The objective of this thesis was to further the validation effort on the SU2 flow solver for non-ideal compressible flows. To do so, an uncertainty quantification infrastructure based on the V&V20 validation standard was developed. To determine whether a model error was present, the different sources of uncertainty were combined and then compared to the error between the SU2 simulations and the experiments. A variety of paradigmatic test cases were proposed which, together, would constitute a robust validation of the SU2 flow solver. Two paradigmatic unit test cases were analyzed; namely, a supersonic inviscid flow and a Mach 2.0 flow over a 2.5 degree wedge. The operating conditions of the experiments were as follows: an inlet total temperature and pressure of T_0=252°C and P_0=18.4 bara, and a back pressure of P_out=2.1 bara, for the first case, and T_0=180°C, P_0=3.5 bara and P_out=1.0 bara, for the second. The solver is able to predict the shock wave angle accurately. This step towards the validation was successful because the calculated total expanded uncertainty of 1.79% is higher the comparison error of 0.57%.In the second part of the thesis, the uncertainty quantification infrastructure was adapted to a test case constituted by more complicated flow features; namely, a supersonic flow through a 5-channel blade row. The computed uncertainties were on the same order of magnitude as in the supersonic inviscid flow test case. The simulations were verified with state-of-the-art CFD software for the quantities pressure and the Mach number. The positive result of the validation represents a step forward in the validation of the flow solver for non-ideal turbomachinery cases. The adaptation of the infrastructure to the cascade blade row paves the way for complex cases and for system response quantities of interest in industrial applications of CFD software, such as efficiency, to be assessed.

A wall-resolved Large Eddy Simulation (LES) of a wing-body junction is performed. The aim is to generate high-fidelity junction flow data to be used in a data-driven turbulence modelling approach, specifically to improve the accuracy of RANS-simulations in junction flows. The simulation is performed on a 61.5 million C-grid body fitted mesh in the pimpleFoam solver of OpenFOAM, with a turbulent channel flow precursor providing the unsteady inlet boundary condition. Analysis of the wall-resolved LES shows that the simulation accurately captures the complex flow phenomena in the wing-body junction flow including intermittency for the present inflow condition. Comparisons of the wall-resolved LES with a coarse-grid RANS simulation and the wall-modelled LES of Srikumar [2019] show that the wall-resolved LES in the present study is an improvement over the other two numerical methods. Most notably, an improvement in terms of the prediction of the location and magnitude of the mean spanwise vorticity and the mean turbulent kinetic energy of the horseshoe vortex systems was observed. Especially the RANS-simulation was unable to accurately capture the complex flow physics in the junction due to the limitations of RANS-methods, which are unable to accurately capture Reynolds stress anisotropy due to the Boussinesq hypothesis. An analysis of the high-fidelity junction flow data was performed to indicate regions where the Boussinesq hypothesis breaks down. The most notable region where the Boussinesq hypothesis was found to be not valid, was the region in close proximity to the wing-body junction upstream of the wing. Due to the breakdown of the Boussinesq hypothesis in the junction region, significant improvements of the accuracy of junction flow RANS-simulations can potentially be achieved by using the high-fidelity data from the present study in a data-driven turbulence modelling approach.

The mimetic spectral element method is a relatively young method in numerical solutions of partial differential equations and actions that describe physical systems. Its advantage is that it takes the geometrical structure of the problem into account which guarantees consistency of the numerical scheme and the conservation of relevant quantities. This prohibits the solution to diverge into non-physical states since the discretization mimics the physical behaviour of the problem. With relation to physics the mimetic method makes use of the stationary action principle which contains all information about the system and reduces the continuous variables to discrete ones that become ready for computation. The stationary action principle requires a formulation of the Lagrangian such that it implies Newton’s laws of motion or equivalent physical laws. However these Lagrangians are only easily formulated for monogenic systems. This work investigates the class of (discrete) non-monogenic physical systems. The goal is to present a general method to numerically mimic non-conservative systems such that energy (loss) is exact. This is done by modifying a newly introduced systematic method that transforms the non-conservative problem to a system with a doubled degrees of freedom that makes it a conservative or monogenic one. In this modification process new degrees of freedom are introduced which are allowed to have arbitrary variations at the initial and final states of the new action. The system can than be mimicked with the spectral element method where algebraic dual polynomials are used as both approximating polynomials and as test functions that serve as arbitrary variations. Furthermore a gauge transformation is introduced by the modified action such that it contains the initial and final states of the system to make the solution unique. The introduced method is generally applicable to Lagrangian mechanics of discrete systems with holonomic constraints. The method is applied to the damped harmonic oscillator (DHO) as a test case. The computational results are clearly accurate and converging. However they deviate with one order from the h-refinement spectral element theory of Gauss-Lobatto-Legendre interpolation basis polynomials. Theoretical analysis of the equations of the mimicked (single and doubled) DHO show that energy (loss) is exact. Nevertheless, the computations show an oscillation in energy that remains closely around the conserved value. This oscillation in energy in time was observed in earlier works that applied the mimetic spectral element method to conservative systems. The reason behind this behavior of energy remains a topic for further research. Furthermore, in a follow up research the method can be extended to the Lagrangian field theory and tested with simplified non-conservative fluid problems to verify it for fields.

Classical RANS (Reynolds-Averaged Navier-Stokes) turbulence models have limited accuracy in the prediction of the flow over the wing-body geometry. Therefore, this work focuses on improving the prediction accuracy of the classical k-ω SST turbulence model for the junction flow by means of the data-driven method SpaRTA. In the SpaRTA methodology higher-fidelity data, in this work LES (Large Eddy Simulation) data, is leveraged to find correction models that enhance the baseline turbulence model. These correction models aim at correcting the Reynolds stress anisotropy and the turbulent kinetic energy (k) equation. The correction models are obtained by sparsely regressing flow features to the correction fields that are found by the k-corrective frozen approach. The results showed that the correction fields determined by the frozen method are very effective. Direct propagation of these fields resulted in a solution very similar to the LES data. In contrast to the model search for the k correction, finding a model for the anisotropy correction by means of sparse regression was difficult. The latter is reflected in the CFD (Computation Fluid Dynamics) propagation of the found models on the same flow case as the training of the models. This model propagation showed an improvement in the placement of the horseshoe vortex, however, the vortex topology in the corner region was unsatisfactory. Which is linked to the poor model fit of the anisotropy correction field. Besides the search for an improved turbulence model, this work also encapsulates an evaluation of the drag reduction by the anti-fairing geometry for the wing-body junction flow. This evaluation was achieved by comparing available wall-resolved LES data sets. The evaluation revealed that the anti-fairing reduces the drag force by a propulsive pressure mechanism over the bottom wall for a specific geometry over which the drag is computed. This geometry included only 5% of the complete wing span, for larger wing spans the drag reduction was not evident.

Accurate measurement of emissions (i.e. volume flow rate multiplied by concentration) is vital to the control and reduction of air pollution. We quantify the uncertainty of flow measurements by S-type pitot tubes in narrow exhaust stacks (diameter < 0.5 m) in support of recent European Union (EU) regulations for emissions from medium-size combustion plants. The contribution of blockage and wall effects to the uncertainty budget is evaluated. The aim of this thesis is to characterize flow uncertainty and to aid the development of more accurate measurement methods for volume flow rates in narrow exhaust stacks. This thesis will provide the scientific basis for implementation of EU regulations concerning emissions from medium-size combustion plants. Characterization of flow fields in narrow exhaust stacks is key to accurate volume flow rate measurements. We use numerical simulations to study the mean velocity profile and turbulence statistics of fully developed turbulent flow. To investigate the impact of wall effects on measurement uncertainty, we simulate the flow eld around an S-type pitot tube in close proximity to the stack wall. In addition, we conduct S-type pitot tube measurements at various wall-distances and bulk velocities. Throughout this thesis, we use several uncertainty quantification techniques to evaluate the uncertainty of numerical simulations and experiments. Studying spatial variation in measurement uncertainty gives insight in the most suitable locations for S-type pitot tube measurements in narrow exhaust stacks. The results of this thesis suggest that wall effects dominate over other sources of measurement uncertainty in the near wall region. Throughout the measurement plane, the combined uncertainty of measurement error sources exceed the impact of blockage. We recommend to determine volume flow rates in narrow exhaust stacks by S-type pitot tube measurements in the stack center. A correction factor can be determined to compute the volume flow rate in a narrow exhaust stack from the maximum flow rate in the stack center.

Unsteady numerical simulation has been proven to be an essential tool for research. The quality of the results can be improved by using mesh adaptation. Mesh adaptation uses error indicators to refine the mesh in regions with high errors. The error indicators used are output errors with the most accurate output error estimation method being adjoint-based error estimation. However, for this method, the primal solution needs to be stored, which is storage intensive, especially for large unsteady simulations. The method proposed in this thesis uses a neural network autoencoder to compress and reconstruct the primal solution. This solution is compared to a reconstructed solution using Proper Orthogonal Decomposition (POD). The one-dimensional unsteady Burgers equation is used as validation for the methods using a manufactured solution while the lid-driven cavity flow is investigated using the proposed method. The manufactured solution of the one-dimensional Burgers case could be exactly reconstructed using two POD modes. For the autoencoder a small latent space was used. For low resolutions, the small latent space did not prove to be a problem as the primal and residual could be captured accurately. However, for higher resolutions, the reconstruction error of the autoencoder became dominant for the residuals and resulted in erroneous adjoint-based error estimates while the primal remained qualitatively similar. For the lid-driven cavity flow, the POD was still able to capture the solution using a low number of modes due to the smoothness of the solution. This resulted in an unfair comparison between the POD and autoencoder reconstructed solutions. The reconstructed autoencoder error estimates for lower resolutions were more accurate due to the latent space being large enough to capture the residual of the discrete primal accurately enough. When moving to higher resolutions, the autoencoder was not able to reconstruct the residual accurately enough leading to erroneous error estimates. Therefore, the latent space of autoencoders should be sufficiently large in order to gain an accurate reconstruction of the residual. If the latent space is large enough, the error estimate is accurate and the local error estimates can be used as a first iteration error indicator for mesh refinement.

Recent years have seen an increase in studies focusing on data-driven techniques to enhance modelling approaches like the two-equation turbulence models of Reynolds-averaged Navier-Stokes (RANS). Different techniques have been implemented to improve the results from these simulations. In particular, the main focus has been on overcoming the limitations implied by the Boussinesq assumption. This has been approached by using machine learning techniques as a way of discovering new formulations that could overperform when compared to traditional models. Despite promising results for steady RANS simulations, little has yet been investigated in URANS applications. In this dissertation, this lack of research will be addressed. The main ideas are then, first, to see how the available information in URANS simulations can be used to improve the anisotropic Reynolds stress tensor prediction, and second if and how this can be done by using a sparse regression technique, whose framework is known as SpaRTA. A procedure involving the triple decomposition of High-Fidelity velocity fields is applied, aiming at finding a model exclusively for the stochastic component of the anisotropy. The test case which is considered is the flow around a cylinder at Re=3900. The High-Fidelity data was collected by running Large Eddy Simulations in OpenFOAM, after which the velocity was split into its different components through Proper Orthogonal Decomposition. A priori results have shown good performance of the trained models, outperforming the Boussinesq assumption both in the prediction of turbulence componentiality and also on the values of the single anisotropy components.

Due to their powerful approximation capabilities, artificial neural networks have seen a wide interest in various fields. A particular application is the use of an artificial neural network to predict solutions of the governing equations for fluid flow i.e. the Navier-Stokes equations. This is done by taking the space and time variables as the inputs and the flow field variables (velocity and pressure) as outputs. Subsequently, the artificial neural network can be trained by using velocity field data on discrete points in a domain. This training is carried out by means of a scalar loss function, which is minimized by the backpropagation algorithm. By taking the gradients of the outputs with respect to the inputs, the Navier-Stokes can be formed and included in the same loss function. By including discrete velocity field data as well as the Navier-Stokes equations physics into the loss function, the trained artificial neural network could leverage its approximation capabilities to approximate the complex solutions of the flow variables. Only data on the velocity field is provided, such that the framework could be used to infer the pressure field, for which no data is given. In order to apply this approach to incompressible flows, the predicted velocity fields are strictly required to be divergence-free. To accomplish this, a divergence-free potential basis was proposed. This Data Physics Fluid Informed Neural Networks (DPFINN) framework was tested on several incompressible fluid flow test cases, for which analytical solutions were available. This work provides an extensive analysis for these test cases. The results of the test cases showed that the balance between the data and the physics loss function proved to be a delicate one, which depended on a number of different choices for the artificial neural network architectures, data and physics informing process as well as the training process. Overall, the flow fields were able to be inferred well, however simultaneous implementation of the data and the physics proved to be difficult. More complex artificial neural networks were shown to not always improve the performance. Thus, the framework results were showed to be largely dependent on the balance between the expressibility and the trainability of the framework.

The GMRES algorithm has been discussed to solve the linear system Ax = b for application in a flow solver. Besides the conventional GMRES algorithm, several approaches are discussed which supposedly reduce the computational costs of the method, and reduce the number of iterations needed to satisfy the convergence criteria. GMRES is one of the most popular methods for solving non-symmetric systems. The method is robust, i.e. the solution of the least squares problem always exists. GMRES does have full recurrences since the basis of the Krylov subspace needs to be stored completely. Also, orthogonalisation of new basis vectors becomes increasingly more expensive when the iterations proceed. GMRES minimizes the residual norm, as a result the residual norm decreases monotonically and the convergence is smooth.

This thesis aims to introduce mesh refinement into the Mimetic Spectral Element Method (MSEM). The concept of mimetic discretizations is to mimic the properties of continuous Partial Differential Equations (PDEs) discretely. In many discretization methods information is lost in the actual discretization step which is detrimental to the physical fidelity of the approximated solution. Mimetic methods try to prevent this, a feat achieved by combining the fields of differential geometry and algebraic topology. Where differential geometry describes the continuous problem, algebraic topology functions as its discrete equivalent. By accounting for the spatial and temporal geometric objects each physical quantity is associated with, mimetic methods preserve as much as possible of the continuous structure.

Today a lot of attention is paid to the development of environmental friendly technologies. In the aircraft industry, emissions need to be reduced as well. Fuel can be saved by ensuring that the flow over the wings of the aircraft is completely or partly laminar. A laminar boundary layer creates less skin friction drag than a turbulent boundary layer. It is however more prone to separation. Nevertheless, at the cruise speeds of transport aircraft a laminar boundary layer can be beneficial (because of the small angle of attack). At these speeds a shock wave can however be present on the airfoil at off-design conditions. This shock wave can cause a local separation of the boundary layer, if the pressure gradient is however not too strong, the flow will reattach after the shock and hence the drag of the aircraft will be reduced. During flight disturbances in the flow can set the aircraft into motion. This motion can be damped or amplified. The boundary between these types of motion is called the flutter boundary. This boundary is an important constraint of the flight envelope of aircraft. The influence of boundary layer transition on the flutter behaviour of transonic aircraft has not been investigated yet, therefore a first step has been taken in this thesis. The objective of this thesis is to: Investigate the influence of laminar to turbulent boundary layer transition on the flutter boundary and damping characteristics of a supercritical laminar airfoil (the CAST-10 airfoil) in transonic flow using numerical simulations. In order to do so numerical simulations are performed with the CAST-10 airfoil with two RANS codes: the DLR TAU code and the ANSYS CFX code. The DLR TAU code uses the eN-method for transition prediction, whereas CFX uses the γ − Re_ transition model. Steady and unsteady flow simulations have been performed with both codes. The steady flow simulations are used to initialise the unsteady flow simulations. In these unsteady flow simulations, the airfoil is forced to perform a sinusoidal pitching or plunging motion. The response of the airfoil to these applied motions serves as input to a flutter program, which uses the k-method to compute the flutter boundary. Simulations with both free and fixed transition are carried out. In case of fixed transition, transition was fixed at the leading edge of the airfoil.

In the present thesis report the author synthetizes 9 months of work at the DSNA department of ONERA in Chatillon, France. The topic of the thesis is goal oriented mesh adaptation with particular application to unstructured grids and to _nite volume methods. The motivation of the present work is the application to unstructured meshes of a novel indicator for mesh adaptation, based on the total derivative of the goal function with respect to mesh nodes coordinates, introduced by Peter et al. in [6], [7] and that has been tested until now on structured grids only. In chapter 1 a brief literature survey is presented, with the aim of introducing the theoretical background of the work and the state of the art of goal oriented mesh adaptation techniques. In chapter 2 the author gives a description of the gradient computation module of the CFD software elsA, developed by ONERA, that has been the main tool used for ow simulations and mesh adaptation and the core of the code development work.

Societies demand for sustainability is one of the main drivers for innovation in the wind energy industry. The increasing size of wind turbines implies increasingly flexible turbine blades. The flexibility of the blades induces the occurrence of fluid-structure interaction (FSI) phenomena. The simulation codes that are used for the design of wind turbines need to be validated with observational data. In the current research a FSI experiment of a wing equipped with a trailing edge flap is performed to provide validation data. Little data is available in literature of free motion FSI experiments. Therefore, a FSI experiment is performed with a freely moving wing equipped with a trailing edge flap.

It is investigated if a Multi Point Inflow Performance Relationship (MIPR) using Uncer- tainty Quantification (UQ) can lead to a step change in completion design. The MIPR model consists of several Kuchuk IPR models coupled with a wellbore network, and is verified against a discretized model. It is found that the assumption of artificial no-flow boundaries for this model is limiting, which leads to an overprediction of the distributed productivity index. A calibration using Markov Chain Monte Carlo (MCMC) is pre- formed, which incorporates both epistemic and aleatory model parameters. In addition, a more efficient ‘partitioned MCMC’ is investigated, which shows promising results. The UQ MIPR model can predict the change in productivity index due to a uniform com- pletion skin profile, and can give an indication of results for a non-uniform change in completion skin. In conjunction with a discretized model, a MIPR model can lead to a step change in completion by providing information to a production technologists earlier on in field development.

Flight testing is an activity that exists since the beginning of flight, and is still seen as an essential part of the aircraft development process. While numerical simulation has been improved significantly over the last decade, flight testing remains attractive in the exploration and evaluation of the flight envelope, especially in the region where numerical simulation lacks prediction accuracy. Free-flying model testing is a potential technique, which is recognized by some of the largest research institutes. Using such relatively cheap scale models aids in the determination of the flight behavior of full-scale aircraft in an early stage in the design process. Furthermore it takes less time to develop a scalemodel and is also safer to fly. A 5.5% scale model of the Boeing 737-700 is created by the TU Delft as a first step towards investigating the potential of dynamically scaled flight testing. This Delft University Unconventional Concept (DUUC) has a conventional tube-wing configuration and features ducted fan engines mounted aft on the fuselage, which replace the conventional empennage. A horizontal and vertical jet vane is placed in the duct, which guide the propeller slipstream and thereby control the attitude of the aircraft. Successful flights have already been performed with this model, which proves the propulsive empennage concept. However, the stability and control behavior has only been assessed at a low level. Flow characteristics entailed by the ducted fan are accounted for by means of approximate functions and empirical relations instead of looking at the physical interaction between the components within the duct. This study focusses on the processing of said interaction by performing a CFD study of which the results are used in the longitudinal equations of motion of the DUUC. This way, certain motion can be simulated from which the effect of the unconventional empennage on the stability and control can be assessed. In order to study the aerodynamics of the propulsive empennage and subsequently its stability, a CADmodel is developed using PARAPY. This is a knowledge based engineering platform written in the Python language and allows for parametrization of the model such that geometric alterations can easily be implemented. Previous studies within the TU Delft have lead to a multi-model generator written in PARAPY. The fuselage and wing generation modules that were developed can readily be applied in the generation of the DUUC model. The propulsive empennage module is created by the author of this thesis, consisting of several sub-parts: (1) outer duct that forms a shielding around the propeller, (2) the pylon that connects the duct to the fuselage, (3) a center body to which the (4) propeller and (5) control vanes are attached. The full aircraft model has been used in a series of CFD simulations with ANSYS FLUENT. Inviscid steady calculations are performed using a pressure-based solver with a SIMPLE pressure-velocity coupling scheme. The propeller is modeled as an actuator disk whereby the thrust is defined by specifying a pressure jump across the thin surface. The thrust produced by the ducted fan is based on the turboprop installed on the ATR72-600 aircraft, which has similar dimensions compared to the Boeing 737-700. Both the full-scaleDUUC as well as the 5.5% scale model are subjected to this aerodynamic analysis. Minor discrepancies were found between the two models, which is a result of the artificial viscosity that is implemented in the Euler scheme to smoothen strong gradients in the solution (numerical dissipation). Due to the location of the engine, the thrust line is located above the center of gravity which contributes to a stabilizing effect. The effect of several parameters on the stability and control has been analysed, such as the center of gravity location, mass and inertia, and geometric scaling factor. The slight difference in aerodynamic performance between the full-scale and subscalemodels, means that the trimmed solution for steady horizontal flight is also slightly different with a maximum error of 2.5 degrees in elevator deflection. More drag is obtained as the aircraft scale goes down, resulting in a higher required thrust to achieve force balance in a trimmed flight condition. Furthermore it is found that by scaling the aircraft down according to Froude scaling theory, the response becomes much faster. This response can in turn be scaled back, which proved to be a very useful estimate of the full-scale simulation.

The most used RANS-model in relation to wind farms, k−epsilon, has significant shortcomings. It over-predicts the eddy viscosity in the near-wake and fails to model the anisotropy of the turbulence quantities. The Sparse Regression of the Turbulence Stress Anisotropy (SpaRTA) method could remedy these shortcomings. This method uses temporally averaged fields from LES data together with sparse regression methods to learn corrections for the anisotropy tensor and turbulence production terms. This thesis extends the SpaRTA method to wind farms in stable atmospheric boundary layer conditions through an additive correction term for the turbulent heat flux. This extended SpaRTA method is applied to a stably stratified parcel of air with and without a turbine. The simulations with implemented correction models show significant improvement over the baseline simulations. However, it is also concluded that the effects of the additional turbulent heat flux correction could be incorporated in the turbulence production correction.

In studies of wind plant designs, wake dynamics are of great interests as wakes affect downstream turbine loading that impacts wind plant efficiency. Recent developments of Tensor Basis Decision Tree (TBDT) based machine learning (ML) models in reconstructing the turbulence anisotropy fields of simple flow cases prompt the motivation in applying such models to the more complex wind plant simulations. A significantly more efficient TBDT framework has been developed to tackle large scale flow domains. By training on the Large Eddy Simulation data of a one-turbine case, the ML models can reconstruct the free shear layers in wind plants of various turbine layouts and atmospheric conditions while predicting the correct shape and orientation of turbine wakes. Subsequently, a data-driven augmentation to Reynolds-averaged Navier-Stokes simulations of wind plants has been employed that results in more accurate turbulent fields and turbine outputs, which demonstrates the potential of efficient wind plant simulations of tomorrow.

A wall resolved LES simulation of the Anti-Fairing wing/body junction introduced by Belligoli et al. [6] to reduce interference drag is performed. The LES mesh is composed of 61.7 million cells with a C-fitted grid around the wing . The simulation is performed using the pimple solver of Open- Foam 4 with a time and space varying inlet boundary condition obtained thanks to a precursor. This simulation will be used to assess the impact of the Anti-Fairing by comparing the result to the wall resolved baseline case of Alberts [2] and to serve as a training data for data driven techniques applied to junctions flows. Using the wall resolved LES we apply the data driven algorithm method Sparse Regression of Turbulent Stress Anisotropy (SpaRTA) developed by Schmelzer et al. [38] in the case of junction flows. It is shown that the first step of the method, namely the k-corrective frozen RANS, is able to produce corrective fields to the Reynolds tensor and the turbulent kinetic energy equations in this case. The corrective fields once added in a k-omega SST simulation make it possible to obtain the exact location, strength and shape of the main horseshoe vortex. The upstream boundary layer is also subject to corrections indicating RANS-LES mismatch in the inflow. Mutual Information (MI) is calculated to identify the relevant tensors, physical features and invariants that correlate with the junction flow data. Finally, algebraic models for the corrective fields are obtained. They are compared to the true values of these fields. It is possible to see that the performance of SpARTA models is good upstream of the wing. However, models found and tested in the vicinity of the wing, where the separation and horseshoe vortex are located, are not fully able to capture the relevant corrections. Additional constraints or steps to the ones performed in the time of this study may be necessary in order to use SpaRTA to generate models giving improved predictions compared to classic RANS turbulence models.

Site analysis to determine the loads experienced by wind turbines based on site-specific environmental conditions is typically done using either coupled aero-servo-elastic simulations for onshore wind turbines or coupled aero-servo-hydroelastic simulations in the case of offshore wind turbines. These simulations become computationally expensive when multiple load cases are needed to be taken into account, together with the numerous possible combinations of turbulence inflow patterns that result in the same mean inflow conditions. Probabilistic surrogate models offer a cheap alternative to these expensive simulations for predicting load statistics. This thesis explores a type of neural network called Conditional Generative Adversarial Networks (CGANs) as a potential candidate for such a surrogate model. Originally developed for image generation, CGANs have seen success in other applications. However, most applications of GANs to date are high-dimensional, with relatively low research focused on low-dimensional problems such as wind turbine load statistics. Multiple experiments are conducted using various multimodal and heteroscedastic datasets to assess its ability to model such characteristics accurately. The conditional log-likelihood andWasserstein-1 distances were used as metrics. The results show that CGANs can indeed model such low-dimensional datasets. Finally, the CGANs are trained on data from simulations of onshore and offshore wind turbines in OpenFAST and compared with predictions from Mixture Density Networks (MDNs). The results from CGANs are comparable to MDNs, showing its potential as another alternative surrogate method, although more research needs to be performed.