Urea Injection Simulation with Adaptive Mesh Refinement for Engine Aftertreatment

Transcription

1 ILASS Americas 26th Annual Conference on Liquid Atomization and Spray Systems, Portland, OR, May 2014 Urea Injection Simulation with Adaptive Mesh Refinement for Engine Aftertreatment Scott A. Drennan Convergent Science, Inc E. Common Street, Suite 1204 New Braunfels, TX USA Abstract Controlling NOx emissions from vehicles is a key aspect of meeting new regulations for cars and trucks across the world. Selective Catalytic Reduction (SCR) is an NOx reduction option that many engine manufacturers are adopting. The performance of urea injection and mixing upstream of an SCR catalyst is critical in obtaining reliable NOx reduction. Computational Fluid Dynamic (CFD) simulations of urea injection systems have become an important development and diagnostic tool for designers. Designers are interested in applying more accurate spray and kinetic models to their CFD simulations and in reducing mesh generation time. This paper presents the application of an automatically generated Cartesian meshing approach to a urea liquid injection system with RANS CFD simulation. Investigations of the impact of injection and mesh refinement strategy with RANS simulation is presented for a commonly accepted urea injected validation case. A modified cut-cell Cartesian method is used that eliminates the need for the computational grid to be morphed with the geometry of interest while still representing the true boundary shape. This approach allows for the use of simple orthogonal grids and completely automates the mesh generation process. The meshing approach also utilizes Adaptive Mesh Refinement (AMR) to resolve the domain near geometric features and in regions near the spray. AMR allows the use of a very fine grid in the vicinity of the spray while keeping the overall cell count relatively low.

2 Introduction Today s global regulations have made engine aftertreatment a necessity rather than an option for on and off-road powertrain systems. Engine aftertreatment for NOx in diesel fueled systems is looking to urea injection and Selective Catalytic Reduction (SCR) to achieve NOx reduction. The urea injection and SCR systems is often combined with a NOx trap. The urea injection/scr system is then operated when the NOx trap flushes itself and a large amount of NOx is emitted. The application of urea injection and SCR for NOx treatment is not new. Power boilers have used both urea injection with and without SCR for more than 10 years. The efficacy of the NOx reduction of a urea injection/scr system is often linked to the uniformity of the distribution of ammonia (NH3) upstream of the catalyst. The engine aftertreatment designer rarely has the ability to influence the surface chemistry inside of the SCR catalyst as these are most often supplied by third party vendors. However, the engine aftertreatment designer has control over the injection and atomization of the urea and how that urea decomposes into ammonia and is mixed prior to entering the SCR. The urea injection/scr system designer is working to balance the competing objectives of generating a uniform ammonia distribution upstream of the SCR catalyst and the overall pressure drop associated with the mixing strategy and the capital, operational and maintenance costs of the unit. Good urea/water injection and mixing is critical to produce the HCNO and NH3 gas phase mixture and then to promote decomposition of the HCNO into NH3. Substantial effort is expended to develop static mixers that are good a mixing the flow at a low pressure drop and are inexpensive to manufacture. Typical urea injection/scr mixers are complicated geometries yet simple in design; such as stamped thin metal plates with variations in all three directions. Background Computational Fluid Dynamics (CFD) offers the designer a valuable tool in the design and analysis of urea injection/scr systems. However, traditional CFD approaches use complex meshes that must be created in advance and can add weeks to the CFD simulation timeframe. It is also imperative the CFD analysis use the proper mesh and atomization models to capture spray behavior correctly. Finally, thin and complicated mixer plate designs add complexity and require more effort to mesh. Mesh generation represents a large portion of the CFD design flow in order to accommodate complex combustor and fuel injector geometries. Additionally, highly skewed non-orthogonal meshes slow CFD simulations while introducing computational errors. It is not uncommon for non-orthogonal, polyhedral meshes to have cells with negative volume. The optimal mesh for CFD simulation numerical accuracy and computational speed is Cartesian. The approach presented in this paper uses a Cartesian mesh that is automatically generated from combustor geometry and uses Adaptive Mesh Refinement (AMR) to deliver the required mesh resolution for the model being used. Modeling Approach In this work, the CONVERGE CFD software package [1] is used as the computational framework for running the spray and combustion simulations. CON- VERGE is a general purpose CFD code for the calculation of three-dimensional, incompressible or compressible, chemically-reacting fluid flows in complex geometries with stationary or moving boundaries. CON- VERGE can handle an arbitrary number of species and chemical reactions, as well as transient liquid sprays, and laminar or turbulent flows. CONVERGE uses an innovative modified cut-cell Cartesian method that eliminates the need for the computational grid to be morphed with the geometry of interest while still precisely representing the true boundary shape. This approach allows for the use of simple orthogonal grids and completely automates the mesh generation process. This section presents a brief overview of the mesh manipulation, numerical algorithms, and physical submodels used in the current work as these elements all contribute to the grid convergence behavior achieved. Fixed Embedding and AMR It is often desirable to add grid resolution locally in critical flow sections of the domain while leaving less critical sections relatively coarse. For example, in the present work, extra grid resolution was added to resolve the complex flow behavior at the nozzle exit, while leaving the remaining grid coarse to minimize simulation time. It is important to note that fixed embedding is specified in a small volume close to the nozzle and is only meant to seed the adaptive mesh refinement described below. In most cases, it is difficult to determine a priori where fixed grid embedding should be added. In these cases, Adaptive Mesh Refinement (AMR) can be applied. Ideally, a good AMR algorithm should add embedding where the flow field is most under-resolved or where the sub-grid field is the largest. The current flow solver estimates the magnitude of the sub-grid field to determine where embedding is added. A cell is embedded if the absolute value of the subgrid is above a user-specified value. Conversely, a cell is released (i.e., the embedding is removed) if the absolute value of the sub-grid is below a user-specified value. To limit the number of embedded cells, a maximum overall number of cells can be specified by the

3 user. With this feature, the user can specify the total number of cells desired in the simulation and AMR will determine where to put the embedding to both best resolve the flow field and meet the target number of cells. Numerical Algorithms In the present CFD solver, all computed values are collocated at the center of the computational cell. To prevent checker-boarding, the Rhie-Chow [2] algorithm is employed. The conservation equations are solved using the finite volume method. In the present study a second-order-accurate spatial discretization scheme is used for the governing conservation equations. In order to maintain stability, time accuracy is set to first order by running fully implicit. The transport equations are solved using the Pressure Implicit with Splitting of Operators (PISO) method. A geometric multigrid solver is used for the pressure solution. Variable time-stepping is used in the current study. The time-step is automatically calculated each computational cycle based on maximum allowed Courant-Friedrichs-Lewy (CFL) numbers for convection, diffusion and the speed of sound. Note that these CFL constraints are maintained for accuracy, and not for stability. In addition, spray and evaporation timestep control methods are used in the present simulations. The calculations in this study are run in parallel on distributed memory machines using the Message Passing Interface (MPI). An automatic domain decomposition technique is employed which allows for efficient load balancing throughout the calculation as the distribution of cells can change significantly due to AMR. Approach to Achieving Grid Convergence This section summarizes the key elements of the computational methodology that result in the grid convergent behavior demonstrated below. Specifically, the following items are critical to achieving grid convergence: 1. Adaptive Mesh Refinement (AMR) Demonstration of grid convergence can only be accomplished if cell sizes below the point of convergence can be simulated. AMR allows the use of a very fine grid in the vicinity of the spray while keeping the overall cell count relatively low. (see Figure 1) 2. Fully implicit momentum coupling Grid convergence cannot be adequately demonstrated if running with a fine mesh causes numerical instabilities. As described in [20], previous studies suffered from such instabilities when the cell size was on the order of the nozzle diameter or smaller. The current methodology utilizes a fully implicit liquid-gas momentum coupling approach to keep the simulations stable in the presence of small cells and high liquid volume fractions. Specifically, an iterative technique is used where the drag is calculated for all drops in a cell and the gas phase velocity is updated accordingly. This updated gasphase velocity is used to calculate drag on all of the drops in the cell which is then used to update the gas-phase velocity. This process is repeated until the drop and gasphase velocities converge to the specified tolerance. 3. Improved liquid-gas coupling The current methodology utilizes a Taylor series expansion to calculate the gas-phase velocity in the liquid-gas coupling calculations. 4. Spray-wall interactions Accounting for spraywall interactions is important in urea/water injection systems. Spray-wall interactions account in the CFD code account for droplet impingement, liquid pooling on the wall, pool evaporation along the wall and droplet rebound from the wall. (see Figure 2) 5. Conjugate Heat Transfer (CHT) As urea/water sprays can impact upon surfaces within the exhaust system, the mixer and the SCR catalyst, CHT must be employed to account for local variations in material temperatures that impact vaporization and reactions. The CFD code employs fully-coupled CHT for such analysis. CFD Model Description The CFD model used in the present study has been used extensively by the community as a validation case for urea/water injection and generation of ammonia upstream of a catalyst. The current model is of a simple pipe (300mm dia. vs. 6m long) with simulated exhaust flow at 673K entering from a circular inlet with a constant pressure outlet at 100kPa (Figure 3). Urea/Water spray (40%/60%) is injected with a Rosin-Rammler size distribution and average SMD of 44µm (Figure 4). Secondary breakup is accounted for with TAB breakup. The simulation results provide gas-phase and liquidphase local compositions, temperatures and exit distributions of NH3, HCNO and dispersed phase. Thermolysis of the urea/water spray is modeled where water and urea are evaporated and urea(l) decomposes generating gas-phase NH3 and HCNO. Subsequently, the HCNO converts to NH3 using a detailed chemical mechanism. Adaptive Mesh Refinement surrounding the spray and fixed mesh embedding are used to provide the proper amount of mesh resolution for the spray model. (see Figure 5). Results The urea/water hollow-cone spray exits from the atomizer and spreads nearly two-thirds of the width of the pipe initially (see Figure 6). As the urea/water spray translates down the pipe, the water is initially evaporated and drop size decreasing leaving urea(l) and gaseous ammonia and HCNO (see Figure 7). The impact of the evaporating water is to reduce the local tem-

4 perature and increase the concentration of urea(l) in the droplets which inhibits urea(l) thermolysis. The gas-phase reaction converting HCNO to NH3 is temperature and local gas composition dependent. Figure 8 shows side by side comparisons of the local gas-phase temperature, water vapor mole fraction and the NH3 mole fraction. Finally, the distribution of ammonia at the exit of the domain is plotted with the droplet size in Figure 9. Summary A urea/water injection system has been simulated using an approach with automatic mesh generation enhanced with Adaptive Mesh Refinement for gridconvergent CFD. The urea/water system simulated was a 40/60 blend (urea/water) injected through a hollowcone spray with heat transfer and detailed chemical kinetics to account for thermolysis of the urea liquid and decomposition to ammonia. Grid-convergent spray modeling in this application is achieved through the use of fixed mesh embedding and Adaptive Mesh Refinement on spray parameters. The Urea/Water spray was shown to evaporate water first and then urea liquid which then decomposed into HCNO and NH3. The resulting ammonia distribution at the exit was shown to be concentrated around the centerline without the benefit of a mixer. The automatic mesh generation approach with Adaptive Mesh Refinement and grid embedding around sprays and geometric feature can be used to simulate full urea/water systems with complex geometries without user-time spent in mesh generation. Conference, ICEF , Vancouver, Canada, Senecal, P. K., Pomraning, E., Richards, K., and Som, S., An Investigation of Grid Convergence for Spray Simulations using an LES Turbulence Model, SAE , Som S., Longman, D. E., Luo, Z., Plomer, M., Lu, T., Senecal, P. K., and Pomraning, E., Simulating Flame Lift-Off Characteristics of Diesel and Biodiesel Fuels Using Detailed Chemical-Kinetic Mechanisms and Large Eddy Simulation Turbulence Model, J. Energy Resour. Technol., 134, Senecal, P. K., Richards, K. J., Pomraning, E., Yang, T., Dai, M. Z., McDavid, R. M., Patterson, M. A., Hou, S., and Shethaji, T., A New Parallel Cut-Cell Cartesian CFD Code for Rapid Grid Generation Applied to In-Cylinder Diesel Engine Simulations, SAE , Pomraning, E. and Rutland, C. J., Dynamic One- Equation Nonviscosity Large-Eddy Simulation Model, AIAA J., 40, No. 4, April Pomraning, E., Development of Large Eddy Simulation Turbulence Models, Ph.D. Thesis, University of Wisconsin-Madison, J.Y. Kim, S.H. Ryu, and J.S. Ha. Numerical Prediction on the characteristics of spray-induced mixing and thermal decomposition of urea solution in SCR system. In Proc Fall Technical Conference of the ASME Internal Combustion Engine Division, Long Beach, California USA, F. Birkhold, et al. Analysis of the Injection of Urea water solution for automotive SCR DeNOx Systems: Modeling of the Two phase Flow and Spray/Wall Interaction SAE Acknowledgements The author wishes to acknowledge the assistance from colleagues Mingjie Wang, Shaoping Quan, Gaurav Kumar and Peter Kelley Senecal. References 1. Richards, K. J., Senecal, P. K., and Pomraning, E., CONVERGE (Version 2.1.0) Manual, Convergent Science, Inc., Middleton, WI, Rhie, C. M. and Chow, W. L., Numerical Study of the Turbulent Flow Past an Airfoil with Trailing Edge Separation, AIAA J., 21, , Senecal, P. K., Pomraning, E., Richards, K., and Som, S., Grid Convergent Spray Models for Internal Combustion Engine CFD Simulations, Proceedings of the ASME 2012 Internal Combustion Engine Division Fall Technical Figure 1: Adaptive Mesh Refinement of a spray for grid-convergent CFD

5 Figure 2: Spray-Wall interactions with Adaptive Mesh Refinement showing impingement, pooling and rebound Figure 3: Urea/Water injection CFD case Figure 4: Closeup of computational domain showing injector location, spray fixed grid embedding and the surface mesh

6 Figure 5: Closeup of the spray region fixed mesh embedding near the atomizer Figure 6: Droplet near-field in the Urea/Water spray

7 Figure 7: Urea/Water injection simulation results for droplet size and ammonia concentration Figure 8: Simulation results for temperature, water vapor and ammonia concentration

8 Figure 9: Visualization of the Urea/Water spray with ammonia concentration