Theses supervised by Prof. Dr. Mehmet Akar
17 theses · Boğaziçi University, Tokat Gaziosmanpaşa University
Novel fault-tolerant distributed algorithms for approximate byzantine consensus
Reaching consensus in a network which contains faulty nodes is a critical problem in the eld of distributed and multi-agent systems. A distributed system which intends to do a certain task needs to display robustness against adverse behavior of some of its faulty nodes, known also as Byzantine nodes. In this thesis, a novel Mean- Select-Reduced (MSR) fault tolerant algorithm is proposed for achieving Approximate Byzantine Consensus. It is shown that the topological condition required for the success of the algorithm is more relaxed compared to the previous results. In contrary to results that appeared in the literature, it is proved that synchronicity of networks and presence of delay on communication paths do not change this condition. Subsequently, the convergence rate and time analysis for the proposed fault-tolerant algorithm is carried out and the results are extended to time-varying networks. In most of the fault-tolerant algorithms that have been introduced for Byzantine networks, it is assumed that each node has knowledge of the maximum number of faulty nodes, ft, in the network. In this thesis, we also propose a new family of algorithms which do not require this a priori information and evaluate their performance facing Byzantine failures.
Radio direction finding & smart antennas
Bu tezde, radyo frekansında yön tespiti (DF) ve Akıllı Antenler konuları incelenmiştir. DF, radyo frekansındaki (RF) telsiz verici kaynakların konumunu tespit etme yöntemidir. DF sistemi, RF bir işaretin yönünü tespit etmek için, işaretin anten dizisindeki her bir antende oluşturduğu faz ve genlik gibi karakteristik özelliklerini kullanır. Pratikte, Watson Watt, interferometri ve yüksek çözünürlük olmak üzere üç temel DF yöntemi tercih edilir. Bu tezde yeni bir yaklaşımla çok kanallı interferometrik DF sistemi üzerinde çalışılmıştır. Önerilen sistemin avantajları, dezavantajları ve problemleri ele alınmıştır. Anten açıklığının artması ile sistem performansının da artması benzetimlerle gösterilmiştir. Anten açıklığının, gelen işaretinin dalga boyunun yarısını geçtiği durumlarda, sistem birden fazla açı için aynı faz farkı değerlerine sahip olmaktadır. İlk kısımda yapılan asıl iş, önerilen yöntem ile bu belirsizliğin önemli bir aralıkta ortadan kaldırılmış olmasıdır. Böylece, anten açıklığı arttırılabildiğinden sistem performansının da artması sağlanmış, bir DF sisteminin tek bir anten dizisi ile çok daha geniş bir bantta çalışması sağlanmıştır. Tezde üzerinde durulan ikinci ana konu ise bir anten dizisindeki anten elemanlarına faz gecikmeleri ekleyerek hüzme yönünü belirleyen "Akıllı Anten" sistemleridir. Mobil haberleşmenin büyümesiyle akıllı antenlere olan ihtiyaç artmakta, akıllı antenlerin kullanımı hızla yaygınlaşmaktadır. Akıllı anten sistemi tasarımlarında sabit, anahtarlamalı ve adaptif olmak üzere üç temel hüzme-yönlendirme yaklaşımı olmakla birlikte bunların birbirleri üzerine artı ve eksileri vardır. Bu tezde, önerdiğimiz anahtarlamalı ve adaptif hüzme-yönlendirme metodlarını birleştiren çoklu adaptif yaklaşım modeli, anahtarlamalı ve adaptif yöntemlerin avantajlarını birleştirmektedir.
Anomaly detection in industrial processes using curriculum-based deep learning
Anomaly detection is increasingly drawing attention in research as anomalies encountered in the industry become commonplace and more complex. In industrial processes, high dimensional multivariate time series (MVTS) data is generated from sensors and actuators. To guarantee the safety of the equipment and high quality products, anomalies recorded by these components have to be detected. In this thesis, the use of deep learning for industrial anomaly detection is studied. Particularly, curriculum learning (CL), a training paradigm for machine learning models is comprehensively analyzed. CL follows the learning progression of humans and animals, in which the model is first trained with "easy" data and more "dıffıcult" data is gradually introduced as training advances. This thesis proposes data-based CL for MVTS anomaly detection and introduces the CL concept to the learner (model). These frameworks are applied to two different case studies, first using a benchmark secure water treatment dataset and then a real world textile dataset. In the data-based method, extensive experiments are conducted to determine the most important MVTS features for improved model performance. In the model-based framework, a simple Long Short-Term Memory Encoder-Decoder model is trained first and then a complex Multi-Scale Convolutional Recurrent Encoder-Decoder model in utilized in the final training round. These techniques are evaluated based on the F1 score, with a data-based approach producing the best results among the proposed designs.
Cluster consensus for networks with antagonistic interactions
Consensus refers to the problem of distributed control protocols that enable all agents in a network to reach an agreement on a certain variable or quantity. Cluster consensus is a more generalized form of consensus where, instead of all agents converging to a single common value, they divide themselves into multiple clusters, and agents within each cluster reach an agreement on a common value. This thesis addresses the cluster consensus problem in multi-agent networks operating over arbitrary directed graphs, both unsigned and signed, without imposing restrictive structural assumptions common in existing literature. The research investigates cluster formation and stability across diverse system dynamics, including higher-order linear, time-delayed, inherent nonlinear, and matrix-weighted systems. A major contribution is the development of a systematic framework, notably utilizing an extended graph representation, which enables the explicit determination of the number of clusters and their members for any given signed directed graph. The thesis provides necessary and sufficient conditions, stability analysis, and control design methods for these general networks and system types. Key findings include the explicit determination of clusters for arbitrary signed networks, handling delays and nonlinearities on general graphs, and analyzing matrix-weighted consensus without topological constraints. By overcoming restrictive structural assumptions and providing specific methods for cluster analysis on general directed networks, this research significantly enhances the understanding and solvability of cluster consensus problems for realistic multi-agent system applications. The theoretical results are supported by numerical examples.
Distributed group consensus in time-varying networks
In past decades, the problem of achieving a common value in a distributed network, or in other words, the distributed consensus problem, has drawn significant attention from researchers due to its rapidly increasing application area, such as mobile robots, clock synchronization, orientation and flight formation of unmanned air vehicles, smart power grids, and rendezvous problem. These applications involve the design and analysis of distributed consensus algorithms regarding the stability and convergence properties of the algorithm. Depending on the topology of the network, consensus can be achieved in the sense of all agents achieving a common state, where complete agreement is beheld, which is a well-studied problem in the literature, as well as consensus can be achieved in the sense of agents form clusters, where the full agreement does not happen. However, we may still claim a partial agreement state has been reached in terms of multiple equilibrium states. This thesis analyzes the multi-equilibria consensus problem for multi-agent networks under switching topologies in discrete time. The main contributions of the thesis are to give sufficient conditions for a network to achieve multi-equilibria consensus under dynamically changing topologies and to express the number of equilibria points in terms of the network's topological properties. It is also demonstrated that the proposed definition of the reduced graph and graph reduction method reduces the size of the network while preserving some critical structural and topological properties that are used to determine the number of clusters forming in the steady-state.
Analysis of vehicular platoon systems under various information flow topologies
Akıllı ulaşım sistemleri, sensör, iletişim ve bilgisayar teknolojilerindeki ilerlemeler sayesinde parlak bir gelecek vadetmektedir. Günümüzde araçlar otomatik park, çarpışma önleme ve adaptif hız sabitleme gibi gelişmiş sürüş destek sistemleri ile donatılmaktadır. Ayrıca, trafikteki diğer araçlara ve yol kenarı birimleri gibi dış sistemlere bağlanmalarını sağlayan gelişmiş iletişim sistemlerine de sahiptirler. Sonuç olarak, araç konvoy sistemleri otomotiv sektöründe öne çıkan konulardan biri haline gelmektedir. Araç konvoy sistemi, bilgi alışverişi yapısına dayalı olarak birbirini takip eden araçlar grubundan oluşmaktadır. Araç konvoy sistemleri, yol güvenliğini arttırmak ve trafik sıkışıklığını azaltmak ile kalmayarak, yakıt ekonomisinde de iyileştirme potansiyeli taşımaktadır. Bu çalışma, farklı bilgi akışı topolojileri altında araç konvoyları koordine etmenin getirdiği zorluk ve fırsatları incelemektedir. Bu bağlamda, araç dinamikleri, merkezi olmayan kontrolörler, bilgi akış topolojileri ve geometrik formasyonlar araştırılmaktadır. Çalışmada, yayılan ağaç varsayımı altında tek ve çift yönlü bilgi akış topolojileri incelenmektedir. Ayrıca, merkezi olmayan kontrolörlerin etkisini anlamak amacıyla araç konvoy sistemlerinde doğrusal bir P-kontrolörü ve doğrusal olmayan bir integral kayan kipli kontrolör uygulanmaktadır. Çeşitli araç konvoy sistemi senaryolarını test etmek için MATLAB yazılımı kullanılmaktadır. Araç konvoy sistemi performansı, kararlılık analizi ve yakınsama süresi ile ölçülmektedir. Ek olarak, yayılan ağaç içermeyen bilgi akışı topolojileri üzerinde çalışılmaktadır. Yayılan ağaç içermeyen bilgi akış topolojileri birincil ve ikincil katman alt grafik algoritmaları kullanılarak incelenmektedir. Bu durumda birden fazla araç konvoyu oluştuğu gösterilmektedir.
Cooperative adaptive cruise control algorithms for vehicular platoons based on distributed model predictive control
Over the past decade, autonomous driving and driver assistance systems have become popular research topics with the aim of reducing traffic congestion, driver labor and rate of accidents. This has changed the requirements and priorities of today's transportation perception and reveals a need for an increased level of complexity. Changes in autonomous driving algorithms as a result of this differentiation have led the platoon system applications turn into a difficult control problem when evaluated together with bidirectional communication topologies and delays. Furthermore, the increased number of vehicles added to the platoon requires higher computational power. Using distributed controllers in such applications where response time is important helps us get more reliable results. The main objective of this thesis is to investigate Cooperative Adaptive Cruise Control Algorithms for Vehicular Platoons using Distributed Model Predictive Control (DMPC) under various communication links including unidirectional and bidirectional and also to propose a solution for a pre-known communication delay. Existing studies show that DMPC provides zero steady-state error with a fast response time under unidirectional communication topologies, however bidirectional links and communication delay effect are not adequately addressed. This thesis presents DMPC algorithms based on the information coming from the follower vehicles and delayed information received from the leader vehicle. It is shown that DMPC is the right tool to be utilized to both model and manage bidirectional communication. Simulation results demonstrate that the steady state error caused by the communication delay is successfully handled.
Multi agent intersection management for autonomous vehicles
Traditional transportation systems cause traffic congestion especially at the intersections as the number of vehicles keeps increasing. This is also the main reason of air pollution and time wasted. Most of the people lose their time and money because of traffic congestion. Thanks to recent research on autonomous vehicles, intelligent transportation and wireless communication systems, efficient traffic management at the intersections with multi-agent scheduling methods will be possible. The main objective of this thesis is intersection coordination for multi-agent systems by using time-based optimization and Model Predictive Control (MPC) methods while considering fuel economy at the intersections. Existing results show that these methods are efficient in comparison to the traditional methods when all the vehicles are autonomous. However, better trajectory planning can improve the total delay of the system. Besides, including fuel economy in the optimization function can also decrease fuel consumption which would be good for both humanity and nature. In this thesis, the effect of trajectory planning and different communication ranges on time-based optimization method is studied. It is shown that a wider communication range and better trajectory planning provide less time delay. Another contribution of this thesis is to propose centralized and decentralized MPC algorithms by including fuel consumption related costs in the objective function. As a result, fuel consumption is decreased at the expense of an increase in the time delay. In simulations, it is also observed that centralized MPC performs better than decentralized MPC.
Joint frequency/power update algorithms for self-organizingfemtocell networks
Heterogenous networks are resorted as one of the most promising ways for meeting the rapidly increasing data demand. As the smallest members of heterogenous networks, femto base stations also carry a huge potential for increasing the service quality in indoor areas. However, their unplanned deployment by the end users increases the possibility of having dense femtocell networks with unknown topologies. Therefore, self–organizing methods have great importance in resource allocation of femtocell networks. In this thesis, power and frequency allocation problems are studied for OFDMA femtocell networks. First, we present a power update algorithm for the general wireless networks as an alternative for a well known power control algorithm from the literature. Then, this algorithm and another power control algorithm from the literature are extended for the OFDMA femtocell networks. As opposed to the previous versions, which are applicable only in networks where each base station can have at most one user, the extended algorithms can be used by base stations that have more than one user. Additionally, a frequency allocation scheme is developed in order to increase the maximum achievable SINR in femtocell networks. Furthermore, by merging this scheme with the proposed power update algorithms, we present two joint frequency/power update algorithms with increased performance. Convergence and optimality analyses of all the proposed algorithms are carried out and illustrated with numerical results.
Distributed consensus based adaptive ramp metering algorithms in freeway systems
In recent years, traffic management has the attention of many researchers due to the public's increasing demand for fast, efficient, and convenient means of travel. In the literature, there has been a vast amount of research and improvement on traffic management control in freeways in various forms such as optimization, consensus protocols, and nonlinear control. The objective of this thesis is to develop a distributed consensus based ramp metering algorithm with the coordination of the traffic network. The applications of the distributed consensus algorithms in the literature indicate the importance of design and analysis of consensus protocols with which the agents in the systems achieve a common objective by exchanging information. Contrary to existing studies on coordinated ramp metering, centralized and decentralized on-ramp flow control algorithms are proposed without assigning priorities to on-ramps with bottlenecks at the upstream cells. Subsequently, the on-ramp flow control algorithm is improved with a consensus based density control algorithm which provides a smooth traffic density in the traffic infrastructure. At the final step, the mainstream inflow control is utilized in order to achieve the decided on-ramp flow and the traffic density. The averaging based and minimum consensus protocols are used to control the mainstream inflow. The performance and the convergence speed analysis of the proposed algorithms are evaluated with numerical simulations.
Vibration based condition monitoring of pumping systems in textile dyehouses
Condition monitoring of pumps is very important due to their critical role in industrial plants. This thesis focuses on vibration based fault detection and diagnosis of the flexible impeller pumps which are mostly used in pharmaceutical, food processing and textile finishing plants. The common problems of this type of positive displacement pumps are associated with the impellers and the cavitation is the main cause of impeller wear and damage. In the literature, there are many studies focused on the condition monitoring of positive displacement pumps. Nevertheless, none of these studies is particularly concentrated on the flexible impeller pumps. In this study, firstly, the data acquisition system has been devised for collecting vibration and motor current signals from the experimental setup. The separation ability of the features extracted from the vibration and motor current signals collected under different pump conditions has been investigated. The analysis regarding motor current signals has shown that they are not that distinctive to separate the conditions experienced by flexible impeller pumps. The classifiers exploiting features based on the time domain, frequency domain and time-frequency domain representations of vibration signals have been trained. The results regarding the performances of trained multi-class support vector machine and feedforward neural network classifiers have been presented as well. The findings of the thesis show that the feedforward neural networks exploiting wavelet variance features perform very well for classifying the flexible impeller pump conditions focused on. Eventually, a digital signal processing chain for fault detection and diagnosis of flexible impeller pumps has been proposed and realized in the embedded hardware.
Model-based and model-free control algorithms for textile processes
Textile processes consist of several control loops that require accurate reference tracking. One of the most crucial control loops is the temperature control where the temperature of the corresponding medium must track the reference value with sufficient accuracy to obtain a high-quality textile product. Even though several studies can be found on designing control algorithms for industrial processes in the literature, none of them focus particularly on the aforementioned textile processes. In this study, in order to achieve successful control, several adaptive control algorithms are developed. In addition, corresponding processes are modelled, and a simulation environment is built to increase the speed and safety of development works. Modelling is realized by dividing the corresponding process into several regions of operation, preparing sub-models for each region and building a composite model by combining these sub-models. A simulation environment is created by examining currently used control algorithms and process dynamics. The simulations of designed models result in significant accuracies. A model-based control algorithm, based on the Model Predictive Control (MPC) approach that utilises previously designed process models, is developed and verified in the simulation environment. Two model-free control algorithms, referred to as Adaptive PI Control and Error Predictive Control (EPC), are developed and verified not only in the simulation environment but also in the field.
Resilient distributed algorithms for solving linear algebraic equations in faulty networks
Various methods have been developed to solve linear algebraic equations distributively over multi-agent networks. Most studies consider that all agents are trustworthy and utilize all the received data from their neighbors throughout the process. Nevertheless, cooperation between non-faulty agents is disrupted if faulty agents intrude into the network. This thesis aims to develop algorithms to detect all faulty agents in the network without prior knowledge of the number of faulty agents. We study four fault models: random-state, fixed-state, single-faced, and double-faced and propose fault detection procedures according to the characteristics of these fault models. First, we introduce a method in which each agent can determine its neighbors' system of equations if it receives sufficient solution estimations from neighboring agents. By utilizing this method, we propose a synchronous discrete-time distributed detection algorithm for the perfectly synchronized agents in terms of their event times. On the other hand, the event time sequences of different agents are not always assumed to be synchronized. Therefore, we also propose an asynchronous discrete-time distributed fault detection algorithm to analyze the effect of the asynchronous event times of agents. Also, we discuss the applicability of our detection algorithm in continuous-time systems. Moreover, complexity analyses for the proposed algorithms are carried out. Theoretical results are also illustrated by numerical examples.
Resource allocation methods for next-generation networks
In 5G and beyond systems, the number of connections and data traffic is expected to grow significantly. To support the ever-increasing requirements, new solutions such as Mobile Edge Computing (MEC), Non-Orthogonal Multiple Access (NOMA), and Heterogeneous Networks (HetNets) are studied extensively. This thesis focuses on the resource allocation methods for uplink Hybrid NOMA for MEC offloading, downlink Hybrid NOMA, and downlink HetNets. First, a joint resource allocation that minimizes the total energy consumption of users for uplink Hybrid NOMA MEC Offloading is proposed. By solving the joint optimization problem, we propose a novel optimal Hybrid NOMA scheme referred to as Switched Hybrid NOMA for power and time allocation. Subsequently, we propose an algorithm to solve the sub-channel allocation (SCA) problem. We demonstrate that the proposed methods outperform the results in the literature analytically and by simulations. Then, we switch to the downlink communication, and study a resource allocation scheme that minimizes the total weighted energy consumption in the network. We propose a novel optimal Hybrid NOMA scheme for two users and then extend this idea to multiple users. Via simulations, we demonstrate that the Hybrid NOMA method outperforms Orthogonal Multiple Access (OMA) and NOMA methods. Afterward, we investigate novel distributed sub-channel and power allocation algorithms for HetNets. We introduce an SCA algorithm that minimizes the effective interference experienced by users in the network. Then, a distributed algorithm for power allocation is proposed, and a joint resource allocation method is constructed by combining the proposed algorithms, which outperforms the existing methods in the literature.
Design & application of low power interior permanent magnet motors
This thesis presents the design, analysis, experimental achievement and performance of an Interior Permanent Magnet (IPM) motor for use in low power range applications. IPMs are suitable for use in household appliance motors, pumps, small electric vehicles, air compressors motor and industrial fan motors. IPMs can significantly reduce power consumption compared to alternative motor types of similar power. In this thesis, an IPM has been designed and tested for use in low power electric applications. The IPM design aims to achieve maximum torque and power density at high efficiency while keeping losses low. The three basic components of the motor design, stator, rotor and permanent magnets, are considered separately with the aim of high efficiency. In order to minimize the cogging torque and thus improve the dynamic performance of the motor, a distributed winding structure is used in the stator windings. Permanent magnets are used in the rotor to provide high torque density in IPM. The magnets used are selected from high strength materials for structural integrity. The motor performance was analyzed by Finite Element Method and the results obtained were verified with the tests performed in the experimental study. According to the results obtained, it is revealed that the IPM shows a good dynamic performance with low cogging torque at high efficiency.
Elektrikli araç uygulamaları için eksenel akılı sabit mıknatıs destekli senkron relüktans motor tasarımı ve analizi
Elektrikli araçlar (EAR), modern ulaşımda emisyonları ve fosil yakıt bağımlılığını azaltmak için önemli bir çözüm olarak ortaya çıkmaktadır. EAR'lerin temelinde yer alan çekiş motorları, çeşitli sürüş koşullarında yüksek performans, verimlilik ve kompaktlık sağlamalıdır. Bu çalışma, üretilebilirlik, yüksek performans, mekanik ve termal kararlılığa odaklanarak EAR uygulamaları için optimize edilmiş bir elektrik motorunun tasarımını ve analizini sunmaktadır. Bu süreçte Eksenel Akılı Sabit Mıknatıs Destekli Senkron Relüktans Motor (EA-SMd-SRM) tasarımı ve analizleri yapılmıştır. Düşük hacimli, yüksek tork ve yüksek güç yoğunluğu için EA çift hava aralıklı motor tasarımı tercih edilmiştir. EA-SMd-SRM model yüksek akım yoğunluğuna sahip cebri sıvı soğutma (soğutma ceketi) yöntemi ile desteklenmiştir. EAR'ler için yapılan bu tasarım çok amaçlı optimizasyon ve çok fizikli analizlerle EA-SMd-SRM modelin elektromanyetik, mekanik ve termal yeterlilik seviyesi belirlenmiştir. EA-SMd-SRM tasarımı 3 000 dev/dk nominal hızda, 96.1 Nm tork, 30 kW güç ve %93.8 verim elde etmiştir. Bu motorun maksimum 10 000 dev/dk hıza, 253.1 Nm torka ve 65 kW güce ulaştığı görülmektedir. Bu çalışmada çift stator (ÇS) EA-SMd-SRM modelin rotoru seri üretime uygun bariyer yapısıyla tasarlanmıştır. EA motorlarda mıknatıs ve bariyerlerin dış çaptan içe doğru azalan konik yapısı yeni bariyer tasarımı ile değiştirilmiş ve hacimsel değişimi önlenmiştir. Yeni bariyerler çok amaçlı genetik algoritma (GA) kullanılarak optimize edilmiştir. Model son aşamada soğutma ceketi yöntemi ile soğutulmuş, nominal 30 kW güç değerinde motor sargılarında ortalama sıcaklığı 73.83 °C ve ortalama mıknatıs sıcaklığı 66.44 °C olarak analiz edilmiştir. EA-SMd-SRM modelin verim haritası çıkarılmış, sabit tork ve sabit güç bölgesi elde edilmiştir.
Eksenel akılı senkron relüktans motor tasarımı, prototip üretimi ve testleri
Bu tez çalışmasında, literatürde çalışması yapılmamış özgün bir çalışma olan, rotorunda çoklu sayıda bariyere sahip Eksenel Akılı Senkron Relüktans Motor (EA-SRM) tasarlanmış, prototip üretimi yapılmış ve son olarak performans testleri tamamlanmıştır. EA-SRM tasarımının ön çalışmasında, rotorunda iki farklı topolojide bariyer tipi, çoklu sayıda bariyer yapısı ve üç farklı yalıtım oranında analitik olarak hesaplamaları yapılmış ve 3D Sonlu Elemanlar Yöntemi (SEY) ile modellenmiştir. Analitik olarak en iyi sonucu veren modelde, Genetik Algoritma (GA) tabanlı optimizasyon analizi ile bariyer geometrisi değiştirilerek tork dalgalanması değeri düşürülmüş, yapısal analizi ve termal analizleri de tamamlanıp prototip motor üretilmiştir. Üretilen prototip IEC TS 60034-30-2 kriterlerine göre test edilmiş ve IEC 60034-30-1:2014 IE4 Süper Premium verim sınıfına sahip olmuştur. Tork, tork dalgalanması, güç faktörü, mil gücü, akım başına tork ve verim parametreleri analiz edilerek benzetim sonuçları ile karşılaştırılmıştır. Sonuç olarak tamamlanmış bu tez çalışması, rotorunda kafes bulunmaması sebebiyle bakır kayıplarının olmaması ve mıknatıssız yapısıyla maliyet-bakım masrafının düşüklüğü, yüksek verim ve tork istenilen uygulamalarda iyi bir alternatif olmuştur. Eksenel yapısı ile boyut problemi yaşanan durumlarda radyal akılı motorların yerini almaya aday hale gelmiştir.