Adjoint-based sensitivity analysis of flames

Simulation of chemically reacting flows using detailed chemistry introduces a large number of chemistry model parameters. While not all significantly affect the target outcomes of a simulation, the parameters that do are not always known a priori. In order to improve simulations for specified target outcomes, termed quantities of interest (QoIs), the sensitivity of these QoIs to the model parameters are needed. However, evaluating the sensitivities is computationally expensive, especially for complex fuels that may involve many parameters. For these simulations, the forward sensitivity method requires the solution of an additional number of governing equations proportional to the number of parameters. Here, an adjoint sensitivity approach is formulated where the computational cost scales as the number of QoIs and not the number of parameters. Specifically, adjoint equations are derived for laminar, incompressible, variable density reacting flow and applied to hydrogen flame simulations. From the solution of the corresponding adjoint equations, sensitivity of the QoIs to chemistry model parameters is calculated. The one-dimensional simulation results show that the adjoint sensitivity results closely match those of forward sensitivity methods, thus providing validation of the adjoint method. The two-dimensional simulation results indicate the most sensitive parameters for two QoIs, flame tip temperature and NOx emission. For these tests, the adjoint method reduces computational expense compared to forward sensitivity methods by a factor proportional to the number of QoIs over the number of parameters, here 2/172. Such savings can be more drastic for cases that involve complex fuels, such as combustion of jet fuel, requiring thousands of chemistry model parameters. Further, this sensitivity information can be used in development of experiments by pointing out which are the critical chemistry model parameters.     

Characteristic boundary conditions for simulations of compressible reacting flows with multi-dimensional,viscous and reaction effects

A generalized formulation of the characteristic boundary conditions for compressible reacting flows is proposed. The new and improved approach resolves a number of lingering issues of spurious solution behaviour encountered in turbulent reacting flow simulations in the past. This is accomplished (a) by accounting for all the relevant terms in the determination of the characteristic wave amplitudes and (b) by accommodating a relaxation treatment for the transverse gradient terms with the relaxation coefficient properly determined by the low Mach number asymptotic expansion. The new boundary conditions are applied to a comprehensive set of test problems including: vortex-convection; turbulent inflow; ignition front propagation; non-reacting and reacting Poiseuille flows; and counterflow cases. It is demonstrated that the improved boundary conditions perform consistently superior to existing approaches, and result in robust and accurate solutions with minimal acoustic wave interactions at the boundary in hostile turbulent combustion simulation conditions.     

A parallel solution – adaptive method for three-dimensional turbulent non-premixed combusting flows

A parallel adaptive mesh refinement (AMR) algorithm is proposed and applied to the prediction of steady turbulent non-premixed compressible combusting flows in three space dimensions. The parallel solution-adaptive algorithm solves the system of partial-differential equations governing turbulent compressible flows of reactive thermally perfect gaseous mixtures using a fully coupled finite-volume formulation on body-fitted multi-block hexahedral meshes. The compressible formulation adopted herein can readily accommodate large density variations and thermo-acoustic phenomena. A flexible block-based hierarchical data structure is used to maintain the connectivity of the solution blocks in the multi-block mesh and to facilitate automatic solution-directed mesh adaptation according to physics-based refinement criteria. For calculations of near-wall turbulence, an automatic near-wall treatment readily accommodates situations during adaptive mesh refinement where the mesh resolution may not be sufficient for directly calculating near-wall turbulence using the low-Reynolds-number formulation. Numerical results for turbulent diffusion flames, including cold- and hot-flow predictions for a bluff-body burner, are described and compared to available experimental data. The numerical results demonstrate the validity and potential of the parallel AMR approach for predicting fine-scale features of complex turbulent non-premixed flames.     

Large eddy simulation of turbulent combustion systems

This paper reviews recent and ongoing work on numerical models for turbulent combustion systems based on a classical LES approach. The work is confined to single-phase reacting flows. First, important physico-chemical features of combustion-LES are discussed along with several aspects of overall LES models. Subsequently, some numerical issues, in particular questions associated with the reliability of LES results, are outlined. The details of chemistry, its reduction, and tabulation are not addressed here. Second, two illustrative applications dealing with non-premixed and premixed flame configurations are presented. The results show that combustion-LES is able to provide predictions very close to measured data for configurations where the flow is governed by large turbulent structures. To meet the future demands, new key challenges in specific modelling areas are suggested, and opportunities for advancements in combustion-LES techniques are highlighted. From a predictive point of view, the main target must be to provide a reliable method to aid combustion safety studies and the design of combustion systems of practical importance.     

Conditional analysis of temperature and strain rate effects on dissipation structure in turbulent non-premixed jet flames

This work presents results from simultaneous high-resolution temperature and velocity measurements in a series of turbulent non-premixed jet flames. The filtered Rayleigh scattering (FRS)-based temperature measurements demonstrate sufficient signal-to-noise (SNR) and spatial resolution to estimate the smallest scalar length scales and accurately determine dissipation rate fields. A comprehensive set of conditional statistics are used to characterize the small-scale structure, including the dependence of dissipation layer widths on Reynolds number, temperature, and dissipation magnitude. In general, the dissipation layer thickness decrease with increasing Reynolds number and increase with increasing temperature. However, dissipation layer widths show two distinct behaviors with respect to dissipation magnitude. For small dissipation values, increases in magnitude results in broadening of the dissipation layer, while for larger magnitude values of dissipation, the layer widths are thinned, highlighting the complexity of small-scale turbulent mixing. Additionally, measured ratios of the dissipation layer width to the Batchelor length scale are consistent across all Reynolds numbers and agree with previous studies in non-reacting flows. The unique aspect about the current set of measurements is the ability to examine the interaction of dissipation structure with turbulent flow parameters for the first time in turbulent non-premixed flames. Particularly, the strain rate/dissipation relationship is examined and compared to previous studies in non-reacting flows. It is found that the dissipation layers tend to align normal to the principal compressive strain axis and this tendency increases with increasing Reynolds number. For the lowest Reynolds number case, no dependence of the dissipation layer width nor dissipation rate magnitude on strain rate is found. However, for higher Reynolds numbers, a strong dependence of the dissipation layer width and dissipation rate magnitude on the principal compressive strain rate is observed. These results indicate the direct role of the compressive strain rate field on small-scale mixing structure in reacting flows.     

Adjoint-based optimization for understanding and suppressing jet noise

Advanced simulation tools, particularly large-eddy simulation techniques, are becoming capable of making quality predictions of jet noise for realistic nozzle geometries and at engineering relevant flow conditions. Increasing computer resources will be a key factor in improving these predictions still further. Quality prediction, however, is only a necessary condition for the use of such simulations in design optimization. Predictions do not themselves lead to quieter designs. They must be interpreted or harnessed in some way that leads to design improvements. As yet, such simulations have not yielded any simplifying principals that offer general design guidance. The turbulence mechanisms leading to jet noise remain poorly described in their complexity. In this light, we have implemented and demonstrated an aeroacoustic adjoint-based optimization technique that automatically calculates gradients that point the direction in which to adjust controls in order to improve designs. This is done with only a single flow solutions and a solution of an adjoint system, which is solved at computational cost comparable to that for the flow. Optimization requires iterations, but having the gradient information provided via the adjoint accelerates convergence in a manner that is insensitive to the number of parameters to be optimized. This paper, which follows from a presentation at the 2010 IUTAM Symposium on Computational Aero-Acoustics for Aircraft Noise Prediction, reviews recent and ongoing efforts by the author and co-workers. It provides a new formulation of the basic approach and demonstrates the approach on a series of model flows, culminating with a preliminary result for a turbulent jet.     

An improved high-order scheme for DNS of low Mach number turbulent reacting flows based on stiff chemistry solver

We present an improved numerical scheme for numerical simulations of low Mach number turbulent reacting flows with detailed chemistry and transport. The method is based on a semi-implicit operator-splitting scheme with a stiff solver for integration of the chemical kinetic rates, developed by Knio et al. [O.M. Knio, H.N. Najm, P.S. Wyckoff, A semi-implicit numerical scheme for reacting flow II. Stiff, operator-split formulation, Journal of Computational Physics 154 (2) (1999) 428–467]. Using the material derivative form of continuity equation, we enhance the scheme to allow for large density ratio in the flow field. The scheme is developed for direct numerical simulation of turbulent reacting flow by employing high-order discretization for the spatial terms. The accuracy of the scheme in space and time is verified by examining the grid/time-step dependency on one-dimensional benchmark cases: a freely propagating premixed flame in an open environment and in an enclosure related to spark-ignition engines. The scheme is then examined in simulations of a two-dimensional laminar flame/vortex-pair interaction. Furthermore, we apply the scheme to direct numerical simulation of a homogeneous charge compression ignition (HCCI) process in an enclosure studied previously in the literature. Satisfactory agreement is found in terms of the overall ignition behavior, local reaction zone structures and statistical quantities. Finally, the scheme is used to study the development of intrinsic flame instabilities in a lean H₂/air premixed flame, where it is shown that the spatial and temporary accuracies of numerical schemes can have great impact on the prediction of the sensitive nonlinear evolution process of flame instability.     

突扩燃烧室湍流预混反应流的数值模拟

对突扩燃烧室这一典型工程燃烧装置内的湍流预混反应流进行了数值模拟。时平均控制方程组的封闭采用k-ε湍流输运模型和EBU-Arhenius湍流反应模型。模拟结果给出了突扩燃烧室内湍流预混反应流的气体时均流场、组分浓度场与温度场的分布。通过数值模拟结果与实验的比较对EBU-Arhenius模型进行了讨论与评价。     

Simulations and experiments on the ignition probability in turbulent premixed bluff-body flames

The ignition characteristics of a premixed bluff-body burner under lean conditions were investigated experimentally and numerically with a physical model focusing on ignition probability. Visualisation of the flame with a 5 kHz OH* chemiluminescence camera confirmed that successful ignitions were those associated with the movement of the kernel upstream, consistent with previous work on non-premixed systems. Performing many separate ignition trials at the same spark position and flow conditions resulted in a quantification of the ignition probability P_ign, which was found to decrease with increasing distance downstream of the bluff body and a decrease in equivalence ratio. Flows corresponding to flames close to the blow-off limit could not be ignited, although such flames were stable if reached from a richer already ignited condition. A detailed comparison with the local Karlovitz number and the mean velocity showed that regions of high P_ign are associated with low Ka and negative bulk velocity (i.e. towards the bluff body), although a direct correlation was not possible. A modelling effort that takes convection and localised flame quenching into account by tracking stochastic virtual flame particles, previously validated for non-premixed and spray ignition, was used to estimate the ignition probability. The applicability of this approach to premixed flows was first evaluated by investigating the model's flame propagation mechanism in a uniform turbulence field, which showed that the model reproduces the bending behaviour of the S_T-versus-u′ curve. Then ignition simulations of the bluff-body burner were carried out. The ignition probability map was computed and it was found that the model reproduces all main trends found in the experimental study.     

Efficient evaluation of the direct and adjoint linearized dynamics from compressible flow solvers

The direct and adjoint operators play an undeniably important role in a vast number of theoretical and practical studies that range from linear stability to flow control and nonlinear optimization. Based on an existing nonlinear flow solver, the design of efficient and straightforward procedures to access these operators is thus highly desirable. In the case of compressible solvers, the use of high-order numerical schemes combined with complicated governing equations makes the derivation of efficient procedures a challenging and often tedious undertaking. In this work, a novel technique for the evaluation of the direct and adjoint operators directly from compressible flow solvers is presented and extended to include nonlinear differentiation schemes and turbulence models. The application to the incompressible counterpart is also discussed. The presented method requires minimal additional programming effort and automatically takes into account subsequent modifications in the governing equations and boundary conditions. The introduced methodology is demonstrated on existing numerical codes, and direct and adjoint global modes are calculated for three typical flow configurations. Implementation issues and the performance measures are also discussed. The proposed algorithm presents an easy-to-implement and efficient technique to extract valuable information for the quantitative analysis of complex flows.     

Evaluation of a flamelet/progress variable approach for pulverized coal combustion in a turbulent mixing layer

A steady flamelet/progress variable (FPV) approach for pulverized coal flames is employed to simulate coal particle burning in a turbulent shear and mixing layer. The configuration consists of a carrier-gas stream of air laden with coal particles that mixes with an oxidizer stream of hot products from lean combustion. Carrier-phase DNS (CP-DNS) are performed, where the turbulent flow field is fully resolved, whereas the coal is represented by Lagrangian point particles. CP-DNS with direct chemistry integration is performed first and provides state-of-the-art validation data for FPV modeling. In a second step the control variables for FPV are extracted from the CP-DNS and used to test if the tabulated manifold can correctly describe the reacting flow (a priorianalysis). Finally a fully coupled a posteriori FPV simulation is performed, where only the FPV control variables are transported, and the chemical state is retrieved from the table and fed back to the flow solver. The a priori results show that the FPV approach is suitable for modeling the complex reacting multiphase flow considered here. The a posteriori data is similarly in good agreement with the reference CP-DNS, although stronger deviations than a priori can be observed. These discrepancies mainly appear in the upper flame (of the present DNS), where premixing and highly unsteady extinction and re-ignition effects play a role, which are difficult to capture by steady non-premixed FPV modeling. However, the present FPV model accurately captures the lower, more stable flame that burns in non-premixed mode.     

Shock capturing with PDE-based artificial viscosity for DGFEM: Part I. Formulation

Artificial viscosity can be combined with a higher-order discontinuous Galerkin finite element discretization to resolve a shock layer within a single cell. However, when a non-smooth artificial viscosity model is employed with an otherwise higher-order approximation, element-to-element variations induce oscillations in state gradients and pollute the downstream flow. To alleviate these difficulties, this work proposes a higher-order, state-based artificial viscosity with an associated governing partial differential equation (PDE). In the governing PDE, a shock indicator acts as a forcing term while grid-based diffusion is added to smooth the resulting artificial viscosity. When applied to heat transfer prediction on unstructured meshes in hypersonic flows, the PDE-based artificial viscosity is less susceptible to errors introduced by grid edges oblique to captured shocks and boundary layers, thereby enabling accurate heat transfer predictions.     

Response of non-premixed flames to bulk flow perturbations

This paper describes the dynamics of non-premixed flames responding to bulk velocity fluctuations, and compares the dynamics of the flame sheet position and spatially integrated heat release to that of a premixed flame. The space–time dynamics of the non-premixed flame sheet in the fast chemistry limit is described by the stoichiometric mixture fraction surface, extracted from the solution of the

-equation. This procedure has some analogies to premixed flames, where the premixed flame sheet location is extracted from the G = 0 surface of the solution of the G-equation. A key difference between the premixed and non-premixed flame dynamics, however, is the fact that the non-premixed flame sheet dynamics are a function of the disturbance field everywhere, and not just at the reaction sheet, as in the premixed flame problem. A second key difference is that the non-premixed flame does not propagate and so flame wrinkles are convected downstream at the axial flow velocity, while wrinkles in premixed flames convect downstream at a vector sum of the flame speed and axial velocity. With the exception of the flame wrinkle propagation speed, however, we show that that the solutions for the space–time dynamics of the premixed and non-premixed reaction sheets in high velocity axial flows are quite similar. In contrast, there are important differences in their spatially integrated unsteady heat release dynamics. Premixed flame heat release fluctuations are dominated by area fluctuations, while non-premixed flames are dominated by mass burning rate fluctuations. At low Strouhal numbers, the resultant sensitivity of both flames to flow disturbances is the same, but the non-premixed flame response rolls off slower with frequency. Hence, this analysis suggests that non-premixed flames are more sensitive to flow perturbations than premixed flames at O(1) Strouhal numbers.     

Transported scalar PDF calculations of autoignition of a hydrogen jet in a heated turbulent co-flow

The joint scalar PDF method, as implemented in FLUENT, was used to simulate the autoignition of a jet of hydrogen in a turbulent co-flow of heated air. While the autoignition phenomenon is intermittent in the experiment, ensemble-averaged data on the effect of the flow on ignition length are available, which enables us to compare them with the steady state calculations.Results of sensitivity tests showed that the choice of chemical mechanism affects the calculation more than the mixing model and model constants. Further calculations for different initial conditions (i.e. temperature and velocity of the jet T _jet and U _jet and the co-flow T _air and U _air) have been done using a set of parameters selected after the sensitivity study. Scatter plots and conditional scalar profiles confirmed that the ignition is always initiated in lean mixture fractions. The ignition length was predicted with good accuracy for the case of U _jet>U _air but not so well for the case of U _jet≈U _air. For the equal velocity case, increasing the velocity resulted in delayed autoignition time (defined as the ignition length divided by the mean velocity), in agreement with the experimental trend. The results give credence to the use of the joint scalar PDF method for autoignition in non-premixed flows.     

Continuous and Discrete Adjoint Approach Based on Lattice Boltzmann Method in Aerodynamic Optimization Part I: Mathematical Derivation of Adjoint Lattice Boltzmann Equations

The significance of flow optimization utilizing the lattice Boltzmann (LB) method becomes obvious regarding its advantages as a novel flow field solution method compared to the other conventional computational fluid dynamics techniques. These unique characteristics of the LB method form the main idea of its application to optimization problems. In this research, for the first time, both continuous and discrete adjoint equations were extracted based on the LB method using a general procedure with low implementation cost. The proposed approach could be performed similarly for any optimization problem with the corresponding cost function and design variables vector. Moreover, this approach was not limited to flow fields and could be employed for steady as well as unsteady flows. Initially, the continuous and discrete adjoint LB equations and the cost function gradient vector were derived mathematically in detail using the continuous and discrete LB equations in space and time, respectively. Meanwhile, new adjoint concepts in lattice space were introduced. Finally, the analytical evaluation of the adjoint distribution functions and the cost function gradients was carried out.     

Mind the gap: Turbulent combustion model validation and future needs

This overview collects a range of well characterized experiments used in the step-wise validation of turbulent combustion models, from gas phase non-premixed jet flames to spray flames, and from simple symmetric jets to real device geometries, focusing primarily on statistically steady state experiments. We discuss how the experiments and models are constructed, approaches to modelling, and the tradeoffs between the level of detail and computational demands. The review highlights a number of experiments used for benchmarking models, selecting a few examples where models have clearly succeeded, as well as some areas where there are clear needs in the experimental database. In particular, the areas of turbulent spray combustion and soot prediction, as well as combustion under high pressures appear as the least developed and present the clearest gaps for both models and experiments. Based on the successful application of advanced methods of uncertainty quantification to a number of problems in reacting flows, we suggest that these methods might be used to advantage in the design of experiments. This would enable an upfront examination of the extent to which comparisons between measurable scalars and velocities allow clear distinction between model features.     

Adjoint wall functions: A new concept for use in aerodynamic shape optimization

The continuous adjoint method for the computation of sensitivity derivatives in aerodynamic optimization problems of steady incompressible flows, modeled through the k–ε turbulence model with wall functions, is presented. The proposed formulation leads to the adjoint equations along with their boundary conditions by introducing the adjoint to the friction velocity. Based on the latter, an adjoint law of the wall that bridges the gap between the solid wall and the first grid node off the wall is proposed and used during the solution of the system of adjoint (to both the mean flow and turbulence) equations. Any high Reynolds turbulence model, other than the k–ε one used in this paper, could also profit from the proposed adjoint wall function technique. In the examined duct flow problems, where the total pressure loss due to viscous effects is used as objective function, emphasis is laid on the accuracy of the computed sensitivity derivatives, rather than the optimization itself. The latter might rely on any descent method, once the objective function gradient has accurately been computed.     

Aerodynamic design optimization by using a continuous adjoint method

This paper presents the fundamentals of a continuous adjoint method and the applications of this method to the aerodynamic design optimization of both external and internal flows.General formulation of the continuous adjoint equations and the corresponding boundary conditions are derived.With the adjoint method,the complete gradient information needed in the design optimization can be obtained by solving the governing flow equations and the corresponding adjoint equations only once for each cost function,regardless of the number of design parameters.An inverse design of airfoil is firstly performed to study the accuracy of the adjoint gradient and the effectiveness of the adjoint method as an inverse design method.Then the method is used to perform a series of single and multiple point design optimization problems involving the drag reduction of airfoil,wing,and wing-body configuration,and the aerodynamic performance improvement of turbine and compressor blade rows.The results demonstrate that the continuous adjoint method can efficiently and significantly improve the aerodynamic performance of the design in a shape optimization problem.     

Coupling of mixing models with manifold based simplified chemistry in PDF modeling of turbulent reacting flows

For general reacting flows the numerical simulation faces two main challenges. One is the high dimensionality and stiffness of the governing conservation equations due to detailed chemistry, which can be solved by using simplified chemical kinetics. The other one is the difficulty of modeling the coupling of turbulence with thermo-chemical source term. The probability density function (PDF) method allows to calculate turbulent reacting flows by solving the thermal-chemical source term in closed form. Usually, the PDF method for turbulent processes such as mixing processes and the reduction method for chemical kinetics are developed separately. However, coupling of both processes plays an important role for the numerical accuracy. To investigate the importance of coupling between turbulence and simplified chemistry, two different coupling strategies for mixing and reduced chemistry are discussed and tested for the well-known Sandia Flames E and F, in which there is a strong interaction between turbulence and chemical kinetics. The EMST mixing model is chosen for turbulent mixing, while the Reaction-Diffusion Manifolds (REDIMs) is used as simplified chemistry. However, the proposed strategies are also valid for other mixing models and manifold based simplified chemistry.