LEE- Uçak ve Uzay Mühendisliği-Yüksek Lisans
Bu koleksiyon için kalıcı URI
Gözat
Son Başvurular
1 - 5 / 102
-
ÖgeReinforcement learning for automatic ground collision avoidance system of fighter jets: sequential maneuver primitive approach(ITU Graduate School, 2025-06-16)Artificial intelligence (AI) technologies have recently made significant advances and have begun to play critical roles in many areas, including aviation. AI's ability to mimic human-like decision-making processes makes it particularly suitable for safety-critical applications such as Automatic Ground Collision Avoidance Systems (AGCAS). However, integrating AI into aviation systems poses challenges due to the industry's strict requirements for explainability, verifiability, and safety. Traditional AGCAS systems primarily rely on rule-based approaches that execute instantaneous primitive maneuvers—typically a wings-level roll followed by a 5G pull-up. While these methods are effective in many scenarios, they can be limited in complex environments such as canyons or mountainous terrains, where pilot-like decision-making and sequential maneuvering are essential for ensuring flight safety. In this study, we propose the integration of reinforcement learning (RL)-based artificial intelligence (AI) into AGCAS, with a particular emphasis on operational safety and explainability. Specifically, we develop a method in which an RL agent generates intelligent, sequential avoidance commands within a high-fidelity, nonlinear 6-degree-of-freedom (6-DOF) simulation environment. Rather than delegating real-time control of the aircraft to AI, the RL agent is used solely to produce candidate avoidance maneuvers in simulation. From these, only successful trajectories are selected. These selected control sequences are then recorded and employed as predefined command sequences during actual operations. It is important to emphasize that reinforcement learning (RL) is not directly employed for real-time aircraft control. Instead, the system relies on avoidance action sets that were generated and verified through offline RL-based simulations. This approach ensures that only feasible and fully interpretable trajectories are utilized, thereby significantly reinforcing the explainability and safety dimensions of AI integration within the AGCAS framework. Because a single simulation run is sufficient to produce a legitimate, collision-free trajectory, this approach eliminates the need for computationally intensive search-based techniques. Furthermore, since the trajectories are generated within a high-fidelity 6-DOF environment, they are dynamically feasible by design. Each recorded control action can be directly traced to its simulated effect, satisfying both safety and explainability requirements. Overall, the proposed framework offers a scalable and interpretable solution that bridges the gap between rule-based rigidity and search-based computational burden, enabling pilot-like sequential avoidance behavior in a tractable and verifiable manner. Comprehensive evaluations were conducted using both benchmark tests and free-flight scenarios, including cases where the aircraft was deliberately placed in imminent collision conditions. The RL-based method demonstrated the capability to perform effective collision avoidance maneuvers across all test cases. It successfully executed intelligent and complex maneuver sequences, particularly in environments where such behavior is crucial. The proposed RL-based AGCAS method successfully prevented collisions in all benchmark tests. In 4 out of the 10 evaluated scenarios, it produced safer trajectories by maintaining a greater distance from the terrain compared to conventional methods. In the remaining tests, although the trajectories flew closer to the ground, this outcome is attributed to the reward structure used during training, which does not explicitly penalize proximity to the terrain. Instead, the agent receives a reward as long as the predefined clearance threshold is met. Therefore, such behavior is entirely expected and consistent with the training objective. Future research could revise the reward function to encourage increased terrain separation, potentially enhancing overall safety margins. Regarding pilot reaction dynamics, the non-intervened flight durations were remarkably consistent across all tested methods, with only millisecond-level deviations. This indicates that the RL-based approach offers comparable performance in terms of giving the pilot adequate time to respond. In addition, the RL method occasionally achieved slightly lower maximum load factors—approximately 0.5 G less—while staying within the operational safety limit of 9~G. This can be beneficial in scenarios where high-G maneuvers are undesirable or infeasible. Overall, although the proposed method is based on classical reinforcement learning algorithms, the results are highly promising. These findings suggest that integrating more advanced RL techniques—such as Hindsight Experience Replay (HER), curiosity-driven exploration, or generative adversarial imitation learning (GAIL) —could further enhance the performance, robustness, and adaptability of AGCAS systems in future research.
-
ÖgeYüksek devirli rulmanlarda açık yağlama sistemi için sprey enjektör tasarımı(Lisansüstü Eğitim Enstitüsü, 2022)Hava araçlarının veya roketlerin itki sistemlerinde kullanılabilen turbo jet motorların kullanıldığı aracın planlanan kullanım ömrüne uygun tasarlanabilmesi mühendislik açısından zorlayıcı bir optimizasyon problemidir. Motorlarda döner elemanların yataklandığı rulmanlar sistemin en kritik noktasını oluşturmakta dolayısıyla tasarlanan motorların kullanım ömrünü de bu kritik noktaları olan rulmanların kullanım ömrü belirlemektedir. Rulmanların kullanım ömrünün uzatılması ve daha pürüzsüz bir şekilde görevini yerine getirebilmesi için tarihi taş devrine dayanan, günümüze kadar ilk kullanılmaya başlandığı andan itibaren önemini yitirmemiş, uygarlığımızın temeli olan aletlerin ve makinelerin sorunsuz çalışmasında ve uzun ömürlü olmasında en büyük katkıya sahip; yağlama sistemi kullanılmaktadır. Yağlama sisteminin kalitesini belirleyen ve sistem gereksinimlerine uygun yağlama yapılabilmesini sağlayan en önemli tasarım noktası doğru enjektörün seçimidir. Bu tez çalışmasında yüksek devirli bir rulmanda açık yağlama sistemi gerekliliklerinin belirlenmesi ve bu gerekliliklere uygun yağlama enjektörü tasarımı çalışması anlatılacaktır. Bu tez kapsamında, yağlama enjektörü tasarımına varan ön tasarım çalışmaları 3 aşamada incelenmiş, bu ön tasarım çalışmaları sonucu elde edilen veriler ışığında enjektör tasarımı yapılmış, tasarlanan enjektörün testleri yapılmış ve deneysel sonuçlar elde edilmiş, bu sonuçlar teorik sonuçlar ile kıyaslanmış ve bu kıyaslamaya dayanarak çıkarımlar oluşturulmuştur. Ön tasarım çalışmasının ilk adımı, yağlama gereksinimlerini karşılamak için rulmanların ürettiği ısının belirlenmesidir. Rulmanlarda üretilen ısı, rulman geometrisine, rulman üzerindeki yüke ve rulmanın dönme hızına bağlıdır. Rulmanlarda ısı oluşumu dönen bilyelerden kaynaklanmaktadır. Bu nedenle hesaplamalarda kullanılacak kritik rulman geometrisi bilgileri bilyelerin bulunduğu bölgede toplanır. Rulmana ait gerekli hız ve boyut bilgileri bu geometri bilgileri kullanılarak hesaplanabilir. Yükler ve bilye hızları hesaplandıktan sonra sürtünmeden kaynaklı kaybedilen enerjinin tamamının ısıya dönüştüğü kabulü ile üretilen ısı hesaplanır. Ön tasarım çalışmasının ikinci adımı olarak rulmana gönderilecek olan yağlayıcının debi hesabı yapılmalıdır çünkü rulmanda üretilen ısının sistemden dışarıya atılması gerekmektedir ki rulman sıcaklığının kritik seviyelere ulaşması önlenebilsin. Rulmanın bu kritik seviyelerdeki sıcaklığa ulaşmasını önleyebilmek için rulmana yağlayıcı sıvı gönderilmektedir. Gönderilecek yağlayıcının debisini belirlemek için ısı transferi hesabı yapılabilir. Bu çalışmanın başlangıcında, gönderilen yağlayıcı ve havanın rulman çıkışında aynı sıcaklığa ulaşacağı kabulüyle, rulmanda üretilen ısının hava-yağ karışımına aktarıldığı ısıl denge kurulmuştur. Böylece karışımın rulman çıkış sıcaklığı tespit edilmiştir. Çıkış sıcaklığı belirlendikten sonra rulman sıcaklığını bulmak için iteratif bir çalışma yapılacaktır. Rulmanlara gönderilecek debi belirlendikten sonra, ön tasarım çalışmasının üçüncü ve son adımı doğru tip enjektör seçimidir. Yağlama için kullanılacak enjektörler sprey ve jet enjektörler olmak üzere iki seçenekte incelenmiştir. Enjektör tipine karar verebilmek için literatür taraması yapılarak sprey ve jet enjektör karşılaştırılmıştır. Literatür taramasından elde edilen bulgular göstermektedir ki, hava akışı hızlandığında jet akış büyük tanecikler halinde parçalandığı için hava akışı ile tam anlamı ile taşınamayıp kendi ağırlığı ile akış içerisinde dağılmaktadır. Bu durum jet akışın istenilen hedefe gitmemesine sebep olmaktadır. Rulmana doğru şekilde ulaşmayan yağlayıcının, rulmandan geçmeden ortamdan atılma olasılığı oldukça yüksektir. Ayrıca kapalı sistemlerde sızdırmazlık elemanları sayesinde rulman dışında gidecek bir yere sahip olmayan yağlayıcı, açık sistemde bu sızdırmazlık elemanlarının bulunmaması sebebi ile doğru taşınma kapasitesine sahip olmaz ise, yağlayıcı rulmana ulaşamayacak ve yağlama görevi başarı ile gerçekleştirilemeyecektir. Dolayısıyla bu tezde söz konusu yüksek devirli rulmanlarda açık yağlama sistemi kullanıldığında yağlayıcının tanecik yapısını küçültmek, dağılımını arttırmak ve ikincil hava akışı ile taşınabilir hale gelmesi için sprey enjektör tercih edilmiştir. Literatür çalışmalarında görülen, enjektör tasarımında kullanılan denklemlerin çoğunun ampirik yaklaşımlar varsayılarak yapılan testler sonucunda oluşturulan denklemler olduğudur. Bu tezde, tasarlanacak sistemin çalışma aralığındaki ve düşük hata oranlarına sahip denklemler referans alınmıştır. Bu tezde bahsedilen enjektör için kullanılan tasarım parametreleri; tahliye katsayısı, dağılma uzunluğu, film kalınlığı, yağ çıkış hızı, dağılma rejimleri, sprey dağılma açısı, Sauter ortalama çapı ve yay hesabıdır. Enjektör tasarımı enjektör gövdesi, döner parça, yay, yay tutucu, filtre ve yüksük olmak üzere 6 parçadan oluşmaktadır. Enjektör boyutları, küçük motorlara uyacak şekilde kasıtlı olarak küçüktür. Her ne kadar bu durum üretimsel bazı sıkıntılara sebep verse de kimi motorlar için küçük boyutlara sahip bir enjektör tasarımı zorunluluk olarak ortaya çıkmaktadır. Bu sebeple küçük boyuttaki enjektörlerin incelenmesi ve bu konuda çalışmalar yapılması bir gereklilik olarak ortadadır. Test sisteminde enjektörlerin akış ve basınç ilişkisinin belirlenmesi amaçlanmaktadır. Ayrıca sprey dağılım açısı farklı debilerde gözlemlenmiş ve sprey kalitesinin bir göstergesi olarak kullanılmıştır. Testler kademeli olarak artırılan farklı debilerde gerçekleştirilmiş ve her adımda kamera kayıtları alınarak sprey formları incelenmiştir. Yapılan testler sonucunda düşük basınçlarda püskürtme kalitesinin kötü olduğu ve istenilen debilerin fazla gönderildiği görülmüştür. Düşük basınçta akış enjektörde girdap şeklinde hızlanamadığı için enjektörden çıkan akışkan saçılarak dışarı çıkamaz. Bu nedenle akış birincil rüzgâr rejiminde gerçekleştiği için yeterli sprey formu elde edilememektedir. Sprey kalitesini artırmak için basıncın artırılması gerektiği açıktır. Ancak bu debiyi artıracağından, debi değeri ile sprey kalitesi arasındaki denge gözetilerek tasarım yapılmalıdır.
-
ÖgeExperimental and numerical study on thermal behavior of space equipment by using two resistor model(Graduate School, 2023)In this study, the qualification level Thermal Balance Test (TBT) performed on the On-Board Computer (OBC) developed within the framework of the satellite programs of the Turkish Aerospace is explained. The numerical solution developed using Simcenter 3D is explained. Test and analysis correlations are made, results and recommendations are explained. The IPS module, the power module of the OBC, has been tested in a vacuum environment. The main reason for choosing the IPS module is to evaluate heat generation and heat dissipation and thermal design. A IPS module (carrier and PCB), cover and baseplate are included in this work. It takes a lot of time to include every component on the PCB in the thermal analysis. For this reason, only components that emit heat above 50 mW are included in the analysis. The aim of this study is to validate the thermal analysis. This study does not contain electrical information. IPS module connections were chosen as AISI 310 stainless steel. According to the data sheets, thermal pads are placed under the components that require thermal pad application. In addition, thermal pad was applied between the carrier and the baseplate. Aluminum bulk structures had to be placed under 2 components due to supply problems. The DC-DC converters, which dissipate a lot of heat, are located in the lower part of the PCB. In order to conduct heat, the upper surface of the PCB under these components is left as copper. The layout of the other components is placed in such a way that they do not heat each other thermally according to the electrical constraints. The carrier design has been meticulously made with ECSS requirements in mind, to ensure mechanical strength and transfer heat to the baseplate by conduction. The baseplate is designed to hold the backplane and supporting all boards, dissipating the equipment heat to the satellite panel. It is also used to support the baseplate cover and carriers, hold them together and provide thermal conduction. There are 8 mounting holes in total for OBC equipment. The equipment is designed with these interfaces to be mounted on the satellite panel and the thermal plate in the thermal vacuum chamber. For this equipment to be used in the satellite system, the space environment is reflected in the thermal vacuum chamber. The qualification temperature range of the equipment is [-30°C, +60°C]. Hot case of qualification level was tested within the scope of this thesis. The thermal balance test was performed for the hot case, because in previous thermal camera measurements and analysis showed that the components were overheated. The thermal balance test was carried out at the Assembly, Integration and Test Center (AIT) of the Turkish Aerospace company. The HVT400 thermal vacuum chamber inside the clean room at AIT is used for equipment testing. Thermocouple placement was done before the equipment was put into the TVAC. Thermocouples are placed on components that emit heat above 50mW, on the PCB 1 mm away from the components, near the bolt connections, on the cover and on the baseplate. A total of 33 thermocouples are placed on the equipment. The IPS module is mounted on the plate in the thermal vacuum chamber using appropriate interfaces. After the necessary checks and connections are made, the door is closed. First, the pressurization process was carried out. The pressure in the chamber was reduced to 10-5 mbar. Tests were performed for 60°C with certain power configurations. Thermal stabilization criteria was determined as 0,5°C/hour for 60°C. The thermocouples data were read, the room temperature and pressure were normalized, and the door was opened. Analysis was performed using Simcenter 3D software. The Thermal/Flow solver was chosen for the solution of the thermal analysis. The Space Systems Thermal Solver ignores convection and creates a suitable solution environment for the space environment. One of the advantages of Simcenter 3D software for electronic equipment is that PCB material properties can be calculated. PCB layers, percentage of dielectric and copper material and layout can be seen, therefore the thermal conductivity is calculated by the software. Thanks to its automatic meshing feature, PCB and components are created automatically in 2D. As the analysis method, two resistor thermal modeling method was chosen among the compact modeling methods. The purpose of choosing this method is that it is simple and the data can be easily accessed. Simcenter 3D offers 2 different methods for modeling with two resistors. First; resistors from junction to case and from junction to board; the second: the resistors from junction to case and from case to board. These resistance values are calculated according to JEDEC standards. These resistors are supplied from component data sheets or manufacturers. In this study, an analysis model was created by using the resistors from junction to case and case to board. These resistance values were obtained from datasheets and manufacturers. Various assumptions were made for the values that could not be found. Other parameters required for modeling are maximum junction temperature, maximum case temperature, and heat dissipation of the components. The maximum junction temperature value and the maximum case temperature value are usually obtained from the datasheet. The heat dissipation values were obtained from the hardware design team working in the Turkish Aerospace company. The heat dissipation value calculation is not included in this thesis. Afterwards, boundary conditions were entered into the analysis. Ambient temperature, thermal plate temperature and radiation are given as boundary conditions. In order to calculate the radiation, the emissivity values of the components, aluminum parts and PCB were included in the analysis. View factors were calculated with the help of enclosure radiation property of Simcenter 3D. Three separate view factor values were assigned for the open sides and the closed side of the equipment. The conductivity value between materials depends on the material to fill the gap, surface roughness, material, surface finishing method and pressure. These values are usually obtained experimentally. As a result of the researches, many conductivity values were found for steel, aluminum and copper. Optimized results were obtained by performing several iterations. Finally, the analysis was reached to an acceptable level as a result of various iterations. The difference between test and analysis results is determined as 10% for acceptable level. This is well accepted value among the companies which conduct equipment thermal analysis. As a result, the correlation of analysis and test results has been completed.
-
ÖgeGyroless attitude estimation algorithm for nanosatellites(Graduate School, 2023)A nanosatellite, or nanosat for short, is a type of small satellite with a mass between 1 and 10 kilograms. They are often built using off-the-shelf components and can be launched relatively inexpensively, making them a popular choice for academic and commercial space missions. Nanosatellites can be used for a variety of applications, including Earth observation, communications, scientific research, and technology demonstration. Nanosatellites face several technical and operational challenges, including limited power, communication capabilities, and computing resources. Due to their small size, they have limited space for equipment and instrumentation, which can make it challenging to implement complex systems. Moreover, they may be more susceptible to radiation and thermal effects in space, which can impact their performance and longevity. Additionally, the short lifespan of some nanosatellites (often just a few years) may limit their usefulness for long-term missions or data collection. Attitude Determination and Control (ADC) subsystem is crucial for nanosatellites to fulfill their mission objectives. Almost 40% of nanosatellites use an active ADC system to control their attitude accurately and reliably. The accuracy of the sensors and the actuator's torque limit determines the ADC subsystem's capabilities. MEMS gyros are preferred for nanosatellites due to their low cost and weight, but they have lower accuracy and stability and may degrade or fail during the mission. Magnetometers and sun sensors are common attitude sensors for nanosatellites, but they face challenges when only one vector measurement is available. The development of higher accuracy and minimal sensor ADC subsystems is a research topic for nanosatellites. Gyroless attitude estimation is a technique used to determine the orientation of a nanosatellite in space without relying on traditional gyroscopes, which are often bulky, expensive, and consume a significant amount of power. Instead, this technique uses a combination of different sensors such as magnetometers, sun sensors, and star trackers, along with advanced algorithms and mathematical models, to estimate the attitude of the nanosatellite. The approach aims to overcome some of the limitations associated with traditional ADC systems and enable more cost-effective and reliable solutions for small satellite missions. There are several advantages of gyroless attitude estimation for nanosatellites, including: • Cost-effectiveness: Gyroless attitude estimation techniques can be less expensive compared to systems that require gyroscopes, which can be relatively expensive and consume significant power. • Reduced size and weight: Gyroscopes can be relatively large and heavy, making them impractical for use on small nanosatellites. Gyroless systems can be smaller and lighter, which is important for satellites with size and weight constraints. • Improved reliability: Since gyroscopes have moving parts, they can be prone to mechanical failure or degradation over time. Gyroless systems are less complex and can be more reliable. Overall, gyroless attitude estimation techniques can provide a practical and cost-effective solution for nanosatellites with limited resources and stringent mission requirements. This thesis presents and evaluates attitude estimation filters that are specifically developed for nanosatellites that do not have gyroscopes. Disturbances can have a significant impact on the accuracy of gyroless attitude estimation for nanosatellites. In particular, non-gravitational disturbances such as residual magnetic dipole moment (RMM), aerodynamic drag, and solar radiation pressure can induce torques on the satellite, causing it to deviate from its desired attitude. These disturbances can affect the accuracy of the attitude estimation algorithm, leading to errors in the estimated attitude. The RMM is the most important disturbance torque for nanosatellites because it interacts with the Earth's magnetic field and causes an external disturbance torque that affects the satellite's attitude. This is particularly important for small satellites like nanosatellites, which have a low moment of inertia and are more susceptible to external disturbances. The RMM can arise from several sources, such as the satellite's magnetic materials, electronic components, or even the solar cells. Modelling and compensating of RMM is essential in the attitude control system to maintain the satellite's stability and accuracy. Accurate modeling of RMM and designing filters capable of precise estimation are crucial in dynamic-based (gyroless) filters, where RMM plays a dominant role. When working with a dynamic-based (gyroless) filter, accurate estimation of the RMM becomes crucial due to its dominant role. As a result, the thesis has focused on developing models that accurately represent the RMM, and designing filters that can precisely estimate it in real-time. The integration of a filtering algorithm with static attitude estimation methods, such as the QUEST algorithm, can improve the accuracy and convergence speed of gyroless attitude estimation for nanosatellites. However, the approach has a major drawback when only one vector measurement is available, such as during eclipse phases when the sun sensor cannot provide a measurement. In these cases, using three-axis magnetometers (TAM) is necessary to ensure accurate attitude estimation. Therefore, TAM only attitude estimation when sun is not available is also proposed.
-
ÖgeNonlinear solver-aided estimation filter based geostationary satellite navigation with available GNSS signals(Graduate School, 2025-04-30)The advancement of geostationary satellite navigation systems is a cornerstone for numerous modern applications, including telecommunications, weather monitoring, and defense systems. Despite their utility, navigating geostationary satellites presents unique challenges due to the nonlinear dynamics of their orbits, orbital perturbations, and the limited geometric diversity of Global Navigation Satellite System (GNSS) signals. Traditional estimation methods, such as the Extended Kalman Filter (EKF), are often inadequate for addressing these challenges, particularly in scenarios involving limited satellite visibility or high noise conditions. This thesis proposes an innovative approach combining nonlinear solvers with adaptive filtering techniques to overcome these limitations, delivering enhanced accuracy and robustness in satellite navigation. The methodology centers around integrating nonlinear solvers, including the Newton-Raphson Method (NRM), Levenberg-Marquardt Algorithm (LMA), and Least Squares Method (LSM), within an Adaptive Extended Kalman Filter (AEKF) framework. This hybrid approach leverages the strengths of both linear and nonlinear estimation methods, incorporating historical data through a configurable window size to balance responsiveness and noise reduction. The adaptive nature of the filter allows dynamic adjustments to varying environmental conditions and measurement noise, making it particularly suitable for constrained scenarios, such as narrow Field of View (FOV) configurations. To evaluate the proposed methodology, extensive simulation scenarios were designed. These scenarios varied in satellite visibility, FOV, and window size parameters, providing a comprehensive assessment of the filter's performance. Key scenarios included wide FOV configurations, narrow FOV setups, and scenarios with varying filter window sizes (N). Performance metrics focused on position and velocity Root Mean Square (RMS) errors, maximum errors, and statistical adherence to the 3-sigma boundary. The results demonstrated significant improvements in both position and velocity estimation accuracy compared to conventional EKF and standalone AEKF methods. For instance, under narrow FOV conditions, the LSM-Aided AEKF consistently maintained errors within acceptable limits, demonstrating robustness against limited satellite geometry and high noise. Specific findings include the following: Position Estimation: The adaptive framework achieved substantial error reductions, with RMS errors significantly lower than those of traditional methods, even in narrow FOV scenarios. For example, under Scenario M, the position RMS errors were approximately 13 meters (X), 20 meters (Y), and 2 meters (Z), highlighting the effectiveness of the adaptive approach. Velocity Estimation: Velocity RMS errors were consistently low, reflecting the filter's capability to track dynamic changes accurately. The integration of nonlinear solvers improved stability during rapid state transitions. Clock Bias Synchronization: The proposed filter maintained robust temporal synchronization, with minimal mean clock bias errors, even under narrow FOV conditions. Additionally, the study explored the impact of filter window size (N) on performance. Scenarios with smaller window sizes provided better responsiveness but higher sensitivity to noise, while larger windows improved noise resilience at the cost of reduced adaptability. A balance was achieved with window sizes around N = 59, which offered a favorable trade-off between accuracy and stability. The thesis also underscores the importance of adaptive capabilities in navigation systems. The integration of nonlinear solvers allows the filter to handle challenging conditions, such as limited satellite visibility and high orbital perturbations. Moreover, the study highlights the potential of machine learning techniques for further enhancing the filter's adaptability, paving the way for future research in intelligent navigation systems. In conclusion, this research contributes to the advancement of geostationary satellite navigation technologies by presenting a robust, adaptive estimation framework. The LSM-Aided AEKF proves to be an effective solution for overcoming the limitations of traditional methods, offering improved accuracy, reliability, and adaptability across a wide range of operational conditions. The findings of this thesis have significant implications for the design and deployment of next-generation satellite navigation systems, enabling more precise and dependable operations in diverse applications.