Search in the TU Delft Repository Collections

TU Delft Repository

210,172
Records
238,705
Files
81,369
Contributors

Recently Added Records

Two-phase flow carries most of the heat in evaporators and condensers, yet its internal structure is hard to measure without disturbing it. Magnetic resonance imaging resolves liquid distribution and velocity inside an opaque pipe without contact, but every published two-phase MRI study uses air and water near atmo-
spheric pressure. Water and steam below atmospheric pressure, where the liquid is twelve thousand times denser than its vapor, have never been imaged.
This thesis covers the design, construction and operation of a thermosyphon loop built for that purpose. The loop is a closed circuit in which water boils, rises, condenses and falls back under gravity, with no pump. A hot test section degrades the image, so the loop runs at 0.12 bar, where water boils at 50 ◦C, with the heater 1.7 m away and nothing ferromagnetic near the magnet. A steady-state pressure balance model set the pipe diameter and the sizing of the heater, the condenser and the expansion vessel that fixes the pressure of the loop. The loop was built and operated on eleven days, at charges between 48 and 70 percent of its volume. The circulation had two states. A freshly charged loop circulates quickly and unsteadily at 170 mL min−1 to 230 mL min−1, and after roughly four hours it changes, with no way back, to a slow and steady 12 mL min−1. Both states repeat across days, but the change between them is governed by the time since charging rather than by any setting, and its cause is not resolved. The model describes only the slow state. The result is a working thermosyphon whose measurement section can go inside a scanner. The fast state is the one with liquid structures worth imaging, but it drifts throughout and the model does not describe it, so the condition being imaged would not be known. In the slow state, which is steady and predictable, no liquid film is visible in the section at all. The changes needed before a scan can be planned are identified. ...
Master thesis (2026) - T.J. Kortekaas, J. Alonso-Mora, L. Ferranti
Autonomous navigation on inland waterways requires local motion planners that can anticipate interactions with other vessels while maintaining safe, efficient, and predictable trajectories. This thesis investigates two fundamentally different approaches to local motion planning for ASVs: MPPI and a proposed trajectory planner based on Frenet.

Both planners are integrated into a common hierarchical navigation framework consisting of global path guidance, encounter assessment, CPA-based path selection, speed governance, and LOS control. The MPPI planner is based on DRA-MPPI, adapted to incorporate longer-term route information. The Frenet planner instead generates a finite set of candidate avoidance trajectories through lateral offsets from the global reference path, using a commitment mechanism to maintain temporal consistency. This shared framework allows the influence of the local planning approach to be evaluated independently of the surrounding navigation architecture.

The planners are evaluated in simulation using representative inland waterway scenarios on the river Merwede, considering both a Damen Waterbus and a CEMT Va M8 cargo vessel. Experiments include single-vessel encounters and more complex scenarios involving multiple TSs. Performance is assessed in terms of collision avoidance, path-following accuracy, navigation efficiency, trajectory quality, and computational performance.

The results demonstrate that both approaches have distinct advantages and limitations. For the Waterbus, both planners achieve comparable success rates, while the Frenet planner significantly reduces mean computation time. However, when vessel dynamics become more restrictive, MPPI achieves higher completion rates and better path-following accuracy. An additional hierarchical experiment combining Frenet-based long-term planning with direct MPPI control demonstrates the potential of combining both approaches, although it does not consistently improve quantitative performance.

Overall, the findings highlight the trade-off between computational efficiency, dynamic feasibility, and planning horizon. The Frenet planner offers an efficient approach to generating interpretable, long-term avoidance trajectories, while MPPI provides greater flexibility when vessel dynamics must be explicitly considered. ...

Redesigning a patient story tool to support compassionate, patient-centred care

Master thesis (2026) - P. Cerrillo Gomez, A. Albayrak, Elif Özcan, Tina van Hemel
Intensive care units save lives through advanced technology, but the intensity of this environment can make it difficult for patients to feel known as individuals. Research shows that being seen, recognised and connected with as a person supports patients’ inner strength and comfort, whereas the isolation, disorientation and communication barriers of intensive care are linked to delirium-related cognitive impairment and post-traumatic stress after discharge. ICU nurses are deeply committed to compassionate, patient-centred care, but extreme cognitive load and time pressure leave little room to connect with patient and understand who they are. At the Leiden University Medical Center (LUMC), the “Maak kennis met” poster is meant to help by letting families share who the patient is, but in practice it is used inconsistently. This graduation project therefore asked: how can the poster be redesigned into a tool that helps nurses recognise and connect with patients as individuals, in a way that fits the existing clinical workflow without adding to nurses’ time pressure?

The project followed the Double Diamond framework. Starting with a literature review on humanised care, an analysis of existing similar patient story boards used in other hospitals, supported by a six-level framework towards humanising the ICU, and a technology mapping against ICU restrictions for possible solutions. This was combined with four weeks of fieldwork at the LUMC ICU: twelve observation sessions, four informal interviews and four workshops with 61 nurses and intensivists. The fieldwork showed that staff value the idea of the poster but that it is often missing, blank or unreadable at the point of care. A poster was absent in seven of twelve sessions and was present for only one of seven patients who could not speak for themselves. Participants asked for specific, short prompts instead of open boxes, with who matters to the patient as the most important thing for them, and for content that staff can act on, such as comfort preferences and music or television choices.

These findings were translated into an information architecture of four tiers (basic information, getting to know the person, taking good care, and journey and goals) and a design vision: make meeting a patient feel less like reading a file and more like getting to know someone. Twelve concept directions were scored on impact and feasibility, and a slideshow profile was iterated with ICU staff. The final design is offered as a three-step implementation path: four colour-coded A3 prints that can start today, a slideshow on the existing unit TVs, and a future app that generates the profile and pairs with the bedside screen. All steps carry the same content and differ in how the information is entered, updated and shown.

The slideshow profile was evaluated in a workshop with eight ICU nurses and in a conversation with an expert in person-centred ICU care. All nurses agreed that the profile helped them connect with the person (mean 4.9 out of 5), seven of eight agreed it would make the ICU feel more human, and seven of eight wrote a conversation opener drawn from it. The tier structure was judged clear in 31 of 32 judgements. Workload was the lowest-rated statement (mean 3.6), and slide pace was the main practical concern. The expert rated the concept as highly relevant and identified privacy and ICU culture as the main barriers to implementation. Both evaluations are indicative only: the sample was small, the profile was an example rather than one written by a family, and patients and families were not involved.

The project contributes an evidence-based content structure, a design system aligned with LUMC’s visual language, and a staged route from paper to digital, together with a clear agenda for the next possible steps: testing with real families and patients, and clarifying privacy, consent and ownership.


...
Urban computational fluid dynamics (CFD) models used for pedestrian wind comfort assessment typically omit vegetation, despite trees having a measurable effect on wind flow. With the Dutch NEN8100 pedestrian wind comfort guideline under review, there is growing interest in automated, scalable methods for incorporating tree geometry into urban CFD workflows. However, reconstructing individual trees from airborne LiDAR remains challenging: point clouds are collected top-down and frequently contain incomplete or overlapping canopy returns, while existing segmentation and reconstruction pipelines have not been systematically evaluated for CFD compatibility or scalability.

This research extends an existing tree reconstruction pipeline (Tree4CFD) along three stages: individual tree segmentation, tree reconstruction, and scalability through an accessible interface. For segmentation, a hybrid approach combining a 2D canopy-height-model (CHM) coarse pass with 3D multi-directional analysis of under-segmented clusters was implemented and benchmarked against the original watershed-only method, using a sensitivity analysis and grid search to tune parameters across densely, moderately, and sparsely built-up areas. This is later combined with a variable window for finding tree peaks in the CHM. For reconstruction, LoD2 and LoD3 tree geometries were compared on computational efficiency, geometric validity (via val3dity), and mesh simplification using CGAL alpha wrapping. Finally, an interactive web interface was developed, allowing users to define an area of interest, select desired output parameters, and automatically trigger data retrieval, segmentation, and reconstruction, producing georeferenced OBJ geometries ready for integration into frameworks such as City4CFD.

Results indicate that cell size, smooth sigma, and minimum peak distance are the most influential segmentation parameters. The multi-directional refinement resulted in a minimal increase in F1-score (average of 1.68%), showing that improvement opportunities are possible but need further evaluation. Reconstruction time was found not to be a bottleneck, with negligible differences between LoD2 and LoD3 processing times, and CGAL alpha wrapping produced watertight, substantially simplified meshes suitable for downstream simulation. The developed interface demonstrates that the full pipeline, from AOI selection to downloadable 3D geometry, can be executed without manual intervention, supporting its practical use by stakeholders such as Haskoning. ...
Master thesis (2026) - D. Feberwee, B. De Schutter, Daan Vis, R.D. McAllister
Imagine being flown offshore by helicopter to retune the tension controller of a crane, because its load tugger system became unstable during operation. Such behaviour is costly and directly affects the safety and workability of offshore heavy-lift operations. Load tugger systems, consisting of a variable frequency drive, a winch and a tugger wire, control the wave induced motion of the payload. Their operating conditions change continuously: the wire length varies, payloads range from light blades to heavy structures, and the disturbances differ over time. The current fixed-gain PI controller works well at its commissioning conditions, but loses performance and can become unstable when these conditions change.

This thesis investigates how Model Predictive Control (MPC) can improve the robustness, adaptability and performance of tension control under varying conditions, and whether it is suitable for industrial deployment. A four-state linear state-space model was derived and validated against a linear spring model of the tugger system. Four controllers were compared: the industrial PI baseline, a linear MPC controller, a MPC controller scheduled on the effective wire length, and this scheduled controller with a disturbance model. All were tuned with the same procedure and evaluated in five test cases on a nonlinear catenary model, which introduces realistic model mismatch.

The scheduled MPC controller with disturbance rejection performed best: compared to the PI controller, it reduced the mean integrated relative error by 75% and the number of failed runs from 27 to 10 out of 54, because it anticipates the payload motion instead of reacting to a tension error. Scheduling alone reduced the error by about 20% under disturbances, whereas MPC with a single model gave no meaningful improvement. The disturbance model, however, assumes a known disturbance period and made the controller the most sensitive to loop delay. Deployment therefore requires that the period can be provided, the loop delay stays small and the optimisation can be solved within the cycle time of the control hardware. Until then, the scheduled MPC controller is the most suitable next step from the current PI controller. ...