Kirigami metasurfaces provide the opportunity for deployable surface modifications that can potentially be used for aerodynamic flow control. An exploratory experimental investigation was performed to assess the flow control capabilities of a triangular and circular kirigami device on an airfoil in an open-jet wind tunnel. 
A finite element model of the kirigami was validated against laser scans of a vinyl-cut prototype, showing strong agreement in deployment height and successfully highlighting non-uniformity effects. Using rounded rectangular cut-outs on the strip outside the kirigami geometry, both the deployment height and the deployment uniformity were increased. The device was placed between x/c = 0.55 to 0.7 on a DU96 airfoil, with a device height ranging h/δ = 0.1 - 0.3. 
PIV measurements showed that at 6–10° angle of attack, separation was aggravated, whereas at 12–14° separation was reduced, with smaller separation regions, weaker reverse flow, and significant drag reduction. At higher angles, the kirigami became submerged in the separation bubble (h/δ = 0.05) and lost effectiveness. Results indicate that the device does not act as a vortex generator. Instead, it is hypothesised that either the kirigami modifies near-surface flow through blockage, causes a reduction of shear stress on the kirigami device, or disrupts backflow structures that drive flow reversal.

This thesis addresses thermoacoustic instabilities and flashback in hydrogen combustion. A simplified model of Ansaldo Energia's GT36 reheat combustor was simulated at 20 bar using Large Eddy Simulation (LES), revealing unsteady flame dynamics driven by strong pressure oscillations. To detect flashback precursors, LES-derived time series were sampled at the combustor wall after an analysis identified suitable monitoring locations, representing a step toward practical sensor placement.

Fourteen thermodynamic, velocity, and species mass fraction signals were reduced via autoencoders with 2–4 latent variables; the three-latent representation emerged as optimal, isolating transition sharpness and mid-frequency modes. Clustering of this space with a modularity-based algorithm consistently identified precursors with maximum lead times of ~42 μs and virtually no false positives. In one case, a flashback was successfully predicted and suppressed. Robustness analyses confirmed generalization across locations and noise levels, demonstrating that wall-based latent clustering advances predictive flashback control toward real-world deployment.

Flow control plays a vital role in improving efficiency in aerospace, maritime, and energy systems by delaying transition to turbulence and suppressing instabilities such as Tollmien–Schlichting waves. This work uses the Kuramoto–Sivashinsky (KS) equation as a model to study convective instabilities in boundary layers. A finite-difference discretization yields a state-space formulation, enabling systematic application of control strategies. Linear Quadratic Regulator (LQR) and adjoint-based optimization methods are developed to minimize flow perturbations. Results show LQR effectively suppresses disturbances, while adjoint-based control scales efficiently to nonlinear cases, highlighting promising avenues for future turbulence management.

This thesis explores the magnetic stabilisation of strained hydrogen flames, focusing on mitigating thermodiffusive instabilities in lean premixed laminar hydrogen flames while preserving their inherent NOx reduction benefits. Utilising comprehensive numerical simulations of counterflow flame configurations, the study reveals that radially decreasing magnetic field gradients modify flame structure, predominantly through indirect mech-
anisms. These mechanisms involve magnetically induced alterations to the bulk flow field, which subsequently couple with differential diffusion effects to influence the distribution of reactants, affecting mixture fraction and the temperature field. Key findings demonstrate that the representative thermal thickness of the flame remains effectively unchanged, with a kernel-density (KDE) mode shift of 0.16 %. Crucially, total kinematic stretch is reduced by approximately 5-13 % near the axis, driven primarily by the curvature-induced contribution. Tangential strain exhibits a distinct two-regime response, decreasing by about 5-8 % in the preheat zone but increasing by 6-8 % in the reaction zone. Furthermore, a two-regime velocity-field redistribution is observed, where the axial component broadly weakens, while the radial velocity decreases for low progress variable (C) near the axis and strengthens for C ≳ 0.5. This research clarifies that direct magnetic effects are intrinsically weak and negligible
for practical control, with effective manipulation relying on these indirect pathways. The study also highlights that the efficacy of magnetic control diminishes with increasing strain rates and richer equivalence ratios. This work provides novel insights into the fundamental mechanisms governing magnetic control in premixed hydrogen flames, offering a framework for advancing sustainable combustion technologies.

This thesis investigates the combined effects of spoiler and split flap deflections on the Flying V aircraft under sideslip conditions. Previous studies examined these surfaces independently under symmetric conditions on an earlier design. To assess their combined performance on the updated Flying V design in sideslip, a full-span wind tunnel model was tested in the TU Delft Low Turbulence Tunnel (LTT) using a Modern Design of Experiments (MDOE) approach. Response Surface Methodology (RSM) was applied to model the aerodynamic forces and moments as functions of surface deflection and model orientation. Both the flaps and spoilers were found to underperform relative to previous results. Statistically significant interactions were identified between flap deflections and inboard/outboard spoiler panels across multiple forces and moments, and undesired lift generation was observed for low spoiler deflections. These findings highlight complex interactions between surfaces in yawed flight and inform future configuration design for the Flying V.

This research, conducted in collaboration with Isar Aerospace SE, presents a heat transfer analysis on a single-element LOx/Propane subscale rocket combustor with reactive-film cooling. The work aims to firstly reconstruct the time/space-resolved wall heat flux distribution using an inverse method, and secondly to analyze how operating parameters influence the thermal loads.
A series of hot-fire experiments were conducted in a capacitively and film-cooled copper subscale chamber, instrumented with thermocouples in multiple axial and azimuthal locations to capture the thermal response. A solver called Roq¤FITT was used to compute the heat flux profiles from temperature measurements, by coupling a transient conduction model with an iterative Jacobian-based optimization loop.
The analysis was focused on the throat region, where thermal stress peaks, and relied on visual graphs and statistical tools. Although a strong collinearity between chamber pressure and film ratio emerged, highlighting the limitation of the experimental set-up to fully isolate the contribution of all variables, meaningful trends emerged. Among them are correlations between heat loads and both chamber pressure and pressure instabilities, as well as an optimal film ratio configuration with minimized heat flux.
By computing the wall heat flux from internal temperature measurements, the study provides thermal insight into combustion-driven heat loading without the need to directly model the complex chemical and turbulent flow dynamics occurring inside the chamber. The obtained results form a valuable foundation for further design decisions on regeneratively cooled chambers and multi-element injector configurations, using the innovative LOx/Propane propellant combination.

Computational Fluid Dynamics has been widely used to model flows for engineering applications. The challenge has been to model or resolve turbulent flows accurately and there exist different approaches such as Direct Numerical Simulations (DNS), Large Eddy Simulations (LES) and Reynolds Averaged Navier Stokes (RANS). With limitations on the available computational power and time, RANS is the favorable choice to model such flows despite significant modeling errors.
With recent advancements in machine learning (ML) algorithms and availability of high-fidelity (accurate) data, the turbulence community has been actively working to improve RANS modeling by informing it from the reference data and developing corrections. This is also known as data driven turbulence modeling. There exist several algorithms that have improved RANS modeling but their computational cost and lack of physical interpretability remains an issue to gain an understanding of flow physics and the relation between different flow quantities.

The present study implements two data driven methods to provide corrections ad improve RANS modeling, focusing on Turbulent Kinetic Energy and the Reynolds Stress Tensor. However, the goal is
to also obtain symbolic models and physically interpretable corrections, while ensuring computational cost remains minimum.

A priori analysis, where only the reference data is used to develop corrections, shows significant re- ductions in the computational cost (function calls to the CFD solver) while converging to acceptable
correlations as well. The superior of the two methods is selected for a posteriori study where it is
coupled with a CFD solver. The correction is now formulated by differentiating the CFD code. This
information is given to the ML algorithm which adjusts its parameters accordingly to optimize the objective function and improve turbulence modeling. Results show that, with regularization, a generalized symbolic formula can be obtained which when tested on different geometries improves the prediction of flow quantities and turbulence modeling, as a whole.

The black-box nature of deep generative models like Variational Autoencoders (VAEs) limits their practical application in engineering design, particularly in aerospace, where interpretability is crucial for reliability and safety. This work investigates the use of Symbolic Regression (SR) to improve the interpretability of VAE latent spaces for airfoil shape optimization. Two approaches are developed: a latent analysis of a trained β-VAE and a novel SR-VAE model that integrates SR directly into VAE training.

The first approach shows that SR can approximate the decoder via analytical equations linking latent variables to geometric airfoil features. This even enabled parametric reconstruction independent of the decoder, though accuracy was limited for airfoils with extreme thickness or uncommon trailing edges.

The second approach investigates several SR integration strategies, with a per-batch method followed by retraining with fixed equations achieving the best balance between generalizability and reconstruction accuracy. The parametric equations of this final SR-VAE show an improvement over those from the latent analysis while preserving, and in some aspects even improving, the generative capability of the decoder. The latent space itself showed limited change due to the use of warm starts, suggesting that interpretability through SR is improved primarily at the output level.

The practical applicability of the decoders and equations obtained in this work are tested and compared to CST parameterization, using inverse design tests, as well as constrained and unconstrained optimization cases. The SR-VAE decoder consistently showed the highest reconstruction fidelity for inverse design, but limited use in practical optimizations due to its poor generalizability far from the training set mean. Although SR-based parameterizations show limited reconstruction fidelity in inverse design, they demonstrate competitive performance and the fastest convergence in optimization tasks.

Overall, this work demonstrates that SR can bridge the gap between black-box generative models and interpretable, equation-based design, opening new pathways for explainable AI in engineering contexts and beyond.

The Air Generation department at Airbus Operations GmbH has the aim to create optimization tools for the performance and sizing of air generation systems, such as the environmental control system. These type of tools will support the development of novel systems for next-generation aircraft, whose designs are driven by environmental and energy efficiency requirements. In support of this goal, this thesis focuses on the centrifugal compressor, one of the key components in aircraft air systems. The work concentrates on the development of a preliminary design and performance prediction tool that is well suited to the broad design space and future integration into a system analysis. The objective is to enable fast and robust design and performance analysis, with support for vaneless, vaned and variable vaned diffuser configurations. The method is designed to be broadly applicable, minimizing the reliance on empirical correlations commonly found in existing models, and accessible to users without extensive experience with turbomachinery design.

The methodology is based on mean-line analysis, with component models implemented in Python to ensure computational efficiency and user accessibility. While the impeller and volute are modeled using established formulations, the diffuser model is developed with a novel structure for the vaned configuration that solves the flow conditions in the channel passage using a strong interaction model for the viscid and inviscid zones. The geometric design of the vaned diffuser uses a throat area matching approach relative to the impeller in order to maximize performance without sacrificing operating range. To ensure the accuracy of the predicted impeller exit conditions, the mixing losses are computed using a two-zone model to estimate the wake area fraction.

The model is validated against six different compressor designs from open literature, four of which include both vaneless and vaned diffuser configurations, resulting in a total of ten validation cases. The validation is based on comparison of the experimental data from open literature and the model’s prediction of the performance maps for total to total pressure ratio and total to total efficiency. The model shows overall good agreement with the experimental data, with a consistent marginal underprediction of the total to total efficiency of about 1-3%.

The design and analysis tool (DACT-A) is sufficiently accurate for its purpose and has a broad applicability. However, the model is expected to have a margin of error and higher fidelity flow simulations are necessary if a higher degree of accuracy is required for the flow analysis. The validation shows good agreement for the computed losses and operating range of the vaned diffuser, but the question remains whether the flow characteristics accurately represent reality and further research and validation in this area is recommended. Furthermore, it was found that adequate flow analysis in the diffuser requires highly accurate prediction of impeller exit conditions, leading to the conclusion that, for vaned diffuser configurations, the impeller model must also possess sufficient fidelity to ensure good results.

Organic Rankine Cycle (ORC) turbines for combined cycle engines offer significant potential for improving fuel efficiency. However, turbine weight remains a critical design factor, as the turbogenerator can account for up to one-third of the power unit mass. The turbine design tool TurboSim, developed by Majer and Pini, is computationally expensive. This study aims to develop an accurate and computationally efficient symbolic regression surrogate model to predict ORC radial-inflow turbine efficiency and weight. While a single multi-output model was initially explored, separate models for each working fluid provided better accuracy. Four fluid categories—refrigerants (R134a), hydrocarbons (butane, cyclopentane, toluene), alcohols (ethanol), and siloxanes (MM)—were analyzed, showing that working fluid has a minor effect on efficiency but a significant impact on turbine weight. A parametric study identified key design variables, with impeller radius ratio and hub-to-tip radius ratio having the greatest impact. While efficiency predictions were generally reliable, weight predictions exhibited greater variability, with lower R² values and higher MSE for some fluids. The results highlight the potential of symbolic regression for rapid and interpretable turbine design optimization. However, further research is needed to improve weight prediction consistency.

This study explores the integration of machine learning with space mapping techniques to enhance the mapping of optimal control sequences between low- and high-fidelity flight mechanic models. Space mapping is a methodology that simplifies the control optimisation process by approximating a high-fidelity model using less computationally demanding low-fidelity models, which are then iteratively corrected to converge towards high-fidelity outputs. The main research question investigates how the integration of space mapping with sequence-to-sequence neural networks can improve control sequence mapping compared to traditional model predictive control (MPC) methods, particularly in managing the trajectory differences in non-linear flight regimes.

In the pursuit of sustainable aviation, with a sharp focus on reducing emissions through innovative designs and enhanced flight mechanics, the computational cost of high-fidelity models becomes a significant limitation. These models, crucial for capturing complex interactions in advanced aircraft designs, often require simplification to reduce computational demands. This research proposes a novel approach by combining the strengths of machine learning, particularly sequence-to-sequence neural networks like Gated Recurrent Units (GRUs) and transformers, with space mapping techniques to bridge the gap between low- and high-fidelity models effectively.

The study delves into two main machine learning architectures: GRUs and transformers. GRUs excel in managing sequences with fewer changes, maintaining stable predictions with minimal error. Transformers on the other hand are well suited at handling complex sequences with frequent changes, thanks to their ability to process entire sequences simultaneously through self-attention mechanisms. This capability makes transformers particularly suitable for dynamic scenarios where anticipating future states is crucial.

A significant contribution of this study is the implementation of the Prior Knowledge Input-Difference (PKI-D) architecture, which uses the low-fidelity model output as a baseline that the neural network corrects, providing a robust framework for the machine learning models to accurately predict trajectory adjustments. This architecture not only enhances the predictive accuracy but also optimises computational efficiency by reducing the dependency on extensive high-fidelity simulations.

Comparative analyses reveal that MPC methods typically provides superior mapping performance for trajectories requiring no anticipation, while the hybrid machine learning-space mapping approach offers improved performance comparably or better in complex scenarios requiring advanced anticipation. This study highlights the critical role of active learning in adapting the machine learning models to new data dynamically, a feature that proves essential in maintaining accuracy over prolonged operational periods.

In conclusion, this research demonstrates that integrating space mapping with machine learning can significantly enhance the mapping of control sequences in aerospace applications. It provides a starting point for future studies to explore tailor made machine learning solutions using extremely small data sets in situations where data availability is sparse. This research could further open up avenues where the advanced capabilities of machine learning can be applied to problems in aerospace engineering previously inaccessible.

The focus of this thesis is the preliminary design optimization and performance analysis of non-cavitating centrifugal turbopumps using a 1D lumped-parameter method. Centrifugal pumps are critical components in many engineering applications due to their high efficiency in converting mechanical energy into fluid pressure. This study presents the development and validation of a computationally efficient model to design and predict the flow properties along different stations of a centrifugal pump. The methodology minimizes the reliance on expensive CFD simulations, making use of multiple design parameters to control the impeller, diffuser, volute, and exit cone configurations to ensure a robust and efficient performance.

A common challenge in turbopump design is the prediction and mitigation of cavitation, which occurs when the fluid static pressure falls below its vapor pressure. This leads to the formation and subsequent collapse of vapor bubbles, which can cause severe damage to pump components. Given the complex physics involved, this work makes use of reduced-order models to predict, and avoid cavitating regimes through design adjustments.

This thesis provides a comprehensive framework for optimizing the design of centrifugal turbopumps by integrating validated loss models and a reduced-order modeling approach. The proposed design methodology is applied to develop a centrifugal pump for the TU Delft ORCHID facility, demonstrating its feasibility and effectiveness. This research contributes a practical modeling tool for centrifugal turbopump optimization and performance analysis.

Accurate fault detection in wind turbines is a key factor for improving their reliability and reducing maintenance cost. This report investigates probabilistic target variable modeling using a conditional Generative Adversarial Network (cGAN) in a Normal Behavior Modeling framework for fault detection on wind turbine drivetrains. The dataset used for the investigation is the open-source Supervisory Control and Data Acquisition (SCADA) dataset provided by Energias de Portugal (EDP). A cGAN and a traditional Recurrent Neural Network (the base model) are developed and trained to model the generator bearing temperature. The performance of the cGAN is compared to that of the base model, to highlight differences associated with the modeling approach. The cGAN’s output is a conditional distribution of the target temperature, given the operational state of the turbine. From this distribution, two methods of deriving a Health Indicator are investigated: (1) By reducing the distribution to a point estimate, and (2) by taking into account the whole distribution. From the base model, a third Health Indicator is derived.
For all three Health Indicators, anomalies are detected by performing a non-parametric hypothesis test designed to detect deviations from the empirical healthy data distribution. The findings indicate an enhanced temperature modeling performance of the cGAN on train, validation and test data. Especially the robustness appears to be significantly improved. While interpreting the outcome of applying the Health Indicators for fault detection remains challenging, the results suggest that the cGAN’s probabilistic modeling improves fault detection by reducing the number of false responses.

Hydrogen, as a carbon-free fuel, offers a promising solution to reduce CO2 emissions in pursuit of cleaner combustion technologies. However, its combustion poses challenges, including increased NOx production due to high adiabatic flame temperatures, along with risks of flashback and thermo-diffusive instabilities. One approach to addressing these challenges is Moderate or Intense Low-Oxygen Dilution (MILD) combustion, which significantly lowers NOx emissions by reducing reaction temperatures. Techniques like exhaust gas recirculation (EGR) can further support this process. Additionally, blending hydrogen with other fuels, such as ammonia and methane, improves flame stability, safety, and flame speed.

This thesis is part of the ACHIEVE project (Advancing the Combustion of Hydrogen-Ammonia Blends for Improved Emissions and Stability), funded by the European Union under the Horizon European Research and Innovation Actions in collaboration with the Clean Hydrogen Partnership. The project aims to advance Technology Readiness Level 4 technologies that reduce pollutants in unconventional hydrogen blends, such as mixtures of NH3, CH4, and H2. The combustion characteristics of these unconventional blends are studied numerically and experimentally in high-stability combustors, with a focus on MILD combustion and the effects of exhaust gas recirculation.

Aligned with these goals, the TU Delft Enclosed-Jet-in-Hot-Coflow (EJHC) burner, designed for MILD combustion, is used to experimentally investigate premixed and non-premixed methane-hydrogen and pure hydrogen flames. The research examines flame morphology, temperature, emissions profiles, and the effects of exhaust gas recirculation on NOx emissions. This serves as a foundational step for future research into ammonia-hydrogen blends. The burner features a central flame auto-ignited by a hot vitiated coflow produced by a secondary burner. Design modifications were implemented to improve coflow uniformity near the jet exit.

The research involved two experimental campaigns. The first, conducted under cold flow conditions using Particle Image Velocimetry (PIV), assessed coflow uniformity at different inlet velocities. The second campaign focused on reactive flow experiments, using thermocouples, chemiluminescence, and gas analyzers to study methane-air coflow and methane-hydrogen blends in premixed and non-premixed conditions. Hydrogen concentration was varied from 0% to 100%, with different power ratios between the central jet and coflow.
The study showed that using a perforated plate as a secondary burner enhanced coflow uniformity, with PIV results showing reduced radial variations of the vertical velocity component. Thermocouple measurements confirmed a uniform temperature profile.
Flame lift-off height and shape in both premixed and non-premixed conditions were influenced by jet velocity, laminar flame speed, and hydrogen content. Hydrogen addition reduced lift-off height. In premixed conditions, flames were compact and narrow, while in diffusion conditions, higher hydrogen content increased mixing with the coflow at greater radial distances. NO emissions remained stable in premixed cases but were affected by coflow composition in diffusion flames, with lower oxygen levels enhancing NO reburning.

This study provides a versatile system for investigating exhaust gas recirculation and hydrogen blends over a wide range of power and equivalence ratios. It lays the foundation for future studies on ammonia-hydrogen combustion.

Boundary Layer Ingestion (BLI) is a promising technology for reducing the impact of aviation on the environment. By placing a propulsor on an aircraft such that it ingests the slower moving air within the boundary layer, a decrease in power consumption can be achieved. Evaluating the performance of BLI configurations remains a challenge though. Many conventional methods for computing the thrust and drag of an aircraft cannot be used due to the interference between the airframe and the propulsor. As a result, researchers have to rely on either low fidelity methods with limited accuracy, or on expensive CFD simulations. In this work, an attempt was made to create a computationally inexpensive method that can evaluate the drag, thrust and propulsive power of an axisymmetric fuselage with a BLI propulsor. The method proposed relies on potential flow modeling and makes use of a boundary layer model and several corrections.