LEE- Uçak ve Uzay Mühendisliği Lisansüstü Programı

Bu topluluk için Kalıcı Uri

http://hdl.handle.net/11527/19473

Gözat

A multiscale approach to understand the effects of design parameters on the elastic behavior of 3D orthogonally woven composites

(Graduate School, 2024-11-04) Erkoç, Hilal ; Cebeci, Hülya ; 511201167 ; Aeronautical and Astronautical Engineering

This study aims to investigate the effect of various parameters on the elastic constants of three-dimensional (3D) orthogonally woven composites. Two-dimensional (2D) laminated composites exhibit high in-plane stiffness and strength; however, they are inadequate in applications subjected to out-of-plane loads, particularly in engine fan blades, aircraft fuselage structures, and wind turbine blades. With an innovative approach, 3D orthogonally woven composites effectively overcome the limitations of traditional 2D laminates. The usage of 3D orthogonally woven composites in these structures can be beneficial because 3D orthogonally woven composites are more resistant to out-of-plane loading than 2D laminates, due to their improved mechanical properties through the thickness. In addition to this, improved impact damage tolerance, higher delamination resistance, and reduced assembly and production costs through single-piece fabric production are advantages of 3D orthogonally woven composites. 3D orthogonally woven composites, in spite of their advantages, present certain challenges in application. One of the significant challenges is the complex nature of their manufacturing process, which demands specialized equipment and skilled personnel, leading to high production costs. Their complex structure can also complicate design, analysis, and simulation, requiring advanced computational models. Additionally, the complex architecture of these composites can present challenges in repair and maintenance procedures. 3D orthogonally woven structures consist of three interwoven sets of yarns arranged in orthogonal directions, where the warp and weft yarns remain straight while the binder yarns interlace them to create a multidimensional architecture. This complex architecture of 3D orthogonally woven composites plays an important role in determining the mechanical properties of the structure. Since differences in cross-section configurations, yarn arrangements, and fiber interactions significantly influence the load-carrying capacity, stiffness, and overall performance of the composite, an in-depth examination of the structural architecture is critical to optimizing the mechanical properties of the material. Several analytical studies have examined the effects of binder-to-weft and binder-to-warp ratios on the elastic properties of 3D orthogonally woven composites. These analyses employ representative volume elements (RVEs) to model the material behaviors. The binder-to-weft ratio characterizes the number of wefts of yarn a binder yarn encircles before reversing direction within the weft layer. Similarly, the binder-to-warp ratio represents the proportion of warp yarns per layer relative to the total number of warps encompassed by the RVE. However, a key limitation of these existing studies is based on the absence of a comparative analysis between analytical solutions and numerical simulations. Furthermore, the impact of RVE thickness on its elastic coefficients has not been thoroughly investigated. Here, the effects of changing thickness on the tensile response of the structure, as obtained through analytical solutions and numerical simulations, are presented. Elastic constants of 3D fiber-reinforced composites were estimated using a multi-scale homogenization technique based on meso-macro homogenization with good correlation. Numerical simulations were performed using ABAQUS software to analyze the behavior of the models. Through the optimization of the geometrical parameters of RVE, 3D orthogonally woven composites can be effectively implemented across a diverse range of engineering applications, especially in the aviation field.
Cavitation performance improvement of engine cooling pump

(Graduate School, 2025-01-22) Karbal, Doğukan ; Edis, Fırat Oğuz ; 511221106 ; Aeronautical and Astronautical Engineering

As part of this study, the cavitation issue observed at the specific operating point of the water pump used in the cooling system of an off-road vehicle was investigated, and an optimization process was conducted to address the problem. The optimization process consisted of three stages. In the first stage, based on the insights gained from a literature review, different designs for the impeller's blade leading-edge profile were tested. The impact of three different blade leading edge configurations (v0, v1, and v2) on the impeller's cavitation performance was investigated using 3D CFD analysis. The initial analyses were performed in steady-state conditions using ANSYS CFX. Regions on the impeller with pressures below the vapor pressure of the fluid were compared, and the optimal design (v2) was selected. Following the selection of the optimal design, cavitation analyses were conducted on the pump model with the baseline impeller and the pump model with the optimized impeller using ANSYS CFX. These analyses employed the Rayleigh-Plesset cavitation model and were solved under time-dependent conditions. The results demonstrated that the optimized design exhibited superior cavitation performance compared to the baseline design. The optimized impeller was then manufactured and subjected to experimental cavitation tests. These tests revealed that the optimized impeller allowed the pump to operate at 30% lower inlet pressure before cavitation occurred compared to the baseline pump, confirming the significant effectiveness of the optimization process. Subsequently, the final impeller design underwent experimental performance testing, and a validation study was carried out by comparing the experimental results with 3D CFD results. The validation analyses were performed using Siemens StarCCM+ in time-dependent conditions. The comparison between 3D CFD results and experimental data showed a maximum deviation of 3%, confirming the validity of the numerical approach. Performance test results were compared both numerically and experimentally using H-Q curves, which are presented for clarity.
Coherent structures and energy transfer in decelerated turbulent boundary layers

(Graduate School, 2023-02-10) Güngür, Taygun Recep ; Güngör, Ayşe Gül ; Maciel, Yvan ; 511162103 ; Aeronautical and Astronautical Engineering

This thesis aims to expand our knowledge about turbulent boundary layers (TBLs) developing under adverse pressure gradients (APG). The main focus of this thesis is coherent structures and energy transfer mechanisms in APG TBLs with small and large velocity defects. For this, two novel non-equilibrium APG TBL direct numerical simulation databases are generated. The first database is a non-equilibrium APG TBL with $Re_\theta$ reaching 8000 and a shape factor spanning between approximately $1.4$ and $3.2$. It is the main database utilized throughout the thesis. The second database has identical domain and boundary conditions to the first one. The difference between them is that turbulence in the inner layer of the second database is artificially eliminated. This second database is generated to examine the effect of the inner layer on the outer layer turbulence. For comparison purposes, a channel flow case, two zero pressure gradient (ZPG) TBLs and two homogeneous shear turbulence (HST) databases from the literature are employed. The energy-carrying and –transferring structures are examined using the spectral distributions and two-point correlations. The analysis reveals that energy-carrying structures in small defect APG TBLs and canonical flows have similar spatial and spectral features. In the large defect case, turbulence in the inner layer, which is the dominant region in canonical flows and small defect APG TBLs, loses its importance and outer-layer turbulence becomes dominant. The inner peak in the $\langle u^2\rangle$ spectra does not exist in the large-defect case. Moreover, two-point correlations show that the spatial organization becomes different in the large-defect case as well. Regarding the energy-transferring structures, production, pressure-strain and dissipation structures behave in a similar fashion to the energy-carrying structures. The spectral distributions show that the canonical flows and small defect APG TBLs behave very similarly. The shape of the spectra is qualitatively similar in both cases. In the large defect case, the wall-normal distributions of production and pressure-strain become different since the outer layer becomes dominant. However, the shape of 2D spectra and the aspect ratio of structures are alike in all cases. The production and pressure-strain structures are analyzed in more detail using the relative size and wall-normal positions with respect to each other and energetic structures using spectral distributions. The results show that production and pressure-strain spectra have similar features in both the inner and outer layers regardless of the velocity defect, despite the differences in energetic structures. In the inner layer, the results suggest that the near-wall cycle or another mechanism with similar spectral features exists in large defect APG. As for the outer layer, an interesting result is that in large-defect APG TBLs it acts more like a free shear layer than in small-defect APG TBLs or canonical flows. Besides that, production and inter-component energy transfer mechanisms are similar in all cases regardless of velocity defect. No inflection point instability in the outer layer of the large-defect APG TBLs was detected. The effect of the near-wall region on the outer-layer layer structures is examined through Reynolds-shear-stress carrying structures' spatial features by detecting individual structures using spatio-temporal volumetric data. The results show that the outer layer is not significantly affected by the inner-layer turbulent activity. The structures' spatial features mostly depend on the mean shear. The aspect ratio of Reynolds-shear-stress carrying structures remains almost identical in the outer layer when the inner-layer turbulence is eliminated. Moreover, the aspect ratio follows a similar trend in both outer layers of APG TBLs and HSTs when the structures' size is normalized with the Corrsin length scale. The overall conclusion is that energy transfer mechanisms remain the same within one layer regardless of the velocity defect. The reason why the wall-normal distribution of energy and energy transfer dramatically changes in the large defect case is probably the change in the mean shear profile due to the increasing velocity defect.
Damage classification of cnt/cnc-reinforced pu foam-cored sandwich panels through acoustic emission testing

(Graduate School, 2024-01-23) Küçükkalfa, Eyüphan ; Yıldız, Kaan ; 511211122 ; Aeronautical and Astronautical Engineering

Aerospace companies are making significant attempts to design highly-efficient lightweight aircraft while satisfying the required mechanical properties and allowing increased volume for the payload. Composite structures, specifically sandwich composites, have widespread use in the aviation industry owing to their high strength-to-weight ratio when compared to standard metal-aluminum construction materials. Although honeycomb sandwich composites have high impact resistance, they have disadvantages such as moisture absorption, water accumulation, and debonding due to the open-cell structure of the honeycombs. Accordingly, rigid polymeric foams with a closed-cell core structure are capable of overcoming these disadvantages and replacing honeycombs in future applications. Closed-cell and rigid polyurethane (PU) foams are valuable since they have an easy manufacturing process, multifunctionality, and capacity for reinforcement studies. Carbon nanotubes (CNTs) frequently serve as reinforcement materials in PU foams owing to their superior strength and light weight. However, when the concentration of CNT reinforcement is relatively high, disadvantages such as agglomeration and nonuniform dispersion may occur in the foam structure, leading to poor mechanical properties of the materials. Therefore, reinforcing CNTs into PU foams by hybridizing them with auxiliary materials such as cellulose nanocrystals (CNCs) might be an efficient alternative approach to the dispersion problem. The hybridization with CNCs can prevent the formation of strong Van der Waals bonds between CNTs. Ensuring passenger safety in aircraft requires prompt detection of any potential damage. Despite several benefits of foam-filled sandwich composites, they also have particular damage mechanisms that may occur under high loadings. The non-destructive testing method called acoustic emission (AE) testing allows real-time identification of transient elastic waves that are released from the defect source by sensors and classification of these failure mechanisms. Unlike other non-destructive testing techniques such as electromagnetic, radiographic, and ultrasonic, AE analyzes the failure tendency and prevents potentially irreversible damage by using data released directly from the defect source. Nevertheless, it is necessary to use a clustering method to categorize AE data into appropriate clusters that are each associated with a possible damage mechanism. Among manual clustering, signal processing, supervised clustering and unsupervised clustering; unsupervised clustering arises due to their great performance in first-time analysis of AE hits without having a historic data of the experiment. The performance of the k-means algorithm, which is one of the most efficient unsupervised clustering methods, heavily depends on the first cluster centers, and it often gets stuck at local minima. Hence, the k-means genetic algorithm can resolve this challenge through improved searching capacity that can provide an optimum solution in a reasonable amount of time. Previous research in the literature mainly studied the AE analysis of sandwich composites with honeycomb and foam cores, but there is a lack of research on sandwich composites with nanomaterial reinforcement of core materials. The thesis investigated the impact of reinforcement studies on the damage mechanisms of CNT/CNC-reinforced foam-cored sandwich composites through the AE test system. Damage mechanisms were classified using peak frequency, amplitude, and cumulative energy of AE data by the k-means genetic algorithm. In addition, as a subsection, the bending properties of CNT-reinforced PU foam-filled 3D woven I-beam composites were studied. Thus, a preliminary study was performed for CNT reinforcement studies on PU foam and its combination with sandwich composites. Furthermore, this preliminary study was beneficial for understanding the influence of PU foams on sustaining structural integrity and avoiding damage mechanisms in composite structures.
Development/ testing of software for a cubesat for high resolution earth observation in a low earth orbit

(Graduate School, 2024-06-24) Azam, Mehreen ; Aslan, Ali Rüstem ; 511221120 ; Aeronautical and Astronautical Engineering

CubeSats, ranging from 1U to 27U, are small satellites many nations pursue for academic and commercial purposes. The success of their missions depends greatly on the design of their software architecture. Beyond merely achieving functionality and optimal performance, the software must also be resilient to faults and shielded from the effects of radiation, potential failures, and errors. As CubeSats accommodates more advanced subsystems, developers worldwide are exploring agile development methods. Consequently, software development must prioritize three essential factors: Modularization, refactoring, and generalization. This study aims to describe the design, implementation, and testing of software modules of a 16U CubeSat, focusing on its onboard computer (OBC) software. A comprehensive software platform has been developed featuring a flexible architecture capable of supporting a multispectral payload and other subsystems. Multiple studies were done to familiarize the current work with experience from past projects, coding standards, and rules. Three fundamental requirements were derived to ensure software development quality: Concurrent documentation, version control for efficient tracking, and Debug tools support. The mission software has been developed using the Free RTOS Real-Time Operating System for real-time scheduling functionality, inter-task communication, timing, and synchronization. SEU/SEL management is considered for relevant subsystems. The development environment of choice was the Eclipse IDE, with code crafted in the C language. The code architecture is structured around creating libraries for individual subsystems, which serve as building blocks for developing higher-level applications specific to each subsystem. Followed by creating subsystem managers and various operating modes (Initialization, idle, Payload operation mode, etc.) ensuring reliable operation. Finally, a mode manager is implemented which acts like a state machine handling decision-making and switching between operating modes. Additional peripherals like packet routing, housekeeping, timekeeping, data logging, and even power management have been designed to match the mission profile in these modes. Following code development, the subsequent phase involves testing the code on actual hardware. The chosen OBC hardware has 03 interfaces; I2C for housekeeping/telemetry, JTAG for programming and debugging, and UART for development and testing. Testing of the developed code is in process for various subsystems. As future work, implementation of developmental changes is an ongoing process to ensure robustness and reliability.
Factional calculus-based modeling of mechanical systems: A case study on inverted pendulum dynamics

(Graduate School, 2023-05-17) Demir, Esra ; Özkol, İbrahim ; 511201123 ; Aeronautical and Astronautical Engineering

Obtaining the mathematical model of a system's behavior and solving the equations of these models using appropriate solution methods is highly important for engineers. Modeling allows the system to be tested in a simulation environment before it is even produced, and it can be checked whether the system meets the requirements. At the same time, performance improvements can be made by changing the parameters of the modeled system. Models created for an airplane, car, or helicopter can provide a training environment for users. For these reasons, modeling the mathematical model of the system to give the closest results to real values is an important requirement. fractional calculus emerges as a different method for obtaining mathematical models for this purpose. Fractional calculus is a branch of mathematics that deals with fractional, complex order derivatives and integrals, in addition to integer-order derivatives. Using this mathematics, mathematical models containing fractional-order derivatives are obtained instead of integer-order derivatives obtained by classical methods. Fractional derivatives have several types such as Riemann-Liouville, Caputo, etc. Equations of motion should be obtained by choosing the appropriate one for modeling. The concepts related to fractional calculus are presented in Chapter 2. In this chapter, the emergence of fractional calculus and the relationship between classical integer-order derivatives and fractional derivatives are explained. In addition, the types of fractional derivatives are briefly given, and special functions used in the calculation of derivatives and integrals using this derivative are described. Chapter 3 provides information about system modeling approaches. The most commonly used classical methods for obtaining the motion equations of a system are Newtonian and Lagrangian mechanics. Newtonian mechanics analyzes the forces and moments that affect the system in order to obtain a mathematical model. These external forces and moments are determined using a free-body diagram. This method is quite easy for simple systems. However, Lagrangian mechanics is preferred for complex systems where determining the forces and moments acting on them is quite difficult. This is because Lagrangian mechanics uses the total kinetic and potential energy of the system to obtain the motion equations. The Lagrangian expression is obtained by subtracting the total potential energy from the total kinetic energy. Then this expression is substituted into the Euler-Lagrange equation to derive the motion equations. Equations obtained using these two methods contain integer derivatives. Different methods are available for obtaining fractional mathematical models. One of these methods is using the fractional Euler-Lagrange equation (FELE). With this equation, calculations are made using the Lagrangian expression in the classical method without obtaining the energy equations of the system in a fractional manner. The Fractional Lagrangian is obtained by replacing the derivatives in the classical Lagrangian with their "Fractional Derivative" counterparts. Fractional motion equations are then obtained using FELE. In Chapter 3, firstly, the equations for the two classical methods used for classical modeling, Newtonian and Lagrangian mechanics, are provided. Then, FELE is explained and its formulas are given. To demonstrate the application of these methods, the modeling of a 4-rotor aerial vehicle (Quadrotor System) is presented. First, the motion equations are obtained using Newtonian mechanics, and then they are obtained using Lagrangian mechanics. In addition, errors made in the open literature when obtaining the rotational motion equations have been corrected in this chapter. Finally, fractional motion equations for the quadrotor system are obtained using FELE. The important issue is the selection of the type of fractional derivative to be used in fractional modeling. The most important priority in modeling a mechanical system is adding initial and boundary conditions. For this reason, Caputo and Caputo-Fabrizio fractional derivatives can be preferred to add initial conditions into the equations in a physically interpretable way. The use of Caputo-Fabrizio fractional derivative in modeling is preferred to avoid non-singular kernel problems in solving the equations. After obtaining the equations, they need to be solved to acquire high accuracy results. Therefore, an appropriate solution method must be determined. These solution methods can be analytical or numerical. Analytical solution methods provide the exact results of the equations and express the system's solution with time-dependent functions. This allows for the derivation and integration of the obtained functions. However, a disadvantage of analytical solutions is that they become more complex and time-consuming as the system becomes more complex. Therefore, numerical solution methods can be preferred for nonlinear and complex systems. Numerical methods, on the other hand, provide approximate solutions to the equations and generally give results faster than analytical methods. Therefore, if the solutions are made with sufficient sensitivity, suitable results can be obtained for engineering purposes. In addition, there are also semi-analytical numerical solution methods. In these methods, the time interval for which the system is studied is divided into a certain time step, and analytical solutions are obtained for each time step to obtain results. In Chapter 4, analytical, numerical, and semi-analytical numerical methods are discussed. First, the differential transform method (DTM), which is a powerful analytical method for solving the classical model, is given, and it is applied to the double pendulum system as an example. It is seen that this method is not suitable for time-dependent systems because the results deviate from the Runge-Kutta (RK) solutions. Then, the improved version of this method, the multi-step differential transform method (MsDTM), is applied to the same system. At the same time, the system is solved using a powerful numerical solution method called RK, and it is seen that the two results are consistent with each other. However, it is also an important output that MsDTM's computation time is very long. In Chapter 4, a numerical method that uses matrix approach for solving the fractional model is also discussed. Since the solution of fractional differential equations is quite complex, numerical solution is preferred instead of analytical solution. At the same time, the fact that multi-step methods cannot be used in fractional differential equations is an important factor that restricts the analytical solution of these equations. Finally, in Chapter 5, all the methods mentioned have been applied to a nonlinear inverted pendulum system, which is a system modeled by two differential equations and exhibits oscillatory behavior. First, the energy equations of this system have been obtained, and then classical motion equations have been derived using Lagrangian mechanics. These equations were then solved using DTM, MsDTM, and RK methods, and the results were compared. It was observed that the results of DTM were quite deviant, but MsDTM gave the same results as RK. Then, the fractional motion equations of this system were obtained, and numerical equations were reached using matrix approach. The obtained equations were solved for different fractional orders, and the results were presented in the thesis. The important issue is that it is not known which fractional derivative order will give the actual model of the system. Therefore, graphs were obtained by trying different orders. Experimental data is required to determine which value is correct.
Hierarchical reinforcement learning in complex wargame environments

(Graduate School, 2024-01-23) Kömürcü, Kubilay Kağan ; Üre, Nazım Kemal ; 511211126 ; Aeronautics and Astronautics Engineering

In recent times, Reinforcement Learning (RL) agents have achieved remarkable success in tackling difficult games, sometimes outperforming human players. This suggests that RL methods are well-suited for wargames, which are characterized by long decision-making periods, infrequent rewards, and extensive sets of possible actions. However, wargames are highly complex, and even with RL, convergence to a near-optimum solution requires an immense amount of experience and makes the solution sample inefficient. To address this inefficiency, we propose dividing the game into simpler sub-games, each focusing on a specific core skill of the overall game. These sub-games have shorter decision horizons and smaller action sets compared to the main game. To guide the learning process, we adopt a curriculum learning approach, employing a hierarchical control structure where the curriculum comprises these simpler sub-games. For my experimentation, we select StarCraft II as the test environment, as it shares common characteristics with wargames and has been extensively used in such scenarios. Through empirical evaluation, we demonstrate that our hierarchical architecture can successfully solve the complex wargame environment based on StarCraft II, whereas a non-hierarchical agent fails to do so. Additionally we plan to conduct an ablation study to investigate the impact of action frequency on training quality, which we believe to be a crucial factor. Since Starcraft II is a real-time strategy game, and not a turn-based strategy game, it is possible to take actions in different time intervals. we believe that taking actions rarely, as well as too frequently, will hinder the training quality. The recent achievements of Reinforcement Learning (RL) methods have garnered widespread attention across various domains. Among the notable applications, wargames stand out, encompassing a diverse array of games like board games such as chess and Go, as well as strategy games like StarCraft and MicroRTS. Despite the broad spectrum within wargames, they share common features that distinguish them from other domains. These include large action spaces, whether discrete or hybrid, branching actions, adversarial opponent dynamics, and notably sparse reward functions. Additionally, wargames exhibit a characteristic where decisions made by an agent have a delayed impact on the game dynamics, posing challenges for optimization. To tackle these obstacles, several methods have been proposed, including Hierarchical Reinforcement Learning (HRL), forward planning, and curriculum learning. In this study, we primarily employ curriculum learning and a straightforward HRL approach in the real-time strategy game StarCraft II, comparing these techniques with a non-hierarchical agent. Also, although Reinforcement Learning (RL) has exhibited remarkable success across various applications, yet the significance of selecting what we term "decision frequency" in constructing Markov Decision Processes (MDPs) remains underappreciated in real-world scenarios. This thesis sheds light on the crucial role of decision frequency and its impact on RL training through a thorough analysis of a toy experiment, followed by the application of this knowledge to solve a complex StarCraft 2 environment. Our findings underscore that finely tuning decision frequency can be pivotal in determining the success or failure of RL training. To illustrate our insights, we propose an intuitive method for decision frequency tuning, showcasing its effectiveness in both a controlled toy experiment and within the context of StarCraft 2 minigames, known as core skills. Utilizing these core skills, we employ a hierarchical approach to train a model for one of the most challenging types of StarCraft 2 games. Benchmarking results highlight the superiority of our method over similar approaches, achieving competitive scores with approximately 30 days of real-time experience, compared to the approximately 30 years required by comparable methods to achieve similar results. In this study, our initial focus involves the dissection of the primary task into sub-tasks, with each sub-task representing a fundamental skill that can be independently addressed by a non-hierarchical agent and is crucial for mastering the primary task. Our contribution encompasses two key aspects: Firstly, we demonstrate that the proposed hierarchical approach effectively resolves complex challenges presented in wargame environments, challenges that remain insurmountable for a non-hierarchical agent. Additionally, our findings reveal that optimizing dedicated agents for each individual sub-task and combining their policies within a hierarchical framework yields commendable performance scores in the StarCraft II environment, even in the absence of optimization for the hierarchical controller. Secondly, we observe that expanding the set of sub-tasks beyond the core-skill set does not yield a substantial improvement in performance. We introduce a curriculum setting and implement hierarchical control for the sub-policies trained in their respective sub-tasks. The overarching objective of the main task is partitioned into a more manageable yet still comprehensive set of sub-tasks. Each sub-task involves an objective function that is simpler to achieve in terms of the required sample size. Despite uniform game mechanics across all sub-tasks, variations in opponent behaviors lead to differences in the transition distribution and the initial state distributio within the Markov Decision Process (MDP) of each sub-task. Owing to distinct objective functions, each sub-task is associated with a unique reward setand a discount factor. To address this complexity, we independently train an agent for each sub-task, with a corresponding policy, subsequently merging them using a hierarchical control structure within the main task. Our experimentation is conducted within the StarCraft II Learning Environments (SC2LE). StarCraft II, operating as a real-time strategy game, features complex military dynamics, presenting an extensive observation space and a diverse set of actions. It has been widely utilized in reinforcement learning (RL) research due to its inherent challenges. Moreover, StarCraft II is recognized as one of the most realistic environments in terms of military dynamics, offering the flexibility of creating custom game scenarios and objectives across a varied selection of maps. In our experiments, we utilize both the provided environments from SC2LE and construct our own custom environments using the game's map editor. We've introduced a Core Skill Decomposition algorithm, a form of Hierarchical Reinforcement Learning. This algorithm learns individual sub-policies for each sub-task and a manager policy that selects among these sub-policies to create a hierarchical agent. Our approach involves decomposing a complex wargame environment into sub-tasks that can be solved by a non-hierarchical A2C algorithm, while retaining the core aspects of the environment. We evaluate our method in three challenging StarCraft II environments and demonstrate that, when these environments are decomposed into sub-tasks, our hierarchical architecture successfully solves the environment, whereas a non-hierarchical agent fails to do so. Additionally, we observe that expanding the core set of skills only results in a marginal increase in performance. This thesis also addresses the often underestimated yet pivotal factor of decision frequency in the construction of Markov Decision Processes (MDPs) in Reinforcement Learning (RL). Our investigation has unveiled that the careful adjustment of decision frequency holds substantial sway over the efficacy of RL training, presenting potential implications across a diverse array of applications.
Launch vehicle navigation system design and comprehensive performance analysis

(Graduate School, 2024-07-02) Ertan, Altuğ ; Hacızade, Cengiz ; 511211108 ; Aeronautical and Astronautical Engineering

Space transportation is one of the most important issues of today because it plays a key role in the advancement of scientific research, economic growth and international relationships. Mainly, it provides to carry satellites which are used for communication, weather forecasting, and navigation into a variety of orbits depend on the need. Moreover, deep space missions can be conducted owing to the reliability of current launch vehicles. Nowadays, there is a big challenge in the space industry which leads to fast technological progress in space transportation technology. Leading companies like SpaceX, Blue Origin, and Rocket Lab are using reusable rockets to greatly lower the cost of reaching space. Examples of these innovations include SpaceX's Falcon 9 and Starship, and Blue Origin's New Shepard and New Glenn rockets. Traditional aerospace leaders such as Boeing and Lockheed Martin are also advancing with their Space Launch System (SLS) and Vulcan Centaur rockets. This competitive environment is fueling new innovations, making space more accessible, and creating more chances for commercial and scientific projects. National launch vehicle programs are essential today to have a say and power in political, economic and military fields. From a political perspective, this type of program makes a country as a key member of the international space community, so its role can be influential to improve space policies and collaborations. Economically, space programs pioneer to technological innovation and industrial development. Due to innovations and inventions in space, related fields like materials science, electronics, and communications are developed as well. Moreover, the capability of launching satellites by itself removes dependency to foreign providers, so projects can be managed as more cost effective and safer in terms of governments. With regard to the military, national launch vehicle programs enhance national security by guaranteeing consistent access to space-based resources that are essential for communication, navigation, and surveillance. It also enhances deterrence by demonstrating advanced technological prowess. Developing its own launch vehicle gives a country complete control over its space activities, protecting strategic interests and promoting national pride and independence in the dynamic areas of space exploration and defense. Launch vehicles are really complex systems that require the working of many subsystems together in a harmony to satisfy designed mission. The main subsystems are the propulsion system, the structural system, avionics, thermal control and guidance, navigation and control (GNC) subsystem. The propulsion system generates the force to elevate the launch vehicle through sky and space. The structural system sustains the integrity of the vehicle under rough environments. Avionics handle the onboard electronics and data processing. Extreme high and low temperatures are serious issues in space, and thermal control system balance the temperature of the system. Lastly, GNC steers the vehicle to the target orbit to ensure that payloads are placed into correct orbits. Among all of subsystems, the navigation subsystem which is an important component of GNC is specifically vital. Navigation subsystem calculates the position, velocity and attitude of the launch vehicle through trajectory to enable guidance and control systems for matching ongoing orbit with target orbit. The launch vehicle may not reach its intended orbit or deliver its cargo accurately if the navigation is unreliable, increasing the chance of mission failure. Thus, robust navigation systems are necessary to ensure that space missions are successful and that the vehicle reaches its target precisely. Designing the navigation subsystem of a launch vehicle is a complex task which requires careful research, trade-off, modelling and testing to assure an system working highly reliable. Evaluating mission requirements coming from the customer, navigation subsystem requirements are derived. Considering these navigation subsystem requirements, a couple of options are determined based on research, and tested are implemented to candidate sensors and algorithms. Generally, inertial measurement units (IMU), global navigation satellite system (GNSS) receivers, and star trackers are used for navigation. Owing to high prices of these components, they are modeled in detail by using modelling environments, and model-in-the-loop (MIL) tests are conducted to analyses their performances. MIL consists of different steps like single run, multiple run and sensitivity analysis. After completing all of these steps of MIL, requirements for navigation subsystem components are derived. Regarding requirements, appropriate sensors are bought or manufactured. Then, hardware-in-the-loop (HIL) test takes place to simulate the response of navigation system under dynamic environments representing launch conditions. This thorough development process ensures that the navigation subsystem can accurately guide the launch vehicle through all stages of the flight, from takeoff to the deployment of the payload, achieving mission success. This thesis focus on designing and analyzing the navigation subsystem for a launch vehicle at MIL level. Analysing various grades of IMUs to achieve precise navigation to a designated orbit is a critical task for improving launch vehicle guidance and control systems. Testing these IMUs with and without the application of a linearized Kalman Filter (LKF) provides a clear framework for assessing the enhancements in navigation accuracy that this filtering technique can offer. LKF is particularly valuable for its ability to integrate sensor data and reduce errors. By applying the LKF to the same set of IMUs, this study methodically quantifies the performance improvements and increased robustness attributable to the filter. Conducting extensive Monte Carlo simulations adds a robust statistical layer to performance analysis, allowing to evaluate the navigation performance across a spectrum of uncertainties. This approach enables effectively gauge the reliability and accuracy of both the basic inertial navigation systems and those augmented by the LKF. Lastly, detailed sensitivity analysis included is crucial. By investigating which specific sensor error parameters—such as biases, scale factors, and misalignments—most significantly influence navigation performance, critical vulnerabilities of the system can be pinpointed. This analysis is key to understanding the relative impact of various errors both in the context of standard inertial navigation and when using the LKF. Overall, this thesis aims to deliver a thorough understanding of the key design considerations for a launch vehicle's navigation subsystem. It explores how different grades of IMUs perform under enhanced filtering techniques and identifies significant factors influencing system accuracy and reliability. Also, it promises to contribute valuable insights into the development of more sophisticated and dependable navigation systems for launch vehicles, aligning with industry needs for precision and robustness in space missions.
Multi agent planning under uncertainty using deep Q-networks

(Graduate School, 2024-04-29) Tarhan, Farabi Ahmed ; Üre, Nazim Kemal ; 511142108 ; Aeronautics and Astronautics Engineering

The extensive popularity of commercial unmanned aerial vehicles has drawn great attention from the e-commerce industry due to their suitability for last-mile delivery. However, the organization of multiple aerial vehicles efficiently to deliver the given set of goods within the existence of no-fly zones, numerous warehouses, limited fuel, and uncertainties are still a problem for traditional algorithms. The main challenge of planning is scalability, since the planning space grows exponentially with respect to the number of agents, and it is not efficient to let human-level supervisors structure the problem for such large-scale settings. With the recent advancements in deep reinforcement learning, algorithms such as Deep Q Networks (DQN), had unprecedented success in solving single-agent decision-making problems. Extension of these algorithms to multi-agent problems such as multi-drone delivery is very limited due to scalability issues. This work proposes an approach that improves the performance of DQN on multi-agent drone delivery problems by utilizing state decompositions for lowering the problem complexity, curriculum learning for handling the exploration complexity of delivery environments, and genetic algorithms (GA) for searching efficient packet-drone matching across the combinatorial solution space. The performance of the proposed method is shown in a multi-agent delivery by drone problem that has $10$ agents and $\approx10^{77}$ state-action pairs. Comparative simulation results are provided to demonstrate the merit of the proposed method. Compared with the conventional DQN schemes, and recently developed utility decomposition techniques, the proposed genetic algorithm-aided multi-agent DRL outperformed the rest in terms of scalability and convergent behavior. The prior techniques become intractable quickly at a large number of agents within the context of delivery by drone problem. The basic DQN algorithm fails to find a solution for three agents in a 10x10 drone delivery scenario within a reasonable number of steps, but the deep correction method successfully converges after approximately 1 million Bellman updates. Furthermore, applying the deep correction method increases the learning capacity to five agents and converges around 35 million Bellman updates. However, using this method does not lead to convergence with ten agents in a manageable way. With powerful computing resources, it becomes clear that while single-agent models set an initial computational standard, increasing the number of agents introduces complexity, as seen through immediate convergence difficulties in a three-agent DQN setup. Although there is promise with three- and five-agent configurations using Deep Correction, the model with ten agents exceeds the threshold for convergence within 24 hours, emphasizing the delicate balance between agent quantity and computational feasibility. The utilization of drone delivery simulation presents intricate challenges, including restricted airspace, fuel limitations, and the pick-and-place scenario. The study demonstrates that employing a method involving packet distribution through genetic algorithms effectively minimizes the complexity in resolving tasks for 10 agents within 5.74 minutes. Subsequently, the reduced problem is handled by deep Q-network inference models with Curriculum Learning and Prioritized Experience Replay, achieving execution times measured in milliseconds. This two-fold approach skillfully learns the dynamic nature of delivery problems without requiring prior domain knowledge input amid uncertain environmental conditions prone to altering actions. Furthermore, visual evidence at various time steps during execution illustrates how integrating GA-based packet distribution empowers the proposed base DQN model with Curriculum Learning and PER framework to tackle scenarios involving 10 agents – an accomplishment deemed unattainable by other explored solutions within reasonable time frames and computational resources. In conclusion, the combination of deep reinforcement learning and genetic algorithms provides a promising approach for efficient and effective delivery with multi-agent drones under uncertainty.
Multi-objective optimization of fiber reinforced laminated hybrid composite plates using particle swarm algorithm

(Graduate School, 2024-07-02) Yegit, Orhan Nuri ; Yıldız, Kaan ; 511201178 ; Aeronautics and Astronautics Engineering

This study investigates the multi-objective optimization of layered hybrid composite structures under axial compressive loads using the particle swarm optimization (PSO) approach. A simply supported composite plate consisting of 64 layers was subjected to constant axial compressive loads in two directions. The classical plate theory (Kirchoff plate theory) was employed to perform the analytical calculation of the plate's critical buckling load. Additionally, since the plate is a layered structure, the necessary formulas were derived using the classical lamination theory. Parameters such as the plate's edge lengths, layer thickness, layer material, total number of layers, magnitude and ratio of the compressive loads in both directions were derived from a previously studied problem in the literature. Thus, the calculations were validated by comparing them with the results obtained in the literature. Extensive studies on hybrid composites have demonstrated their superior design flexibility and mechanical properties compared to non-hybrid composites. In this purpose, a hybrid plate was created by employing carbon/epoxy and glass/epoxy materials. The positioning of plies with different material properties was carefully considered during the hybridization process and taking into account its impact on the plate's bending rigidity. Therefore, carbon/epoxy layers exhibiting higher bending rigidity were placed farther from the symmetry axis, whereas less rigid glass/epoxy layers were positioned closer to the symmetry axis. By employing this strategy, the elements of the bending rigidity matrix, which directly impact the critical buckling coefficient, were effectively maximized. The cost function was determined by utilizing previous works in the literature. Instead of using the actual material prices, the cost ratio of the two materials was calculated and incorporated into the equation. Population optimization algorithms are powerful tools employed for solving complex problems. Different types of population optimization algorithms such as genetic algorithm, ant colony optimization (ACO) algorithm, differential evaluation (DE) algorithm. In this study, particle swarm optimization (PSO) was utilized. The primary motivation for selecting this algorithm arises from the demonstrated evidence from previous studies indicating its superior convergence speed and performance compared to other stochastic algorithms. The objective functions were defined as maximizing the buckling load, minimizing the cost function, and maximizing the difference between the first two natural frequency values. The weighted sum method and the $\epsilon$ constraint method were employed to solve the multi-objective optimization problem. Since the objective functions were conflicting, Pareto optimal solutions were obtained. The optimization was repeated with different load and aspect ratios to investigate the effects of these parameters on the results.
One dimensional case studies with picfoam solver

(Graduate School, 2024-01-05) Yılmaz, Fatma Tuğçe ; Edis, Fırat Oğuz ; 511191197 ; Aeronautical and Astronautical Engineering

Since the electrical field can produce a rocket thrust idea was given by pioneers, electrical propulsion became widely used in space missions. As humanity's interest in exploring space increased, the idea of developing efficient and relatively affordable spacecraft contributed to the development of electric propulsion thrusters. Hall thrusters which produce thrust by ionization of the plasma gases with electric field and magnetic field have become one of the interesting and attractive thrusters means of space propulsion; since, they are promising devices that can be used in deep space exploration, space mining satellite systems, etc. On the other hand, to be used efficiently and provide deeper distance in space, it needs to be studied for some problems regarding the event in the discharge chamber. More specifically, E ⃗×B ⃗ drift instability or anomalous electron transport is one of the most important and still needs more clarification and explanations about the physical phenomena inside the Hall thruster. To achieve this, it is needed that new codes or solvers that can model those physical problems more accurately and less costly. Although there are fluid or hybrid approaches that solve the simulation relatively cost-efficiently, to obtain a more comprehensive understanding of those physical problems, fully kinetic codes are needed. Therefore, in this thesis, an attempt was made to implement the new picFoam solver which is developed as a fully kinetic and electrostatic solver based on OpenFoam. Two benchmark cases which are capacitively coupled benchmarks for the low-pressure plasma and the other one is one-dimensional azimuthal particles in cell simulation were chosen. In the first case, some collision models which include elastic scattering, excitation, and ionization were implemented. The results from the first benchmark case show that the picFoam solver can simulate collisions in a plasma environment, emphasizing circuit design parameters for more accurate modeling, and improvement of some collision model applied in the picFoam solver is needed to implement both excitation states at the same time, more study is required to better accomplish. In the second benchmark case, a one-dimensional azimuthal E ⃗×B ⃗ drift case which includes the main problems in the Hall thruster physics, namely the E ⃗×B ⃗ drift instability and anomalous electron transport, was applied. However, instead of going deeper into those physical phenomena, an attempt was made to implement the picfoam solver for the specified benchmark case 2. Following tests, a simulation can produce similar results; however, it does not have same solutions. Therefore, this work will be continued for better outcomes.
Optimal solution of orbital facility location problem utilizing optimum rocket staging and Q-Law orbit transfer

(Graduate School, 2025-01-17) Çam, Hasan Hüseyin ; Özkol, İbrahim ; 511201147 ; Aeronautics and Astronautics Engineering

The rapid growth of satellite constellations has made satellite servicing a crucial aspect of sustaining and enhancing their functionality over extended periods. This thesis focuses on developing an optimization framework to address the Orbital Facility Location Problem (OFLP), which involves determining the optimal placement of service facilities in orbit to minimize the combined costs of launching and servicing satellite constellations. The research emphasizes the importance of aligning servicing strategies with cost efficiency while adhering to operational and logistical constraints. A comprehensive methodology is adopted to tackle the OFLP, combining advanced techniques such as rocket staging optimization and low-thrust orbit transfer analysis. The launch cost is modeled based on the maximum payload capacity of launch vehicles and the characteristics of candidate orbits. Service costs, on the other hand, are determined by calculating the fuel consumption and payload requirements for servicing trips between service facilities and client satellites. These costs are integrated into an optimization model formulated using Binary Linear Programming (BLP), which enables the identification of cost-efficient orbital configurations. The study explores multiple scenarios defined by varying numbers of launch vehicles and servicing trips. For each scenario, the model identifies the optimal orbits for deploying service facilities, ensuring that the launch vehicles operate within payload capacity limits and that client satellites are serviced efficiently. The results reveal critical insights into the interplay between launch and servicing costs, highlighting the significance of parameters such as semi-major axis, eccentricity, and right ascension of ascending node in achieving cost-effective solutions. The thesis also presents detailed numerical analyses and visualizations that illustrate the spatial distribution of service facilities, the alignment between service orbits and client satellite constellations, and the trade-offs between resource allocation and operational efficiency. These findings underline the importance of precise orbit selection and resource management in reducing mission costs. In conclusion, this research provides a robust framework for solving the OFLP and offers practical guidelines for designing satellite servicing missions. The insights gained from this work contribute to the field of aerospace engineering by promoting cost-effective strategies for managing satellite constellations. This framework can serve as a foundation for future research and operational planning in orbital servicing and satellite maintenance.
Parametric investigation of mechanical properties of auxhex unit cell sandwich structures

(Graduate School, 2024-07-01) Saygı, Kadircan ; Yıldız, Kaan ; 511201169 ; Aeronautics and Astronautics Engineering

In this study, the mechanical properties of the AuxHex structure, which consists of hybrid unit cells obtained by combining the classical hexagonal honeycomb unit cell structure with the auxetic unit cell structure, are investigated under axial compressive loads by changing the geometric parameters. In the studies carried out, it has been observed that the AuxHex structure, which is obtained by hybridizing two different cell structures, has a better strength value against in-plane compressive loads. Before investigating the effect of geometric parameters on the mechanical properties of the AuxHex structure, a validation study was performed by analyzing the AuxHex structure using the ABAQUS 2020 commercial finite element program with an appropriate study in the literature and based on the results of this validation study, the boundary and loading conditions, material properties to be defined, and other variables for finite element models and analyses of other AuxHex structures have been determined. Subsequently, the mechanical properties, including Young's modulus, yield strength, energy absorption capacity under different strain values, and toughness were investigated for nine (9) different geometries of AuxHex sandwich structures by varying geometric parameters such as inclined and non-inclined wall cell lengths, angles, and wall thicknesses that characterize the AuxHex unit cell structure. To analyze the Young's modulus and yield strength among these mechanical properties, both theoretical formulas used in the literature and finite element analysis results were utilized. The results obtained from the finite element analyses were employed to determine and compare the absorbed energy under certain strains and toughness values of nine (9) different AuxHex structures. The force-displacement graphs obtained from the finite element analyses were converted into stress-strain graphs. The resulting stress-strain graphs were integrated by using a curve fitting function in the MATLAB program to obtain the absorbed energies and toughness values at different strains values. These graphs were also useful for determining the Young's modulus and strengths. As a result of all these studies, it was observed that the elastic modulus, yield strength, energy absorption capacity, and toughness of the sandwich structure decreased with increasing inclined and non-inclined wall lengths of AuxHex unit cells. However, it was concluded that all these mechanical properties increased significantly and had a positive effect on the structure with increasing AuxHex unit cell wall thicknesses. Finally, it was concluded that the elastic modulus, yield strength, energy absorption capacity, and toughness of the AuxHex sandwich structure decreased with increasing angle values used to design the AuxHex unit cells, and lastly, the combination of all AuxHex structures' stress-strain graphs in one shows that the collapse behaviour of structure appears to be largely related to the change in the length of non-inclined wall.
Rocket engine altitude test facility design and 1D altitude simulation of IoX/LH2 propellant rocket engine

(Graduate School, 2024-07-28) Özcan, İsmail ; Edis, Fırat Oğuz ; 511201101 ; Aeronautical and Astronautical Engineering

The interest in space and development studies started in B.C. for humanity. Astronomical observations began to understand the dynamics between planets and stars. Reaching space and flying around the world have been the most considered areas over the years. Before the 20th century, theoretical and experimental studies were conducted; however, scientific research increased after the 20th century. Tsiolkovsky and Goddard contributed to rocketry both theoretically and experimentally. During World War II, the Germans developed V2 rockets, which were the predecessors of space rockets in the following years. The Saturn V rocket was launched in 1967, and it remained the most powerful rocket for humankind up to the 21st century before Starship Heavy. In today's world, the development of liquid-fueled rocket engines is progressing rapidly. Many companies and agencies are designing launch vehicles, payloads, and rocket engines all over the world. Today, NASA, ESA, SpaceX, JAXA, and others have numerous studies on rocket engines and their components. Nowadays, research and development of rocket engines have reached the most powerful rocket engine, the Raptor Engine by SpaceX. Test system designs and test facility setups are constructed even for the most powerful rocket engine designs. Various test setups and infrastructures are being established for testing numerous subcomponents and equipment as part of pre-flight verification for liquid-fueled rocket engines. It is worth noting that rocket engines with turbopump feed systems undergo ground tests and altitude simulation tests before launch in many industrial and academic studies. Today, there is ongoing development of various test infrastructures for the feed systems of turbopumps, including valve characterization test setups, cryogenic valve test setups, injector test setups, turbine test setups, pump test setups for oxidizer and fuel, leak test setups, gas generator and combustion chamber test setups, and more. Test setups for altitude testing of first-stage and second-stage rocket engines have been developed and are used to verify thrust and turbopump efficiency during the rocket's flight. Altitude test setups for rocket engines are available in various countries such as France, the United Kingdom, the United States, Germany, and South Korea, and they are used for altitude tests of rocket engines with different thrust capacities. Due to various factors such as different types of oxidizer and fuel feed systems, thrust capacities, rocket engine sizes, exit temperatures, and flow rates, unique system developments have been made and continue to be pursued. The primary sub-components of the rocket altitude test setup, as described in the literature, include motor feed tanks, rocket engines, vacuum chambers, diffusers, cooling water injectors, deflectors, ejector structures, and condensing equipment. In addition to these components, instruments such as pressure sensors, flow meters, temperature sensors, accelerometers, and load cells are used during performance testing to measure and verify performance values. Design considerations for the 1D modeling of the rocket engine followed established processes from the literature in this study. The most common programs for 1D modeling are EcosimPro and Simcenter Amesim. Simcenter Amesim was preferred for studying 1D system modeling for both the rocket engine and the altitude test facility design. In the first step, a 1D model of the rocket engine was created. A 1D model of the LOX/LH2 liquid-fueled Expander Cycle Rocket Engine VINCI's full model was developed, and a simplification method was used to reduce system unknowns. For the simplification method, the boundary conditions at the system inlet were kept the same, while the outlet conditions of the pump and turbine became the new inlet conditions for the thrust chamber. The pressurized oxidizer and fuel conditions remained stable with the addition of orifices, and the performance of the combustion chamber and nozzle was monitored. After completing the rocket engine design, the design of the altitude test facility was undertaken. Piping, chambers, divergent-convergent structures, steam ejectors, and steam feeding boundary conditions were added to the system to control the results. A vacuum system model was created to observe the VINCI rocket engine's performance under vacuum conditions lower than 50 mbarA. In this thesis study, the performance of the VINCI rocket engine and the altitude test facility were considered. The desired version of the VINCI was the 180 kN thrust in vacuum. It was observed that the thrust levels and performance data produced by the VINCI rocket engine were satisfactory during simulation. The thrust of the 1D modeled VINCI engine was found to be nearly 178 kN. The combustion chamber pressure manufacturer data was given as 60 barA, and the 1D model of the VINCI simulation result was found to be 59.78 barA. The feeding lines of the fuel and oxidizer mass flow rates were given as 5.59 kg/s and 34.11 kg/s according to the manufacturer. Simulation results for the fuel and oxidizer were found to be 5.5 kg/s and 34.31 kg/s, respectively. The manufacturer's specific impulse data was given as 457.2 s, and the 1D model of the VINCI simulation result was found to be 456.17 s. The main consideration for the vacuum conditions was set at 50 mbarA, and the 1D simulation results were found to be 42.93 mbarA with an increased steam inlet mass flow rate of the second-stage steam ejector. For the first stage of the steam ejector, literature information was followed, and a 110 kg/s steam condition was satisfied. For the second ejector, 141 kg/s of steam was supplied, creating a vacuum of 42.93 mbarA. The manufacturer's second-stage steam mass flow rate was given as 118 kg/s, suggesting that better vacuum conditions could be achieved according to the simulation results. Future work includes creating a digital twin with the 1D model, conducting 3D model-based CFD analysis for the altitude test facility design, comparing real-time test data between altitude test facilities and 1D simulation models, and running performance scenarios in the 1D altitude test model before conducting tests.

Gözat

Sustainable Development Goal "Goal 7: Affordable and Clean Energy" ile LEE- Uçak ve Uzay Mühendisliği Lisansüstü Programı'a göz atma

Sayfa başına sonuç

Sıralama Seçenekleri