzy logic controllers as the primary controllers in the flight control system, which are built on the concept of differential flatness. Today, quadcopters are used for a wide variety of applications and purposes such as aerial photography, search and rescue operations, surveying and mapping, inspection, agriculture, and emergency response. Quadcopters are getting more and more well-liked in the commercial and consumer markets as a result of the rising demand for their usage areas. Additionally, the dimensions of quadcopters have significantly changed along with the rapid development in contemporary technology. As a result, we can discuss quadcopter types such as mini, micro, or nano. Nano quadcopters, the smallest ones, are lightweight, more portable, and easier to operate and maneuver with high agility since they are constructed with small-scale rotors and frames.
  • Öge
    Deep reinforcement learning for partially observable markov decision processes
    (Graduate School, 2022-07-19) Haklıdır, Mehmet ; Temeltaş, Hakan ; 504102110 ; Control and Automation Engineering
    Deep reinforcement learning has recently gained popularity owing to its many successful real-world applications in robotics and games. Conventional reinforcement learning faces a substantial challenge in developing effective algorithms for high-dimensional environments. The use of deep learning as a function approximator in reinforcement learning is a viable solution to overcome this challenge. Furthermore, in deep reinforcement learning, the environment is often thought to be fully observable, meaning that the agent can perceive the true state of the environment and so act appropriately in the current state. Most real-world problems are partially observable and the environmental models are unknown. Therefore, there is a significant need for reinforcement learning approaches to solve these problems, in which the agent perceives the state of the environment partially and noisily. Guided reinforcement learning methods solve this issue by providing additional state knowledge to reinforcement learning algorithms during the learning process, thereby allowing them to solve a partially observable Markov decision process (POMDP) more effectively. However, these guided approaches are relatively rare in the literature, and most existing approaches are model-based, which means that they require learning an appropriate model of the environment first. In this thesis, we present a novel model-free approach that combines the soft actor-critic method and supervised learning concept to solve real-world problems, formulating them as POMDPs. We evaluated our approach using modified partially observable MuJoCo tasks. In experiments performed on OpenAI Gym, an open-source simulation platform, our guided soft actor-critic approach outperformed other baseline algorithms, gaining 7∼20% more maximum average return on five partially observable tasks constructed based on continuous control problems and simulated in MuJoCo. To solve the autonomous driving problem, we focused on decision making under uncertainty, as a partially observable Markov decision process, using our guided soft actor-critic approach. A self-driving car was trained in a simulation environment, created using MATLAB/SIMULINK, for a scenario in which it encountered a pedestrian crossing the road. Experiments demonstrate that the agent exhibits desirable control behavior and performs close to the fully observable state under various uncertainty situations.
  • Öge
    2-step indoor localization for "smart AGVs"
    (Graduate School, 2022-06-17) Yılmaz, Abdurrahman ; Temeltaş, Hakan ; 504142101 ; Control and Automation Engineering
    With the fourth industrial revolution, in other words, Industry 4.0 (I4.0), the transition from traditional mass production to personalized production started in factories. One of the components of the next-generation factories compatible with I4.0 is cyber-physical systems (CPSs). Smart manufacturing islands, smart warehouses, and smart material-handling vehicles are examples of CPSs. The material handling vehicles employed in today's factories, such as automated guided vehicles (AGVs), are not ready for use in smart factories, as the digital transformation has not been completed and the vehicles are not equipped with software to perform fully autonomous operations. In smart factories, it is aimed that the new generation AGVs will do all the planning themselves while performing a given task. Thus smart AGVs will be able to use the whole free space in the factory instead of being restricted to the routes reserved for them. With this development, it will be possible to increase flexibility and efficiency in production. There may be no physical difference between the traditional and smart AGVs, but thanks to the capabilities of the embedded software, smart AGVs will be able to operate autonomously. One challenging problem to be overcome for smart AGVs to effectively realize an assigned logistic task is localization. Although localization is an extensively studied topic for both indoor and outdoor environments, there are still open problems. Considering the logistics problem, the localization problem can be divided into three in the general sense. The first is global localization, which means determining where the smart AGV is in the environment at the time the vehicle wakes up. The second problem is position tracking, which means updating the pose information depending on the movements of the robot, while the instantaneous pose of the robot is known. The third and last problem is the kidnapped robot problem, which occurs when the robot is moved from one place to another without informing. Cases that reduce the reliability of the calculated pose, such as instantaneous skidding, slipping, and crashing an object, can also be addressed under this problem. The localization approach to be utilized in smart factories is supposed to overcome these three sub-problems. There are two main tasks in a logistic operation. The first is the docking stage, which covers the cases of taking a load to the smart AGV or dropping the load of the smart AGV. At this stage, the aim is to reach the target (destination) where the load will be taken or left with industrial standards. With I4.0, reaching the target with sub-centimeter precision has become a goal. Therefore, estimating the pose with high accuracy and precision is expected from the docking localization algorithm. The second is the delivery stage, which covers carrying the load to the destination in the fastest and safest way in the parts outside the docking region. It is not essential to follow the planned route exactly in this stage, so rather than the high accuracy of the localization approach, showing similar positioning performance in the entire operating field is more important. Within the scope of this thesis, different localization algorithms have been proposed for the delivery and docking stages. In addition, a probabilistic decision mechanism that determines the boundary between the delivery and docking stages is designed. A variant of the particle filter-based Monte Carlo Localization (MCL) approach, Self-Adaptive MCL (SA-MCL), is taken as the basis localization method for the delivery stage. The main reason for choosing SA-MCL is that it can solve all aforementioned sub-problems of localization. While performing the traditional SA-MCL global localization task, it uses energy maps and assumes that all range sensors are uniformly placed on the robot in energy map generation. However, this assumption is not valid for many real applications, such as AGVs with two-dimensional (2D) laser scanners front and rear. Moreover, three-dimensional (3D) sensing technology is developing day by day with the widespread use of autonomous vehicle technology. With the ellipse-based energy model proposed in this thesis, the energy map-generating part of the traditional SA-MCL has been updated to overcome both of these constraints. The pose estimation accuracy of the SA-MCL approach performs more or less the same across the entire environment, making it suitable for the delivery stage. However, since the pose estimation accuracy level is proportional to the grid dimensions of the occupancy map, it may not be possible to reach the expected sub-centimeter precision within the docking region in large areas such as factories. Therefore, it was decided to use a scan matching-based precise localization algorithm in the docking region, and for this purpose, the affine iterative closest point (ICP) algorithm was adapted to the localization problem. To make the developed method robust against factors such as noises, disturbances, and/or outliers, the correntropy criterion was utilized while constructing the cost function of affine ICP. As a result, an updated SA-MCL method with an ellipse-based energy model is proposed for the solution of global localization, position tracking, and kidnapped robot problems in the delivery stage. On the other hand, an affine ICP-based precise localization approach is presented for position tracking in the docking stage. However, the boundary between the delivery stage and the docking stage may not be clear. For example, limiting the docking stage to a zone very close to the target may require extra maneuvers to tolerate positioning errors during the delivery stage due to the physical constraints of smart AGVs. If a larger area is specified as a docking stage, it may not meet the expectations since the performance of the precise localization approach may decrease further away from the target. For this reason, there is a need for a switching mechanism that can be adapted specifically to the application and decides whether to switch from the delivery stage to the docking stage. Since the pose estimation performance of the SA-MCL-based localization approach is roughly similar on the entire map, the deciding factor in the transition to the docking stage is the performance of the precise localization method used in the docking stage. In the literature, it is emphasized that the amount of overlap between matched point sets is supposed to be above 50% for the scan-matching-based methods to yield successful results. Within the scope of the thesis, a correntropy-based similarity rate definition, which gives better results than the overlap ratio calculation methods in the literature, is presented and utilized as the decision parameter of the switching approach. To avoid instabilities, a gap is left according to Hysteresis curve behavior while switching from the delivery stage to the docking stage or vice versa. Within the scope of the thesis, the two-stage localization method developed for the next-generation AGVs to be used in smart factories has been experimentally tested on a differential drive mobile robot. First, the ellipse-based energy model addition to the SA-MCL method has been verified by field tests, and its superiority in global localization has been demonstrated. Then, the affine ICP-based localization method used in the docking stage has been tested over nine separate real-world scenarios and it has been shown that it is possible to compute pose with sub-centimeter precision and reach the target at industrial standards. In addition, an affine ICP method, which is not available in the literature, was proposed, and the point set matching performance was demonstrated over synthetic point sets. After validating its performance in point set registration, it was also used for precise localization. Finally, the whole system was tested together. The delivery was carried out with improved SA-MCL, and the switching point from delivery to the docking stage was determined by the decision mechanism. As seen through three different scenarios, it is possible to complete the localization tasks in the delivery and docking stages in the smart factories by using the proposed methods.
  • Öge
    Controller design methodologies for fractional order system models
    (Graduate School, 2022-01-25) Yumuk, Erhan ; Güzelkaya, Müjde ; 504152101 ; Control and Automation Engineering
    Fractional order calculation deals with cases where the derivative and integral order is non-integer. Although the notion of fractional order was introduced at the end of the 17th century, this concept in engineering was employed after the first quarter of the 19th century. Its first application to control engineering areas was made after the second quarter of the 20th century. Since fractional calculus is a generalized version of integer order calculus, it provides great flexibility in system modeling and controller design. In other words, fractional calculus offers three different combinations in terms of the controller and system types: Fractional order control for integer order system, Fractional order control for fractional order system, and Integer order control for fractional order system. In this respect, fractional calculus is an excellent tool to describe a control system compared to integer order calculus. Besides the flexibility, the notion brings more complexity to system modeling and controller tuning. Therefore, many studies over the last half-century have been trying to overcome these difficulties. Numerous real-time systems have nonlinear characteristics and high-order system dynamics. In literature, simple integer-order models, i.e. the first and second order with or without time delay, are used to represent system dynamics.
  • Öge
    Doğrusal olmayan sistemler için model öngörülü kontrol yöntemine ters optimal kontrol yapısının katılması
    (Lisansüstü Eğitim Enstitüsü, 2021-08-02) Ulusoy, Lütfi ; Güzelkaya, Müjde ; 504122103 ; Kontrol ve Otomasyon Mühendisliği
    Optimal kontrol probleminin amacı, bazı kontrol ve durum kısıtlamalarını sağlayacak ve bir başarım kriterini optimize edecek şekilde bir kontrol giriş fonksiyonu veya kontrol kuralı elde etmektir. Buna rağmen, optimal kontrol kuralı, kısıtsız ve doğrusal durumlarda bile oldukça kolay ve analitik olarak bulunamaz. Optimal kontrol kuralının çözümünün Hamilton-Jacobi-Bellman (HJB) denklemini çözmeyi gerektirdiği iyi bilinen bir gerçektir ki bu son derece zordur. Dahası, doğrusal olmayan sistemlerin çoğu için analitik bir HJB çözümü mevcut değildir. Sistem doğrusal olduğunda ve başarım kriteri ikinci dereceden olduğunda, HJB, belirli durumlarda analitik olarak çözülmesi zor olabilen bir Riccati denklemi olarak ortaya çıkar. Bu zorlukların üstesinden gelmek amacıyla önceden belirlenmiş bir sonlu ufuk için mevcut sistem durumunu, başlangıç durumu olarak atayarak, sistem modeli yardımıyla optimal kontrol problemini tekrar tekrar ve ardışıl olarak çözmek düşünülmüştür. Bu stratejiyi kullanan kontrol yaklaşımları, Model Öngörülü Kontrol (MÖK) olarak adlandırılır. Bu yaklaşımda, sistemin gelecekteki davranışı, sistem modeli kullanılarak tahmin edilir ve kontrol işareti, anlık sistem durumlarına göre her kontrol ufku için tekrar tekrar yenilenir. Öte yandan, HJB problemini çözmek yerine bize farklı bir bakış açısı sağlayan bir başka yaklaşım ise Ters Optimal Kontrol (TOK) teorisidir. TOK, HJB denklemini çözmenin zahmetli görevinden kaçınarak, doğrusal olmayan optimal kontrol problemini çözmek için alternatif bir yaklaşımdır. Son yıllarda, birçok gerçek zamanlı uygulamada doğrusal olmayan optimal kontrol problemlerini çözmek için ters optimizasyon yaklaşımı giderek daha fazla kullanılmaktadır. Tezde, ilk olarak model öngörülü kontrol yaklaşımının optimal kontrol problemini ele alış biçimi anlatılmıştır. Önerilecek yöntem ile karşılaştırabilmek amacıyla, klasik model öngörülü yaklaşımlarından, doğrusal sistem modelini kullanan gradyant tabanlı MÖK ve doğrusal olmayan sistem modeli Runge-Kutta tabanlı MÖK (RKMÖK) yaklaşımları verilmiştir. Daha sonra ters optimal kontrol (TOK) yaklaşımları incelenmiş ve ayrık-zamanlı girişte-afin doğrusal olmayan sistemler için TOK problemini Kontrol Lyapunov Fonksiyonu (KLF) bulma problemine dönüştürerek çözen TOK yaklaşımı anlatılmıştır. TOK yaklaşımı için takip probleminde karşılaşılabilecek sorunlar üzerinde durulmuştur. Bu tezde ilk olarak, takip problemi sorunlarını çözebilmek amacıyla kontrol işareti ağırlık matrisinin her bir elemanı için sistem durum değişkenlerine bağlı bir sigmoid fonksiyon önerilmiştir. Önerilen yaklaşımın başarımını gösterebilmek için klasik TOK yaklaşımıyla karşılaştırma yapılmıştır. Bu tez çalışmasında, ayrıca girişte-afin doğrusal olmayan sistemler için MÖK ve TOK yaklaşımları birleştirilerek yeni bir optimal kontrol yöntemi önerilmektedir. Gerçek hayatta ve literatürde karşılaşılan doğrusal olmayan sistemlerin ve sistem modellerinin çoğu, bazı doğrusal olmayan azaltma yöntemleri ile girişte-afin biçime dönüştürülebilir. Önerilen yöntemin temel özelliği, her kayan ufuk ve sonuç olarak yeni bir başlangıç koşulu için çözülmesi gereken MÖK optimizasyon problemini TOK problemi olarak ele alıp, bu TOK problemini tekrar tekrar çözmesidir. Bu yaklaşımda, sistemin gelecekteki davranışının tahminini elde etmek için sistem modeli kullanılır ve önceden belirlenmiş bir kontrol ufku için TOK yönteminden elde edilen kontrol işareti sisteme uygulanır. TOK probleminin çözümü aşamasında, belirlenmesi gereken aday kontrol Lyapunov fonksiyon matrisinin parametreleri, evrimsel Büyük Patlama-Büyük Çöküş (BP-BÇ) optimizasyon arama algoritması kullanılarak çevrim içi bir şekilde tahmin edilir. Önerilen kontrol yapısında, MÖK yaklaşımında her kontrol ufku için uygun bir KLF matrisinin aranması ile optimal kontrol problemi çözülmektedir. Diğer bir bakış açısından ise, MÖK yapısı TOK problemine dahil edilerek TOK problemi, her kayan ufkun başlangıcındaki farklı başlangıç koşulları kullanılarak tekrar tekrar çözülmekte ve böylece, TOK için çevrim içi bir düzeltme mekanizması elde edilmektedir. Bu yaklaşım ve literatürdeki diğer yöntemler kullanılarak top ve çubuk kontrol sistemi üzerinde benzetim çalışmaları ve gerçek zamanlı uygulama yapılmıştır. Elde edilen sonuçlar bazı kontrol başarım ölçütlerine karşılaştırılmış ve önerilen yaklaşımın başarımı değerlendirilmiştir.