The training process of machine learning models for self-driving applications suffers from bottlenecks during loading and processing of LiDAR point clouds with large storage complexity.
Many studies aim to remedy this problem from an implementation perspective by developing efficient data loading and processing pipelines.
This study, on the other hand, explores an alternative approach by augmenting data representations to achieve lower storage complexity known as point cloud compression.

A broad analysis is presented on novel point cloud compression codecs using tensor decomposition methods.
Several point cloud representations and tensor decomposition methods are considered over a range of hyperparameter choices and compression values.
In order to assess the performance of the presented codecs: the compression rate, quality of the reconstruction, and time complexity is compared to the octree-based baseline model: TMC13.
Compared to the baseline model, the performance of the presented tensor decomposition-based codecs falls short. One of the presented codecs does notably outperform the others. This codec uses synthetic tensorization followed by sorting using z-location and decomposition using the TT-SVD algorithm.
Sorting by z-value isolates the ground plane, which is a dominant low-rank feature, which can effectively be decomposed using the TT-SVD algorithm yielding adequate results.

Several limitations of the presented tensor decomposition-based codecs are: the omission of bitwise compression on the factor matrices, and the trade-off between bitwise precision and truncation due to tensor decomposition.
Future work could improve in these areas along with considering the use of different heuristics and optimizing the tensor network topology.

The thesis presents a novel approach to optimizing input computation for minimizing classification error in image classification tasks. It leverages the capabilities of the Soft Actor-Critic (SAC) algorithm, a reinforcement learning method tailored for continuous action spaces. The focus is on developing a real-time adaptable feedback loop that continuously learns and adjusts inputs based on classifier output probabilities. Key to this approach is the incorporation of a Gated Recurrent Unit (GRU) architecture within the SAC framework to capture temporal dependencies, addressing the challenge of ever-increasing state dimensions.

Automotive radars are increasingly used in automated driving systems due to their cost effectiveness, ease of integration, and ability to withstand adverse weather conditions. Semantic segmentation for radar point clouds is a crucial step in radar pre-processing, which can be used on almost all downstream tasks of radars, such as detection and tracking. Radar point clouds are noisier compared to LiDAR point clouds due to sensor noise and the multi-path propagation, which makes the segmentation task for radar more challenging. In this paper, we address the problem of segmentation in noisy radar point clouds in terms of ghost target vs. real detection, moving vs. static objects as well as semantic segmentation of moving road users. We demonstrate how these three tasks can be performed in a single, unified pipeline using an auto-labeled radar dataset. Our approach, called Real, Moving, and Semantic Segmentation Network (RMSNet), is able to output point-wise labels for all three tasks simultaneously. On our dataset, RMSNet attains an IoU of 82.5% for real detection segmentation, 66.9% IoU for moving object segmentation, and 64.9% mIoU for semantic segmentation. We also did a live demonstration, using the model on two different radars. The result shows that the model is capable of running inference in real-time and demonstrates good ability in model generalization.

Radiation pattern measurements is a critical step in characterizing antennas before they are used in any system for a specific application. To identify any defects or to acquire the true radiation characteristics of the antenna under test, densely sampled measurements are desired. However, this results in a time and cost expensive measurement process. Compressed sensing allows accurate reconstruction of radiation patterns using a reduced number of measurements. To ensure exact recovery, it is necessary to select an optimal sampling strategy as well as an effective reconstruction method.

In this thesis, the discrete Fourier transform and spherical harmonic expansion of the electric field are used to obtain a sparse representation of radiation patterns. As a variation from the basis pursuit optimization problem which is widely used in compressed sensing for antenna measurements, a sparsity enhancing weighted l1-norm minimization problem is considered. The weights are determined from prior information on antenna from electromagnetic simulations. The proposed method, after investigation with various antennas and comparison with existing benchmark results in a further reduction of number of required measurements. A near-optimal sampling technique is adopted to acquire measurement in an incoherent manner for exact recovery of the pattern. The performance of the method has been evaluated using error metrics specific to important parameters of the radiation pattern such as the gain, peak side lobe level and half power beam width. Radiation patterns with non-idealities and distortions have also been recovered with high accuracy from a small number of measurements using the proposed method.

Increasing the efficiency of wind farms is important for speeding up the transition from fossil fuels to renewable energy sources. Current wind farm control relies on maximization of power generation of individual turbines. However, research has demonstrated that plant-wide wind farm control could optimize the performance of a wind farm. Wind farm simulation tools are crucial in designing, testing, and validating wind farm controllers. Fatigue, Aerodynamics, Structures, and Turbulence Farm tool (FAST.Farm) is a recently developed multi-physics engineering tool for modeling power performance and structural loads by solving the aero-hydro-servoelastic dynamics of each individual turbine within a farm. FAST.Farm aims to balance the need for accurate modeling of the relevant physics while maintaining low computational costs. However, designing controllers in FAST.Farm lacks flexibility and interactivity compared with MATLAB/Simulink. The capabilities of FAST.Farm for control design purposes can be extended through a co-simulation with MATLAB/Simulink. Therefore this thesis presents a FAST.Farm and MATLAB/Simulink interface. Consequently, this interface was used to implement and simulate wind farm controllers in FAST.Farm. FAST.Farm and MATLAB/Simulink are coupled by linking the individual Open Fatigue, Aerodynamics, Structures, and Turbulence tool (OpenFAST) instances in FAST.Farm to MATLAB with the use of an Message Passing Interface (MPI) and MATLAB Executable (MEX) functions. An Active Power Controller (APC) was implemented and simulated in FAST.Farm with the use of this interface. The APC responds to grid requirements through the control of wind farm power output. Comparing FAST.Farm and Simulator fOr Wind Farm Applications (SOWFA) simulation results of the APC shows that FAST.Farm reduces the computation time for a 10-minute simulation from 24 hours to 15 minutes, with little detriment to the accuracy of the simulation results. Although SOWFA remains the preferred validation tool, the FAST.Farm and MATLAB/Simulink interface supports developing and accelerates testing advanced closed-loop control at the wind turbine and wind farm levels.

Macro X-ray fluorescence (MA-XRF) is a recently developed technology allowing to obtain elemental information from cultural heritage objects. This information can, for example, be used to identify pigments used in a painting. Yet, the extended period of time it takes to scan an object is a major issue within MA-XRf. For instance, it took about 60 days to scan the Ghent Altarpiece. The long scanning time is a consequence of the necessary dwell time per pixel to create a robustly interpretable spectrum: the higher the dwell time, the higher the signalto-noise ratio (SNR), hence, the easier to detect elements. This thesis explores a possible solution for this problem using a denoising algorithm that increases the signal-to-noise ratio post-acquisition by exploiting the similarity between neighbouring pixels and spectra. To this end, a customized method of wavelet filter bank denoising is proposed. Current thresholding methods used in wavelet filter bank denoising are not suitable for filtering MA-XRF data, therefore, a novel thresholding method is introduced. Here, the widely used universal thresholding method is used as a basis, for which the formula for calculating the standard deviation of the detail coefficients of a channel is altered. Several design parameters of wavelet filter bank denoising were evaluated using a synthetic dataset, for which the performance quality indicators root mean square error (RMSE), mean absolute error (MAE) and SNR were determined. The parameters for which we optimized were the mother wavelet, the number of decomposition levels, and the number of neighbouring channels used for determining the standard deviation σ for thresholding. Good performance was obtained with the haar, db2, and coif1 wavelets, all at 3 levels of decomposition. A suitable number of neighbouring channels depended on the decomposition level and was determined to be 3 (on each side of the channel). Herewith, the signal-to-noise ratio was improved for both the average pixel spectra and the sum spectrum. The filtered synthetic dataset simulated to have a dwell time of 0.5 seconds had a SNR approximately equal to the raw synthetic dataset simulated to have a dwell time of 0.75 seconds. Hence, the algorithm succeeded in lowering the necessary dwell time. A case study of a daguerreotype was used to test the proposed denoising algorithm.

This project develops a channel estimation technique for millimeter wave (mmWave) communication systems. Our method exploits the sparse structure in mmWave channels for low training overhead and accounts for the phase errors in the channel measurements due to phase noise at the oscillator. Specifically, in IEEE 802.11ad/ay-based mmWave systems, the phase errors within a beam refinement protocol packet are almost the same, while the errors across different packets are substantially different.

We show that standard compressed sensing algorithms that treat phase noise as a constant fail when channel measurements are acquired over multiple beam refinement protocol packets. Most of the methods that have addressed this problem treat phase noise as purely random, missing the inherent structure within the measurement packets. We present a novel algorithm called partially coherent matching pursuit for sparse channel estimation under practical phase noise perturbations. The proposed approach leverages this partially coherent structure in the phase errors across multiple packets. Our algorithm iteratively detects the support of sparse signal and employs alternating minimization to jointly estimate the signal and the phase errors.

We numerically show that our algorithm can reconstruct the channel accurately at a lower complexity than the benchmarks, and derive a preliminary support detection bound as a performance guarantee.

Advancements in radar technology such as phased array antennas, digital beamforming, and adaptable waveform generation have led to the flexibility in controlling radar resources such as scan time, beamwidth and bandwidth during a radar mission. This new flexibility has led to a new research topic, radar resource management. Radar resource management involves the allocation of radar resources in order to achieve the highest performance in a radar mission. This thesis focuses on a specific scenario where a radar is tasked with deciding multiple object presence decisions located at multiple scan directions. The resource considered is the scan time allocated to each scan direction. To optimally allocate the scan time over the scan directions, a cost function is formulated, where the expected performance of an individual decision of an object being present or absent is formulated as the expected Bayes risk. The expected performance of all individual decisions in the same scan direction, are summed to obtain an expected task performance. The global performance at the system level is formulated using two approaches. The Sum approach formulates a cost function at the system level as the sum of the expected cost of each task. The Max approach formulates a cost function that minimizes the maximum of all expected task costs. Simulations have been performed to demonstrate the flexibility of the Sum and Max approach to adapt the scan time allocation based on different scenarios, including multiple object presence decisions per scan direction, sequential measurements, and a birth-death process regarding the presence of an object over time. Simulations demonstrate that using the Sum and Max approaches for the allocation of scan time results in an improved performance compared to the uniform distribution of the scan time resource over all scan directions.

This study delves into the application of coded covers in enhancing Acoustic Vector Sensor (AVS) performance for sound source localization. We initially explored the use of a coded mask inspired by ultrasound imaging. However, our analysis indicated that the coded mask primarily acts as a scaling factor without substantial benefits for direction of arrival (DOA) estimation.
In response, we investigated an alternative approach using a larger cover with spaced channels, positioned above the AVS sensor. In this setup, each channel operates independently, and their combined acoustic outputs are analyzed upon reaching the sensor. Importantly, the upper surface channels can be viewed as a virtual uniform linear array (ULA), allowing the application of compressed analysis methods.
Two distinct DOA estimation approaches were developed: the compressive sampling (CS) method and the compressive covariance sensing (CCS) method. Both methods were validated, showing improved accuracy in angle estimation. Notably, the CCS method exhibited the potential to expand the number of detectable sound sources using a single AVS.
In summary, while the coded mask did not offer significant advantages, the coded cover design, along with the CS and CCS methods, improved DOA accuracy and extended the detection capabilities of single AVS.

Hydrogen is a promising fuel for both reducing greenhouse gas emissions and being used as an energy carrier for renewable energy sources. The adaption of hydrogen instead of natural gas as a fuel in gas turbine combustors introduces additional challenges regarding flame stability. One instabilities is boundary layer flashback. This phenomenon occurs if the flame speed exceeds the local flow velocity. Then the flame can propagate upstream, which can result in equipment failure. Lean premixed hydrogen mixtures are more prone to boundary layer flashback because hydrogen flames have both a smaller quenching distance and higher flame speed compared to natural gas. Therefore, operating at 100% hydrogen fuel content is an enormous challenge. Active control strategies are a promising strategy to prevent and withstand possible occurrences of boundary layer flashback in gas turbine combustors. In this work, the detection, prevention, and counteraction of flashback are investigated in unconfined Bunsen burners. After an experimental analysis of suitable sensors, a tracking controller is therefore designed and different control design approaches evaluated.

Both the increase in flame fluctuations and in flame angle are reasonable as indicators for a flame moving towards flashback as the flow velocity is reduced. However, the complex nature of a flame makes it challenging to measure these effects in a flame. Four types of sensors, i.e. an ion sensor, a thermocouple, a photo-detector and a microphone, were investigated for their ability to detect flashback and to find precursors that indicate the onset of flashback. A Bunsen burner equipped with the different types of sensors was used in an experiment to determine the most suitable sensor type. Of the investigated sensors, the thermocouple is the most promising sensor for both flashback detection and use the temperature signal as a control variable. The temperature is a precursor for the onset of flashback since the temperature increases with increasing flame angle. Also, by placing multiple thermocouples in $360^oC$ configuration on the burner rim, it is possible to estimate the position, where the flame entered the burner.

A complete fault-tolerant framework to control a flashback event is proposed. This framework increases the flashback resistance of the system and it counteracts possible flashback events, which is considered as a fault in the system. A supervisor block has two functions. Firstly, the supervisor uses a fault detection method, known as trend checking, to detect flashback. Secondly, it decides which strategy should be applied, based on the temperature measurements. The fault-tolerant control strategy is threefold. First, the prevention controller closes the loop, so it can steer away from flashback conditions by rejecting disturbances and tracking a desired reference path using the temperature data. Different tuning methodologies based on the traditional proportional and integral (PI), linear-quadratic-integral (LQI) and pole placement regulator (PPR) are explored for this purpose. The robustness of the proposed controller was verified through loop shaping interpretation. However, flashback could be consider as an stochastic process that still could be triggered. Therefore, the second task is to detect present flashback events and apply counteraction to oppose the fault. A design is proposed by the injection of pressurized air upstream of the burner rim to push back the flame by diluting the boundary layer and increasing the flow speed. A proof-of-concept is validated in the experimental set-up as a possible and feasible counteraction. The third task is a final safety measure. A safety switch would be applied to shut off the fuel supply to fail safe after a potential unsuccessful counteraction attempt. The fault-tolerant control framework was simulated in Simulink for a demonstration purpose, where the models were identified from experimental data.

The proposed fault tolerant control framework is intended to increases the flashback resistance of the system by being more robust towards disturbances and flashback events that would normally require a restart of the system. Its potential to do so has been confirmed by simulations. However, further research on the prediction of flashback and controller implementation are still required to run burners on 100% hydrogen without flashback.

When objects are imaged, often aberrations slip into the captured images. Many techniques exist to remove part of those aberrations using hardware, software or a combination of those. This research will focus on Phase Diversity for Blind Multi-Frame Deconvolution. In Blind Multi-Frame Deconvolution both the object and the aberrations are estimated from multiple images of the same object. Previous research has shown, Phase Diversity can be used to improve the quality of the reconstructed object in combination with Blind Multi-Frame Deconvolution. Using Defocus Phase Diversity is currently mainly used in literature.

In this research new Phase Diversity methods are proposed. When there is no knowledge on the object or aberrations, Phase Diversities are proposed to improve the object reconstruction compared to Defocus Phase Diversity.

We also propose a new method to optimize the Phase Diversity for the N + 1th image, when already N images have been taken. This method first determines the low values in the frequency spectra of the images. Next it estimates a Phase Diversity that results in a Modulation Transfer Function to enhances those frequencies. This method leads to an improved restoration quality in terms of Structural Similarity and Peak Signal to Noise Ratio when there is Gaussian noise or Poisson noise with a higher Signal to Noise Ratio. Recommendations are
done to improve the method for lower Signal to Noise Ratios.

Settels Savenije van Amelsvoort is interested in investigating the use of piezoelectric shunt damping for a measurement and calibrations setup. The goal is to damp the remaining vibrations from previous vibration isolation stages to sub-nanometre amplitude. There was interest in piezoelectric shunt damping as a damping mechanism. Although piezoelectric shunt damping has previously been achieved at nanometre level and high frequency, the use of piezoelectric shunt damping at (sub)nanometre amplitude and low resonance frequency is still an open challenge and a novel application.

In order to test the feasibility of this application, a test setup is constructed. The experiment is performed on several vibrations isolation stages and with two different shunts. The final setup is able to damp the vibrations at 100 nm and indicates that it can damp vibrations at 50 nm. However, the results at 50 nm are inconclusive due to the resolution of the sensor and vibrations induced by the ground motion. For future research it is recommended to increase the vibration isolation and to use a sensor with higher resolution, this might show that damping at sub-nanometre amplitudes is feasible.

In the field of Adaptive Optics (AO) a new development has been introduced recently: the Wavefront Sensorless Adaptive Optics (WFSless AO) approach, which only uses camera measurements to optimize image quality and does not use a Shack-Hartman (SH) Wavefront Sensor (WFS) [Booth, 2006, Débarre et al., 2009, Hinnen et al., 2008]. In order to optimize the image quality of a point source or an extended object, the residual wavefront should be minimised as much as possible. The Model Based Approach [Booth, 2007, Linhai and Rao, 2011] is a very promising approach which can be used for this optimisation, resulting in a significant reduction in the amount of measurements. This approach uses an initial measurement and then excites all the modes that need to be optimised separately and takes a measurement of the intensity distribution after each excitation. Then, the Second-Moment (SM)’s of the measured intensity distributions are computed and used for the computation of the optimal control input, which is the coefficient vector in a Zonal representation of the wavefront. The control vector can be used to minimise the residual wavefront and thus optimise the image quality. However, in this approach it is assumed that the incoming wavefront does not change between the first and last measurement. This assumption is called the frozen window time and it is not valid for real-time problems. In [Lianghua et al., 2017] a solution was proposed for this problem, by using a precomputed independent set of modes, which allows a decoupling of the modes and computations and reduces the frozen window time. Unfortunately, this research introduces more measurements which leads to more delay. In this thesis a new approach will be presented, the Continuous Model Based Approach. This approach tries to solve both problems simultaneously: a reduction of the number of measurements and a reduction of the frozen window time. The number of measurements will be reduced by combining the excitation and optimisation step and the frozen window time will be reduced by using a two step optimisation, based on [Lianghua et al., 2017]. First, the validity of this approach will be proven mathematically and with simulations, then it will be tested in experiments. In order to make the step to real-time experiments, also an online calibration method will be introduced, proven and tested. This is also a new technique, based on the Model Based Approach [Linhai and Rao, 2011] and can be used to obtain the correlation matrix for any set of modes, focussed on actuator responses.

Parasitic diseases such as malaria remain a mayor burden on global health. One of the biggest challenges still to be overcome is that of inadequate diagnoses. This research explores the opportunities that Hyperspectral Imaging yields in this field. The first goal is to estimate the spectral signature of Malaria parasites in non-stained or Giemsa-stained thin smear blood samples and of Schistosoma parasite eggs in urine samples. For this different endmember extraction algorithms are combined with various methods of pre-processing and dimensionallity reduction. The used endmember extraction methods are pure pixel index (PPI), NFINDR, Statistics Based and simplex identification via split augmented Lagrangian (SISAL). For denoising Savitzky Golay and 3 dimensional gaussian filtering is used and the dimensionallity reduction is done with PCA, ICA or HySime. The resulting spectral signatures of the algorithms are validated by inspecting the endmember locations, spectra and abundance maps. They have furthermore been compared by the classification performance where the spectral signatures are used in the feature derivation. This is done by deriving a detection map using OSP or CEM detection and then using the SVM or random forest classifiers to classify cells as being infected or not. These performances are furthermore compared to RGB image based classification.

In case of the stained Malaria sample the four endmember extraction methods are shown to be applicable to various degrees. Firstly, the PPI method is shown to be inconsistent, resulting in different spectra each run. Secondly, the statistics based method unable to separate the spectral signatures of the red blood cells and thirdly the background. Thirdly, The NFINDR method seems to work well considering the endmember locations, spectra and abundance maps, but leads to a low classification performance. The research concludes that Sisal made the most accurate estimations of the spectral signature of the parasite. The results from all the validation methods are in line with expectations. Furthermore, the use of this spectral signature in the feature derivation process results in the highest classification performance. This performance is also shown to be significantly higher compared to using either the first principal component of the full hyperspectral data or the RGB images. Applying the same methods to the Schistosoma sample it is found that some of the methods, though interestingly not Sisal, are able to to create an abundance map in which the egg is separated from the background. However, none of them are able to separate the egg and the white blood cell and detection maps using these signatures did not show the egg more clearly than the first principal component did. None of the methods are found to be able to extract the spectral signature of the unstained Malaria parasite.

Finally, a hypothetical multispectral microscope is proposed which images at the wavelengths where the spectral signature of the parasite in a stained sample has the biggest difference in light transmittance to the other endmembers. This setup is simulated from the available hyperspectral data and its classification performance is compared to classification performance using the full hyperspectral data and using the RGB images which are simulated from the same data. The classification using the discriminative wavelengths is found to outperform both in terms of sensitivity and specificity. This implies that the images at these specific wavelengths provide more discriminative power and such a multispectral setup could provide a significant advantage over RGB imaging.