Non-intrusive aerodynamic load estimation is attractive for applications where force balances or pressure instrumentation are impractical, such as the aerodynamic testing of transiting athletes and other full-scale objects.
This thesis investigates whether volumetric particle tracking velocimetry can reconstruct the drag of a freely falling sphere using wake-rake and full control-volume formulations. Spheres of 15 and 25 cm diameter were repeatedly released through a seeded measurement volume of approximately 750 × 750 × 750 mm^3. Tracer-particle motion was recorded by six high-speed cameras and reconstructed using object-aware Shake-The-Box particle tracking.
Measurements from repeated releases were transformed into a sphere-fixed reference frame and ensemble-averaged.
Independent wind-tunnel load-cell measurements provided reference drag coefficients. The reconstructed velocity fields capture the expected bluff-body wake structure, but pressure reconstruction is more demanding.
Two ensemble strategies were compared. In the first, one frame per release was selected at a comparable sphere velocity, producing dynamically similar samples but limited particle support. These velocity-matched datasets required bin sizes of order 0.6D, which is too coarse to resolve the near-wake pressure gradients.
In the second, multiple frames per release were normalised by the instantaneous sphere velocity and combined, increasing particle support and allowing bin sizes of 0.23D–0.24D, close to the sphere-pressure criterion Δb < 0.25D.
This improved the pressure reconstruction but introduced a non-physical downstream decay of the momentum contribution due to velocity-dependent sampling bias.
For the matched-frame 15 cm sphere case, the wake-rake formulation gives the most robust drag reconstruction. Once the downstream integration plane is placed outside the near wake, around y/D ≳ 2, and the lateral integration width is at least approximately ±1.5D, the reconstructed drag coefficient reaches the closest agreement with the wind-tunnel reference. Near-wake planes remain unreliable because pressure and momentum contributions vary rapidly inside and close to the recirculation region.
The full control-volume formulation is highly sensitive to residual mass-flow imbalance. Without mass-conservation correction, small non-cancelling velocity errors over the control-volume faces lead to substantial drag underprediction. The correction reduces this sensitivity and shifts the reconstructed drag towards the reference value but is reliable only for wide control volumes where lateral-face contributions have nearly vanished. In this regime, the full control-volume result effectively reduces to a wake-rake estimate. The study concludes that drag
reconstruction from volumetric PTV measurements of a transiting sphere is feasible using a properly placed wake-rake formulation, but that the full control-volume method does not provide a practical accuracy advantage under
the present experimental conditions. Successful application of the full control-volume method requires a dynamically repeatable experiment with lower mass-flow imbalance, finer spatial resolution, and retained-particle densities
above the lower-bound estimate n_min D^3≈ 2.6×10^3, with additional margin required for outlier rejection, nonuniform seeding and object shadowing. ... Read more

In order to retrofit an existing aircraft with hydrogen propulsion, the nacelle size has to increase to house all propulsion components. This thesis investigates the aerodynamic effects of increasing the nacelle size through load balance measurements and particle tracking on a simplified geometry.
Based on these measurements, it can be found that the nacelle enlargement causes additional drag through more frontal area as well as additional induced drag through vortices. It also highlights changes in effective angle of attack at the wing sections right next to the geometry. ... Read more

This thesis presents a framework for generating Design of Experiments within mixed variable, constrained design spaces representative of aircraft operational flight envelopes. By explicitly modelling the feasible design space using simplicial meshes, the proposed methodology eliminates inefficient sampling outside the valid region while supporting continuous, discrete, and categorical variables. The framework was validated on an aircraft loads assessment case and evaluated using different meshing approaches. The results demonstrate improved computational efficiency and scalability, while highlighting the limitations of Monte Carlo sampling for accurately capturing extreme structural loads. ... Read more

Aerodynamics plays a crucial role in cycling, as most of the resistance a cyclist must overcome is drag. The aerodynamic performance of cyclists has been investigated in multiple experiments all around the world to understand the aerodynamics and reduce the drag. To improve the cycling aerodynamic knowledge and understanding, a Generic Cyclist Model (GCM) is created. The original Generic Cyclist Model is placed in a time trial with an asymmetrical leg position. This work broadens the aerodynamic knowledge by investigating the GCM with a symmetrical leg position. The aerodynamic sensitivity of the Symmetric Generic Cyclist Model and the differences compared to the Asymmetric GCM are analysed based on the influences of the leg position, Reynolds number variation, crosswind and the addition of flow control devices.
The wind tunnel experiment used force balances and 3D particle image velocimetry to analyse the aerodynamic performances; drag area, and flow fields of the cyclist.
The experiment indicates that the leg position has a significant influence on the aerodynamic performance, with the symmetric leg position yielding a lower drag area. This indicates that the legs have a great influence on the overall aerodynamic performance of the cyclist. The velocity and vorticity flow fields show a leg wake reduction and weakened repositioned vortices. The Reynolds number variations and the introduction of crosswind have a great influence on the drag area and flow fields of the cyclist. Showing that high velocities and crosswind are beneficial for drag area reductions for the Symmetrical GCM. Furthermore, the addition of flow control devices is beneficial for drag area reduction.
... Read more

Aerodynamic drag reduction of heavy-duty vehicles is essential for lowering fuel consumption, with the front wheels contributing significantly to overall drag. Truck aerodynamics is typically investigated using Computational Fluid Dynamics (CFD) and small-scale wind tunnel testing. However, validating CFD results in the highly complex and unsteady front wheel wake region remains challenging. Conventional validation methods, such as wind tunnel measurements or wake-rake techniques, are often limited to simplified models or planar data.

This study addresses this limitation by applying the non-intrusive “Ring of Fire” measurement technique to reconstruct the three-dimensional front wheel wake of a full-scale truck. The objective was to design and implement a Ring of Fire setup capable of resolving the wheel wake region to support CFD validation, while also evaluating the conventional wake-rake method.

The Ring of Fire measurement technique is based on Particle Image Velocimetry (PIV). Helium-Filled Soap Bubbles (HFSBs) served as tracer particles, illuminated by high-power LEDs and recorded by high-speed cameras. A Shake-The-Box (STB) Lagrangian Particle Tracking (LPT) algorithm reconstructed three-dimensional particle trajectories, from which velocity fields were derived. For comparison, a wake-rake equipped with Kiel probes measured a two-dimensional total pressure field in the wheel wake.

Experiments were conducted on a test track using a full-scale European cab-over-engine tractor–trailer combination. Two configurations were tested: a baseline (Variant A) and a reduced-aero configuration (Variant B). Containing the HFSBs in an outdoor environment was a major challenge and was addressed using a foldable dome tent housing three vertically oriented LED units and a seeding rake. Four high speed cameras recorded the motion of the HFSBs within a measurement domain extending 1 meter from the truck surface and up to 1 meter in height, covering the longitudinal distance of 2.2 meter. A dedicated run procedure ensured synchronization between dome opening and data acquisition during truck passage.

A total of 100 runs were performed over two days. 8 valid runs for Variant A and 21 for Variant B were processed after excluding invalid measurements. Crosswind effects were assessed using conditional averaging and were found to be negligible within measurement uncertainty. Convergence analysis indicated that at least 16 combined runs were required to achieve a stable, time-averaged flow field with over 95% spatial data coverage.

For both configurations, the wheel wake region, characterized by significant reduced velocity magnitude, emerged downstream of the footstep and expanded further downstream. Overall, Variant A showed a larger wake region.

Comparison with wake-rake results revealed similar wake topology, but the Ring of Fire provided higher spatial resolution and full three-dimensional reconstruction. The study demonstrates that the Ring of Fire technique successfully captures the complete front wheel wake of a full-scale truck and offers substantial advantages over conventional planar wake-rake measurements. ... Read more

As the need to reduce the carbon footprint of the energy sector grows, wind energy has emerged as a leading renewable source for sustainable power generation. In particular, the offshore sector has gained significant attention due to its higher wind speeds as well as its reduced noise impact. TU Delft's Faculty of Aerospace Engineering has identified accelerating the energy transition towards renewable energy sources as one of its main goals. To achieve this, offshore wind turbines must be tested under realistic Atmospheric Boundary Layer (ABL) conditions.

Wind-tunnel ABL simulations typically use spires, barriers and roughness elements as passive devices. However, this configuration requires a long downstream length for the boundary layer to develop. Therefore, an alternative approach developed by Cowdrey (1967) is used, which employs a non-uniform rod grid to produce a prescribed power-law mean velocity profile. Moreover, unlike spire-based configurations, this method allows for variations in the turbulence levels of the resulting boundary layer through the grid design parameters.

This thesis investigated the capability of Cowdrey's method to reproduce an offshore ABL in a short open-jet test section. Particle Image Velocimetry and Hot-Wire Anemometry were used to measure the mean velocity profile, turbulence intensity, integral length scales and power spectral density. Cowdrey grids were evaluated over a test matrix with four rod diameters, two cross-sectional shapes, two turbulence design settings, two target boundary layer thicknesses at three free-stream velocities.

Within the tested velocity range, the use of Cowdrey grids increased the boundary layer height by up to 26 times compared with the naturally developing layer within the same fetch length. The fitted power-law exponents for most configurations were slightly larger than the design value but remained close to the target value overall. Moreover, most grids exhibited a good agreement with the logarithmic law in the inner layer and within the available downstream distance, attained the characteristic aerodynamic roughness length of an offshore Atmospheric Boundary Layer. The measured turbulence intensities exhibited elevated near-wall levels relative to semi-empirical relations but qualitatively followed the predicted offshore ABL trends. The power spectral density conformed to the von Kármán model across all sampled heights, with noticeable departures only near the top of the outer layer at the furthest downstream station. Lastly, the longitudinal integral length scales tended to be consistently lower than predicted by semi-empirical relations.

The study provided valuable insights into the applicability of Cowdrey’s method in reproducing a scaled offshore ABL. It laid the foundations for further research regarding ABL testing in TU Delft's experimental facilities. Further improvements should focus on bringing the fitted power law closer to the target value and refining the turbulence characteristics to minimize deviations from accepted standards. ... Read more

In a world with an urgent demand for sustainable energy, the wind energy industry plays a key role in accelerating this transition. Over the past decades, wind turbines have evolved from expensive, relatively inefficient machines into increasingly cost-effective and highly efficient technologies, supported and promoted by many countries worldwide. However, this rapid progress has come at a cost: larger and more efficient turbines require substantial investments in raw materials, research and development, component standardization, and supply chain optimization.

In recent years, the curtailment of wind power in Europe has increased, largely due to insufficient grid capacity and limited energy storage. As a result, wind turbines are more frequently operated in parked conditions, with their rotors brought to a standstill. Under these circumstances, one of the key challenges in scaling up turbine size is the risk of vortex-induced vibrations (VIV) in the blades. In parked conditions, the blades are often pitched to very high angles of attack (close to 90°) to cut out of the wind. If the vortex shedding frequency approaches the blade’s natural frequency, a lock-in phenomenon may occur, leading to strong vibrations. This vibration in the long term can contribute to the overall fatigue load of the wind turbine and reduce the structural life.

Although increasing attention has been given to VIV in wind turbine blades, significant gaps remain in understanding the fundamental flow physics that govern these vibrations, specifically the unsteady aerodynamics of airfoils at high angles of attack. This dissertation therefore investigates the unsteady aerodynamics of both static and oscillating airfoils under such conditions, with the aim of building a detailed physical understanding of VIV from an aerodynamic perspective.

The research was carried out through a series of wind tunnel measurements. First, a campaign on a static airfoil examined unsteady aerodynamics across a wide range of angles of attack (up to 310°). Aerodynamic forces, vortex shedding patterns, and shedding frequencies were compared between forward flow (leading edge upwind) and reverse flow (trailing edge upwind) conditions. Although reverse flow is uncommon in normal operation, it can occur during parked or installation phases; the insights gained in this research therefore form a critical foundation for subsequent studies on oscillating airfoils.

The main focus of the dissertation is the unsteady aerodynamics of oscillating airfoils, studied using the forced motion method to mimic VIV. Three motion types, namely surging, plunging, and pitching, were investigated. Particle Image Velocimetry (PIV) was employed to capture the flow fields, while surface pressure measurements provided aerodynamic forces. By correlating vortex dynamics with force responses, the study reveals how the mean angle of attack and motion parameters (such as frequency and amplitude) influence the overall unsteady aerodynamics of the airfoil and how lock-in is triggered under different motion kinematics. Comparisons between forward and reverse flow conditions further enrich the findings, where the reverse flow dynamic stall was thoroughly discussed—from vortex dynamics and aerodynamic forces to a newly proposed dynamic stall vortex and trailing edge vortex onset determination method.

Overall, the comprehensive experimental dataset and resulting conclusions advance the fundamental understanding of unsteady airfoil aerodynamics at large angles of attack. These findings not only clarify the underlying mechanisms causing VIV from the perspective of vortex dynamics and aerodynamic forces, but also provide a valuable basis for future aeroelastic VIV studies and the development of engineering models. ... Read more

Volumetric Particle Image Velocimetry (PIV) is a state of the art technique for quantitative flow diagnostics. Its ability to measure the velocity field around the typically complex objects, as needed in the field of aeronautics, makes it a valuable tool for designers and engineers. Wind tunnel experiments making use of PIV as a diagnostic tool are used to gain physical insight into the flow field organization, generate data for validation of computational methods, and perform optimization in various domains, most notably in the field of aerospace and wind engineering.

Despite continuous advancements in the measurement technique, performing a PIV experiment in industrial wind tunnel environments remains challenging, making its use rather limited to niche applications within aeronautics or the automotive industry. In this thesis the recent advancements in helium-filled soap bubbles (HFSB) technology and 3D particle tracking algorithms are synthesized, to demonstrate the impact of PIV in industrial environments and extend its utility to a wider range of aeronautical applications.

Two limitations currently preventing the broad adoption of PIV are addressed: i) the limited spatial coverage of full-field volumetric velocimetry due to shadows and blocked optical access; ii) the limited measurement accuracy and the accessible velocity ranges by conventional two-pulse PIV systems.

Both these problems are introduced and treated in Part I of this thesis. Multi-directional illumination and imaging systems with redundancy are introduced for the study of volumetric flows around complex geometries. A volumetric loss parameter is defined, that can be used as a guideline in the phase of experimental setup design of these systems. Additionally the logics of the combinations of the multiple cameras that work in a single system are examined and the results of the different combinations are presented.

The technical limitations in hardware technology of the cameras’ frame rates have inspired the revisiting of original PIV methods of multi-exposure imaging. This is investigated as a way to increase the dynamic velocity ranges of more common and practical two-pulse systems used in industrial testing. Methodology and initial results are presented for the workings of a novel concept.

Part II introduces specific application experiments in two fields. The first is part of integrated propulsion, the study of the flow around a thrust-reverser at- tached to a complete aircraft; and the second, the flow around the top side of the superstructure of a ship with helicopter and drone landing capabilities in the deck, investigated across a spectrum of incoming wind directions.

The study around the thrust reverser has demonstrated the feasibility and added value of PIV even in challenging complex industrial wind tunnel experiments, by providing insight into the mechanisms of jet reversal and re-ingestion, as well as a rich database that is in good agreement with, and complementary to, more traditional wind tunnel re-ingestion measurement methods.

The study of the redundant multi-illumination and camera systems has proven to increase the spatial coverage of measurements with these setups, providing more complete data for numerical low-fidelity turbulence models validation, as well as practically proving to increase the robustness of PIV systems against reflections. This has been demonstrated in the final chapter of the thesis.
... Read more

Particle Image Velocimetry (PIV) constitutes the state-of-the-art for quantitative flow diagnostics. Its volumetric variant is nowadays also well established, able to provide a full description of the three-dimensional flow field. Despite the maturity of the technique, the use of volumetric PIV in industrial facilities is seldom considered, due to the challenges associated with the high Reynolds numbers, the presence of complex wind tunnel models in the domain of interest or the long distances with respect to the PIV instrumentation. In an attempt to aid the dissemination of volumetric PIV in industrial large-scale facilities and promote its use for engineering design and validation of flow simulations, four elements of the PIV working principle are critically reviewed in this dissertation, namely: the scalability of tracer particles for experiments in air flows; image preprocessing to deal with complex light reflections; the recording strategy and particle tracking algorithm; and data reduction techniques that exploit the modal decomposition of the velocity field.

The scalability of experiments using PIV relies upon several parameters: primarily the tracers scattering cross section and their concentration, the power and distribution of illumination; and the imagers sensor size and their amount. Given their larger cross section, helium-filled soap bubbles (HFSB) allow measurements in air flows over a significantly larger domain compared to traditional oil or water droplets. Controlling their diameter translates into scalability of the experiment. In chapter 3, a technique is presented to extend the control of HFSB diameter by geometrical variations of the generator. A theoretical model predicts the bubble size and production rate, which is verified experimentally by high-speed shadow visualization. The overall range of HFSB produced in a stable regime can be varied from approximately 150 μm, targeting experiments in research facilities at high spatial resolution, to a few millimetres, sufficient for large-scale measurements in industrial facilities as well as full-scale on-site experiments. Imaging by light scattering of such tracers is also investigated, in view of controversies in the literature on whether diffraction or geometrical imaging dominate the imaging regime. For large-scale volumetric applications, it is shown that varying the bubble diameter allows increasing both the measurement domain as well as the working distance of the imagers at 10 m and beyond.

Measuring the velocity field around a complex object by volumetric PIV is hindered by shadow formation (illumination), camera occlusion (imaging) and light reflections from the object surface. The former have been recently dealt with by multiplying illumination and imaging directions (redundancy) and by the integration of ray-tracing techniques to include the effect of visual blockage caused by the object. Instead, the problem of light reflections blinding regions of the images has not been afforded yet. The latter pertains to interactions between illumination and imaging through the object surface and it poses additional challenges to ghost particle formation, particle detection and tracking in general, increasing the computational cost and reducing the accuracy of the measured velocity field. In chapter 4, a method is proposed to effectively detect such regions, and measures to modify the particle triangulation algorithm are devised. The viability of this novel approach is examined by application to two experiments of increasing complexity. The first case is the flow around a stationary wall-mounted cube as imaged with a redundant number of cameras. The second experiment tackles an elite runner sprinting across the measurement region obtained with the Ring-of-Fire technique. A considerable reduction of ghost particles (false positives) is attained, while the formation of voids (false negatives) is also minimized. The overall result of the method maximizes the measurement region around and in proximity of the object of interest.

Multiple-exposure (ME) recording is a variant of PIV whereby more than two samples of the particle position are obtained to overcome some limitations of single-exposure dual-frame recordings, such as accelerometry, pressure from instantaneous PIV data, or to further extend the dynamic velocity range. Compared to time-resolved systems, ME lowers system requirements in terms of laser power and camera frame rate, thus making it more suitable for applications involving a redundant number of cameras and higher flow velocities. In chapter 5, the reliability and accuracy of volumetric particle tracking in ME recordings comprising up to 5 exposures with one or two frames is first scrutinized on a synthetic particle field motion based on a Taylor-Green vortex lattice, yielding viable results. The measurement accuracy in terms of dynamic velocity and acceleration ranges is reported, as a function of particle image density, number of pulses and timing sequence. Besides, ME recordings are simulated from a time-resolved experiment around a wall-mounted cube, which yield equivalence between ME and time-resolved conditions. A demonstration of volumetric ME for accelerometry and pressure from PIV is also included, with experiments in the turbulent wake of a circular cylinder.

Modal decomposition of PIV measurements is a common approach to reduce data complexity and aid interpretability. In chapter 6, a method to reconstruct the dense velocity field from relatively sparse particle tracks, as obtained for instance from volumetric ME recordings, is introduced. The goal is to provide a representation of the 3D flow field on a Cartesian grid, for inspection of derived flow quantities, at high spatial resolution. The approach leverages the properties of proper orthogonal decomposition (POD) and it iteratively reconstructs the detailed spatial modes from a first, coarse estimation thereof. The initially coarse Cartesian representation of the velocity field is obtained by local data averaging, where POD is applied. The spatial resolution of the POD modes is enhanced by reprojecting them onto the sparse particles velocity to iteratively improve the reconstruction of the temporal coefficients. Finally, the enhanced velocity field is represented at high-resolution with a reduced order model using the dominant POD modes. Experiments in the wake of a cylinder at Re_D = 27,000 are used to assess the suitability of the method to resolve the turbulent Kármán-Benard wake. The approach is benchmarked against traditional as well as state-of-the-art reconstruction methods, illustrating the capability of enhancing the spatial resolution of sparse velocity data. ... Read more

This thesis project examines the drag crisis trigger mechanism for double layer fabrics on cylindrical cross-flow. The primary goal is to discover why the double layer fabric is able to trigger the drag crisis much sooner than conventional surface roughness.

The methodology employed in this research follows an experimental approach, using balance measurements to determine the aerodynamic drag at varying Reynolds numbers for different configurations. Particle Image Velocimetry (PIV) measurements are performed to examine the boundary layer, flow separation point and other flow phenomena occurring near the cylinder surface.

Over the course of this research, eight different double layer configurations have been the subject of study, as well as six single fabric configurations and two reference configurations consisting of a bare cylinder and a cylinder with zigzag trips. Balance measurements have been performed on all of the configurations to determine which configurations are deemed relevant to be studied with PIV techniques. Thus, PIV measurements have been performed on two double layer configurations with a varying underlayer and the same overlayer, as well as the study of the individual fabrics employed to make up the two-fabric construction. That is to say, the two underlayers and one overlayer used have been studied on their own.

The balance results uncover that the minimum drag coefficient across all double layer configurations is achieved with the smallest rib spacing. Conversely, this minimum drag coefficient is located at the highest critical Reynolds number across all configurations. Additionally, a relationship between the critical Reynolds number and the underlayer rib spacing has been determined for the studied configurations. Furthermore, the PIV measurements provide insight into the normalized velocity fields and reconstructed pressure fields. With these results, the development of the boundary layer on the foreside of the cylinder with double layer fabrics can be studied. Examining the flow near the surface, it can be seen that the presence of the ribs results in a localized flow convergence (upstream of the rib) and divergence (downstream of the rib), these geometric effects accelerate and decelerate the flow locally, causing static pressure oscillations on the foreside of the cylinder. With the appearance of localized adverse pressure gradients on the foreside of the cylinder, flow instabilities are seeded eventually trigger the transition of the boundary layer to a turbulent state, thus allowing the flow to remain attached to the cylinder surface for longer, ultimately delaying separation and reducing pressure drag.

While the study has provided valuable insights regarding the trigger mechanism for the drag crisis on double layer fabrics on cylinders, it has also paved the way for further research regarding this topic and the specific effects of rib height, behavior and performance in unsteady flows and whether the two fabric construction is strictly necessary. These considerations are addressed at the end of the conclusions chapter. ... Read more

Static pressure is a scalar magnitude that expresses the force per unit area exerted by a fluid at rest. As such, it constitutes one of the two mechanisms through which fluid flows generate forces on bodies. Moreover, static pressure is not only relevant in the definition of surface loads, as it plays a key role in a number of fields, such as turbulence research due to its impact on the amplification or damping of turbulent flow instabilities, or medical research, provided its key paper on cardiovascular disease.

Accordingly, different approaches exist that allow to obtain pressure information in Fluid Dynamics applications. Among these, simplified analytical models, Computational Fluid Dynamics and experimental measurements stand out given their extensive use. While each of these comes with its own advantages and limitations, the latter typically offers the advantage of being conducted with real flows, hence providing a reliable source of information if proper similarity parameters and set-up are achieved.

While there exist different techniques to measure pressure experimentally, among which pressure tapping and pressure sensitive paint stand out, these present major limitations, such as the limited spatial resolution that can be achieved without intrusion effects or the challenges encountered during calibration, respectively. Consequently, more recent methods have been developed that allow to reconstruct pressure from velocity fields obtained via Particle Image Velocimetry. An instance of such algorithms is the Poisson Solver, which is based on the application of the incompressible relation to the momentum conservation equations, yielding a boundary-value problem for pressure. Nonetheless, while this approach benefits from the instantaneous and simultaneous nature of PIV measurements, it presents its own challenges, among which the propagation of noise from the velocity field into the reconstructed pressure field stands out.

More recently, with the proliferation of Machine Learning, the number of applications in Fluid Mechanics has grown. In particular, an approach that stands out are Physics-Informed Neural Networks, which optimize Deep NN models minimizing a loss function with contributions from labeled flow data variables and residuals from physical equations, thus learning the flow field variables. While diverse use cases have been reported for PINNs, such as the generation of reduced order models or the direct simulation of flows, especial emphasis has been placed in research on their ability to infer unsteady or mean pressure fields from velocity measurements, via the application of the Navier-Stokes equations.

Even if research has shown PINNs offer key advantages with respect to traditional pressure reconstruction methods, such as robustness to Gaussian noise or lack of discretization errors, analysis of the research available highlights key areas that require further exploration in the establishment of PINNs as a reliable alternative to traditional methods. Specifically, the study of PINNs performance with real experimental data and its comparison with solvers as the Poisson against direct experimental measurements is of paramount relevance, provided that the vast majority of publications concern the use of artificial experimental data from CFD simulations.

In accordance, a PINN framework has been developed and its accuracy in the reconstruction of surface pressure has been tested using time-averaged data from both CFD simulations and experimental tests of the two-dimensional flow around a cylinder. Particularly, comparison of the PINNs and the Poisson surface pressure reconstructions with pressure tap data showed superior performance of the former, with respective MSE reductions of -1% and -21% for the flows around a smooth cylinder and one fitted with zig-zag strips at θ = ±45 º.

Additionally, sensitivity studies to understand the effect of various parameters in the PINN training process has resulted in the identification of various trends. Among these, especial attention is required by the ability of PINNs to add regularization in areas affected by correlated noise such as reflections via the addition of collocation points, where the PDE loss is evaluated. Further noteworthy findings concern the benefit of using physical boundary conditions at solid surfaces in the form of the no-slip, no-penetration and no-fluctuations constraints. In this study, it has been proven that these allow not only to bypass non-physical pressure fluctuations that derive from spatially-correlated noise, but also to reduce surface pressure reconstruction error when data gaps exist close to the surface, achieving reductions of up to -92% for a radial data gap of 75% of the cylinder radius from the cylinder surface. Finally, it has been shown that, provided that the dataset contains points that allow to define a reference pressure value, the provision of sparse pressure tap data to the NN during training results in local, rather than generalized, error reductions.

In conclusion, the sensitivity studies carried out on the smooth cylinder dataset have resulted in pressure MSE reductions as substantial as -50\% with respect to the Poisson solver when both are compared with static pressure tap data, supporting the establishment of PINNs as an alternative method to conventional pressure reconstruction algorithms as the Poisson solver, despite the time penalty that these can represent given the instantaneous nature of the latter. On the qualitative side, it has been proven that PINNs can provide a flexible framework to embed prior knowledge of the solution, such as the positive nature of normal Reynolds Stress components or boundary conditions. Along the same lines, it is shown throughout the thesis how PINNs can be used to perform debugging steps that allow to identify sources of error. ... Read more

The development of offshore wind farms has demonstrated significant value in harnessing renewable energy sources. However, the offshore wind energy sector faces increasing challenges, making the realization of new wind farms a greater financial risk. This has reached the point where several high-profile offshore wind projects have been halted. A key contributor to these struggles is the aerodynamic effect of wind farm wake losses. As wind turbine sizes increase, costs rise, but the wind farm concept limits energy yield. This is due to slow wake recovery behind turbines, causing downstream turbines to only be able to harvest energy from low-momentum flow. Based on the literature review conducted, the methods to improve wind farm efficiency by reducing wake losses, such as wind farm layout optimization and rotor yawing or tilting, only partially address the problem.

The concept of regenerative wind farming, proposed by Ferreira, seeks to provide a solution. This involves integrating lifting devices into wind energy harvesting systems that can redirect the wake vertically, enabling the flow from higher atmospheric layers to contribute more to the wake recovery process. Such interactions can dramatically enhance the energy that is replenished into the wind farm, offering a promising approach to overcoming these challenges.

This research aims to evaluate the potential of the novel wind farm concept, regenerative wind farm, through a scaled wind farm experiment. In the scaled wind farm, there are nine wind energy harvesting systems, which are aerodynamically modeled by using porous disks and wings. The study focuses on the far wake and the performance of downstream turbines. The experiment was conducted in the Open Jet Facility (OJF) at TU Delft. The flow field was measured using Particle Tracking Velocimetry (PTV) with Helium-Filled Soap Bubbles (HFSB), and load measurements were accompanied.

Load measurements revealed that thrust values for downstream turbines increased by more than three times when lifting devices were attached to the actuator surfaces. Also, flow field data showed significantly higher wake velocities, as potent vertical flows in the wake regions were induced by the tip-vortices of the wings, which enhanced vertical energy entrainment. This vertical motion actively entrains the flow above the wind farm, allowing high-momentum air to enter the wind farm layer, something that does not occur without the wings. Additionally, the lifting devices reduce turbulence intensity in the rotor projection area at the downstream end of the wind farm, helping to lower the fatigue loading of downstream systems. An investigation of a misaligned row confirmed that the concept remains effective even under such conditions.

This work demonstrates that the regenerative wind farm concept holds great potential to enhance wind farm power output while reducing the required wind farm area. With this concept, wind turbine wake losses are effectively mitigated by entraining high-momentum flow into the wind farm. ... Read more

This thesis presents an innovative approach to investigating automotive underbody aerodynamics through the development and application of an on-site 3D Lagrangian Particle Tracking (LPT) system. Automotive performance, particularly in high-speed racing applications, is significantly influenced by the aerodynamic efficiency of vehicle diffusers. The work addresses the challenges associated with accurately capturing complex three dimensional flow structures beneath a moving vehicle, where traditional flow measurement techniques struggle to capture underbody flows, especially in an experimental setting.

The research builds upon a previous study on a diffuser equipped radio-controlled car, which uses a measurement technique known as the Ring of Fire. By improving the camera setup, as well as creating better seeding and illumination, the setup allowed for successful particle tracking of neutrally buoyant Helium Filled Soap Bubbles underneath a car model, driving at around 7.5 m/s. The particle tracks captured in a measurement domain the size of (300 x 150 x 200) mm3 allow for the reconstruction of a velocity field around three tested car geometries, by combining data from multiple runs of the car driving through the measurement domain. These geometries are a flat floor car model, a car model fitted with a 15◦ planar diffuser, and a car model with the same diffuser, but also an additional strip of vortex generator fins placed ahead of the diffuser leading edge.

Using a pressure gradient integration method, a pressure field around the car models was obtained. Looking at both the velocity and pressure distribution around the models, the setup was able to capture the difference in peak velocity underneath the car, where the diffuser equipped model showed a maximum velocity of around 1.4 times the freestream velocity. Velocity and pressure coefficient profiles measured along the car’s centerline closely match those reported in the literature, confirming that the diffuser primarily impacts the rear region of the vehicle.

Streamwise vortices introduced into the diffuser by the vortex generator strip showed to be primarily moving high momentum flow closer to the diffuser surface, while potentially resolving a laminar separation bubble near the diffuser leading edge, observed for the plain diffuser case. Difference in local velocity magnitude and pressure coefficient measured at the diffuser leading edge between the flat floor and diffuser equipped models proved to be large enough to be statistically significant. An estimated 25 runs was needed to reach a velocity convergence inside the diffuser within 1% of the mean car velocity, where only 4 or 5 runs would be enough for the flat floor regions upstream of the diffuser.

This work shows the improvement made to the Ring of Fire setup developed to measure on-site automotive underbody aerodynamics. It proves the capabilities of applying 3D LPT to quantify underbody flows, and the potential to apply this setup on larger and faster vehicles.
... Read more

Aerodynamic drag plays a critical role in high-speed sports, including running, where small performance margins can be decisive for victory in elite competitions. Despite its importance, research on running aerodynamics remains underexplored, with most studies relying on stationary mannequins or simulations that do not capture the dynamic behavior of flow around a moving runner. In other words, there is a significant research gap in experimentally visualizing the wake flow and accurately measuring the drag in real-world running scenarios. This study aims to address this research gap.

In recent years, the Ring of Fire measurement technique has emerged as a feasible option to visualise and analyse flow structures of transiting objects based on particle image velocimetry.
The technique has already been proven in sports like cycling and ice skating, but has not yet been applied to running. This study adapts the Ring of Fire measurement method for sprinting athletes, and the raw images are processed by Shake-the-Box (STB) Lagrangian Particle Tracking. This results in a three-dimensional, time-resolved velocity field in the wake of a runner with an uncertainty of less than 5\% of the runner's speed. This velocity field is used for qualitative flow visualisations, as well as for drag estimations, which are computed from a control volume approach by utilising the flow field before and after the passage of the athlete.

The experimental methodology involved nine junior athletes, wearing both standard sprint suits and aerodynamic suits, sprinting at constant speeds through a measurement setup including three high speed cameras, four LED arrays and Helium Filled Soap Bubbles (HFSB).

The results of the investigation include the visualisation of the full wake,
quantification of the velocity deficit, a qualitative vorticity analysis, some lateral velocity findings and the measured values for drag of one of the athletes.

There are multiple interesting findings about the flow around a dynamic runner in this report. One of them is the splitting of the wake into two side-by-side stream tubes in the far wake. The velocity deficit in a relevant region of the dynamic wake has also been quantified, and it is shown to vary with the inverse of the distance behind the runner, which is useful information for trailing runners.
Another interesting observation is that the vorticity field in the near wake generally follows the rules of finite cylinder flow, where the body parts of the athlete are finite cylinders with variable diameter.
The hypothesis that the flow should have a lateral oscillation related to the frequency of the runner's steps has also been confirmed.

This study demonstrates the applicability of the Ring of Fire system to running aerodynamics and offers the full visualisation of the three dimensional flow in the wake of a moving runner, bridging the research gap to pave the way for further advancements in the field. ... Read more

Understanding atmospheric motion is crucial for analyzing wind-structure interactions, particularly within the Atmospheric Boundary Layer (ABL). Traditionally, wind tunnels have simulated unsteady ABL conditions using active and passive devices. Recently, multi-fan wind generators have emerged as a flexible, cost-effective alternative, allowing independent fan control to replicate wind conditions the wind turbines, airborne devices or civil structures are subjected to.

This thesis investigates the capability of a newly manufactured 3 x 3 multi-fan wind generator to produce idealized uniform flow, linear shear flow profiles, and streamwise sinusoidal gusts. The system consists of off-the-shelf computer fans, controlled via open-source software, offering a low-cost and adaptable approach. Particle Image Velocimetry (PIV) measurements show a 96% uniform core region with 11.5% turbulence intensity 1 m downstream from the system. There, the area of the core is reduced to 1/9 of the system’s area due to the outer shear layer. The system struggles to replicate linear wind shear, but can generate oscillatory gusts at 0.2 Hz, 0.4 Hz and 0.8 Hz, corresponding to the operating frequency of the fans. Further improvements should focus on extending the configuration to maintain a larger core region, and on reducing swirl dynamics and recirculation to enhance flow quality for wind engineering applications. ... Read more

This thesis presents a novel methodology for investigating the aerodynamics of a competitive racing vehicle through the application of Lagrangian Particle Tracking (LPT) combined with the Shake-the-Box (StB) algorithm. Applied to the DUT24, developed by Formula Student Team Delft (FSTD), the approach enables detailed reconstruction of the three-dimensional velocity field around a fully aerodynamically equipped race car. Particular emphasis is placed on two flow regions of critical aerodynamic importance: the underbody diffuser and the
wakes shed by the rotating tyres. The study addresses the challenges of adapting the experimental set-up to full scale and highlights the advantages of advanced LPT compared with conventional flow measurement techniques.

The experiments were conducted on a 60 m test track at approximately 12 m/s, using Helium-Filled Soap Bubbles (HFSB) as neutrally buoyant tracers. For the diffuser study, three cameras and four LEDs were installed inside a ditch beneath the car, yielding a measurement volume of 300 × 400 × 350 mm. The tyre-wake study employed a lateral arrangement of cameras and LEDs, achieving a volume of 400 × 800 × 450 mm. Statistical convergence was ensured by repeating the tests multiple times, with 40 runs for the underbody and 30 for the tyre wakes.

The results revealed strong flow acceleration beneath the front wing and diffuser, with peak velocities exceeding twice the free-stream (u/U∞ > 2.2) and suction pressures around Cp ≈ −3. Diffuser strakes generated coherent streamwise vortices that promoted flow attachment and contributed to downforce, with no evidence of vortex breakdown. The tyre wakes were shown to be highly three-dimensional and unsteady, with the front tyre producing larger wakes than the rear. Vortical structures shed from the front wing and underbody mitigated and
reshaped these wakes, highlighting the strong coupling between wheel aerodynamics and upstream devices.

Because pressure cannot be directly obtained from LPT, a reconstruction algorithm based on omnidirectional integration with an irrotational boundary condition was applied. The method reproduced physically consistent pressure distributions, such as diffuser recovery and tyre-wake stagnation zones, but absolute discrepancies remained when compared with CFD, particularly in regions of strong adverse gradients or near reflective surfaces.

Comparison with CFD showed good agreement in overall flow topology and acceleration. Both approaches captured the diffuser acceleration and recovery, though CFD underpredicted vortex strength and momentum conservation. In the tyre wakes, CFD resolved similar structures but lacked rotation effects, producing smoother and less energetic vortices. Pressure fields agreed on overall trends but diverged in magnitude.

A convergence analysis indicated that 25 runs were required to reach 1% velocity uncertainty in the diffuser, while the unsteady tyre wakes demanded up to 70 runs for the same threshold, though 30 were sufficient to achieve a convergence of 2% of the velocity flowfield.

In conclusion, the Ring of Fire methodology, scaled to full vehicle dimensions, successfully captured the complex aerodynamic mechanisms of a modern race car. Despite limitations in pressure reconstruction and CFD comparison, the system proved robust and capable of resolving high-velocity underbody flows, coherent diffuser vortices, and unsteady tyre wakes, confirming its value as a powerful tool for experimental vehicle aerodynamics. ... Read more

Offshore wind energy has gained prominence due to its favorable wind resources, turbine scaling potential, and vast installation areas. However, wake losses caused by closely spaced wind turbines remain a significant challenge, reducing overall wind farm efficiency. This thesis explores the application of vertical-axis wind turbines (VAWTs) geometries for maximizing energy density through passive wake recovery techniques. In addition to traditional Darrieus-type VAWTs, the efficacy of passive wake control techniques for novel multi-rotor systems (MRS), namely the X-Rotor and block design concepts, is assessed through experimental studies and proof-of-concept demonstrations.

The thesis begins with a large-scale experimental investigation into the wake dynamics of a high-energy-density VAWT wind farm, providing the first comprehensive dataset of three-dimensional, time-averaged flowfield measurements. A dense grid of nine Htype VAWTs with fixed spacing was analyzed, exploring a passive wake control strategy known as the "vortex generator" mode, where blade pitch is modified to accelerate wake re-energization. Two pitch configurations were tested: positive (pitched-in) and negative (pitched-out). The positive pitch case exhibited significant momentum influx from above and below the rotor, along with lateral wake deflection. In contrast, the negative pitch case induced upwash while injecting high-momentum flow from the sides. Wake recovery was quantified by assessing available power, showing a maximum of 72.4% recovery three diameters downstream in the positive pitch case, 6.4 times higher than the baseline. The negative pitch case reached a 53% recovery four diameters downstream, a 2.1-fold improvement over the baseline. These findings highlight the potential of passive wake control strategies to enhance wind farm energy density.

The X-Rotor introduces an innovative design featuring an X-shaped VAWT, referred to as the "primary rotor," and blade-tip-mounted HAWTs, known as the "secondary rotors." This design employs an "aerodynamic gearbox" mechanism, where the primary rotor extracts mechanical power while the secondary rotors drive electrical generators at the blade tips. This thesis presents the first experimental wake measurements of the X-Rotor, revealing that its wake remains concentrated within its projected frontal area, shaped by the coned blades. The shed vorticity follows an elliptical pattern, inducing crossflow components along the height. This dataset provides a baseline for evaluating the secondary rotors’ impact on wake evolution near the bottom blade tips and demonstrating the aerodynamic gearbox mechanism. ... Read more

This thesis proposes techniques to produce wall-shear stress estimates from three-dimensional Lagrangian particle tracking (LPT). Several works have already faced the problem of determining near-wall velocity and in particular skin friction for the case of a flat surface. Here, the problem is translated to generic three-dimensional objects. The work makes use of an experimental database, recently produced, with LPT (Hendriksen et al. 2024), that provides in-situ registration of the object, for three shapes of increasing complexity: a cube, an airfoil and a cyclist. Four techniques are examined and compared, including interpolation method as well as local data regression. The uncertainty is evaluated a-posteriori, comparing the results with a local coin-stacking technique, where applicable. All techniques yield accurate and robust representations of the skin friction lines around three-dimensional objects, allowing for an insightful inspection of the near-surface flow topology. Instead, distinct differences are found when the skin-friction magnitude is estimated. ... Read more

This report presents the preliminary design process of an aerial drone aimed at exploring the surface of Mars. The aim of this project is to help determine whether the Red Planet ever hosted alien life. Building on recent breakthroughs in the field, the team set out to develop a drone that can sustain 30 minutes of continuous flight, travel up to 20 kilometres, and identify, collect and transport 3 kilograms of rocks to a Martian ground station. The design process involved planning and integration of various subsystems, including propulsion, structural, electrical and payload systems, all tailored to overcome the unique challenges posed by Mars’ environment. This document provides a thorough overview of the design methodology, subsystem integration and performance analysis, underscoring key design decisions and providing recommendations for future development and optimisation of the drone. ... Read more

Laser light reflection mitigation in Particle Image Velocimetry (PIV) is crucial for accurate flow
field measurements. While numerous methods exist for planar PIV, fewer have been developed for volumetric PIV systems, especially for coaxial setups like Robotic PIV. Light reflections in volumetric PIV experiments result in high-intensity regions that corrupt particle detection and analysis.

This study presents three novel approaches for treating light reflections in Robotic PIV experiments. The first and second methods use image filtering and masking techniques in the
wavenumber space to separate particle images from reflection regions. The first technique called Spatial Fourier Filter involves decomposing the image signal into low- and high-wavenumber components using the 2D discrete Fourier transform (DFT). A high-pass filter is then applied to attenuate the intensity of reflection regions. Then, the second methodology Spatial Fourier Filter + Mask takes the resulting image from the first method and performs a step of automated adaptive masking to remove residual reflection areas that the filtering approach is not able to eliminate. The third methodology named 3D-based Particle Concentration Mask acts in a later stage of the processing pipeline, creating a 3D mask on the instantaneous processed Shake-the-Box data by analysing the particle concentration distribution over the flow domain.

The proposed methods are tested on experimental data obtained from experiments performed with Robotic PIV on three different geometries: a side-view mirror, Formula 1 car and a propeller. The tests were conducted at one of TU Delft Aerospace Engineering Faculty’s facilities, the W-tunnel in the High-Speed Laboratory (HSL). Comparison between raw and pre-processed images, as well as particle tracking results, is presented.

The results from this data comparison show unsatisfactory outcomes from both Spatial
Fourier Filter and 3D-based Particle Concentration Mask, which fail to fully remove the spurious regions. Nevertheless, the results confirm the successful removal of reflection-induced artifacts in instantaneous images by using the spatial Fourier filter automated masking approach. The developed image pre-processing strategy effectively removes reflection regions in Robotic PIV images, preventing the appearance of spurious particle tracks. The method shows promising results mitigating unsteady light reflections in Robotic PIV, improving the accuracy of flow field measurements. Additional attention is required in the PIV sequence creation step to ensure an adequate level of overlap between measurement volumes. This facilitates addressing the spatial gaps introduced by the masking procedure, that have been proven to robustly be filled in by the multi-view advantage offered by Robotic PIV. ... Read more