The pulmonary route presents an attractive delivery pathway for topical treatment of lung diseases. While significant progress has been achieved in understanding the physical underpinnings of aerosol deposition in the lungs, our ability to target or confine the deposition of inhalation aerosols to specific lung regions remains meagre. Here, we present a novel inhalation proof-of-concept for regional targeting in the upper airways, quantitatively supported by computational fluid dynamics (CFD) simulations of inhaled micron-sized particles (i.e. 1-10 μm) using an intubated, anatomically-realistic, multi-generation airway tree model. Our targeting strategy relies on selecting the particle release time, whereby a short-pulsed bolus of aerosols is injected into the airways and the inhaled volume of clean air behind the bolus is tracked to reach a desired inhalation depth (i.e. airway generations). A breath hold maneuver then follows to facilitate deposition, via sedimentation, before exhalation resumes and remaining airborne particles are expelled. Our numerical findings showcase how particles in the range 5-10 μm combined with such inhalation methodology are best suited to deposit in the upper airways, with deposition fractions between 0.68 and unity. In contrast, smaller (< 2 μm) particles are less than optimal due to their slow sedimentation rates. We illustrate further how modulating the volume inhaled behind the pulsed bolus, prior to breath hold, may be leveraged to vary the targeted airway sites. We discuss the feasibility of the proposed inhalation framework and how it may help pave the way for specialized topical lung treatments.

Particulate flow in closed space is involved in many engineering applications. In this paper, the prediction of particle removal is investigated in a thermally driven 3D cavity at turbulent Rayleigh number Ra = 10 using Coarse Large Eddy Simulation (CLES). The depletion dynamics of SiO aerosol with aerodynamic diameters between 1.4 and 14 µm is reported in an Euler/Lagrange framework. The main focus of this work is therefore to assess the effect of the subgrid-scale motions on the prediction of the particulate flow in a buoyancy driven 3D cavity flow when the mesh resolution is coarse and below optimal LES standards. The research is motivated by the feasibility of modeling more complex particulate flows with reduced CPU cost. The cubical cavity of 0.7 m side-length is set to have a temperature difference of 39 K between the two facing cold and hot vertical walls. As a first step, the carrier fluid flow was validated by comparing the first and second-moment statistics against both previous well-resolved LES and experimental databases [Kalilainen (J. Aero Sci. 100:73-87, 2016); Dehbi (J. Aero. Sci. 103:67-82, 2017)]. First moment Eulerian statistics show a very good match with the reference data both qualitatively and quantitatively, whereas higher moments show underprediction due to the lesser spatial resolution. In a second step, six particle swarms spanning a wide range of particle Stokes numbers were computed to predict particle depletion. In particular, predictions of 1.4 and 3.5 µm particles were compared to LES and available experimental data. Particles of low inertia i.e. dp < 3.5 µm are more affected by the SGS effects, while bigger ones i.e. dp = 3.5-14 µm exhibit much less grid-dependency. Lagrangian statistics reported in both qualitative and quantitative fashions show globally a very good agreement with reference LES and experimental databases at a fraction of the CPU power needed for optimal LES.

Ignition and combustion behavior in the second stage of a sequential combustor are investigated numerically at atmospheric pressure for pure fueling and for two - fuel blends in 24:1 and 49:1 mass ratios , respectively, using Large Eddy Simulation (LES). Pure fueling results in a turbulent propagating flame anchored by the hot gas recirculation zones developed near the inlet of the sequential combustion chamber. As the content increases, the combustion process changes drastically, with multiple auto-ignition kernels produced upstream of the main flame brush. Analysis of the explosive modes indicates that, for the highest amount investigated, flame stabilization in the combustion chamber is strongly supported by auto-ignition chemistry. The analysis of fuel decomposition pathways highlights that radicals advected from the first stage flame, in particular OH, induce a rapid fuel decomposition and cause the reactivity enhancement that leads to auto-ignition upstream of the sequential flame. This behavior is promoted by the relatively large mass fraction of OH radicals found in the flow reaching the second stage, which is approximately one order of magnitude greater than it would be at chemical equilibrium. The importance of the out-of-equilibrium vitiated air on the ignition behavior is proven via an additional LES that features weak auto-ignition kernel formation when equilibrium is artificially imposed. It is therefore concluded that parameters affecting the relaxation towards chemical equilibrium of the vitiated flow can have an important influence on the operability of sequential combustors fueled with varying fractions of blending.

Three secondary flows, namely the inward radial flow along the cyclone lid, the downward axial flow along the external surface of the vortex finder, and the radial inward flow below the vortex finder (lip flow) have been studied at a wide range of flow rate 0.22-7.54 LPM using the LES simulations. To evaluate these flows the corresponding methods were originally proposed. The highly significant effect of the Reynolds number on these secondary flows has been described by equations. The main finding is that the magnitude of all secondary flows decrease with increasing Reynolds number. The secondary inward radial flow along the cyclone lid is not constant and reaches its maximum value at the central radial position between the vortex finder external wall and the cyclone wall. The secondary downward axial flow along the external surface of the vortex finder significantly increases at the lowest part of the vortex finder and it is much larger than the secondary flow along the cyclone lid. The lip flow is much larger than the secondary inward radial flow along the cyclone lid, which was assumed in cyclone models to be equal to the lip flow, and the ratio of these two secondary flows is practically independent of the Reynolds number.

A three-dimensional compressible Direct Numerical Simulation (DNS) analysis has been carried out for head-on quenching of a statistically planar stoichiometric methane-air flame by an isothermal inert wall. A multi-step chemical mechanism for methane-air combustion is used for the purpose of detailed chemistry DNS. For head-on quenching of stoichiometric methane-air flames, the mass fractions of major reactant species such as methane and oxygen tend to vanish at the wall during flame quenching. The absence of at the wall gives rise to accumulation of carbon monoxide during flame quenching because cannot be oxidised anymore. Furthermore, it has been found that low-temperature reactions give rise to accumulation of and at the wall during flame quenching. Moreover, these low temperature reactions are responsible for non-zero heat release rate at the wall during flame-wall interaction. In order to perform an in-depth comparison between simple and detailed chemistry DNS results, a corresponding simulation has been carried out for the same turbulence parameters for a representative single-step Arrhenius type irreversible chemical mechanism. In the corresponding simple chemistry simulation, heat release rate vanishes once the flame reaches a threshold distance from the wall. The distributions of reaction progress variable and non-dimensional temperature are found to be identical to each other away from the wall for the simple chemistry simulation but this equality does not hold during head-on quenching. The inequality between (defined based on mass fraction) and holds both away from and close to the wall for the detailed chemistry simulation but it becomes particularly prominent in the near-wall region. The temporal evolutions of wall heat flux and wall Peclet number (i.e. normalised wall-normal distance of isosurface) for both simple and detailed chemistry laminar and turbulent cases have been found to be qualitatively similar. However, small differences have been observed in the numerical values of the maximum normalised wall heat flux magnitude and the minimum Peclet number obtained from simple and detailed chemistry based laminar head-on quenching calculations. Detailed explanations have been provided for the observed differences in behaviours of and . The usual Flame Surface Density (FSD) and scalar dissipation rate (SDR) based reaction rate closures do not adequately predict the mean reaction rate of reaction progress variable in the near-wall region for both simple and detailed chemistry simulations. It has been found that recently proposed FSD and SDR based reaction rate closures based on DNS analysis of simple chemistry data perform satisfactorily also for the detailed chemistry case both away from and close to the wall without any adjustment to the model parameters.

The in wall-bounded turbulent flows is a complex turbulence regeneration mechanism that remains not fully understood. This study explores the potential of deep reinforcement learning (DRL) for managing the wall regeneration cycle to achieve desired flow dynamics. To create a robust framework for DRL-based flow control, we have integrated the DRL libraries with the open-source direct numerical simulation (DNS) solver . The DRL agent interacts with the DNS environment, learning policies that modify wall boundary conditions to optimise objectives such as the reduction of the skin-friction coefficient or the enhancement of certain coherent structures' features. The implementation makes use of the message-passing-interface (MPI) wrappers for efficient communication between the Python-based DRL agent and the DNS solver, ensuring scalability on high-performance computing architectures. Initial experiments demonstrate the capability of DRL to achieve drag reduction rates comparable with those achieved via traditional methods, although limited to short time intervals. We also propose a strategy to enhance the coherence of velocity streaks, assuming that maintaining straight streaks can inhibit instability and further reduce skin-friction. Our results highlight the promise of DRL in flow-control applications and underscore the need for more advanced control laws and objective functions. Future work will focus on optimising actuation intervals and exploring new computational architectures to extend the applicability and the efficiency of DRL in turbulent flow management.

The statistical behaviours of wall heat flux and wall shear stress and their interdependence during unsteady head-on quenching of statistically planar turbulent premixed flames within turbulent boundary layers due to heat loss through the cold wall have been analysed using three-dimensional Direct Numerical Simulation data with friction Reynolds numbers of and 180. In both cases, the mean wall shear stress decreases during flame-wall interaction, whereas the mean wall heat flux magnitude increases with time as the flame approaches the wall and eventually assumes a maximum value before decreasing with the progress of flame quenching. The integral length scales of wall heat flux in both streamwise and spanwise directions have been found to grow with time after the maximum mean heat flux magnitude is obtained for the two cases considered. However, the integral length scale of wall shear stress in the streamwise direction grows but the integral length scale of wall shear stress in the spanwise direction decreases with time after the maximum mean heat flux magnitude is reached. Moreover, the correlation coefficient between the wall heat flux magnitude and wall shear stress becomes increasingly negative while the mean wall heat flux increases with time, but this negative correlation weakens with the progress of flame quenching. The first few (i.e., most energetic) Proper Orthogonal Decomposition (POD) modes of wall shear stress and the wall heat flux magnitude have been found to capture the qualitative nature of the correlation between these quantities and their spatial variations. It is found that tens of most energetic POD modes are needed to capture the mean and variances of wall heat flux and wall shear stress. The number of most energetic modes, which contribute significantly to the statistics of both wall heat flux and wall shear stress, decreases with decreasing and also with the progress of flame quenching due to the weakening of turbulence effects.

The spatial resolution requirements of the Stochastic Fields probability density function approach are investigated in the context of turbulent premixed combustion simulation. The Stochastic Fields approach is an attractive way to implement a transported Probability Density Function modelling framework into Large Eddy Simulations of turbulent combustion. In premixed combustion LES, the numerical grid should resolve flame-like structures that arise from solution of the Stochastic Fields equation. Through analysis of Stochastic Fields simulations of a freely-propagating planar turbulent premixed flame, it is shown that the flame-like structures in the Stochastic Fields simulations can be orders of magnitude narrower than the LES filter length scale. The under-resolution is worst for low Karlovitz number combustion, where the thickness of the Stochastic Fields flame structures is on the order of the laminar flame thickness. The effect of resolution on LES predictions is then assessed by performing LES of a laboratory Bunsen flame and comparing the effect of refining the grid spacing and filter length scale independently. The usual practice of setting the LES filter length scale equal to grid spacing leads to severe under-resolution and numerical thickening of the flame, and to substantial error in the turbulent flame speed. The numerical resolution required for accurate solution of the Stochastic Fields equations is prohibitive for many practical applications involving high-pressure premixed combustion. This motivates development of a Thickened Stochastic Fields approach (Picciani et al. Flow Turbul. Combust. , YYY (2018) in order to ensure the numerical accuracy of Stochastic Fields simulations.

A scaling relation has been derived to link the fractal dimension of a flame surface with the ratio of the normalised 3D flame surface area to its 2D counterpart. This derivation assumes an isotropic distribution of angles between the measurement plane and the flame's normal vector, as well as a uniform distribution of angles between the principal direction and the flame's tangent vector. The validity of the newly derived relation was assessed using an existing Direct Numerical Simulation (DNS) database of statistically planar turbulent premixed flames, encompassing a range of different Karlovitz numbers. The DNS data-based assessment revealed that the newly derived relations are reasonably accurate for the thin reaction zones regime flames, with the precision of predictions based on isotropy improving, as the Karlovitz number increases. Moreover, 2D measurements of the flame surface fractal dimension and the flame wrinkling factor can be effectively used to predict the actual 3D flame wrinkling factor for flames with Karlovitz numbers much greater than unity. Alternatively, the ratio of the 3D wrinkling factor to its 2D counterpart can provide a reasonable estimate of the 3D fractal dimension for flames in the thin reaction zones regime. The newly derived relations provide an estimation for the value of fractal dimension in the limit of high Karlovitz number using an alternative route.

This study presents novel drag reduction active-flow-control (AFC) strategies for a three-dimensional cylinder immersed in a flow at a Reynolds number based on freestream velocity and cylinder diameter of . The cylinder in this subcritical flow regime has been extensively studied in the literature and is considered a classic case of turbulent flow arising from a bluff body. The strategies presented are explored through the use of deep reinforcement learning. The cylinder is equipped with 10 independent zero-net-mass-flux jet pairs, distributed on the top and bottom surfaces, which define the AFC setup. The method is based on the coupling between a computational-fluid-dynamics solver and a multi-agent reinforcement-learning (MARL) framework using the proximal-policy-optimization algorithm. This work introduces a multi-stage training approach to expand the exploration space and enhance drag reduction stabilization. By accelerating training through the exploitation of local invariants with MARL, a drag reduction of approximately is achieved. The cooperative closed-loop strategy developed by the agents is sophisticated, as it utilizes a wide bandwidth of mass-flow-rate frequencies, which classical control methods are unable to match. Notably, the mass cost efficiency is demonstrated to be two orders of magnitude lower than that of classical control methods reported in the literature. These developments represent a significant advancement in active flow control in turbulent regimes, critical for industrial applications.

Deep reinforcement learning (DRL) has found application in numerous use-cases pertaining to flow control. Multi-agent RL (MARL), a variant of DRL, has shown to be more effective than single-agent RL in controlling flows exhibiting locality and translational invariance. We present, for the first time, an implementation of MARL-based control of three-dimensional Rayleigh-Bénard convection (RBC). Control is executed by modifying the temperature distribution along the bottom wall divided into multiple control segments, each of which acts as an independent agent. Two regimes of RBC are considered at Rayleigh numbers and 750. Evaluation of the learned control policy reveals a reduction in convection intensity by and at and 750, respectively. The MARL controller converts irregularly shaped convective patterns to regular straight rolls with lower convection that resemble flow in a relatively more stable regime. We draw comparisons with proportional control at both and show that MARL is able to outperform the proportional controller. The learned control strategy is complex, featuring different non-linear segment-wise actuator delays and actuation magnitudes. We also perform successful evaluations on a larger domain than used for training, demonstrating that the invariant property of MARL allows direct transfer of the learnt policy.

Control of chaotic systems has far-reaching implications in engineering, including fluid-based energy and transport systems, among many other fields. In real-world applications, control algorithms typically operate only with partial information about the system () due to limited sensing, which leads to sub-optimal performance when compared to the case where a controller has access to the full system state (). While it is well-known that the effect of partial observability can be mediated by introducing a memory component, which allows the controller to keep track of the system's partial state history, the effect of the type of memory on performance in chaotic regimes is poorly understood. In this study we investigate the use of reinforcement learning for controlling chaotic flows using only partial observations. We use the chaotic Kuramoto-Sivashinsky equation with a forcing term as a model system. In contrast to previous studies, we consider the flow in a variety of dynamic regimes, ranging from mildly to strongly chaotic. We evaluate the loss of performance as the number of sensors available to the controller decreases. We then compare two different frameworks to incorporate memory into the controller, one based on recurrent neural networks and another novel mechanism based on transformers. We demonstrate that the attention-based framework robustly outperforms the alternatives in a range of dynamic regimes. In particular, our method yields improved control in highly chaotic environments, suggesting that attention-based mechanisms may be better suited to the control of chaotic systems.

The extrapolation performance of Convolutional Neural Network (CNN)-based models for Large-Eddy Simulations (LES) has been investigated in the context of turbulent premixed combustion. The study utilises a series of Direct Numerical Simulation (DNS) datasets of turbulent premixed methane/air and hydrogen/air jet flames to train the CNN models. The methane/air flames, which are characterised by increasing Reynolds numbers, are used to model the subgrid-scale flame wrinkling. The hydrogen/air flame, exhibiting complex thermodiffusive instability, is employed to test the ability of the CNN-based combustion models to predict the filtered progress variable source term. This study focuses on the influence of varying training Reynolds numbers, filter sizes, and filter kernels to evaluate the performance of the CNN models to out-of-sample conditions, i.e., not seen during training. The objectives of this study are threefold: (i) analyse the performance of CNN models at different Reynolds numbers compared to the one trained with; (ii) analyse the performance of CNN models at different filter sizes compared to the one trained with; (iii) assess the influence of using different filter kernels (i.e., Gaussian and box filter kernels) between training and testing, to emulate applications. The results demonstrate that the CNN models show good extrapolation performance when the training Reynolds number is sufficiently high. Vice versa, when CNN models are trained on low-Reynolds-number flame data, their performance degrades as they are applied to flames with progressively higher Reynolds numbers. When these CNN models are tested on datasets with filter sizes not included in the training process, they exhibit sufficient interpolation capabilities, the extrapolation performance is less precise but still satisfactory overall. This indicates that CNN models can be effectively trained using data filtered with a limited range of filter sizes and then successfully applied across a broader spectrum of filter sizes. Furthermore, when CNNs trained on box-filtered data are applied to Gaussian-filtered data, or vice versa, the models perform well for smaller filter sizes. However, as the filter size increases, the accuracy of the predictions diminishes. Interestingly, increasing the quantity of training data does not significantly enhance model performance. Yet, when training data are distributed with greater weighting towards larger filter sizes, the model's overall performance improves. This suggests that the strategic selection and weighting of training data can lead to more robust generalization across different filter conditions.

We present large-eddy simulation (LES) of a high-pressure gas jet that is injecting into a quiescent inert environment. The injection is through a nozzle with a diameter of 1.35 mm. Four injection strategies are considered in which the results of a single continuous injection case are compared with those of double injection cases with different injection splitting timing. In all double injection cases, the injection pulsing interval is kept the same, and the total injected mass is equal to that of the single injection case. On the other hand, the splitting timing is varied to investigate the effects of various injection splitting strategies on the mixture formation and the penetration length of the jet. Results show that the jet penetration length is not so sensitive to the splitting timing whereas the mixing quality can significantly change as a result of shifting the onset of injection splitting toward the end of injection. Especially, it is found that by adopting a post-injection strategy where a single injection splits into the main injection and late small injection near the end of injection period the mixing between the injected gas and ambient air is significantly improved. This trend is not as obvious when the injection splitting timing shifts toward the beginning or even in the middle of injection period. The increase of entrainment in the tail of each injection is one of the underlying physics in the mixing improvement in double injection cases. In addition to that, splitting a single injection into two smaller injections increases the surrounding area of the jet and also stretches it along the axial direction. It can potentially increase the mixing of injected gas with the ambient air.

Direct numerical simulations (DNS) have been utilised to investigate the impact of different thermal wall boundary conditions on premixed V-flames interacting with walls in a turbulent channel flow configuration. Two boundary conditions are considered: isothermal walls, where the wall temperature is set either equal to the unburned mixture temperature or an elevated temperature, and adiabatic walls. An increase in wall temperature has been found to decrease the minimum flame quenching distance and increase the maximum wall heat flux magnitude. The analysis reveals notable differences in mean behaviours of the progress variable and non-dimensional temperature in response to thermal boundary conditions. At the upstream of the flame-wall interaction location, higher mean friction velocity values are observed for the case with elevated wall temperature compared to the other cases. However, during flame-wall interaction, friction velocity values decrease for isothermal walls but initially rise before decreasing for adiabatic walls, persisting at levels surpassing isothermal conditions. For all thermal wall boundary conditions, the mean scalar dissipation rates of the progress variable and non-dimensional temperature exhibit a decreasing trend towards the wall. Notably, in the case of isothermal wall boundary condition, a higher scalar dissipation rate for the non-dimensional temperature is observed in comparison to the scalar dissipation rate for the progress variable. Thermal boundary condition also has a significant impact on Reynolds stress components, turbulent kinetic energy, and dissipation rates, showing the highest magnitudes with isothermal case with elevated wall temperature and the lowest magnitude for the isothermal wall with unburned gas temperature. The findings of the current analysis suggest that thermal boundary conditions can potentially significantly affect trubulence closures in the context of Reynolds averaged Navier-Stokes simulations of premixed flame-wall interaction.

The Stochastic Fields approach is an effective way to implement transported Probability Density Function modelling into Large Eddy Simulation of turbulent combustion. In premixed turbulent combustion however, thin flame-like structures arise in the solution of the Stochastic Fields equations that require grid spacing much finer than the filter scale used for the Large Eddy Simulation. The conventional approach of using grid spacing equal to the filter scale yields substantial numerical error, whereas using grid spacing much finer than the filter length scale is computationally-unaffordable for most industrially-relevant combustion systems. A Thickened Stochastic Fields approach is developed in this study in order to provide physically-accurate and numerically-converged solutions of the Stochastic Fields equations with reduced compute time. The Thickened Stochastic Fields formulation bridges between the conventional Stochastic Fields and conventional Thickened-Flame approaches depending on the numerical grid spacing utilised. One-dimensional Stochastic Fields simulations of freely-propagating turbulent premixed flames are used in order to obtain criteria for the thickening factor required, as a function of relevant physical and numerical parameters, and to obtain a model for an efficiency function that accounts for the loss of resolved flame surface area caused by applying the thickening transformation to the Stochastic Fields equations. The Thickened Stochastic Fields formulation is tested by performing LES of a laboratory premixed Bunsen flame. The results demonstrate that the Thickened Stochastic Fields method produces accurate predictions even when using a grid spacing equal to the filter scale. The present development therefore facilitates the accurate application of the Stochastic Fields approach to industrially-relevant combustion systems.

Improving mixing between two coaxial swirled jets is a subject of interest for the development of next generations of fuel injectors. This is particularly crucial for hydrogen injectors, where the separate introduction of fuel and oxidizer is preferred to mitigate the risk of flashback. Raman scattering is used to measure the mean compositions and to examine how mixing between fuel and air streams evolves along the axial direction in the near-field of the injector outlet. The parameters kept constant include the swirl level in the annular channel, the injector dimensions, and the composition of the oxidizer stream, which is air. Experiments are carried out in cold flow conditions for different compositions of the central stream, including hydrogen and methane but also helium and argon. Three dimensionless mixing parameters are identified, the velocity ratio between the external stream and internal stream, the density ratio between the two fluids, and the inner swirl level in the central channel. Adding swirl to the central jet significantly enhances mixing between the two streams very close to the injector outlet. Mixing also increases with higher velocity ratios , independently of the inner swirl. Additionally, higher density ratios enhance mixing between the two streams only in the case without swirl conferred to the central flow. A model is proposed for coaxial swirled jets, yielding a dimensionless mixing progress parameter that only depends on the velocity ratio and geometrical features of the swirling flow that can be determined by examining the structure of the velocity field. Comparing the model with experiments, it is shown to perform effectively across the entire range of velocity ratios , density ratios , and inner swirl levels . This law may be used to facilitate the design of coaxial swirled injectors.

The structure of autoignition in a mixing layer between fully-burnt or partially-burnt combustion products from a methane-air flame at = 0.85 and a methane-air mixture of a leaner equivalence ratio has been studied with transient diffusion flamelet calculations. This configuration is relevant to scavenged pre-chamber natural-gas engines, where the turbulent jet ejected from the pre-chamber may be quenched or may be composed of fully-burnt products. The degree of reaction in the jet fluid is described by a progress variable ( = taking values 0.5, 0.8, and 1.0) and the mixing by a mixture fraction ( = 1 in the jet fluid and 0 in the CH-air mixture to be ignited). At high scalar dissipation rates, , ignition does not occur and a chemically-frozen steady-state condition emerges at long times. At scalar dissipation rates below a critical value, ignition occurs at a time that increases with . The flame reaches the = 0 boundary at a finite time that decreases with . The results help identify overall timescales of the jet-ignition problem and suggest a methodology by which estimates of ignition times in real engines may be made.

Spatiotemporal wall temperature (T) distributions resulting from flame-wall interactions of lean H-air and CH-air flames are measured using phosphor thermometry. Such measurements are important to understand transient heat transfer and wall heat flux associated with various flame features. This is particularly true for hydrogen, which can exhibit a range of unique flame features associated with combustion instabilities. Experiments are performed within a two-wall passage, in an optically accessible chamber. The phosphor ScVO:Bi is used to measure T in a 22 × 22 mm region with 180 µm/pixel resolution and repetition rate of 1 kHz. Chemiluminescence imaging is combined with phosphor thermometry to correlate the spatiotemporal dynamics of the flame with the heat signatures imposed on the wall. Measurements are performed for lean H-air flames with equivalence ratio Φ = 0.56 and compared to CH-air flames with Φ = 1. T signatures for H-air Φ = 0.56 exhibit alternating high and low-temperature vertical streaks associated with finger-like flame structures, while CH-air flames exhibit larger scale wrinkling with identifiable crest/cusp regions that exhibit higher/lower wall temperatures, respectively. The underlying differences in flame morphology and T distributions observed between the CH-air and lean H-air mixtures are attributed to the differences in their Lewis number (CH-air Φ = 1: Le = 0.94; H-air Φ = 0.56: Le = 0.39). Findings are presented at two different passage spacings to study the increased wall heat loss with larger surface-area-to-volume ratios. Additional experiments are conducted for H-air mixtures with Φ = 0.45, where flame propagation was slower and was more suitable to resolve the wall heat signatures associated with thermodiffusive instabilities. These unstable flame features impose similar wall heat fluxes as flames with 2-3 times greater flame power. In this study, these flame instabilities occur within a small space/time domain, but demonstrate the capability to impose appreciable heat fluxes on surfaces.

There is significant interest in the gasoline direct-injection engine due to its potential for improvements in fuel consumption but it still remains an area of active research due to a number of challenges including the effect of cycle-by-cycle variations. The current paper presents the use of a 3D-CFD model using both the RANS and LES turbulence modelling approaches, and a Lagrangian DDM to model an early fuel injection event, to evaluate the regimes of combustion in a gasoline direct-injection engine. The velocity fluctuations were investigated as an average value across the cylinder and in the region between the spark plug electrodes. The velocity fluctuations near the spark plug electrodes were seen to be of lower magnitude than the globally averaged fluctuations but exhibited higher levels of cyclic variation due to the influence of the spark plug electrode and the pent-roof geometry on the in-cylinder flow field. Differences in the predicted flame structure due to differences in the predicted velocity fluctuations between RANS and LES modelling approaches were seen as a consequence of the inherently higher dissipation levels present in the RANS methodology. The increased cyclic variation in velocity fluctuations near the spark plug electrodes in the LES predictions suggested significant variation in the relative strength of the in-cylinder turbulence and that may subsequently result in a thickening of the propagating flame front from cycle-to-cycle in this region. Throughout this paper, the numerical results were validated against published experimental data of the same engine geometry under investigation.

Improving the power output from wind farms is vital in transitioning to renewable electricity generation. However, in wind farms, wind turbines often operate in the wake of other turbines, leading to a reduction in the wind speed and the resulting power output whilst also increasing fatigue. By using wake steering strategies to control the wake behind each turbine, the total wind farm power output can be increased. To find optimal yaw configurations, typically analytical wake models have been utilised to model the interactions between the wind turbines through the flow field. In this work we show that, for full wind farms, higher-fidelity computational fluid dynamics simulations, in the form of large eddy simulations, are able to find more optimal yaw configurations than analytical wake models. This is because they capture and exploit more of the physics involved in the interactions between the multiple turbine wakes and the atmospheric boundary layer. As large eddy simulations are much more expensive to run than analytical wake models, a multi-fidelity Bayesian optimisation framework is introduced. This implements a multi-fidelity surrogate model, that is able to capture the non-linear relationship between the analytical wake models and the large eddy simulations, and a multi-fidelity acquisition function to determine the configuration and fidelity of each optimisation iteration. This allows for fewer configurations to be evaluated with the more expensive large eddy simulations than a single-fidelity optimisation, whilst producing comparable optimisation results. The same total wind farm power improvements can then be found for a reduced computational cost.