120
Archived Theses
1
DOIs Assigned
1%
DOI Rate
Departman Tezleri
indexedOpen AccessTRAkıllı üretim sistemlerinde çoklu hata teşhisi için hibrit derin öğrenme yaklaşımları
Endüstriyel üretim süreçlerinde akıllı sistemlerindeki ürünün yüzey arızalarının erkenden teşhis edilmesi, kalite kontrol süreçleri otomasyonu, maliyetlerin azaltılması ve kalite standartlarının korunması açısından kritiktir. Akıllı üretim sistemlerinde çoklu hata teşhisi ile birbirinden farklı türdeki arızaların saptanması sağlanmaktadır. Endüstri 4.0 ile üretim hatlarında veri miktarının artması sonucunda, klasik makine öğrenmesi yaklaşımları ile arızanın teşhis edilmesi yetersiz olması sonucunu doğurmuştur. Bu durumda, daha gelişmiş analiz yöntemlerine ihtiyaç artmıştır. Bu kapsamda bu tezde, hibrit derin öğrenme mimarileri ile farklı türdeki arızalı veri örneklerinin sınıflandırılması yüksek doğruluklarla sağlanmıştır. Bu tezde, metal ve kumaş yüzeylerde meydana gelen çoklu kusurların belirlenmesi için hibrit derin öğrenme yaklaşımları incelenmiştir. Önerilen yöntemde, farklı derin öğrenme mimarileri birlikte kullanılarak hem nesne tespiti hem de segmentasyon işlemleri gerçekleştirilmiştir. Bu çalışmada, arızalı görüntülerden elde edilen özellikler konvolüsyonel sinir ağları ile çıkarılmış, ardından bu özellikler farklı sınıflandırıcılar ile analiz edilmiştir. Sınıflandırma aşamasında kullanılan derin mimariler Vision Transformer (ViT) ve dikkat tabanlı derin sinir ağları gibi yenilikçi yaklaşımlardır. Gerçekleştirilen deneysel çalışmalar sonucunda görülmüştür ki önerilen yöntem farklı tipteki arızaları yüksek performans elde edilmiştir. Bu çalışma, akıllı üretim sistemlerinde kalite kontrol süreçlerinin iyileştirilmesine katkı sağlamaktadır.
Karmaşık olay işleme sistemleri için gerçek zamanlı olay ilişkilendirme ve alarm kural madenciliği modelleri
World is creating the same quantity of data every two days, as it created from up until 2003. Evolving data streams are key factor for the growth of data created over the last few years. Streaming data analysis in real-time is becoming the fastest and most effective way to get useful information from what is happening right now, thus allowing organizations to take action quickly when problems occur or to detect new trends to improve their performance. Data stream analytics is needed to manage the data currently produced from applications such as sensor networks, measurements in network monitoring, mobile traffic management, web click streams, mobile call detail records,social media posts/blogs and many others. Stream data analytics is hard because data are temporally ordered, fast changing, massive and potentially infinite. In order to cope with the challenges of data stream mining, in this thesis two main contributions are discussed. Both of them summarize the high volume streaming data and present meaningful, actionable information to end users. The first one is finding ?event correlations? over the data stream pairs on real GPS data of public transportation buses. The second one is alarm sequence rule mining, with a new parameter called ?time confidence?, that helps automatically set time-window values for registered rules and also reduces the generated alarm rule count.
Üç boyutlu konumlandırma problemleri için yeni bir veri eşleştirme çözüm yöntemi: One-poınt ransac wıth epıpolar constraınt
The problem of Localization or Simultaneous Localization and Mapping has received a great deal of attention within the robotics literature, and the importance of the solutions to this problem has been well documented for successful operation of autonomous agents in a number of environments. Of the numerous solutions that have been developed for solving the problems, many of the most successful approaches continue to either rely on, or stem from noise ltering techniques, especially the Extended Kalman Filter method or Particle Filtering methods. Localization problems are downgraded to a data association problem after using mentioned lters. This topic has also received a great deal of attention in the robotics literature in recent years, and various solutions have been proposed. In the thesis, rst mostly studied methods, such as Joint Compatibility, Sequential Compatibility Nearest Neighbor, Joint Maximum Likelihood, one point RANSAC and epipolar consistency, are studied. As the second part of the thesis a new method is presented. One-Point RANSAC with Epipolar Constraint (OPRF) is based on RANSAC and epipolar geometry. Later the performance and consistency of the method will be compared to epipolar consistency solution.
Bir yabancı dilde bilgi çıkarımı sonuçlarının semantik temsilciliğinin sağlanması ve bir uygulama
Son yıllarda İnternet'in hızla gelişmesi ve yaygın kullanımı ile Web, dünyada erişilebilir en geniş veri kaynağı haline gelmiştir. Internet'teki bilgi yığınları aşırı şekilde artarken, Web ziyaretçilerinin isteklerine uygun hizmetlerin sağlanabilmesi, Web site yapısının iyileştirilebilmesi, geliştirilebilmesi ve etkin olarak kullanılabilmesi gibi amaçları sağlamak için kullanılan Bilgi Çıkarım Metodu, gün geçtikçe daha çok ilgi çeken bir konu haline gelmiştir.En büyük bilgi kaynağı olarak görülen İnternet'te, bilgiye ulaşabilmenin önündeki en büyük engel, içindeki bilgilerin %90'ının doğal dille oluşturulmuş olmasıdır. Bilgisayarların doğal dili henüz yeterli bir şekilde anlayamıyor olmalarından dolayı sayfalardaki bilgileri bulup çıkarmak yine insanlara düşmektedir.Bu çalışma, doğal dillerden bilgiyi elde etmek için kullanılan metotlardan biri olan Bilgi Çıkarımı'na (Information Extraction) dayanarak geliştirilmektedir. Bu kapsamda doğal dillerin bileşenleri olan sıfatlar, isimler, fiiller, zarflar vb. kaynaklar kullanılarak nitel bir sınıflandırma çıktısı elde edilmes amaçlanmaktadır. Bu çalışmada odaklanılan yabancı dil Arapça olarak seçilmiş ve teknoloji olarak Java diliyle geliştirilmiş olan GATE platformu tercih edilmiştir.
Tümce kökenli konu modelleme
Fast augmentation of large text collections in digital world makes inevitable to automatically extract short descriptions of those texts. Even if a lot of studies have been done on detecting hidden topics in text corpora, almost all models follow the bag-of-words assumption. This study presents a new unsupervised learning method that reveals topics in a text corpora and the topic distribution of each text in the corpora. The texts in the corpora are described by a generative graphical model, in which each sentence is generated by a single topic and the topics of consecutive sentences follow a hidden Markov chain. In contrast to bag-of-words paradigm, the model assumes each sentence as a unit block and builds on a memory of topics slowly changing in a meaningful way as the text flows. The results are evaluated both qualitatively by examining topic keywords from particular text collections and quantitatively by means of perplexity, a measure of generalization of the model. Keywords: probabilistic graphical model, topic model, hidden Markov model, Markov chain Monte Carlo.
Contextproxy: Ağ tabanlı bağlamdan-haberdar hizmetleri ve uygulamaları desteklemek için yerden-haberdar HTTP proxy sunucusu
ABSTRACT CONTEXTPROXY: A LOCATION-AWARE HTTP PROXY SERVER TO SUPPORT WEB BASED CONTEXT-AWARE SERVICES AND APPLICATIONS Alper R. Uluçınar M.S. in Computer Engineering Supervisor: Asst Prof. Dr. David Davenport January, 2005 The pervasion of computing in our physical world promises more than the ubiquitous availability of computing resources; totally new and exciting interaction schemes are to be explored. Context-awareness, one of the most important aspects of ubiquitous computing, enables applications that make use of their users' context to provide dynamically adapting information and services to their users or to other applications. Although the technological infrastructure to support ubiquitous and context-aware applications is being deployed rapidly, the standards and the best practices for the interactions of various components in a context-aware application are still missing. In our work we have developed a location-aware HTTP proxy server, called ContextProxy that runs on the popular Symbian platform. ContextProxy acts as a standard HTTP proxy server from the client application's perspective but it augments the service request of the client with the available location information while submitting the request to the service provider. This allows the existing nomadic applications to immediately become location- aware if they can be configured to make use of a standard HTTP proxy which is a common scheme for web based applications. And also it is possible to write new nomadic applications without considering the context-awareness aspect at the service requestor level. The contextual information added by ContextProxy can then be utilized by the service provider to dynamically adapt its services according to the service requestor's context. Keywords: Context- Aware/Nomadic Computing, Symbian, Bluetooth, GSM m
Tasarsız ve algılayıcı ağlarda yer belirleme algoritmalarının değerlendirilmesinde kullanılan metrikler üzerine bir araştırma
In ad-hoc and sensor networks, the location of a sensor node making an observationis a vital piece of information to allow accurate data analysis. GPS is anestablished technology to enable precise position information. Yet, resource constraintsand size issues prohibit its use in small sensor nodes that are designed tobe cost efficient. Instead, most positions are estimated by a number of algorithms.Such estimates, inevitably introduce errors in the information collected from thefield, and it is very important to determine the error in cases where they leadto inaccurate data analysis. After all, many components of the application relyon the reported locations including decision making processes. It is, therefore,vital to understand the impact of errors from the applications? point of view. Todate, the focus on location estimation was on individual accuracy of each sensor?sposition in isolation to the complete network. In this thesis, we point out theproblems with such an approach that does not consider the complete networktopology and the relative positions of nodes in comparison to each other. Wethen describe the existing metrics, which are used in the literature, and also proposesome novel metrics that can be used in this area of research. Furthermore,we run simulations to understand the behavior of the existing and proposed metrics.After having discussed the simulation results, we suggest a metric selectionmethodology that can be used for wireless sensor network applications.
Yapısal olmayan eşler arası bilgisayar ağlarında katkısız katılımı engelleme
The peer-to-peer (P2P) network paradigm has attracted a significant amount ofinterest as a popular and successful alternative to traditional client-server modelfor resource sharing and content distribution. However, researchers have observedthe existence of high degrees of free riding in P2P networks which poses a seriousthreat to effectiveness and efficient operation of these networks, and hence totheir future. Therefore, eliminating or reducing the impact of free riding on P2Pnetworks has become an important issue to investigate and a considerable amountof research has been conducted on it.In this thesis, we propose two novel solutions to reduce the adverse effects of freeriding on P2P networks and to motivate peers to contribute to P2P networks.These solutions are also intended to lead to performance gains for contributingpeers and to penalize free riders. As the first solution, we propose a distributedand localized scheme, called Detect and Punish Method (DPM), which dependson detection and punishment of free riders. Our second solution to the free ridingproblem is a connection-time protocol, called P2P Connection Management Protocol(PCMP), which is based on controlling and managing link establishmentsamong peers according to their contributions.To evaluate the proposed solutions and compare them with other alternatives,we developed a new P2P network simulator and conducted extensive simulationexperiments. Our simulation results show that employing our solutions in a P2Pnetwork considerably reduces the adverse effects of free riding and improves theoverall performance of the network. Furthermore, we observed that P2P networksutilizing the proposed solutions become more robust and scalable.
3 boyutlu çizim ile gerçek zamanlı saç modellemesi ve dinamik simülasyonlar
Hair has been an active research area in computer graphics society for over a decade. Different approaches have been proposed for different aspects of hair research such as modeling, simulating, animating and rendering. In this thesis, we introduce a sketch-based tool making use of direct manipulation interfaces to create hair models and furthermore simulate the created hair models underphysically based constraints in real-time. Throughout the thesis, the created tool will be analyzed with respect to dierent aspects of the problem such as hair modeling, hair simulation, hair sketching and hair rendering. Finally, the results of a usability test of the tool are provided.
Mekanlar arasında rotalar sunmak için katılımcı bir sistem
Many systems, such as in-car GPS devices and airline company web sites, provide route information between locations. Although such systems are used widely and can provide route information successfully, users of these systems cannot contribute to the data entry process. In these systems, data is entered by the administrators and these systems cannot take advantage of the route expertise of their users.In this work, we present a collaborative system, which provides routes between locations upon user queries. The data in the system is entered by the users of the system. We present a model which is containing locations, links between locations and relationships between locations (containment, neighborhood and intersection) in order to store the data. For the route finding purpose, we present a customized version of the A* search algorithm. This customized version, named A*CD (A* for Collaborative Data), uses heuristics for estimating the cost remaining to the target location while processing the nodes. A*CD can also provide alternative routes, exclude certain link types in the searches according to user preferences and handle the problems associated with multiple stop transportation lines. As the cost models, we use duration and financial cost.We also present the intuitive connections concept. Even if a route does not exist between the selected locations, the system can provide a route with missing links. The gap(s) between the disconnected locations are filled by the help of the relationships between locations.In order to evaluate the performance of the A*CD algorithm, we present automated tests. These tests show that the costs of the routes that are provided by the A*CD algorithm are close to the actual shortest routes. In order to demonstrate the intuitive connections concept, we also present manual test queries.