The growing intensity and frequency of fires and volcanic activity have heightened the demand for real-time monitoring of thermal events. This Thesis presents the first comprehensive end-to-end processing pipeline for real-time segmentation of thermal hotspots in raw multispectral imagery. The pipeline combines an onboard Deep Learning (DL) design with the use of raw imagery to achieve real-time performance. The processing pipeline was thoroughly tested on CubeSat edge computing hardware, demonstrating its feasibility for CubeSats with limited computing resources. The DL architecture used was specifically designed in this project to address onboard constraints such as model size, complexity and inference time. The satellite imagery used in this work is from the Sentinel-2 mission, and a key contribution of this project is the creation of SegTHRawS, the first dataset for thermal hotspot segmentation in raw multispectral imagery.

Given the rapid growth in popularity of mega-constellations for telecommunication purposes, this thesis aims to identify methods for designing orbital layouts of such constellations. Traditional optimisation methods for regularly sized constellations do not scale due to the number of satellites involved. Therefore, this thesis investigates the design of mega-constellations and how to achieve an optimal layout within a reasonable timeframe.

Initially, the figures of merit most commonly used in mega-constellation design were identified. The most important figure of merit used in all types of missions, except for Earth observation, is visibility, i.e., the number of satellites visible from a point on the ground. For Earth observation missions, the revisit time is more relevant than visibility. Thus, the thesis focused on mission objectives where visibility is the main figure of merit, such as Satcom, Satnav, IoT, etc. There is also more literature and data available on such constellations, including Starlink, OneWeb, Project Kuiper, and others. Consequently, two closely related figures of merit were selected for this study: minimum visibility and mean visibility. The former ensures an N-fold uninterrupted coverage, while the latter provides a general metric of how many satellites are visible over time. Also, the main focus was on Walker constellations as this is the most common geometry used for mega-constellations.

The visibility computation begins with constellation propagation. Due to the large number of satellites in a mega-constellation, existing tools used at ESOC, such as Godot, are insufficient. Therefore, a self-written tool was developed, named mcdo (Mega-Constellation Design Optimisation), which utilises the basic functionality of Godot and takes advantage of NumPy by applying vectorisation to notoriously slow Python. This approach resulted in a decrease in computational time by a factor of 100x with respect to godot.cosmos.BallisticPropagator.

Moreover, the visibility computation was accelerated with the utilisation of graphic processing units (GPU) and simplifications such as North-South symmetry, longitude averaging, or estimating visibility for a single time instance. Depending on the methods and simulation setups used, a reduction of 400x-120,000x in computational time was achieved for visibility computation.

Additionally, the parametric analysis of large Walker constellations yielded some valuable discoveries. For instance, the number of planes P and phasing parameter F do not influence the mean visibility but may have a significant effect on minimum visibility. This means that P and F can be omitted when designing for the mean visibility. It was also revealed that the mean visibility curve scales linearly with the number of satellites N, allowing to avoid propagation of very large constellations, and to scale up the mean visibility curve from smaller constellations instead. Parametric analysis of other parameters provided a general insight into their effect on visibility.

The improved computational efficiency of visibility computation for large Walker constellations enabled the application of multi-shell constellation design. A case study was set up with a requirement of uninterrupted coverage of 50 satellites over European latitudes (35-70 deg), assuming that all shells were at the same altitude of 700 km with a goal to minimise N. The analysis revealed that for two- and three-shell layouts, the minimum N was attained when higher-inclination shells had more satellites than lower-inclination ones. Moreover, methods for design acceleration were discussed.

The "building blocks" method was proposed for designing mega-constellations with more than three shells. This method starts from placing shells at high inclinations and then gradually lowers the inclination of the subsequent shells. It takes advantage of visibility properties of Walker constellations: the higher-inclination shells can cover both high and low latitudes, while the lower-inclination shells can only cover low latitudes. This method reduces the computational time drastically: by a factor of 150x for three-shell layouts and even more for larger numbers of shells.

The presented research offered valuable insights into initial phases of design of the orbital layout of mega-constellations. The mcdo tool and the obtained results were already applied for internal projects at ESA, demonstrating their relevance and usefulness.

In recent years, researchers have proposed a variety of approaches to tackle the problem of orbital debris. Debris targets are diverse, and prior knowledge may be limited, with unknown uncooperative debris targets being the most challenging category. A crucial portion of any debris capture scenario is the observation phase during the approach. During this phase, the chaser spacecraft attempts to learn as much as possible about the target by using remote sensing, of which relative pose is of particular interest, to enable the advancement of the mission towards eventual capture.

This research utilizes the fusion of data from a scanning LiDAR with a long-wavelength infrared camera to estimate the relative pose of an unknown uncooperative target. Two separate bespoke pose estimation algorithms, color-ICP and Feature Matching, were developed and tested with laboratory experiments mimicking the close-approach phase with a target under various lighting conditions and relative motion rates. The color-ICP algorithm uses a thermal infrared-infused color-assisted Generalized Iterative Closest Points method, while the Feature Matching algorithm uses computer vision on LiDAR point-infused thermal images to track BRISK feature points in each frame to estimate pose.

In general, the color-ICP algorithm delivered more accurate results throughout the range of experiments, though the fusion was slightly detrimental while the target is being heated or cooled. The Feature Matching algorithm contains a large amount of tunable parameters, making the estimation highly sensitive yet
versatile, demonstrating that harsh lighting conditions can be mitigated with accurate features tracked after the implementation of image processing techniques. Overall, the end product shows promise as a light-agnostic remote sensing and pose estimation solution.

This research contributes to the advancement of active debris removal theory and explores two promising avenues for LiDAR-infrared sensor fusion for pose estimation, laying the groundwork for further iterations exploring this sensor pairing. The resulting use case is a conceivable scenario in which these sensors work together to supplement individual strengths and mitigate disadvantages throughout the approach phase of a debris removal mission.

Inaccuracies in robotic arms can significantly hinder their performance in tasks where precision is critical. This thesis focuses on the kinematic calibration and elasticity compensation of a 6 degrees of freedom robotic arm integrated into the Lunar Rover Mini, developed in collaboration with the Robotics and Mechatronics Institute of the German Aerospace Center (DLR), Wessling. The arm, constructed using 3D-printed components and driven by affordable RC servo motors, experiences notable inaccuracies in end-effector positioning due to joint flexibility and structural deformation, especially under load. This research aims to enhance the arm's accuracy through a model-based calibration approach that compensates for both elastic deformations and geometric misalignments, addressing the absence of feedback sensors. This cost-effective approach, which requires only 3D measurements of the end-effector's position, has resulted in an approximately 80% reduction in the robotic arm's average position error, proving advanced robotic technologies can be more accessible for educational and research applications.

On-orbit servicing and active debris removal missions are becoming increasingly important to limit the growth of space debris in Earth’s orbit. SmallSats offer a promising option for reducing the cost and risk of these missions, allowing for their expanded use on a wider scale. For these missions to be successful, SmallSats are required to operate at very-close ranges to a target object. Relative visual navigation at this distance still presents a hurdle that needs to be overcome prior to implementing these platforms. This work analyses the suitability of Simultaneous Localization and Mapping (SLAM) for navigation at very-close ranges and proposes a modified algorithm to increase performance. Characterization of this algorithm is performed against a custom generated lab-based image dataset simulating final approach to a satellite. Algorithm performance shows promising results for SLAM use in very-close range environments and presents a first step in the development of this SmallSat navigation technology.

Nowadays, many space missions require highly accurate pointing for Earth observation or cosmic vision purposes. However, the vibration environment from a spacecraft's structure and reaction wheels can cause disturbances in its line-of-sight stability and severely impact image quality. Additionally, these effects are not known precisely due to limitations in ground testing, and this uncertainty leads to significant challenges in control. This thesis tackles these problems by creating a control design and verification framework. Modern scientific literature explores three solutions: high-fidelity nonlinear modelling, advanced control design methods, and optimized verification campaigns. Respectively, they provide a reliable testing environment, directly handle structural dynamics, and guarantee system stability. While the three approaches are usually studied separately, the thesis proposes a novel combination using a payload isolation platform. The synergy of the framework offers a robust solution for precise pointing in flexible spacecraft and enables further mass reductions in the future of space exploration.

The continued increase in the number of satellites in low Earth orbit has led to a growing threat of collisions between space objects. On-orbit servicing and active debris removal missions can alleviate this threat by extending the lifetime of active satellites and deorbiting inactive ones, but this requires advanced guidance and control algorithms for the rendezvous phase. Recently, various control policies based on machine learning have been proposed to leverage the advantages of neural networks. One notable technique that has shown much potential in asteroid and planetary landing scenarios is reinforcement meta-learning. This technique consists of training recurrent neural networks in uncertain scenarios in order to develop highly robust control policies that can adapt to unknown conditions in real-time. The goal of this thesis was to apply the meta-learning technique to a rendezvous scenario.
Thus, throughout this project a recurrent neural network was trained via reinforcement meta-learning to generate a control policy that can perform the final approach maneuver of a chaser spacecraft towards a rotating target. A feedforward network was also trained for comparison. The learning algorithm used to train the policy is the Proximal Policy Optimization algorithm, which is a modern actor-critic method that has shown good performance in several continuous control settings. A virtual environment was developed in Python to simulate the rendezvous scenario and collect data to train the policy.
Before beginning with the training process, the hyperparameters of the model were tuned to ensure a smooth and efficient learning process. The three components that required tuning were the learning algorithm, the architecture of the neural networks, and the reward function. Each of these components was tuned in turn, primarily through trial and error. This process required the learning algorithm to be executed multiple times, using a different combination of hyperparameters on each iteration. By repeating this process over a large search space, suitable hyperparameters were found for the learning algorithm and the neural networks. The hyperparameters were chosen to maximize the amount of reward achieved by the policy while maintaining a reasonable training runtime. The reward function was split into several components to guide the policy towards its objective, thereby improving the speed at which the policy learns. Each of the components of the reward function represented some partial goal that the controller had to accomplish. Tuning the relative weight of these components was a challenging process since it often leads to trade-offs between different policy behaviors. Once the tuning process was completed, a sensitivity study was performed to ensure that the model can be used for different kinds of rendezvous trajectories. The sensitivity analysis was performed by training the model on different scenarios, including different orbit altitudes, different distances from the target, different target sizes, and different target rotation speeds. The results of this study showed that the model could be applied to most of these scenarios without the need for any major changes.
After completing the tuning and the sensitivity analysis, the recurrent and the feedforward policies were each trained on a partially observable environment, and their performance was evaluated using a Monte Carlo simulation for a total of one thousand trajectories. The results showed that the recurrent policy was able to learn how to infer hidden information from the environment, which led it to have a much better performance than the feedforward policy. However, the recurrent policy was not without its limitations, since it could not always generate collision-free trajectories, especially when the target rotated at a faster rate. Overall, this thesis showed that reinforcement meta-learning can be a valuable tool for executing complex rendezvous maneuvers, which may be useful now that active debris removal missions are becoming a reality. Furthermore, this thesis also presented a description of how the model was designed and tuned, so that other machine learning practitioners may apply the technique to different scenarios.

The Geosynchronous Large Debris Reorbiter concept is an active debris removal method proposed for defunct satellites in geostationary orbit. This method uses the electrostatic tractor (or tug), which is a spacecraft that controls a mutual electrostatic force on a target and uses this force to slowly accelerate the target towards or away from the tug. The challenging task of station-keeping a few dozen meters away between the tug and the debris spacecraft must be done with thrusters that can only intermittently fire. In between the firing, the electron gun is used to control the tug and debris potential. This research investigates different thruster options and discrete control algorithms for this application of the electrostatic tractor. The objective is to devise and simulate a reliable propulsion system for the geosynchronous large debris reorbiter concept by simulating and analyzing different micro-propulsion thrusting options and developing a discrete thrusting solution using pulsed thrust and active charging control.

Laser communication provides numerous benefits over typical Radio Frequency communication, such as lower power, not occupying regulated frequency bands, possibilities for much higher data rates and resistance to jamming. Combined with satellite constellations, Laser Inter-satellite Links (LISL) can enable global connectivity. However, satellites move at fast relative velocities, while optical beam divergence angles are in micro-radian levels. Thus, low-latency and precise position data between linking satellites is crucial. This thesis investigates the LISL conditions in a combined LEO/MEO constellation and the applicability of on-board GNSS-based Orbit Determination (OD) and Orbit Prediction (OP). Novel methods, such as Preprocessing Extended and Single-propagation Unscented Kalman Filters are tested and compared to typical GNSS-OD methods. Analyzing Pointing Uncertainty contributions in various link cases, results indicated that fully-kinematic methods could support LISL for 100-s periods, whereas reduced-dynamic OD-OP methods performed more consistently.

Mitigation strategies which eliminate existing space debris, such as with Active Space Debris Removal (ASDR) missions, are now more important than ever regarding the ever-growing space debris population problem. One of the considered ASDR approaches uses a net as a capturing strategy. The benefits of such strategy are to allow for a large capturing distance, high compatibility for different space debris sizes and reduced accuracy requirements.

A key requirement of any ASDR missions is that during capture, no new space debris is to be generated during the process. However, when simulating net capturing in the literature, the potential to break of vulnerable structures, like antennas or solar panels is often neglected. Such elements may show an enhanced risk of failure, especially if these are already damaged, potentially contributing to even more space debris.

A discrete Multi-Spring-Damper net model was used to simulate the 20 m/s-frontal impact of a 30 m x 30 m net onto an ESA Envisat mock-up. The Envisat was modelled as a two rigid-body system with a Single-Degree-of-Freedom hinge connection. A sequential modelling strategy was implemented, which de-coupled all the necessary dynamic and structural models. More than two large sub-structures (the Ka-band antenna dish and solar array) were found to have a high likelihood of breaking, leading to the recommendation of several design mitigation strategies using two types of sensitivity analysis. With secondary space debris being generated, net capturing is found to be riskier than originally assumed throughout the literature.

To enable greater access to space, this study aims to develop a cost-effective PocketQube Deployment System capable of deploying multiple PocketQubes from various launch vehicles. Mission cost, of which launch cost is a significant driver, can be reduced by minimizing deployment system mass. This can be achieved by structural analyses of single and multi pod configurations using finite element method. A metric called ‘deployer ratio’, defined as the ratio of payload and deployer mass, is introduced to compare the effectiveness of CubeSat and PocketQube deployers. From the perspective of deployer mass and mission cost, it was observed that it is more efficient to launch PocketQubes from a multi pod deployer rather than a single pod deployer. Although the estimated launch cost for PocketQubes is currently equivalent to commercial manufacturers, further cost reduction is possible by optimizing the mass of such a multi pod deployer.

Traditional spacecraft design has reached the limits of size and volume that current launchers allow. This is especially true for space telescopes (James Webb), where size is directly correlated with performance. A solution to this is on-orbit assembly, however some enabling technologies are insufficiently developed. One of the most important factors is the design of standardized interfaces for modular designs (only three existing designs). This thesis proposes a mechanical interface design - including data and electrical transfer - meant to allow the assembly of a modular platform (including all spacecraft subsystems except the mirror segments) upon which a support layer and a mirror layer can be built/assembled. The design presents a unique mechanism allowing for assembly outside of the platform plane direction, thus allowing disassembly of a minimal amount of modules in the event of servicing. Three different iterations of the design were considered, and simulations for both the assembly and assembled state have been performed in ANSYS. To compare performance with existing interfaces, a scaled down model was simulated. The results of the preliminary simulations show a performance increase of at least 200% for each load case. A physical prototype was manufactured and tested, however destructive testing was not performed.

This project aimed to develop and verify a numerical model of a brush. This model is meant to aid the study and design of brushes in Active Space Debris Capture. The model is based on the Discrete Element Method and uses the YADE software package to construct the brush model. This model simulates the dynamics of a bristle, their collisions among themselves and other bodies. The behaviour of the brush predicted by the DEM model was compared to observations recorded from three different experiments. The first test compared the dynamic behaviour of bristle, the second focused on the brush's static behaviour, and the final test studied its dynamic behaviour. The experiment observations fall within the behaviour predicted by the model, within the limits of the experiment.

For several decades, humanity has worked on cataloguing the population of near-Earth asteroids (NEA's). Knowledge of these small members of the Solar system not only helps in defending Earth from asteroid impacts; study of the population might also provide valuable scientific insights, and open economic possibilities through developments such as asteroid mining. Surveys from Earth currently face challenges in detecting the smallest NEA's: day- and night cycles, weather patterns, and atmospheric distortion have encouraged several recent studies into the possibility of NEA surveys from deep space.

In this thesis, an extension to these proposals is studied: using multiple spacecraft co-operating on the survey. Next to the immediate increase in the data gathering capabilities, such systems offer synergistic benefits: Firstly, spacecraft can be placed in such a way as to cover each others blind spots caused by e.g. Solar glare. Secondly, imaging from multiple directions allows for triangulation to more quickly determine the orbit. Lastly, such a system could feature implementation of advanced search strategies, for example utilizing a part of the system as follow-up telescopes.

As one of the first works on predicting and optimizing the performance of such a multi-spacecraft NEA survey system, the work aims to provide a foundation on how to compose such a system, and what orbital configurations to select for its operation. A simulation tool was developed which explicitly models a NEA survey, and this model was validated against other research works and surveys. Using this tool, the behavior of the system is studied under changing of various parameters such as the number of spacecraft, thermal infrared or visual light telescopes, and various orbital elements. Following this, numerical optimization was performed to obtain conclusions with regards to the optimal composition and position of the system. The findings are supported by an investigation into the underlying principles driving the performance of the survey. Ultimately, results are obtained which can be used for more detailed studies into design of future NEA survey missions, or trade-offs against other concepts.

Spacecraft are under constant threat from hypervelocity impacts by micrometeoroids and space debris. Due to their large surface area and hence increased probability of being impacted by debris, solar arrays are of particular concern. For larger, trackable debris objects, collision avoidance manoeuvres can be performed so as the minimise the damage caused to the spacecraft structure. However, for untrackable debris, particularly among the 1-3mm size range, strikes can occur at any stage and without any prior warning. During these collisions, momentum transfer takes place whereby the initial projectile momentum is imparted to the target. Due to the Momentum Enhancement Factor (MEF) caused by ejecta being expelled from the target, the actual value of transferred momentum can vary greatly depending on the product of initial projectile momentum and this factor. In this thesis, MEF values for a range of different impact conditions and scenarios are obtained. Hypervelocity impact simulations using smoothed particle hydrodynamics discretisation form the basis of these studies, enabling high impact velocities to be simulated which would otherwise be impossible using existing test equipment.

Simplified models of Hubble Space Telescope solar arrays are implemented in order to gather large amounts of data and analyse the general trends and parameters which influence the MEF. On the whole it is shown that when the projectile momentum is corrected with the MEF to yield the actual momentum transferred, it can map to a whole range of potential projectile mass/velocity/shape combinations. This data alone is not enough to distinguish between impact types, but by reducing the very coarse MEF assumptions used in prior work, the uncertainties surrounding in situ perturbation data to validate debris environment models can begin to be reduced.

Reaction wheels are often the main contributor to microvibrations on satellites. These vibrations are a source of pointing errors and therefore need to be limited as much as possible. The miniaturization trend in the space industry and universities fostered the production of very small reaction wheels, suited to Cubesat and PocketQube scales. A novel method to measure microvibrations for these kinds of wheels is proposed in this thesis. Due to budget and resolution constraints, a standardized method is indeed not available for this
scale yet. The main idea behind this method is to amplify vibrations through a flexible mounting bracket on which the wheels are fixed and later calculate the source disturbance via a transmissibility function. This measurement method will be studied for validation against some initial requirements. Two types of tests are proposed for this purpose: In the first, a MEMS accelerometer is attached to the mounting bracket to measure accelerations, while in the second a laser is pointed to a mirror on the mounting bracket to measure displacement. The tests will be compared with two models: the empirical model, which is a parameter estimation model with test results as input; and the unified model, which combines the empirical model and an analytical model, simulating the equations of motion of the system with input parameters from a CAD model. The main results will include microvibration waterfall plots and static and dynamic imbalance values. Although it was not possible to entirely validate the test bench prototype within the scope of this thesis, room for improvement is left, with some hints for possible future work.

At the Technical University of Delft(TU Delft), a Kirigami inspired method of spirally deploying a coiled up band into a parabolic reflector is being studied for its use in space.
This thesis performs a thermo-elastic analysis on the spiral dish Antenna (SDA) using a combination of ESATAN-TMS, Abaqus, and python scripting. After analysing a reflector for different low earth orbital cases in ESATAN, the heat fluxes are transferred to Abaqus to perform a second thermal analysis. A base SDA is modelled with an Aluminum zipper and Carbon-fiber-reinforced polymers (CFRP) band. The SDA is shown to experience temperatures ranging from 175K to 375K with temperature changes of 100K in less than 100s.
The reflector can experiences maximum temperature deltas of 93K across the reflector at a single moment during orbit. These temperature changes cause the SDA to have a maximum displacement of 20mm with a maximum root mean square (RMS) of 6mm. This thesis also shows the creation of a cross pattern on the SDA most likely created by the coefficient of thermal expansion (CTE) mismatch between the zipper and band material. An improved SDA is modelled after an initial parametrization, showing an improvement to the displacement field with the max RMS being around 1mm.
This thesis shows how the base model of the SDA deforms into an cross pattern and experience high displacement regions near the end of the spiral interface. This version of the SDA is shown to not be able to perform in space. A short parametrization analysis with regards to materials choice, spiral geometry, and
thickness does show a path in which the SDA can perform similarly to a normal reflector with the same design parameters.

There is no denying that the ever-increasing demand in space endeavours calls for more sustainable exploitation of the space environment. Approaches to mitigate space debris, such as ADR, are frequently suggested. Most of these proposals suggest a single sensor approach to provide accurate and continuous shape and pose estimation of the target. However, a more robust system can be developed by relying on input from multiple sensors with different modalities.

Compared to earlier literature, several alternative multimodal methods using visual-Lidar data have been researched during this thesis, of which the most promising method has been investigated in more detail. The method suggests to determine the 3D location of 2D features by projecting these on detected 3D planes, thereby fusing the visual-Lidar data at feature level. The visual-Lidar data has been acquired through both simulation in Blender using Blensor and through experimentation using an visual camera, a scanning Lidar and a robotic arm.

To verify and validate the proposed multimodal feature detection method, the detected 3D features are compared to the ground truth directly. Next to that, the method is also verified through analysis of the end-to-end process to estimate the relative pose of an unknown target. Where the resulting 3D features serve as input into an particle filter, combined with an EKF, based on the FastSLAM algorithm. The proposed method showed promising results, encouraging further research to determine the pose based on plane normal vectors for the relative pose estimation to operate during adverse illumination conditions.

Drones need to be able to detect and localize each other if they are to collaborate in multi-robot teams or swarms. Typically, computer vision methods based on visual appearance are investigated to this end. In contrast, in this work, a method based on dense optical flow (OF) is developed that detects dynamic objects. This is achieved by comparing the flow vectors of dense OF with the direction to the Focus of Expansion (FoE) in the image plane. A simulation in AirSim is developed to validate this approach and to create datasets for motion-based object detection of MAVs. This simulation includes ground-truth FoE, depth, OF and IMU data. The results show that this method performs well if the OF vector's magnitude is large enough and its angle is sufficiently different from those of static world points. We expect that the presented method will serve as a useful baseline for deep learning methods that use dense optical flow as input.