The rapid growth of e-commerce has led to escalating product return volumes, generating
substantial economic costs and environmental impact. Existing research largely focuses on either
advanced predictive modelling or process optimisation, yet it commonly overlooks the role of
stakeholders in designing actionable return reduction strategies. This study introduces a practical
and interpretable framework that clusters products and orders features, finding the return risk to support targeted interventions. Using a Design Science Research (DSR) approach, the framework integrates expert interviews, a structured literature review, and extensive data analysis. Statistical tests reveal significant return rate variation across product attributes (category, color, size, price) and order characteristics (order value, quantity, shipping carrier). Three clustering techniques, K-Prototypes, CAVE, and LCC, were evaluated, with LCC providing the most distinctive segmentation of high-risk groups. High-risk products were predominantly found in the Fashion Category, particularly in black and red variants and in sizes other than “One Size”. Furthermore, high-risk orders were associated with large, high-value purchases shipped via DHLDE. The resulting framework enables retailers to implement validated strategies such as improved product information, category-specific return policies, and targeted marketing adjustments. Overall, the approach offers a scalable, stakeholder-aligned foundation for reducing return flows.

Federated learning (FL) has emerged as an important paradigm in distributed machine learning, enabling collaborative model training across decentralized devices while preserving data privacy. FL’s privacy-preserving nature – where raw data remains on local devices and only model updates are shared – has made it suitable in sensitive domains like healthcare and finance. However, the decentralized framework introduces fundamental challenges that threaten its reliability and adoption. Data heterogeneity, security threats, and privacy leakage risks create critical vulnerabilities that demand robust solutions.
To study such vulnerabilities, this thesis considers two kinds of parties: the clients and the servers. Clients act as data owners that perform localized computations and share only model parameters, thereby preserving raw data privacy, yet they introduce vulnerabilities through potential malicious behaviors (e.g., data/model poisoning attacks) or unreliable contributions due to data quality. In contrast, the server, while facilitating model convergence through aggregation, poses inherent privacy risks by potentially inferring sensitive client information from shared gradients, even without direct data access. These two parties create a dual-threat landscape: clients may compromise model performance through adversarial manipulations, while servers break confidentiality via reconstruction methods....

In this review, we have digitized and georeferenced over 7000 Low-Temperature Thermochronology (LTT) data points and 750 Time-Temperature Modelling (TTM) results from 252 published works. The study area includes the continental crusts adjacent to the rifted margins (~Late Triassic to Early Jurassic) of the Central Atlantic Ocean and its direct neighbours.

Our main intention is to map out the thermal cooling events as recorded by LTT data and as illustrated by TTM results. The time interval targeted in this review is the Phanerozoic (i.e., 540 to 0Ma), which is possible thanks to LTT ages spanning this entire period in the study area. It allows us to investigate the thermal evolution of the continental rims of the Central Atlantic Ocean at an unprecedented scale. In rifted margins and their shoulders, a debate exists whether the LTT-recorded cooling is the results of post-rift erosional exhumation or post-heating thermal relaxation, especially for the area directly in the vicinity of the paleo-rift zone. We therefore devised a short workflow to examine these propositions by filtering out the LTT dataset and spatially plotting the LTT ages. Furthermore, we investigate the relationship between LTT ages and distance from the Continent-Ocean Boundary/Transition Zone.

LTT ages alone have often been described as bearing little geological meaning, thus requiring to run TTM in order to reconstruct the thermal/geological history, as several factors are to be taken into account in the thermal history reconstruction. Here, we examine whether a statistically significant LTT dataset can serve as a proxy in the reconstruction of cooling events. To this end, we compare peaks of LTT cooling ages and of TTM cooling event.

Our investigation reveals that i) generalised cooling occurred in the pre-, syn-, and post-rift phases of the Central Atlantic, ii) there is a clear LTT age oceanward youngening trend, iii) the lack of LTT age with a syn-rift signal within ~500km along the shorelines suggests erosional exhumation (i.e., vertical movements) as main driver of the cooling, and iv) large LTT datasets bear meaning on the cooling events and thus on vertical movements, at least in this case studies in the rims of the Central Atlantic Ocean.

The increasing frequency of extreme weather events intensifies the demand for accurate monitoring of nearshore wave dynamics, as these dynamics directly affect shoreline stability, flood risks, and coastal operations. While operational wave models provide essential forecasts, they often underestimate significant wave height (SWH) during energetic conditions and omit infragravity (IG) waves. The latter can substantially amplify coastal impacts such as flooding, dune erosion, and harbor seiching. These limitations are further exacerbated by the limited spatial coverage of in-situ buoys, making model validation in coastal waters particularly challenging.

The launch of the Surface Water and Ocean Topography (SWOT) mission presents a promising yet underutilized opportunity to improve coastal monitoring and support the validation of hydrodynamic models. Using Ka-band Radar Interferometry (KaRIn), SWOT provides high-resolution, two-dimensional measurements of sea surface height (SSH), offering spatial detail that exceeds the capabilities of buoys and traditional altimeters. While initial validations in the open ocean have confirmed SWOT’s capacity to retrieve SWH and resolve long-period waves, its performance in shallow, morphologically complex coastal seas remains largely untested.

This research presents a novel approach comprising (i) a multi-pixel match-up strategy for SWH retrieval, and (ii) the first satellite-based spectral analysis for detecting IG wave energy in the Southern North Sea. Through three targeted case studies, the study investigates SWOT’s two-dimensional SSH and SWH patterns, retrieval accuracy, and sensitivity to key processing parameters across varying sea states and bathymetric regimes. Validation is performed using in-situ buoy observations, platform-mounted radar measurements and numerical wave models to assess consistency, spatial performance, and retrieval accuracy.

By applying both single- and multi-pixel match-up strategies, our findings revealed the trade-off between statistical variance reduction through spatial averaging and the preservation of local wave variability. Similarly, for spectrally derived IG wave energy, different analysis box sizes were evaluated against platform-mounted radar observations in the Southern North Sea. Results show that SWOT-derived SWH exhibits strong agreement with buoy data (bias < 7 cm) and outperforms operational wave models during energetic events. IG wave height estimates also demonstrate good correspondence (MAE $\approx$ 0.9 cm), provided that residual noise amplification is carefully managed. This finding underscores the need for improved noise modelling in future spectral retrieval frameworks. A key uncertainty identified is the role of spatial heterogeneity, which induces representation errors due to the inherent mismatch in spatial and temporal scales between SWOT observations, in-situ buoys, and wave models.

Despite these uncertainties and other limitations, the results confirm SWOT’s capacity to observe nearshore wave dynamics with high spatial detail. The proposed configurations and filtering strategies offer a transferable framework for future applications. These insights support SWOT’s integration into coastal wave model validation, boundary condition improvement, and data assimilation schemes across coastal scales. Ultimately, this thesis advances high-resolution coastal wave monitoring by demonstrating how SWOT’s two-dimensional observations can enhance our understanding of spatial sea state variability, particularly during energetic conditions in morphologically complex environments.

This thesis investigates the potential of coffee-based granulate for Large Scale Additive Manufacturing (LSAM) in sustainable outdoor furniture design. The research is a collaboration between Coffee Based and 10XL, supported by Het Ooievaarsfonds. The project aims to determine whether materials from Coffee Based, made from spent coffee grounds (SCG), can be used to create large-scale 3D-printed furniture, and how their material behavior influences design and manufacturing decisions.

The research attempts to fill a clear gap in the literature. While SCG-based composites have been used in small-scale filament printing, their performance in Fused Granulate Fabrication (FGF) using LSAM has not yet been studied. The project uses a research-through-design approach and follows a double diamond process with the following clear parts: literature review and problem definition, material and printer exploration, design development, and final design and validation.

Three SCG-based compounds provided by Coffee Based were tested: SCG with recycled polypropylene (rPP), bio-based high-density polyethylene (bioHDPE), and thermoplastic starch (TPS). Their sustainable properties are different: rPP contributes to technical circularity through recycling within the technical cycle, bioHDPE is bio-based and derived from renewable feedstock, and TPS is bio-based, biodegradable, and compostable within the biological cycle. The material exploration phase is mainly focused on improving print quality rather than the mechanical properties of the compounds themselves.

Many print tests were performed at 10XL, mostly on a smaller extruder, and in the end, once on the large extruder. The main test variables were flow/extrusion rate, temperature, bed adhesion, print speed, and minimum layer time. Print quality was mainly focused on shape fidelity (overhang, bridging, radii, corner angles, and shrinkage), extrusion consistency, and surface finish. Performance was measured both qualitatively and quantitatively.

Shrinkage turned out to be a major challenge in this research. The TPS compound showed the lowest and most consistent shrinkage (below 0.5%) and had proper adhesion to the print bed without needing additional fixation. In contrast, the rPP and bioHDPE compound showed significant shrinkage levels, and therefore also needed additional fixation to the print bed. While adding chalk to these compounds reduced shrinkage, because it was manually mixed, it led to inhomogeneity and inconsistencies in print quality. Using the TPS compound, successful overhang angles reached up to 40 degrees, while bridging was mainly unsuccessful, due to the continuous flow nature of LSAM. Corner angles smaller than 40 degrees turned out to be unreliable. However, small radii were printable and reliable.

These findings led to the design of an XL 3D-printed prototype that combines the potential of the coffee-based material with the preferences of the stakeholders, within the found constraints of the printing technology.

Overall, this research shows that the provided granulates by Coffee Based have the potential to be used in LSAM with FGF, provided that further development is performed to minimize shrinkage and improve extrusion consistency on the large extruder. Design choices also have to align with material properties and printer possibilities.
Thus, while much is still unknown and challenges towards print quality remain, the project lays the groundwork for further development on using circular, biobased (or waste-based) materials in LSAM.