Accelerating multi-dimensional combustion simulations using GPU and hybrid explicit/implicit ODE integration

Simulating multi-dimensional combustion with detailed kinetics often requires solving a large number of ordinary differential equation (ODE) problems at each global time step. In many cases, the ODE integrations account for the bulk of the total wall-clock time for the simulation. This paper introduces CHEMEQ2-GPU – a new explicit stiff ODE solver (based on the existing CHEMEQ2 solver) that exploits the parallel architecture of the modern graphics processing unit (GPU) to accelerate ODE integration in multi-dimensional combustion simulations. We also demonstrate efficient application of the CPU and GPU as co-processors, for further speedup. We describe a hybrid explicit/implicit ODE solver approach that combines the strengths of both solver types running simultaneously on the GPU and CPU, respectively. A dynamic load balancing scheme was used to assign the kinetics ODE integrations over all grid points to either the CPU-based implicit solver DVODE (which is the more efficient solver for highly stiff grid points) or CHEMEQ2-GPU (more efficient for moderately stiff or non-stiff grid points). We demonstrate CHEMEQ2-GPU and the hybrid approach in 3-D simulations of homogeneous charge compression ignition (HCCI) engines. The test cases applied two different n-heptane reaction mechanisms (a large detailed model and a small skeletal model) and three different mesh sizes. Engine simulations were performed using KIVA-CHEMKIN. CHEMEQ2 was about 2–3 times faster than DVODE, with similar prediction accuracy. The CHEMEQ2-GPU speedup relative to CHEMEQ2 increased linearly with the number of grid points for the range of meshes tested in this work. Assuming ideal linear scaling of simulation time with number of processors, the speed of CHEMEQ2-GPU on the Tesla C2050 GPU was equivalent to CHEMEQ2 running on approximately 13 parallel 2.8 GHz CPU processors for the finest mesh; and the hybrid solver approach was equivalent to CHEMEQ2 on ～15 such CPU processors. In summary, CHEMEQ2-GPU provided the additional computing power of 14 parallel CPU processors (for the finest mesh tested) and the hybrid solver approach demonstrated a method to efficiently apply these additional co-processors with existing CPU cores for combustion simulations. CHEMEQ2-GPU scales favorably with the number of grid points and is available by request to the authors. This work presents opportunities for further development, particularly in CPU/GPU load balancing algorithms.     

Linear eddy mixing based tabulation and artificial neural networks for large eddy simulations of turbulent flames

A large eddy simulation (LES) sub-grid model is developed based on the artificial neural network (ANN) approach to calculate the species instantaneous reaction rates for multi-step, multi-species chemical kinetics mechanisms. The proposed methodology depends on training the ANNs off-line on a thermo-chemical database representative of the actual composition and turbulence (but not the actual geometrical problem) of interest, and later using them to replace the stiff ODE solver (direct integration (DI)) to calculate the reaction rates in the sub-grid. The thermo-chemical database is tabulated with respect to the thermodynamic state vector without any reduction in the number of state variables. The thermo-chemistry is evolved by stand-alone linear eddy mixing (LEM) model simulations under both premixed and non-premixed conditions, where the unsteady interaction of turbulence with chemical kinetics is included as a part of the training database. The proposed methodology is tested in LES and in stand-alone LEM studies of three distinct test cases with different reduced mechanisms and conditions. LES of premixed flame–turbulence–vortex interaction provides direct comparison of the proposed ANN method against DI and ANNs trained on thermo-chemical database created using another type of tabulation method. It is shown that the ANN trained on the LEM database can capture the correct flame physics with accuracy comparable to DI, which cannot be achieved by ANN trained on a laminar premix flame database. A priori evaluation of the ANN generality within and outside its training domain is carried out using stand-alone LEM simulations as well. Results in general are satisfactory, and it is shown that the ANN provides considerable amount of memory saving and speed-up with reasonable and reliable accuracy. The speed-up is strongly affected by the stiffness of the reduced mechanism used for the computations, whereas the memory saving is considerable regardless.     

The use of dynamic adaptive chemistry and tabulation in reactive flow simulations

Detailed chemical kinetics is an integral component for predictive simulation of turbulent flames and is important for reliable prediction of flames and emissions. Major challenges of incorporation of detailed chemistry in flame simulations are induced by the large number of chemical species and the wide range of timescales involved in detailed kinetics. In this work, dynamic adaptive chemistry (DAC) and in situ adaptive tabulation (ISAT) for efficient chemistry calculations in calculating turbulent reactive flows with detailed chemistry are studied in iso-octane/air homogeneous charge compression ignition (HCCI) and methane/air combustion in a partially-stirred reactor (PaSR). Chemistry calculations are accelerated by DAC via expediting the integration of ordinary differential equations (ODEs) governing chemical kinetics with local skeletal mechanisms obtained on-the-fly using the directed relation graph (DRG) method, and by ISAT via reducing the number of ODE integrations through tabulating and re-using the ODE solutions. It is shown that, in contrast to ISAT, the performance of DAC is mostly independent of the nature of combustion simulations, e.g., steady or unsteady, premixed or non-premixed combustion, and its efficiency increases with the size of chemical kinetic mechanisms. DAC is particularly suitable for transient combustion simulations with large mechanisms containing hundreds of species or more, such as those for gasoline or diesel fuels. A speedup factor of about 30 is achieved for HCCI combustion of iso-octane/air with good agreements in the histories of temperature and species concentrations. In contrast, ISAT performs better for simulations where chemistry calculations can be predominantly resolved by retrieving from the ISAT table, i.e., re-using the ODE solutions. It is shown that ISAT achieves speedup factors of about 100 with only about 10%, 0.1% and 0.01% incurred errors in NO, CO, and temperature, respectively, for the premixed methane/air PaSR simulations. Moreover, a coupled DAC and ISAT approach, namely ISAT–DAC, has been developed and demonstrated in this study to accelerate chemistry evaluation. It is shown that the incurred errors in temperature and species concentrations in ISAT–DAC are well controlled, and it can significantly enhance the performance of ISAT, when the fraction of direct ODE integration is significant, via accelerating the ODE integrations by DAC.     

On preconditioned BiCGSTAB solver for MLPG method applied to heat conduction in complex geometry

Meshless local Petrov–Galerkin (MLPG) method is a promising meshfree method for continuum problems in complex domains, especially for large deformation, moving boundary and phase change problems. For large-scale problems, iterative methods for solving the discretized equations are more suitable than direct methods. Krylov subspace solvers of conjugate gradient type are the most preferred iterative solvers. The convergence rate of these methods depends on preconditioner used. Recently, proposed schedule relaxation Jacobi (SRJ) method can be used as a stand-alone solver and as a preconditioner. In the present work, the SRJ method is tested as a stand-alone solver and as a preconditioner for BiCGSTAB solver in the MLPG method, and its performance has been compared with successive overrelaxation (k) preconditioner. Two-dimensional linear steady-state heat conduction in complex shape geometry has been used as the model test problem.     

Redesigning combustion modeling algorithms for the Graphics Processing Unit (GPU): Chemical kinetic rate evaluation and ordinary differential equation integration

Detailed modeling of complex combustion kinetics remains challenging and often intractable, due to prohibitive computational costs incurred when solving the associated large kinetic mechanisms. The Graphics Processing Unit (GPU), originally designed for graphics rendering on computer and gaming systems, has recently emerged as a powerful, cost-effective supplement to the Central Processing Unit (CPU) for dramatically accelerating scientific computations. Complex scientific computations are now being performed on the GPU in several research fields, such as quantum chemistry, molecular dynamics, and atmospheric modeling. Here, we present methods for exploiting the highly parallel structure of GPUs for combustion modeling. This paper outlines simple algorithm revisions that can be applied to the majority of existing combustion modeling algorithms for GPU computations. Significant simulation acceleration and predictive capability enhancements were obtained by using these GPU-enhanced algorithms for reaction rate evaluation and in ODE integration. For the demonstrations, we implemented the rate evaluation revisions in CHEMKIN and the ODE integration revisions in DASAC and DVODE and we tested the performance for simulating constant-volume ignition using SENKIN. The simulations using the revised algorithms are more than an order of magnitude faster than the corresponding CPU-only simulations, even for a low-end (double-precision) graphics card. Additionally, the computational time scales less than quadratically with the number of chemical species in the kinetic mechanism when using the GPU, as compared to the super-quadratic scaling normally seen with CPU-only chemical kinetics computations; and the GPU-based revisions do not involve approximations to the detailed kinetics. An analysis of the growth rates of combustion mechanism sizes versus computational capabilities of CPUs and GPUs further reveals the important role that GPUs are expected to play in the future of combustion modeling. Finally, we briefly outline practical steps for effectively transitioning from CPU-only to GPU-enhanced combustion modeling.     

Dynamic stiffness removal for direct numerical simulations

A systematic approach was developed to derive non-stiff reduced mechanisms for direct numerical simulations (DNS) with explicit integration solvers. The stiffness reduction was achieved through on-the-fly elimination of short time-scales induced by two features of fast chemical reactivity, namely quasi-steady-state (QSS) species and partial-equilibrium (PE) reactions. The sparse algebraic equations resulting from QSS and PE approximations were utilized such that the efficiency of the dynamic stiffness reduction is high compared with general methods of time-scale reduction based on Jacobian decomposition. Using the dimension reduction strategies developed in our previous work, a reduced mechanism with 52 species was first derived from a detailed mechanism with 561 species. The reduced mechanism was validated for ignition and extinction applications over the parameter range of equivalence ratio between 0.5 and 1.5, pressure between 10 and 50 atm, and initial temperature between 700 and 1600 K for ignition, and worst-case errors of approximately 30% were observed. The reduced mechanism with dynamic stiffness removal was then applied in homogeneous and 1-D ignition applications, as well as a 2-D direct numerical simulation of ignition with temperature inhomogeneities at constant volume with integration time-steps of 5-10 ns. The integration was numerically stable and good accuracy was achieved.     

Preconditioned Solver Performance with Compressed Banded Data Format in Three-Dimensional Convective Heat Transfer Simulations

ABSTRACT

A new data storage format called compressed banded data (CBD) is developed for sparse banded matrices generated by hybrid finite-element/volume methods in numerical heat transfer. The platform of the new CBD structure permits dynamic switching between various solvers. The performance of various Krylov techniques, including GMRES(m) (Generalized Minimal RESidual), Bi-CGSTAB (Bi-Conjugate Gradient STABilized), Bi-CG (Bi-Conjugate Gradient), CG (Conjugate Gradient), and CGS (Conjugate Gradient Squared) with an ILU(0) preconditioner, are compared in three test problems. The performance of each preconditioned iterative solver is compared with a direct solver, particularly in terms of memory storage requirements. It is shown that the new CBD format provides useful benefits with respect to both reduction of storage requirements and CPU runtime.     

Two-dimensional numerical simulations of detonation cellular structures in H2O2Ar mixtures with OpenFOAM®

In the present work, the solver rhoCentralRfFoam, developed using the finite volume framework provided by OpenFOAM^®, is employed to perform numerical simulations of two-dimensional detonations. This solver uses the central scheme of Kurganov, Noelle, and Petrova for dealing with convective terms. Also, the detailed kinetic model for hydrogen oxidation of Marinov, Westbrook, and Pitz was used for properly defining chemically induced source terms, and the semi-implicit Bulirsh Stöer (SIBS) method was employed for solving the stiff ODE system required to compute the species' rates. The present study intends to investigate the solver's capability for computing cellular structures, which develop when non-planar detonations are propagating in confined mixtures. Interactions between waves, resulting from several ignition points, are used as perturbation sources for the onset of cellular structures. Numerical simulations allowed us to identify a well-shaped cellular structure and other different structures that are not clearly defined, close to the ignition sources. However, after extending the computational domain, convergence towards a unique cellular pattern is attained. Such cellular pattern compares with most of the available data. Also, in order to improve the presentation of cellular structures and their dynamic behavior, a numerical schlieren technique is utilized for some flow variables (e.g. vorticity and density).     

T2SOLV: an enhanced package of solvers for the TOUGH2 family of reservoir simulation codes

T2SOLV is an enhanced package of matrix solvers for the TOUGH2 family of codes. T2SOLV includes all the Preconditioned Conjugate Gradient (PCG) solvers used in T2CG1, the current solver package, as well as LUBAND, a new direct solver and DLUSTB, a PCG solver based which implements the BiCGSTAB(m) method. LUBAND, the new direct solver, is faster, more reliable and capable of solving problems orders of magnitude larger than the MA28 routine which it replaces. A significant capability of T2VOC is two types of matrix preprocessing, which make possible the solution of a class of challenging numerical problems (previously tractable only with direct solvers) using the PCG routines. For regular grids, the size of problems tractable with direct solvers is doubled by implementing a D4 alternative diagonal gridblock ordering option.     

On upgrading the numerics in combustion chemistry codes

A method of updating and reusing legacy FORTRAN codes for combustion simulations is presented using the DAEPACK software package. The procedure is demonstrated on two codes that come with the CHEMKIN-II package, CONP and SENKIN, for the constant-pressure batch reactor simulation. Using DAEPACK generated code, analytical derivative calculations, sparsity pattern information, and hidden discontinuity information can be obtained for the models of interest. This information can be easily integrated with different solvers giving the modeler great flexibility in selecting the best solution procedure. Using the generated code, the CONP code was connected to three different solvers, and the SENKIN code was connected to two different solvers. The effect of model formulation, analytical derivatives, sparsity, and sensitivity equation solution method were analyzed for three large kinetic mechanisms for methane, acetylene, and n-heptane. For the n-heptane model, with 544 species and 2446 reactions, a factor of 10-speed improvement over the original solution procedure was found using analytical derivatives and sparse linear algebra. For sensitivity calculations, for a small number of parameters, a factor of 55 improvement over the original solution procedure was found for the n-heptane problem. Upon closer examination of results, no one method is found to always be superior to other methods, and selection of the appropriate solution procedure requires an examination of the specific kinetic mechanism, which is easily conducted using DAEPACK generated code.     

A dynamic multi-timescale method for combustion modeling with detailed and reduced chemical kinetic mechanisms

A new on-grid dynamic multi-timescale (MTS) method is presented to increase significantly the computation efficiency involving multi-physical and chemical processes using detailed and reduced kinetic mechanisms. The methodology of the MTS method using the instantaneous timescales of different species is introduced. The definition of the characteristic time for species is examined and compared with that of the computational singular perturbation (CSP) and frozen reaction rate methods by using a simple reaction system. A hybrid multi-timescale (HMTS) algorithm is constructed by integrating the MTS method with an implicit Euler scheme, respectively, for species with and without the requirement of accurate time histories at sub-base timescales. The efficiency and the robustness of the MTS and HMTS methods are demonstrated by comparing with the Euler and VODE solvers for homogenous ignition and unsteady flame propagation of hydrogen, methane, and n-decane-air mixtures. The results show that both MTS and HMTS reproduce well the species and temperature histories and are able to decrease computation time by about one-order with the same kinetic mechanism. Compared to MTS, HMTS has slightly better computation efficiency but scarifies the stability at large base time steps. The results also show that with the increase of mechanism size and the decrease of time step, the computation efficiency of multi-timescale method increases compared to the VODE solver. In addition, it is shown that the integration of the multi-timescale method with the path flux analysis based mechanism reduction approach can further increase the computation efficiency. Unsteady simulations of outwardly propagating spherical n-decane-air premixed flames demonstrate that the multi-timescale method is rigorous for direct numerical simulations with both detailed and reduced chemistry and can dramatically improve the computation efficiency.     

Numerical simulation of the influence of confined multi-point ignited H2–O2 mixture on the propagation of shock waves towards a deformable plate

The study of shock wave propagation in a detonation chamber is of great importance as a part of the plate forming process. Investigations related to the effects of premixed gas detonation on the deflection of a plate require in-depth examination. An Eulerian-Lagrangian numerical simulation is conducted using the space-time conservation element and solution element method of LS-DYNA software to study the effect of confined multi-point ignited gaseous mixture on the dynamic response of thin plates clamped at the end of a combustion chamber. The FSI couples a Lagrangian finite element solver with a Eulerian fluid solver in a 2D space with detailed chemistry of H₂–O₂ mixture. The solution contains the detonation wave propagation through the combustion chamber and its interaction with the plate. The influence of variation in the multi-point ignition locations and combustion chamber dimensions on the pressure history and plate deflection is studied. To verify the model, a comparison with the experimental study is carried out using an adjustable model representative of the real experiment. The verified model is used to link the evolution of plate shape with the arrival time and intensity of shock waves within the chamber. It is found that a longer distance between the ignition point and the plate intensifies the ultimate deflection of the plate. In addition, a fairly large combustion area employed in a direction rather than transverse to the plate surface is unable to influence the ultimate deformation of the plate.     

On a Wavenumber Optimized Streamline Upwinding Method for Solving Steady Incompressible Navier-Stokes Equations

A stabilized upwind finite-element model is developed to solve the three-dimensional incompressible steady Navier-Stokes equations. The test function is constructed to have a larger weight on the upstream side. This has been achieved by adding a stabilized term to the shape function so as to optimize the numerical wavenumber for convection terms. To avoid Lanczos or pivoting breakdown while solving the resulting unsymmetric and indefinite mixed finite-element matrix equations iteratively, the finite-element equation has been modified by pre-multiplying it with its transpose. The resulting normalized matrix equation becomes symmetric and positive-definite. We can therefore apply a computationally efficient conjugate gradient Krylov iterative solver to get an unconditionally convergent solution. However, the condition number of the new system becomes the square of the original unsymmetric indefinite system. To fully exploit excellent convergence nature of the conjugate gradient iterative solver, an element-by-element strategy is adopted to avoid assembling of all the stiffness matrices obtained at element level. We alleviate the drawback of slower convergence of the conjugate gradient method due to the increased condition number by preconditioning the positive-definite matrix. The resulting preconditioned matrix equation is solved in a matrix-free manner using the preconditioned conjugate gradient iterative solver. The developed finite-element code is first verified by solving a problem amenable to analytical solution. The benchmark lid-driven cavity problem is also solved in a cube for assessing the three chosen iterative solvers.     

A combustion model for IC engine combustion simulations with multi-component fuels

Reduced chemical kinetic mechanisms for the oxidation of representative surrogate components of a typical multi-component automotive fuel have been developed and applied to model internal combustion engines. Starting from an existing reduced mechanism for primary reference fuel (PRF) oxidation, further improvement was made by including additional reactions and by optimizing reaction rate constants of selected reactions. Using a similar approach to that used to develop the reduced PRF mechanism, reduced mechanisms for the oxidation of n-tetradecane, toluene, cyclohexane, dimethyl ether (DME), ethanol, and methyl butanoate (MB) were built and combined with the PRF mechanism to form a multi-surrogate fuel chemistry (MultiChem) mechanism. The final version of the MultiChem mechanism consists of 113 species and 487 reactions. Validation of the present MultiChem mechanism was performed with ignition delay time measurements from shock tube tests and predictions by comprehensive mechanisms available in the literature.A combustion model was developed to simulate engine combustion with multi-component fuels using the present MultiChem mechanism, and the model was applied to simulate HCCI and DI engine combustion. The results show that the present multi-component combustion model gives reliable performance for combustion predictions, as well as computational efficiency improvements through the use of reduced mechanism for multi-dimensional CFD simulations.     

Application of spray combustion simulation in DI diesel engine

This work presents the development and implementation of combustion model for DI diesel engines by using the PDF-Chemical Equilibrium combustion model. The key concept of this approach is to predict the thermochemical variables (e.g., temperature, species mass fractions) and then the average scalars of these variables are evaluated by a probability density function (PDF) averaging approach. To realize flame propagation, the reaction time scale is employed to relax the infinitely fast chemistry of chemical equilibrium. The PDF-Eddy Break Up ignition model is adopted in the auto-ignition calculation. With regard to the comparison results, the simulation results are in good agreement with the experimental results in both ignition and combustion modes. In addition, the predicted lift-off length also corresponds to a power-law scaling of Siebers et al.     

A numerical study of the combustion and jet characteristics of a hydrogen fueled turbulent hot-jet ignition (THJI) chamber

Turbulent hot-jet ignition (THJI) is an advanced ignition enhancement technology which can potentially overcome the problem associated with lean burn combustion. The present study makes an effort on the comprehensive understanding of a hydrogen fueled THJI chamber with various pre-chamber spark locations. Computational fluid dynamics (CFD) simulations are performed using an in-house code based on the KIVA-3V release 2 program coupled with an in-house chemical solver. A detailed chemical kinetics mechanism with 10 species and 19 reversible reactions is used for the H₂/air mixture in both the pre-chamber and the main chamber. The results show that moving the spark ignition location farther from the orifice significantly reduces the $0?10\text{\%}$ mass fraction burn period. By analyzing the local Mach number, turbulence kinetic energy and turbulence length scale, the compressibility and turbulence level of the jet flow are evaluated. Further analysis of the OH mass fraction distribution identifies three regions in the hot jet, i.e. extinction region, just-igniting region and combustion region. A critical Damköhler number of 0.3 is determined to separate the extinction region from the other regions. Meanwhile, transition Damköhler numbers ranging from 0.3 to 0.6 are determined in the just-igniting region.     

A numerical assessment of the novel concept of crevice containment in a rapid compression machine

Rapid compression machines (RCMs) typically incorporate creviced pistons to suppress the formation of the roll-up vortex. The use of a creviced piston, however, can enhance other multi-dimensional effects inside the RCM due to the crevice zone being at lower temperature than the main reaction chamber. In this work, such undesirable effects of a creviced piston are highlighted through computational fluid dynamics simulations of n-heptane ignition in RCM. Specifically, the results show that in an RCM with a creviced piston, additional flow of mass takes place from the main combustion chamber to the crevice zone during the first-stage of the two-stage ignition. This phenomenon is not captured by the zero-dimensional modeling approaches that are currently adopted. Consequently, a novel approach of ‘crevice containment’ is introduced and computationally evaluated in this paper. In order to avoid the undesirable effects of creviced piston, the crevice zone is separated from the main reaction chamber at the end of compression. The results with ‘crevice containment’ show significant improvement in the fidelity of zero-dimensional modeling in terms of predicting the overall ignition delay and pressure rise in the first-stage of ignition. Although the implementation of ‘crevice containment’ requires a modification in RCM design, in practice there are significant advantages to be gained through a reduction in the rate of pressure drop in the RCM combustion chamber and a quantitative improvement in the data obtained from the species sampling experiments.     

Impact of aerodynamic modelling on seakeeping performance of a floating vertical axis wind turbine

This study focuses on the impact of the aerodynamic model on the dynamic response of a floating vertical axis wind turbine (VAWT). It compares a state‐of‐the‐art quasi‐steady double multiple streamtube (DMS) solver, a prescribed vortex wake (PVW), and a free vortex wake (FVW) solver. The aerodynamic loads acting on a bottom‐fixed VAWT and computed with the three aerodynamic solvers are compared, then the dynamic responses of the floating turbine in irregular waves and turbulent wind with the different aerodynamic solvers are compared. Differences are observed, particularly in the mean motions of the platform. Eventually, the aerodynamic damping computed by the solvers are estimated with aerodynamic simulations on the turbine with imposed surge and pitch motions. The estimated damping can then be correlated with the dynamic response amplitude of the VAWT. Substantial discrepancies are observed between the three solvers at high tip speed ratio, when the rotor is highly loaded. It is shown that the quasi‐steady DMS solver seems to give greater amplitude of motions for the floating VAWT because of strong rotor/wake interaction that are not correctly accounted for.     

Toward accommodating realistic fuel chemistry in large-scale computations

The need and prospect of incorporating realistic fuel chemistry in large-scale simulations of combustion phenomena and combustor performance are reviewed. The review first demonstrates the intricacies of chemical kinetics in homogeneous and diffusive systems, and emphasizes the essential importance of the comprehensiveness of chemical fidelity for mechanisms at the detailed and reduced levels. A systematic approach towards developing detailed reaction mechanisms is then outlined, followed by an extensive discussion on the development of reduced mechanisms and the associated strategies towards facilitated computation. Topics covered include skeletal reduction especially through directed relation graph; time-scale reduction based on the concepts of quasi-steady species enabled through computational singular perturbation; the lumping of isomers and of species with similar diffusivities; on-the-fly stiffness removal; the relative merits of implicit versus explicit solvers; and computation cost minimization achieved through tabulation and the judicious re-sequencing of the computational steps in arithmetic evaluations. Examples are given for laminar flames and direct numerical simulations of turbulent combustion to demonstrate the utility of the integrated strategy and the component methods in incorporating realistic chemistry of practical fuels in large-scale simulations, recognizing that the detailed mechanisms of these fuels may consist of hundreds to thousands of species and thousands to tens of thousands of reactions. Directions for further research are suggested.     

A level-set based adaptive-grid method for premixed combustion

A numerical method to simulate premixed combustion is analyzed. It consists of a Cartesian cut-cell flow solver for compressible viscous flows coupled with a level-set method which solves the G-equation to describe the kinematics of the premixed flame. The coupling of the two solvers is achieved via a dual hierarchical dynamic adaptive-mesh framework. Both solvers operate on different Cartesian hierarchical meshes sharing a common background grid level through which they are connected. For the flow solver, feature- and G-based adaptive mesh refinement is taken advantage of, while a uniform high-resolution grid is used for the level-set solver. The heat release due to combustion is described by a source-term formulation by which the reaction rate profile of the premixed flame can be attached to the flame front, the motion of which is governed by the G-equation. A flame–vortex interaction problem is discussed in detail to validate the proposed methodology and to demonstrate the benefits of solution-adaptive mesh refinement in the context of the level-set approach for premixed combustion. After a forced laminar Bunsen flame is considered as an example for attached flames, the coalescence of two spherical flame kernels is simulated to assess the performance of the method and the potential savings in terms of computational costs for three-dimensional problems. The results of the test problems show the artificial thickening of the flame and numerical errors in the level-set solution on coarser grids to possess a comparatively small impact on the overall accuracy. The best findings in the sense of efficiency and physical quality are achieved by the combined feature-/G-based adaptation method.