Theses supervised by Prof. Dr. Ömer Morgül
19 theses · İhsan Doğramacı Bilkent University
DA/DA buck ve boost dönüştürücülerin ikinci derece kayan kipli kontrolünde DAD-optimize doğal anahtarlama yüzeylerinin kullanılması ve akım limitleme metotları
Second-order sliding mode control utilizing natural switching surfaces has demonstrated robust dynamic performance in the control of switching buck and boost converters for start-up, load changes, and large signal disturbances. However, practical implementation often encounters challenges such as the chattering phenomenon and large inductor currents during start-up or loading transients. The primary focus of this research is proposing potential solutions for these practical issues. Assuming the parasitic resistances, such as the on-state resistances of switches and the equivalent series resistances of output capacitors, are relatively low, the direct current resistance (DCR) of the inductor becomes the primary parasitic resistance that causes chattering due to its damping effect. For this reason, the DCR of the inductor is taken into account in the system dynamics from which natural switching surfaces are derived. By incorporating these surfaces into the control laws, the chattering effect caused by losses is mitigated while maintaining effective control. Furthermore, current limiting techniques are proposed for converter protection and reliability. These techniques are integrated into the control laws to manage large inductor currents effectively during transients, thereby preventing potential damage and enhancing the overall performance of the converter. Through theoretical analysis, geometrical representations, and simulation studies, a comprehensive framework for minimizing chattering and managing large inductor currents in the second-order sliding mode control of buck and converters integrating natural switching surfaces is established.
Basit doğrusal olmayan mekanik sistemler için yeni gözlemci tabanlı sürtünme tahmini ve kontrol yöntemleri
Friction is a common nonlinear phenomenon in inertial mechanical systems, often leading to undesirable effects such as stick-slip motion, hysteresis, and reduced tracking accuracy. Effective friction compensation is essential for enhancing robustness and achieving high precision in controlling such systems. Both model-based and non-model-based approaches have been widely utilized for friction compensation, with friction observers playing a significant role in estimating the friction acting on the system. This study focuses on observer-based adaptive estimation techniques, specifically employing the Friedland-Park observer to estimate the parameters of a friction model in an inertial system. The proposed approach aims to mitigate stick-slip motion and enhance tracking performance. The research evaluates various applications of the Friedland-Park observer for friction compensation and its implementation in high-order systems. Both two state and single-state observers are examined, alongside different system models and controllers, to ensure the robustness of the proposed methods. Performance metrics include velocity reference tracking accuracy and the compensator's responsiveness during velocity sign reversals, assessed under diverse reference input conditions. Simulation results demonstrate a significant improvement in velocity tracking accuracy, with up to a 65% reduction in tracking error. The proposed friction compensation methods effectively handle varying friction conditions, ensuring system robustness and precision. This study contributes a comprehensive approach to friction compensation, offering various design options for inertial mechanical systems. Future work will extend this research by incorporating a gyroscope model to address the noisy nature of inertial sensors and performing the hardware implementation of the proposed methodology on gyro-stabilized platforms to validate its practical applicability. This work can also be extended within the scope of system identification by developing a novel approach for estimating plant parameters through the adaptation of Friedland-Park observer equations.
Öğrenilebilir kesirli Fourier dönüşümü ile insan aktivitesi sınıflandırması
The human activity classification regarding micro-Doppler characteristics has been a captivating research area since the introduction of the micro-Doppler concept in various fields such as security surveillance, healthcare monitoring/diagnostics, and gait analysis. The extraction of micro-Doppler signatures through radar sensing and subsequent time-frequency analysis of humans or other targets enables precise and reliable characterization of their motion dynamics, allowing for robust classification even in cluttered or low-resolution environments. Compared to conventional vision-based or wearable sensing methods, radar systems are favored due to their inherent ability to preserve target privacy without capturing identifiable visual features. To extract the meaningful micro-Doppler signatures embedded in the radar data, assorted time-frequency analysis approaches have been employed in the literature, especially the Short-Time Fourier Transform (STFT), due to simplicity and computational efficiency. The Fourier transform-based approach might be broadened by utilizing the Fractional Fourier Transform (FrFT), which generalizes the classical Fourier transform by considering the continuum of infinitely many representations between the time and frequency domains. Thanks to the opportunity to work in intermediate domains provided by the Fractional Fourier transform, more distinguishable representations of the micro-Doppler signatures might be obtained. Motivated by the limitations of conventional approaches that depend on fixed time-frequency representations, this thesis proposes a deep learning-based classification methodology for the micro-Doppler signature classification of human activities based on the Fractional Fourier transform. Instead of using the singular usage of the fractional domain representations by searching for optimal transform order empirically, trainable Fractional Fourier transform blocks are engaged in different deep learning architectures. Spectrograms representing the time-frequency characteristics are constructed within the deep-learning architectures, aiming for optimum representations via trainable fractional Fourier transform blocks dynamically. The proposed approach is evaluated with multiple radar-based datasets, including both simulated and real-world measurements with different radars and configurations. A comprehensive set of experiments is conducted with various deep learning architectures, including single-branch CNN, LSTM, and GRU-based architectures, with the multi-branch configurations including the simultaneous use of different time-frequency representations obtained via fractional Fourier transform and multi-input Siamase-based models. The proposed approach is compared in detail with the conventional Fourier transform approach in distinct experiments, and the behavior of the fractional representations across the models and datasets is examined. As far as we are aware, this study is the first to combine a trainable Fractional Fourier Transform with micro-Doppler signature-based human activity classification. The experimental results demonstrate that the proposed approach consistently outperforms traditional Fourier Transform-based methods. Moreover, the simultaneous utilization of multiple time-frequency representations with distinct FrFT orders is, to the best of our knowledge, a novel contribution to the literature, yielding considerable performance improvements. It is believed that the findings of this study pave the way for future research on adaptive time-frequency analysis in radar-based sensing applications.
SLIP ve TD-SLIP modelleri için yeni bir adım planlaması
Spring Loaded Inverted Pendulum (SLIP) is a well-known model and an accurate descriptive tool, which can scientifically represent the dynamics of the legged locomotion. Torque actuated Dissipative SLIP (TD-SLIP), on the other hand, is fundamentally an enhanced version of the SLIP model. Inclusion of more realistic damping model and the hip torque actuation has led the researchers to develop a sufficiently better analytic approximation. This thesis proposes a new methodology to achieve footstep planning on the SLIP and TD-SLIP models, distinctly. It contributes a novel planning algorithm with respect to the constructed touchdown-to-touchdown map, and a novel recursive function to plan and execute the planning. The thesis provides a background information about the modelling and simulation of both of the used models, and an auxiliary function, which administers a derivative-free method to calculate the minimum of an input function. After defining the problems and the corresponding proposed solutions, the foundations of the preparation phase is established. This phase is fundamentally constructed to accumulate required information for the algorithm implementation and simulation phase. The main phase consists of subsections, which can be composed from the combination of following properties; planning type, as online and offline, policy type; as forward and backwards and output type; as based on distance or based on minimum step count. According to the stated problem, the planning is successfully realised not only for a single desired distance, but also an array of waypoints. In addition to this, the presented illustrations of different initial states show that the planning can also be constructed via any different initial touchdown state. Therefore, the obtained results are quite promising, since all of the cases and their combinations successfully reach the destinations with a negligible error value, which is less than 1%. Although, the offline planning type provides the results in a rapid way, the obtained data to use the plan requires much more space, which also increases dramatically when the step count (level) is incremented. In addition to this, the forward planning is faster than the backwards one, but they both generate very similar results.
Evrişimli yapay sinir ağları için bellek verimli filtreleme algoritmaları
Deployment of state of the art CNN architectures like Xception, ResNet and GoogleNet in resource limited devices is a big challenge. These architectures consist of many layers and millions of parameters. Moreover, they require billions of floating point operations to inference just an image. Therefore, memory space needed to store parameters and to execute them are the main constraints for efficient convolutional neural network architectures. In this thesis, we examine Winograd's minimal filtering algorithms to reduce number of floating point operations performed in convolutional layers. We reduce the number of multiplications x2.25 times without any accuracy loss. Moreover, we investigate, sparse and quantized Winograd's algorithms so that we can make conventional Winograd algorithms more memory efficient. We propose a linear quantization scheme to quantize weights of the networks more than 1-bit. We use ReLU activation function and Targeted Dropout which is a variant of Dropout to prune transformed inputs of Winograd algorithm. We binarize weights so that most arithmetic operations are converted to bit-wise operations. We conduct several experiments on CIFAR10 and CIFAR100 datasets and discuss the classification performances of both conventional and modified Winograd minimal filtering algorithms. We achieve less than 1.9% classification error with ReLU-ed Winograd CNN compared to conventional Winograd. We reduce memory requirements up to x32 times by binarizing weights of ReLU-ed Winograd CNN, and in return we incur around 2% accuracy loss. Lastly, for applications which are less tolerant to accuracy loss, rather than binarizing weights we quantize them to 2-bit, 4-bit and 8-bit. Our quantized ReLU-ed Winograd CNNs reach same accuracy levels as ReLU-ed Winograd CNN.
DA/DA buck ve boost dönüştürücülerin sınır kontrolünde doğal anahtarlamanın kullanılması
DC-DC converters are extensively used in many power electronics applications such as photovoltaic systems, wind energy systems, DC motor drives, mobile devices, electric vehicles, etc. Fundamental performance criteria in these applications include tight line and load regulation, low output voltage ripple, high efficiency and fast response to load uncertainties. Also, the trade-off between high performance and component sizes must be considered. In order to meet these requirements, a boundary control method is developed for the resistive loaded buck and boost DC-DC converters. First, normalized plant models are obtained for both converters. The normalization generalizes the controller design by making it independent of the circuit parameters. Then, natural phase plane trajectories of the systems are derived in the normalized domain. Using the natural trajectories of the converters as switching surfaces, special boundary control laws are defined. Switches in the systems are driven by control inputs generated according to the control laws. Via this boundary control method, the fast dynamic response is provided by utilizing passive components that take up the most space, namely inductor and capacitor, at their theoretical limits. This allows the overall circuit size to be kept small. Finally, the control laws are altered by a small factor so that in steady state, finite and controlled frequency operation and known ripple magnitudes of system states are obtained. In this way, a common problem in boundary control applications called chattering is eliminated. It is shown via simulations that the proposed controllers manage to recover from load and start-up transients by single switching action for both converters.
Zaman gecikmeli basit mekanik sistemlerin pozisyon kontrolünde sürtünme kestirimi için adaptif gözlemci tasarımları
Friction force/torque is a well known natural effect that can cause performance degradation or even instability in mechanical systems, although it sometimes can be disregarded in closed loop feedback design phase. Hence, friction modeling and cancellation methods can be vital to achieve desired robustness and performance criteria in position control problems. Basically, the topic of friction cancellation is divided into two main categories named model based and non-model based methods. Friction modeling is a broad area of research and there are lots of different modeling approaches in various complexities. Among these approaches, Coulomb Model is one the simplest yet fundamental models. Nevertheless, in some cases, being a classical static model, it is inadequate to exhibit the dominant friction components occurring at different motion stages such as break-away force, stick-slip motion, pre-sliding behavior or friction lag. Generally, dynamical models, i.e. LuGre Model, are more advanced as a result, they are better to describe such friction effects. Unfortunately, for these cases, the number of friction parameters are increased. In fact, there is a trade-o_ between model complexity and parameter identification. A desired system response may not be achieved when model parameters do not coincide with the existing friction coefficients. In this manner, precise identification of each parameter can be challenging when there are many of them. Besides, some of these parameters might be time varying due to environment, temperature, material properties, position, etc. Therefore, non-model based adaptive schemes are prevalent in the literature since these methods do not require any parameter identification. In this study, we focus on adaptive observer based friction compensation techniques and provide some stability conditions. First, we consider simple second order mechanical systems with or without time delay under Coulomb friction. To estimate the Coulomb friction, we first consider Friedland-Park observer. Then, some necessary conditions are stated to extend the estimation function in the observer structure to a larger class of functions. Especially measurement delay can be significant since observers estimate friction based on the velocity measurements. Therefore, it is proposed to employ a velocity predictor either based on numerical differential equation solvers or inverse Pade approximant when the existing time delay is large. What is more, a new observer design that considers friction and velocity error dynamics together is proposed as a novel contribution. Extensive MATLAB simulations are conducted to investigate the performances of proposed observers in a closed loop position control system with and without delay. To this end, Smith predictor and ITAE index-based designs are considered to utilize a position controller. In some of these simulations, LuGre model is preferred to mimic the actual friction instead of Coulomb friction in order to observe the effects of dynamic parameters. Moreover, some experiments are performed on DC motor platform driven by Arduino Uno microcontroller. Under the light of acquired results, observer based friction compensation improves the system performance even existing friction cannot be confined to Coulomb coefficient, especially when the implemented controller has low bandwidth. Also, in terms of practicability, it is an advantage that these observer structures do not require any parameter identification.
FMCW radar datası kullanılarak derin öğrenme ile insan etkinliği sınıflandırılması
Human Activity Recognition (HAR) has recently attracted academic research attention and is used for purposes such as healthcare systems, surveillance-based security, sports activities, and entertainment. Deep Learning is also frequently used in Human Activity Recognition, as it shows superior performance in subjects such as Computer Vision and Natural Language Processing. FMCW radar data is a good choice for Human Activity Recognition as it works better than cameras under challenging situations such as rainy and foggy conditions. However, the work in this field does not progress as dynamically as in the camera-based area. This can be attributed to radar-based models that do not perform as well as camera-based models. This thesis proposes four new models to improve HAR performance using FMCW radar data. These models are CNN-based, LSTM-based, LSTM- and GRU-based, and Siamese-based. For feature extraction, the CNN-based model uses CNN blocks, the LSTM-based model uses LSTM blocks, and the LSTM- and GRU-based model uses LSTM and GRU blocks in parallel. Furthermore, the Siamese-based model is fed in parallel from three different radars (multi-input). Due to the Siamese Network nature, parallel paths will have the same weight. On the other hand, after feature extraction, all models use dense layers to classify human motion. To our best knowledge, the Siamese-based model is used for the first time in multi-input data for the classification of human movement. This model outperforms the state-of-the-art models by using various features of radars operating at different frequencies in terms of classification accuracy.
Öğrenme temelli melez hareket kontrol sistem mimarisi ile yeniden yapılandırılabilir modüler yılan robot hareket kabiliyeti
Snake robots propose significant advantages especially for indeterminate, chaotic environments through their robustness and versatility against unforeseen conditions and scenarios. In addition to distinct locomotion characteristics of snake robots, their redundant structure provides also fault tolerant operation capacity. However, sophisticated and versatile locomotion characteristics and redundant body structure also bring difficulty for dynamic modelling and motion control of snake robots and, for this reason, generation of snake locomotion patterns have been an ongoing challenge. To address this point; a reconfigurable, modular snake robot is designed and modelled with fundamental electromechanical structure, joint and actuation subsystem and contact force modellings based on minimal requirements which are determined by presented mathematical analysis for snake locomotion. A hybrid motion control system architecture which is constituted with state-of-the-art reinforcement learning based algorithms and a cascaded PID controller which comprises command shaper and gain scheduling components is presented. While reinforcement learning based algorithms which indicate promising potential for generation of sophisticated behaviours are employed for generation of 2D snake gait patterns with corresponding reward function terms, locomotion capabilities are expanded to 3D space with the proposed cascaded PID architecture and possible high level planners. Various experimentations that cover comparison of different reinforcement learning algorithms, individual effects of specified reward function terms, locomotion of snakes which are composed by different number of modules, fault tolerant locomotion trainings for defective snake robots and realization of 3D operation scenarios are investigated. In this content, from a holistic perspective, future directions are drawn for potential physical realization of the electro-mechanical structure, mechanical design details for self-assembly mechanisms, further improvements of the training process and curriculum, and learning based multi-robot scenarios which cover swarms of differently configured snakes to realize collaborative tasks.
Mekanik sistemlerde gözlemci tabanlı sürtünme giderimi
In real life feedback control applications of mechanical systems, friction and time delays are two important issues that might have direct e ects on the performance of systems. Hence, an adaptive nonlinear observer based friction compensation for a special time delayed system is presented in this thesis. Considering existing delay, an available Coulomb observer is modi ed and closed loop system is formed by using a Smith predictor based controller as if the process is delay free. Implemented hierarchical feedback system structure provides two-degree of freedom and controls both velocity and position separately. For this purpose, controller parametrization method is used to extend Smith predictor structure to the position control loop for di erent types of inputs and disturbance attenuation. Simulation results demonstrate that without requiring much information about friction force, the method can signi cantly improve the performance of a control system in which it is applied.
Derin sinir ağlarında oto-kodlayıcının düzenlenmesi için terkinim ve seyreklik kullanımı
Deep learning has emerged as an effective pre-training technique for neural networks with many hidden layers. To overcome the over-fi tting issue, usually large capacity models are used. In this thesis, two methodologies which are frequently utilized in deep neural network literature have been considered. Firstly, for pre-training the performance of sparse autoencoder has been improved by adding p-norm of the sparse penalty term to an over-complete case. This effi ciently induces sparsity to the hidden layers of a deep network to overcome over- fitting issues. At the end of the training, features constructed for each layer end up with a variety of useful information to initialize a deep network. The accuracy obtained is comparable to the conventional sparse autoencoder technique. Secondly, the large capacity networks suff er from complex co-adaptations between the hidden layers by combining the predictions of each unit in the previous layer to generate the features of the next layer. This results to certain redundant features. So, the idea we propose is to induce a threshold level on the hidden activations to allow only the highest active units to participate in the reconstruction of the features and suppressing the e ffect of less active units in the optimization. This is implemented by dropping out k-lowest hidden units while retaining the rest. Our simulations confi rm the hypothesis that the k-lowest dropouts help the optimization in both the pre-training and fi ne-tuning phases giving rise to the internal distributed representations for better generalization. Moreover, this model gives quick convergence than the conventional dropout method. In classi fication task on MNIST dataset, the proposed idea gives the comparable results with the previous regularization techniques such as denoising autoencoders, use of rectifi er linear units combined with standard regularizations. The deep networks constructed from the combination of our models achieve favorably the similar state of the art results obtained by dropout idea with less time complexity making them well suited to large problem sizes.
Tekrarlı sinir ağı öğrenimi ile bacaklı hareket kontrolüne uygulanması
Use of robots for real life applications has an increasing trend in today's industry and military. The robot platforms are capable of performing dangerous and difficult tasks, which are not efficient when carried out by human beings. Most of these tasks require high motion ability. There are various robotic platform and locomotion algorithms which may solve a given task. Among these, the biped robot platforms promise high performance in realizing difficult maneuver due to their morphological similarity to legged animals. Thus, legged locomotion is highly desirable in order to perform difficult maneuvers in rough terrain environments. However both modeling and control of such structures are quite difficult due to highly nonlinear structure of the resulting equations of motion and computational load of inverse kinematic equations. Central nervous systems and spinal cords of animals take role in control of locomotion of animals together. For controlling such biped robotic platforms frequently used control algorithms are based on so-called Central Pattern Generators (CPG). The controllers based on CPG's can be realized in different ways which includes the utilization of neural networks. However CPG is only capable of imitating spinal cord type of reflex-based motions in locomotion because of their restricted parameter space to sustain stable oscillation. Fully recurrent neural networks have capability of controlling locomotion with a higher conscious level such as central nervous system, hence motion space can be enlarged. Unfortunately, training of recurrent neural networks (RNN) takes long time. Moreover, their behaviors may be unpredictable against untrained inputs and training process may encounter with instability related problems easily. In order to solve these problems, various acceleration and regularization techniques are tested in the neural network training and their successes were compared with each other. Furthermore, time constant and error gradient limitation methods are employed to sustain stable training and their benefits are discussed. Finally leg angles of walking biped robot are taught to a group of RNNs with different configurations by benefiting from training stability enhancing methods. The resulting RNNs are then used in biped locomotion by using a classical PD controller. After that, performance of resulting RNNs and their stable locomotion generation capabilities are evaluated and effects of configuration parameters are discussed in detail.
Mekanik sistemlerde sürtünme giderimi için deadbeat denetleyici ve kutup yerleştirme denetimi uygulaması
Friction is an almost unavoidable component of many mechanical systems. When not taken into account in designing control systems, the effect of friction may result in the degradation of controlled system performance. This thesis deals with the problem of designing a control system, for friction compensation in mechanical systems, via pole placement and deadbeat methodologies. Pole placement design is based on different performance measures and indices such as settling time, overshoot and ITAE. Deadbeat controller design is based on parameterization of Diophantine equations which depend on the reference signal to be tracked. System performance is analyzed on simulation level by the application of the two methodologies in a hierarchical feedback system structure, which provides both position and velocity control separately. Simulation results show that both methodologies provide acceptable performance as compared to the existing compensation schemes in literature and control performances are improved with respect to their accuracy of tracking. In addition, deadbeat controller is observed to be more promising in terms of minimum settling time.
Doğrusal ve zamanla periyodik olarak değişen sistemlerin harmonik transfer fonksiyonlar yoluyla tanılanması, kararlılık analizi ve kontrolü
Many important systems encountered in nature such as wind turbines, helicopter rotors, power networks or nonlinear systems which are linearized around periodic orbit can be modeled as linear time periodic (LTP) systems. Such systems have been analyzed and discussed from analytical viewpoint extensively in the literature. However, only a few method are available in the literature for the identification of LTP systems which utilize input/output measurements. Especially, due to obtaining analytical solutions for LTP systems are quite challenging, utilization of experimental data to identify, analyze and stabilize such systems may be preferable. To achieve this aim, the utilization of harmonic transfer functions (HTFs) of LTP systems can be quite helpful. In the first part of this thesis, we aim to obtain harmonic transfer functions (HTFs) of LTP systems via data-driven approach by using only input and output data of the system. In this respect, we first present the identification procedure of HTFs by using single cosine input signal with a specific frequency. However, because of the fact that this method requires multiple experiments in order to cover desired frequency range, we propose a formula for the sum of cosine input signal including different frequencies which their output components do not coincide. Then, we present the prediction performance of the estimated HTFs by using single cosine and sum of cosine input signals according to analytical solution of HTFs. In the second part of the thesis, our goal is to utilize harmonic transfer functions in order to analyze and design controllers which stabilize and enhance the performance of LTP systems. In this regard, we implement well known Nyquist stability criterion which is based on eigenloci of HTFs. As an illustrative example, we consider the well-known (unstable) damped Mathieu equation and design P, PD and PID controllers by using obtained Nyquist diagram. Finally, for the unknown LTP systems whose state space model may not be available, we seek to design a novel methodology, where we can obtain Nyquist plots of unknown LTP systems via input-output data analysis using the concept of HTFs. Then, we design PD controllers for the unknown LTP system by using Nyquist diagram in order to enhance the performance and increase the robustness. We illustrate the performance results of these controllers in time domain simulations.
Bacaklı robotlar için periyodik yürüme davranışlarının analizi ve kontrolü
The analysis, identification and control of legged locomotion have been an interest for various researchers towards building legged robots that move like the animals do in nature. The extensive studies on understanding legged locomotion led to some mathematical models, such as the Spring-Loaded Inverted Pendulum (SLIP) template (and its various derivatives), that can be used to identify, analyze and control legged locomotor systems. Despite their seemingly simple nature, as being a simple point mass attached to a massless spring from dynamics perspective, the SLIP model constitutes a restricted three-body problem formulation, whose non-integrability has been proven long before. Thus, researchers came up with approximate analytical solutions or they used some other different techniques such as partial feedback linearization for the sake of obtaining analytical Poincaré return maps that govern the motion of the desired legged locomotor system. In the first part of this thesis, we consider a SLIP-based legged locomotion model, which we call as Multi-Actuated Dissipative SLIP (MD-SLIP) that extends the simple SLIP model with two additional actuators. The first one is a linear actuator attached serially to the leg spring to ensure direct control on the compression and decompression of the leg spring. The second actuator is a rotatory one that is attached to hip, which provides ability to inject some torque inputs to the system dynamics, which is mainly inspired by biological legged locomotor systems. Following the analysis of MD-SLIP model, we utilize a partial feedback linearization strategy by which we can cancel some nonlinear dynamics of the legged locomotion model and obtain exact analytical solutions without needing any approximation. Having exact analytical solutions is crucial to investigate stability characteristics of the MD-SLIP model during its hopping gait behavior. We illustrate and compare the applicability of our solutions with open-loop and closed-loop hopping performances on various rough terrain simulations. Finally, we show how the MD-SLIP model can be anchored to bipedal legged locomotion models, where we assign two independent MD-SLIP models to each leg and investigate the system performance under their simultaneous but independent control. The proposed bipedal legged locomotion model is called as Multi-Actuated Dissipative Bipedal SLIP (MDB-SLIP) model. The key idea here is that we can still utilize the partial feedback linearization concept that we applied for the original MD-SLIP model and ensure exact analytical solutions for the MDB-SLIP model as well. We also provide detailed investigations for open-loop and closed-loop walking gait performance of the MDB-SLIP model on different noisy terrain profiles.
Tüm-kutuplu ve enküçük evreli doğrusal zamanda bağımsız ve doğrusal zamanla değişen sistemlerin çıkış regülasyonu
In this thesis, the problem of enabling the output of a system to track the referencesignals and reject the disturbances created by the same exogenous system isconsidered. This problem is widely known as Output Regulation Problem. Firstly,we propose a method for all-pole LTI systems by using relative degree propertyand then we apply the same method for minimum phase LTI systems along withsome modifications. In order to obtain controllers for a minimum phase LTIcase, the system is converted into an all-pole system by employing the inversesystem as the first part of the controller. Then using the method that we usedin all-pole cases, we obtain the second part of the controller. Combining thesetwo controllers gives us an overall controller which solves the output regulationproblem. This method for LTI systems is then extended to all-pole and minimumphase LTV systems. However, in order to apply the same methodology wehave to make some assumptions on LTV systems. For minimum phase cases, thenormal form is obtained by applying certain Lyapunov transformations and thenminimum phaseness is defined in accordance with the normal form. Furthermorewe show that, similar to minimum phase LTI cases, pole / zero cancelations occurbetween the inverse system and the original system in minimum phase LTVcases. The method that we develop depends on analytical calculation of thecontroller and gives a certain degree of freedom to change the transient behaviorof the system by only changing some controller parameters.
Doğrusal olmayan bazı sistemlerin en küçük kareli destek vektör makineleriyle tanılanması
The well-knownWiener and Hammerstein type nonlinear systems and their various combinations arefrequently used both in the modeling and the control of various electrical, physical, biological, chemical,etc... systems. In this thesis we will concentrate on the parametric identification and control ofthese type of systems. In literature, various identification methods are proposed for the identificationof Hammerstein and Wiener type of systems. Recently, Least Squares-Support Vector Machines(LS-SVM) are also applied in the identification of Hammerstein type systems. In the majority ofthese works, the nonlinear part of Hammerstein system is assumed to be algebraic, i.e. memoryless.In this thesis, by using LS-SVM we propose a method to identify Hammerstein systems where thenonlinear part has a finite memory. For the identification of Wiener type systems, although variousmethods are also available in the literature, one approach which is proposed in some works would beto use a method for the identification of Hammerstein type systems by changing the roles of inputand output. Through some simulations it was observed that this approach may yield poor estimationresults. Instead, by using LS-SVM we proposed a novel methodology for the identification ofWiener type systems. We also proposed various modifications of this methodology and utilized it forsome control problems associated with Wiener type systems. We also proposed a novel methodologyfor identification of NARX (Nonlinear Auto-Regressive with eXogenous inputs) systems. We utilizeLS-SVM in our methodology and we presented some results which indicate that our methodologymay yield better results as compared to the Neural Network approximators and the usual SupportVector Regression (SVR) formulations. We also extended our methodology to the identification ofWiener-Hammerstein type systems. In many applications the orders of the filter, which represents thelinear part of the Wiener and Hammerstein systems, are assumed to be known. Based on LS-SVR,we proposed a methodology to estimate true orders.Keywords: System Identification,Wiener Systems, Hammerstein Systems,Wiener-Hammerstein Systems,Nonlinear Auto-Regressive with eXogenous inputs (NARX), Least-Squares Support VectorMachines (LS-SVM), Least-Squares Support Vector Regression (LS-SVR), Control.
Sarkaç benzeri sistemlerin eşzamanlaması
Synchronization is a phenomenon that is widely encountered in nature, life sciences and engineering. There exist various synchronization definitions in various research fields. The general definition for synchronization is the adjustment of rhythms of oscillating systems due to their weak interaction. Synchronization problem depends on the type of applications that require suitable properties and comparison functions. Different applications require different properties and comparison functions. Throughout our study, we choose the comparison function to be the difference of the states variables of the systems in hand.In this thesis, we will present types and methods of synchronization which has practical applications, i.e. mechanical systems. Then, we will investigate the passive controlled in-phase synchronization of spring-damper coupled single and double pendulum systems by using various stability analysis for both the system in hand and its appropriately defined error dynamics. We mostly achieved in-phase synchronization in these coupled pendulum systems with a few exceptions which are based on several conditions. Finally, we will explain the master-slave synchronization of two ball hoppers using two different gait controllers, namely, fully-actuated and under-actuated controllers. By using fully-actuated controller for the slave hopper, we achieved apex state synchronization and by using under-actuated controller for the slave hopper, we achieved apex position synchronization between these two hoppers in master-slave configuration.
Tekrarlayan sinir ağları ile bacaklı lokomosyonun kontrolü ve sistem tanımlanması
In recent years, robotic systems have gained massive popularity in the industry, military, and daily use for various purposes, thanks to advancements in artificial intelligence and control theory. As an exciting sub-branch of robotics with their differences and opportunities, legged robots have the potential to diversify and spread the use of robotic systems to new fields. Especially, legged locomotion is a desirable ability for mechanical systems where agile mobility and a wide range of motions are required to fulfill the designated task. On the other hand, unlike wheeled robots, legged robot platforms have a hybrid dynamical structure consisting of the flight and contact phases of the legs. Since the hybrid dynamical structure and nonlinear dynamics in the robot model make it challenging to apply control and perform system identification for them, various methods are proposed to solve these problems in the literature. This thesis focuses on developing new neural network-based techniques to apply control and system identification to legged locomotion so that robotic platforms can be designed to move efficiently as animal counterparts do in nature. In the first part of this thesis, we present our works on neural network-based controller development and evaluation studies for bipedal locomotion. In detail, neural controllers, in which long short-term memory (LSTM) type of neuron models are employed at recurrent layers, are utilized in the feedback and feedforward paths. Supervised learning data sets are produced using a biped robot platform controlled by a central pattern generator to train these neural networks. Then, the ability of the neural networks to perform stable gait by controlling the robot platform is assessed under various ground conditions in the simulation environment. After that, the stable walking generation capacity of the neural networks and the central pattern generators are compared with each other. It is shown that the proposed neural networks are more successful gait controllers than the central pattern generator, which is employed to generate data sets used in training. In the second part, we present our studies on the end-to-end usage of neural networks in system identification for bipedal locomotion. To this end, supervised learning data sets are produced using a biped robot model controlled by a central pattern generator. After that, neural networks are trained under series-parallel and parallel system identification schemes to approximate the input-output relations of the biped robot model. In detail, different neural models and neural network architectures are trained and tested in an end-to-end manner. Among neuron models, LeakyReLU and LSTM are found as the most suitable feedforward and recurrent neuron types for system identification, respectively. Moreover, neural network architecture consisting of recurrent and feedforward layers is found to be efficient in terms of learnable parameter numbers for system identification of the biped robot model. The last part discusses the results obtained in the control and system identification studies using neural networks. In the light of acquired results, neural networks with recurrent layers can apply control and systems identification in an end-to-end manner. Finally, the thesis is completed by discussing possible future research directions with the obtained results.