• Energy Research

Abstract Large-eddy simulation is conducted to study the turbulent combustion processes of hydrogen and syngas flames. The subgrid scale momentum transport is performed with a one-equation model using the subgrid scale turbulent kinetic energy, while the linear-eddy model is employed to represent the scalar transport. Reduced chemical kinetics is used to describe the flame chemistry with the extended Zeldovich mechanism to account for the thermal formation of NOx. Results show the effects of the hydrogen content and of the Reynolds number on the characteristics of the flames. Higher hydrogen content contributes to increase the heat released after combustion leading to higher temperature peaks and thicker shear layers. The presence of CO in the fuel stream affects the flame dynamics with the development of a more vortical and wrinkled flow field. The vorticity field and turbulence/chemistry interactions are also discussed. Scalar profiles and comparison against experimental data are presented where a reasonable agreement is observed.

Abstract Fuel variability plays a major role in biogas combustion utilisation, where modelling and simulation can be used to provide guidelines. The construction of a flamelet generated manifold (FGM) is an important step for the simulation of complex turbulent flames. The FGM is made from premixed laminar flames and the limited literature on biogas combustion has led to this study on the effects of the fuel composition on biogas combustion in a freely propagating laminar premixed flame. The results have shown that methane concentration has a significant impact on the combustion process. There was no major difference observed among fuel mixtures with a fixed CH 4 volume concentration of 50%, with CO 2 varied from 40%, 49% to 50% and completed with either 10% of N 2 or 1% H 2 . The study can be further extended for numerical simulations under more realistic and practical conditions in order to develop guidelines for industrial combustors.

Large-eddy simulation of the reacting flow field in a combustion-based mitigation system to reduce the emissions of methane contained in ventilation air methane is presented. The application is based on the preheating and combustion of ventilation air methane. Effects of preheating and methane concentration are examined in five computational cases. The results indicate that the oxidation of the ventilation air methane can take place in a co-annular jet configuration provided that the preheating temperature is as high as 500 K for mixtures containing a low methane concentration of 0.5%. It is found that the oxidation process that eventually leads to reaction and combustion is controlled by the methane concentration and the level of preheating. The authors would like to thank the European Union’s Research Programme of the Research Fund for Coal and Steel (RFCS) research programme under Grant agreement Number RFCR-CT-2010-00004 and EPSRC Grant EP/G062714/2 for funding the activities under- taken for this study. Peer Reviewed

Une détermination précise des propriétés du carburant des mélanges complexes dans une large gamme de conditions de pression et de température est essentielle à l'utilisation de carburants de substitution. Le présent travail vise à construire des modèles d'apprentissage automatique (ML) bon marché pour agir comme des équations de fermeture pour prédire les propriétés physiques des carburants alternatifs. Ces modèles peuvent être formés à l'aide de la base de données à partir de simulations MD et/ou de mesures expérimentales dans une approche de fusion-fidélité des données. Ici, le processus gaussien (GP) et les modèles probabilistes génératifs sont adoptés. La GP est une approche bayésienne non paramétrique populaire pour construire des modèles de substitution principalement en raison de sa capacité à gérer les incertitudes aléatoires et épistémiques. Les modèles génératifs ont montré la capacité des réseaux de neurones profonds employés avec la même intention. Dans ce travail, l'analyse ML se concentre sur deux propriétés particulières, la densité et la diffusion du carburant, mais elle peut également être étendue à d'autres propriétés physico-chimiques. Cette étude explore la polyvalence des modèles ML pour gérer les données multi-fidélité. Les résultats montrent que les modèles ML peuvent prédire avec précision les propriétés du carburant dans une large gamme de conditions de pression et de température. La determinación precisa de las propiedades del combustible de mezclas complejas en una amplia gama de condiciones de presión y temperatura es esencial para utilizar combustibles alternativos. El presente trabajo tiene como objetivo construir modelos de aprendizaje automático (ML) de bajo costo para actuar como ecuaciones de cierre para predecir las propiedades físicas de los combustibles alternativos. Esos modelos se pueden entrenar utilizando la base de datos de simulaciones MD y/o mediciones experimentales en un enfoque de fidelidad de fusión de datos. Aquí, se adoptan el proceso gaussiano (GP) y los modelos generativos probabilísticos. GP es un enfoque bayesiano no paramétrico popular para construir modelos sustitutos principalmente debido a su capacidad para manejar las incertidumbres aleatorias y epistémicas. Los modelos generativos han demostrado la capacidad de las redes neuronales profundas empleadas con la misma intención. En este trabajo, el análisis de ML se centra en dos propiedades particulares, la densidad del combustible y la difusión, pero también se puede extender a otras propiedades fisicoquímicas. Este estudio explora la versatilidad de los modelos de ML para manejar datos de fidelidad múltiple. Los resultados muestran que los modelos ML pueden predecir con precisión las propiedades del combustible de una amplia gama de condiciones de presión y temperatura. Accurate determination of fuel properties of complex mixtures over a wide range of pressure and temperature conditions is essential to utilizing alternative fuels. The present work aims to construct cheap-to-compute machine learning (ML) models to act as closure equations for predicting the physical properties of alternative fuels. Those models can be trained using the database from MD simulations and/or experimental measurements in a data-fusion-fidelity approach. Here, Gaussian Process (GP) and probabilistic generative models are adopted. GP is a popular non-parametric Bayesian approach to build surrogate models mainly due to its capacity to handle the aleatory and epistemic uncertainties. Generative models have shown the ability of deep neural networks employed with the same intent. In this work, ML analysis is focused on two particular properties, the fuel density and diffusion, but it can also be extended to other physicochemical properties. This study explores the versatility of the ML models to handle multi-fidelity data. The results show that ML models can predict accurately the fuel properties of a wide range of pressure and temperature conditions. يعد التحديد الدقيق لخصائص الوقود للمخاليط المعقدة على نطاق واسع من ظروف الضغط ودرجة الحرارة أمرًا ضروريًا لاستخدام الوقود البديل. يهدف العمل الحالي إلى بناء نماذج رخيصة للتعلم الآلي (ML) لتكون بمثابة معادلات إغلاق للتنبؤ بالخصائص الفيزيائية للوقود البديل. يمكن تدريب هذه النماذج باستخدام قاعدة البيانات من محاكاة MD و/أو القياسات التجريبية في نهج دقة دمج البيانات. هنا، يتم اعتماد العملية الغاوسية (GP) والنماذج التوليدية الاحتمالية. الممارس العام هو نهج شعبي غير باراميتري لبناء نماذج بديلة ويرجع ذلك أساسا إلى قدرته على التعامل مع الشكوك الغريبة والمعرفية. أظهرت النماذج التوليدية قدرة الشبكات العصبية العميقة المستخدمة بنفس القصد. في هذا العمل، يركز تحليل التعلم الآلي على خاصيتين معينتين، كثافة الوقود وانتشاره، ولكن يمكن أن يمتد أيضًا إلى خصائص فيزيائية كيميائية أخرى. تستكشف هذه الدراسة تنوع نماذج التعلم الآلي للتعامل مع البيانات متعددة الدقة. تظهر النتائج أن نماذج التعلم الآلي يمكنها التنبؤ بدقة بخصائص الوقود لمجموعة واسعة من ظروف الضغط ودرجة الحرارة.

Transport property prediction of fatty acid methyl esters (FAMEs) is essential to its utilisation as biodiesel and biolubricant which can work under high-pressure conditions. Equilibrium molecular simulation is performed to study the viscosity, diffusivity, density and molecular structure dynamics at conditions up to 300 MPa. Among the transport properties, convergence of the viscosity needs a sufficiently large number of independent replications of the simulation. The system size effect on diffusion coefficient should be taken into consideration in fitting the Stokes-Einstein relation. The capability of three different force fields on predicting transport properties is evaluated in terms of the united-atom molecular model and all-atom molecular model. The solidification of FAMEs under high pressure occurs with parallel molecular alignment. The spatial inhomogeneity results in the breakdown of Stokes-Einstein relation. A hybrid effective hydrodynamic radius is established on the linear relation between experimental viscosity and diffusion coefficient in molecular simulation. This provides a predictive method to estimate viscosity from molecular diffusion coefficient over a broad range of conditions provided that Stokes-Einstein relation applies. Peer Reviewed

A comparison of two different models addressing the scalar transport in large-eddy simulations is conducted for a non-reacting jet and an experimental flame. A simple approach based on a gradient diffusion closure is compared against the linear-eddy model in the context of hydrogen-enriched non-reacting fuel jets and flames burning hydrogen-enriched mixtures. The results show that the gradient diffusion model is not valid as a subgrid scale model for large-eddy simulations of mixtures containing hydrogen. It produces unphysical scalar fields with unrealistic temperature distributions. Approaches based on the linear-eddy model can be used instead to obtain appropriate representation of the scalar field and more accurate predictions of the scalar transport and the temperature field.

The current study presents a numerical investigation of the flow field of a swirl-stabilized burner featuring a non-swirling axial air jet on the central axis of the mixing tube. The system has been designed and optimized to burn hydrogen at the Technische Universität Berlin over the last 6 years in the context of the EU-funded projects GREENEST and AHEAD. As the burner design was based on experimental work, high-fidelity large-eddy simulations (LES) are used to provide deeper understanding on the non-reacting and reacting flow fields to elucidate the occurrence of flashback under certain operating conditions. The experimental measurements suggest that flashback is produced by a velocity deficit at the mixing tube outlet and these conditions are analyzed here using LES. The work includes code validation for non-reacting and reacting conditions by comparison to water tunnel and combustion test rig data, and aims to evaluate the accuracy of LES with a combustion model based on premixed flamelets to predict the reacting flow field under conditions close to flashback.

The present work addresses the coupling of a flamelet database that can accurately represent the flame structure in composition space with a low-Mach approximation of the Navier-Stokes equations. An advancement of the CFI combustion model, which is currently based on laminar premixed flamelets, is used for chemistry tabulation. This model can be applied to different combustion regimes from premixed to non-premixed combustion, although this work is concentrated on turbulent premixed flames for Reynolds-averaged Navier-Stokes (RANS) and large-eddy simulations (LES). A premixed confined jet flame, which has been investigated experimentally at the German Aerospace Center (DLR) is used for validation in adiabatic conditions showing satisfactory agreement.