Wouter Van Hoey

• 09/02/2026
  
• 5 p.m.
• Venue: Campus Middelheim, G.010
• ​Online PhD defence
• Supervisor: Pegie Cool
• Department of Chemistry

Abstract

At present, the majority of all intricate global challenges are related to sustainability. Therefore, the urge for sustainable development is bigger than ever. To date, the chemical sector is already a major driving force for innovation and essential for sustainable growth across all sectors. However, much more scientific contributions will be needed to tackle the difficult task ahead, which is to create a sustainable world whereby the needs of the present are met without compromising the ability of future generations to meet their own needs. Hence, the aim of this thesis is to make a positive contribution to sustainability by applying heterogeneous catalysis in various green chemistry related applications. The research in this PhD thesis is mainly focussed on the development of innovative CeO2-based catalysts and can be divided into four different topics: catalytic combustion of VOCs, plasma-catalytic conversion of CO2, photo-induced reduction of nitrobenzene to aniline, and the selective hydrodeoxygenation of aromatic carbonates. In more detail, chapters 2, 3 and 4 focus on catalyst optimisation to improve the catalytic combustion of toluene, which is a model compound for aromatic VOCs. Based on the obtained results it can be stated that combining noble and transition metals offers a unique approach to limit the costs and consumption of noble metals for industrial large-scale combustion of VOCs. Unfortunately, while intriguing results are obtained regarding the catalytic combustion of VOCs, it cannot be ignored that combustion of carbonaceous compounds results in the production of CO2. Therefore, the fifth chapter tackles the plasma-catalytic conversion of CO2 using CeO2. More specifically, it is investigated if increasing the amount of oxygen vacancies at the surface of CeO2 enhances the dissociation of the stable CO2 molecule, which facilitates the conversion of CO2 into value-added products. Next, in chapter 6 the development of a more sustainable photochemical process to convert nitroarenes into anilines at room temperature and in open atmosphere without the addition of hydrogen is discussed. Additionally, it is examined if the presence of a heterogeneous catalyst can further improve the obtained results. Finally, in chapter 7 a thorough characterisation is provided to gain a better understanding on the selective hydrodeoxygenation of aromatic carbonates when commercially available heterogeneous Ni-SiO2 catalysts are applied. Unravelling the reason behind the remarkable activity is trivial as the successful hydrodeoxygenation provides a next step in the development of important bio-based compounds from wood.

Burak Karaduman

• 05/02/2026
  
• 10 a.m.
• Venue: Campus Middelheim, G.006
• Supervisor: Moharram Challenger
• Department of Computer Science

Abstract

As embedded system-originated paradigms grow in complexity and lack of autonomy, traditional control architectures struggle to manage the uncertainty and dynamism of real-world environments. Intelligent agent architectures, particularly those grounded in the Belief–Desire–Intention (BDI) model, offer promising cognitive capabilities for Cyber–Physical System (CPS). However, their integration into resource-constrained and (soft) real-time embedded platforms remains challenging. To address this, eventually, we propose a Model-driven Engineering (MDE) approach for deploying fuzzy-BDI agents into CPS. This approach combines fuzzy logic with BDI reasoning to manage imprecision and context shifts, while also enabling platform-independent system design through model-based abstraction. Despite the growing relevance of such hybrid architectures, existing literature lacks comprehensive methods that link cognitive agent behaviours to deployable embedded code in complex systems settings. This thesis fills this gap by introducing an integrated approach that bridges high-level agent modelling and low-level system implementation. The results demonstrate that fuzzy-BDI agents, supported by model-based analysis and engineering, can enhance adaptability, reduce development complexity, and improve robustness in uncertain environments. The central challenge addressed in this thesis lies in bridging the gap between traditional embedded control and intelligent decision-making within CPS. While intelligent agent architectures, particularly those grounded in the BDI model, provide a promising cognitive framework for autonomous reasoning, their practical integration with enhanced capabilities into resourceconstrained embedded platforms remains an unresolved problem. The BDI model enables agents to reason about their environment, goals, and actions proactively, maintaining beliefs about the world, forming desires representing objectives, and committing to intentions that guide action. However, classical BDI systems are based on crisp logic, which severely limits their applicability under uncertainty and continuous change. Without mechanisms to interpret ambiguous sensory data or adapt to fluctuating contexts, BDI agents fall short of the robustness required in CPS deployments. To address these shortcomings, this thesis introduces an integrated fuzzy-BDI and MDE framework for the development and deployment of intelligent CPS. The core idea is to combine fuzzy logic, which allows reasoning with degrees of truth, with the structured deliberation of BDI agents, thereby enabling systems to handle uncertainty natively at every stage from perception and planning to action execution. Complementing this reasoning innovation, the research employs MDE to elevate the abstraction level of system development. The proposed framework bridges high-level agent design models with low-level embedded implementations, achieving explainability, reusability, and platform-independent deployment. This dual integration of fuzzy and BDI reasoning with model-driven design forms the foundation of a new engineering methodology for building adaptive and smart CPS.

Arno Raes

• 02/02/2026
  
• 4 p.m.
• Venue: Stadscampus, Klooster van de Grauwzusters, Promotiezaal
• Supervisor: Sammy Verbruggen
• Department of Bioscience Engineering

Abstract

Titanium dioxide is widely used in photocatalysis because it is abundant, stable, and inexpensive, yet pristine anatase and rutile absorb mainly ultraviolet light. One strategy to extend activity into the visible region is the introduction of sub-gap states associated with oxygen vacancies and Ti³⁺ species. These are typically generated through hydrogenation, high-temperature reduction, plasma treatment, or strong chemical reductants, routes that often raise energy concerns. This thesis therefore addresses a practical question: can TiO2 be reduced by mild and potentially scalable methods that still deliver useful functionality under realistic conditions, and where do such approaches succeed or fail?

High-intensity ultrasound was first investigated as a potential reduction route. Using calorimetrically calibrated power delivery and systematic probe integrity checks, any apparent darkening of TiO2 during aqueous sonication was traced to metallic debris from horn erosion rather than defect formation. UV–Vis diffuse reflectance showed no band-edge shift or sub-gap absorption, and no evidence for stable oxygen vacancies or Ti3+ species was found. Under these conditions, cavitation hot spots do not provide a sustained reducing environment, and any transient defects are rapidly re-oxidised by radical species and dissolved oxygen. Ultrasound-induced “blackening” is therefore an artefact rather than a viable reduction strategy.

Despite this negative outcome, ultrasound proved valuable as a synthesis and processing tool. During sol–gel synthesis it yielded high-surface-area, open-mesoporous amorphous TiO2 with interconnected nanoparticulate networks. Controlled ultrasonic crystallisation produced predominantly anatase while retaining about 50% of the surface area, compared to the 10% of conventional calcination. These materials showed rapid uptake of volatile organic compounds, consistent with their preserved mesostructure.

Vacuum annealing was then explored as a hydrogen-free route to defective TiO2. Mild vacuum treatment of P25 darkened the powder, increased visible-light absorption without shifting the band edge, and shifted the Ti:O ratio away from perfect stoichiometry. Functionally, this resulted in an approximately 25% increase in methane formation during CO2 photoreduction compared with pristine P25. Applying the same treatment to Au@P25 preserved plasmonic properties but provided only marginal additional activity beyond that already conferred by gold.

Overall, this work shows that mild routes give mild but reliable outcomes. They are effective for preserving structure, adsorption capacity, and stability, and can deliver meaningful functional gains in selected cases.

Arthur Gestels

• 02/02/2026
  
• 2 p.m.
• Venue: Campus Middelheim, A.143
• ​Online PhD defence
• Supervisors: Koen Janssens & Gunther Steenackers
• Department of Physics

Abstract

Scientific research in cultural heritage has increasingly shifted toward non-invasive imaging methods that can reveal material information across entire artworks. Because paintings often exhibit complex stratigraphies and heterogeneous compositions, traditional point-based techniques and small-scale sampling frequently fail to provide representative insight into the full material structure of an artwork. In this context, hyperspectral imaging techniques have become essential tools for conservators, researchers, and art historians.

Macroscopic X-ray powder diffraction (MAXRPD) is one such hyperspectral technique. It enables direct identification of crystalline compounds, making it highly valuable for the study of pigments and degradation products in painted artworks. Despite its high chemical specificity, MAXRPD is limited by low spatial resolution and long acquisition times, restricting its applicability for large-scale investigations. To address these limitations, this thesis investigates the combination of MAXRPD with faster hyperspectral imaging modalities, including reflectance imaging spectroscopy (RIS) and macroscopic X-ray fluorescence (MAXRF), using machine learning approaches to reduce acquisition time and improve efficiency. ​ ​

The research begins with a case study on Rembrandt’s The Night Watch, where MAXRPD was used to assess the material impact of an acid vandalism attack. The results revealed the formation of anglesite as a crystalline degradation product in the upper paint layers, demonstrating the value of compound-specific imaging for conservation assessment and treatment planning.

Next, a proof-of-concept study on corroded metal plates showed that quantitative MAXRPD data could be combined with RIS data to train machine learning models capable of predicting material distributions with reasonable accuracy. This demonstrated the feasibility of extrapolating material-specific information from faster imaging techniques.

The methodology was then adapted to cultural heritage objects, enabling the generation of high-resolution, material-specific images of artworks. Case studies on an illuminated manuscript and oil paintings showed that fusing MAXRPD with RIS or MAXRF data produces reliable predictions. Finally, the approach was applied to more complex historical paintings to assess its generalizability, demonstrating that accurate compound mapping is possible even in unscanned areas when representative training data are available. Together, these results establish a scalable framework for accelerating MA-XRPD-based imaging while preserving its analytical power.

Donja Baetens

• 30/01/2026
  
• 2 p.m.
• Venue: Campus Drie Eiken, O.5
• Supervisor: Siegfried Denys
• Department of Bioscience Engineering

Abstract

Indoor air quality (IAQ) is increasingly recognised as a major public health concern, as people spend most of their time indoors and are therefore continuously exposed to pollutants present in indoor environments. Among indoor contaminants, particulate matter (PM) and volatile organic compounds (VOCs) are of particular concern due to their health impacts and prevalence in buildings. While air purification technologies can reduce exposure, most systems target only one type of pollutant, i.e. particulate or gaseous. This dissertation explores the integration of electrostatic precipitation (ESP) and photocatalytic oxidation (PCO) into a single device to enable simultaneous removal of PM and VOCs. It is hypothesised that incorporating PCO into the collector section of a two-stage ESP reactor enables simultaneous removal and may reduce the need for cleaning the ESP by degrading deposited particles.

A combined ESP-PCO reactor was developed with a photocatalytic coating immobilised on the collector plates and UV lamps positioned between them, alongside multiphysics models describing the underlying processes. First, the standalone ESP functionality was investigated to determine total and fractional particle collection efficiencies, the influence of operating conditions, and ozone emission. Next, the photocatalytic coating was examined independently for photocatalytic VOC (acetaldehyde) removal, comparing metallic substrates and studying the effects of coating layers, inlet concentration, and relative humidity. Photocatalytic soot degradation experiments were also performed to evaluate removal of deposited PM.

Subsequently, acetaldehyde removal was studied in the full two-stage ESP-PCO reactor to assess PCO and ESP functionality in terms of clean air delivery rate (CADR) and single-pass removal efficiency (SPRE). Activation of the ionisation section resulted in acetaldehyde removal comparable to photocatalytic removal, while combining ESP and PCO yielded the highest removal observed, although synergistic effects were not clearly identified.

Finally, multiphysics models for ESP and PCO were developed to describe air flow, electric field, ion transport, particle charging and motion, acetaldehyde transport, radiation distribution, and photocatalytic kinetics. The models were verified against experiments and provide insight into the influence of plate spacing, irradiance distribution, and electric field effects while identifying opportunities for performance optimisation.

Overall, this dissertation demonstrates the feasibility of integrating photocatalytic oxidation into a two-stage electrostatic precipitator for indoor air purification, showing that the combined system can remove both PM and VOCs and has the potential to reduce particle accumulation. However, the results highlight design, material, and modelling challenges requiring further development.

Jan Morez

• 29/01/2026
  
• 4 p.m.
• Venue: Campus Drie Eiken, O.005
• ​Online PhD defence​
• Supervisors: Jan Sijbers & Ben Jeurissen
• Department of Physics

Abstract

Diffusion magnetic resonance imaging (dMRI) offers a unique way of imaging the human brain noninvasively. By carefully controlling various acquisition parameters, the dMRI signal can be used to probe the movement of water molecules inside biological tissues, revealing important information about tissue structure. However, this signal is subject to noise and other undesired electromagnetic effects that reduce the signal-to-noise ratio. Efficient acquisition schemes and robust estimators are required to obtain accurate and precise tissue maps of the human brain. In this thesis, we have aimed our efforts towards improving the accuracy, precision, generalizability and applicability of several dMRI analysis methods. ​ ​

Constrained spherical deconvolution (CSD) is a popular dMRI technique that can be used to infer the local tissue densities and their orientations in the human brain, which consists of cerebrospinal fluid, anisotropic white matter and isotropic gray matter. This is achieved by densely sampling q-space, the space of q-vectors representing the direction and strength of diffusion weighting. After sampling q-space in multiple shells, tissue densities and orientations can be estimated by deconvolving this multi-shell dMRI signal with the respective response functions for white matter, gray matter and cerebrospinal fluid. These response functions are typically represented using spherical harmonics (SH) basis functions. However, when sampling is nonspherical, either due to inhomogeneous gradients or by design such as Cartesian sampling, the estimated tissue maps suffer from biases. To counter this issue, we adopted a compact response function model that accounts for nonspherical sampling. On multi-shell data, our approach provides fiber orientation density functions and tissue densities indistinguishable from those estimated using SH. On Cartesian data, estimates are on par with those obtained from shell-wise data, significantly broadening the range of data sets analyzable using CSD. In addition, inhomogeneous gradients can be accounted for, resulting in more accurate apparent tissue densities and connectivity metrics. ​ ​

Q-space trajectory imaging is a dMRI technique that uses time-varying q-vectors to sensitize the dMRI signal to microscopic variations in heterogeneous tissues. By modelling the dMRI signal with a diffusion tensor distribution (DTD), this approach allows teasing apart variations in diffusivity from microscopic anisotropy, orientation dispersion, and mixtures of multiple isotropic diffusivities. To improve the estimation of the DTD parameters, we propose an efficient acquisition scheme optimized for the most used QTI-derived microstructural parameters. A constrained iteratively reweighted least squares estimator is used to further improve the bias and precision of the DTD parameters.

Jonathan Sanctorum

• 13/01/2026
  
• 4 p.m.
• Venue: Campus Drie Eiken, Q.001
• ​Online PhD defence
• Supervisors: Jan Sijbers & Jan De Beenhouwer
• Department of Physics

Abstract

X-ray phase contrast imaging is known for its potential to yield high contrast in soft tissue and light materials compared to conventional transmission contrast, making it a valuable tool for (bio)medical applications and non-destructive testing. In addition to transmission and phase contrast, dark field contrast is a third contrast type that has received increased interest recently due to its unique ability to detect the presence of unresolved microstructures. Phase and dark field contrast are both related to the material’s X-ray refractive index, but as opposed to phase contrast, dark field contrast results from refractive index fluctuations that cannot be resolved by the imaging system. Remarkably, all three contrast types can be measured in a single experiment using dedicated X-ray phase contrast imaging systems. The introduction of compact lab-based systems for phase sensitive X-ray imaging has resulted in a growing number of applications. Optimization of conventional X-ray imaging setups often relies on Monte Carlo simulations, where stochastic models are used to simulate the physics processes. In this work, a collection of tools is presented that aim to meet the requirements associated with Monte Carlo simulations for the design of X-ray phase contrast imaging systems. Here, two X-ray phase contrast imaging methods are mainly of interest: grating-based interferometry and edge illumination, each of which relies on the use of gratings. First, the realization of X-ray phase contrast simulations for grating-based interferometry is demonstrated using the GATE Monte Carlo framework, hereby relying on a hybrid simulation approach combining Monte Carlo simulations with wave optics calculations. Although this simulation framework is suitable for edge illumination simulations as well, Monte Carlo simulations are known to be very time consuming, certainly for parameter studies. To address this limitation, the concept of virtual gratings is introduced. By replacing the gratings in the simulation with virtual gratings, the parameters of the gratings can be changed after the simulation, thereby significantly reducing the overall simulation time. This concept is subsequently used for the design of edge illumination gratings for the augmentation of the FleXCT micro-CT scanner to a phase sensitive X-ray imaging system. Finally, the benchmarking of multi-contrast X-ray imaging simulations is addressed. The reference values required for benchmarking are particularly difficult to determine for the dark field contrast. Here, a practical method based on the virtual grating approach is presented to directly estimate reference values for all three contrast types from the simulated X-ray trajectories, allowing for efficient benchmarking.

Salma Haidar

• 13/01/2026
  
• 4 p.m.
• Venue: Campus Middelheim, A.143
• ​Online PhD defence
• Supervisor: José Oramas Mogrovejo
• Department of Computer Science

Abstract

Hyperspectral imaging (HSI) combines digital imaging with spectroscopy, capturing hundreds of contiguous wavelength bands per pixel. This produces a three-dimensional “hypercube” that integrates spatial and spectral information, offering powerful opportunities for material analysis, environmental monitoring, and beyond. Yet, HSI also poses challenges: high dimensionality, limited annotated data, and heavy computational demands.

My thesis addresses these challenges through three complementary approaches. First, I develop a deep learning framework for multi-label classification that demonstrates strong performance in complex, mixed-material scenes and highlights limitations in common annotation practices. Second, I explore self-supervised contrastive learning to enhance classification accuracy under limited supervision, demonstrating substantial gains across diverse datasets. Third, I apply explainability-driven dimensionality reduction to identify the most informative spectral bands, reducing redundancy while maintaining or even improving accuracy.

Together, these contributions demonstrate that tailored representation learning strategies and explainability-driven dimensionality reduction can deliver hyperspectral classifiers that are accurate, computationally efficient, and adaptable to challenging data conditions. The results highlight practical pathways for improving hyperspectral image analysis and open opportunities for further exploration across diverse application domains.