International Multidimensional Engine Modeling User's Group Meeting April 20, 2015, Detroit, Michigan Progress in Predicting Soot Particle Numbers in CFD Simulations of GDI and Diesel Engines Abstract Karthik V. Puduppakkam, Abhijit U. Modak, Chitralkumar V. Naik, Long Liang and Ellen Meeks Reaction Design, a wholly owned subsidiary of ANSYS, Inc. 5930 Cornerstone Court West, Suite 230, San Diego, CA 92121 Tracking soot evolution in diesel and GDI engines using CFD simulations can provide valuable insights for engine design. ANSYS Forte CFD [1] includes a particle tracking capability based on the Method of Moments, thereby allowing calculation of spatially resolved soot volume fraction, number density and average particle diameter. Here we report recent progress in predicting soot in diesel and GDI engines, over a wide range of conditions. The cases include a BMW high-performance diesel engine over six operating conditions; an IFPEN diesel engine over four operating conditions; and an IFPEN GDI engine. We briefly present our approach to including more fundamental science knowledge to accurately capture the effects of varying fuel and engine conditions. This includes the use of multicomponent surrogates for gasoline and diesel fuels to capture fuel effects. For the gas-phase chemistry of the surrogates, we employ chemistry mechanisms with several hundred species that include soot precursors as large as pyrene. A 29-species soot-surface chemistry was developed that could be used for both diesel and GDI engine simulation. The particle-tracking model considers the statistical particle-size distribution on a cell-by-cell basis at each timestep and includes the effects of particle coagulation. Modeling fuel effects and chemistry Towards the goal of accurately predicting soot emissions, it is important to use a multicomponent fuel surrogate to mimic real fuel properties, such as aromatic content, and to accurately model the chemistry occurring in the gasphase and on the surface of soot particles. For the GDI case, a gasoline surrogate was assembled using the Surrogate Blend Optimizer [2]: 25/23.3/3.3/21.5/ 11.5/9.6/5.8 liquid volume% of iso-octane/n-pentane/n-butane/1,2,4-trimethyl benzene/toluene/methyl cyclohexane/ 1-hexene. This surrogate represents the gasoline properties well, as shown in Table 1. Table 1: Properties of the European gasoline tested and the surrogate fuel used in the simulation Properties Gasoline Surrogate RON 96.7 96 MON 86.3 90 H/C molar ratio 1.86 1.85 Lower heating value (MJ/kg) 43.0 43.4 Temperature 10% mass evaporated, K 324 340 Temperature 50% mass evaporated, K 367 380 Temperature 90% mass evaporated, K 437 435
For this gasoline surrogate, a detailed gas-phase chemistry consisting of 230 species and 1740 reactions, and a surface chemistry consisting of 11 species and 29 reactions were used in the simulations. Reaction Workbench software [2] was used to reduce the chemistry from a much larger and extensively validated master mechanism [3]. The same source of master gas-phase kinetics (the Model Fuel Library [3]) was used for all simulations to derive the reduced chemistry. The same soot surface mechanism was used for gasoline and diesel simulations; this points to fundamental soot surface chemistry steps being invariant of the fuel. A diesel surrogate blend was assembled to mimic the real fuel properties used in the BMW and IFPEN diesel engines, using the Surrogate Blend Optimizer [2]. The surrogate composition was 39/36.5/16.4/8.1 liquid volume% of n-hexadecane/decalin/hmn/amn. Table 2 shows that the resulting surrogate blend's properties agrees well with the real-fuel properties used in the BMW case [4], for cetane number, sooting tendency (H/C ratio), chemical class distribution, lower heating value and liquid density. Table 2: Comparison of surrogate properties with diesel fuel properties Diesel fuel Property Measured in this study Typical values from other sources Surrogate fuel Cetane number 52.2 54.3 H/C molar ratio ~1.8 [5] 1.87 Lower heating value (MJ/kg) ~43 [5] 43.7 Threshold sooting index 25.7 Liquid density (kg/m 3 ) 835 827 Temperature 10% mass evaporated (K) 490 487 Temperature 50% mass evaporated (K) 547 515 Temperature 95% mass evaporated (K) 622 560 Aromatics content (vol%) 15-40 [5] 8.1 n- and iso-alkanes content (vol%) 25-50 [5] 55.4 Naphthenes content (vol%) 15-60 [5] 36.5 Carbon number range 10-17 (9-24) [5] 10-16 For this surrogate, an accurate and detailed gas-phase chemistry consisting of 548 species and 3892 reactions, and a surface chemistry consisting of 11 species and 29 reactions were used for the BMW and IFPEN simulations. As with the gasoline surrogate mechanism, the diesel mechanism was extracted from the Model Fuels Library [3] using the CHEMKIN Reaction Workbench mechanism-reduction facility. Results and Discussion BMW diesel engine A high-performance BMW diesel engine was previously modeled using ANSYS Forte CFD, using a pseudo-gas soot model [4], where soot mass fractions and other emissions were well predicted by the model. The current work considers the same engine cases, but includes tracking of soot particle size distributions using the method of moments. The engine had a seven hole nozzle injector. A 51.4-degree sector mesh was used to represent the engine cylinder after intake valve closure and before exhaust valve opening. Details of ANSYS Forte CFD and its submodels can be found elsewhere [6]. Validation has been performed using these sub-models for high-pressure lifted flames [7]. In this modeling work, turbulence was modeled using the RNG (Re-Normalization Group) k-ε model. The gas-jet model was used that reduces grid dependency for the spray breakup model by directly modeling the gas entrainment into the spray jet. The solid-cone spray-injection model considers droplet-breakup governed by KH-RT sub-models and employs the radius-of-influence collision model. A discrete multi-component spray-vaporization model
considers the vaporization properties of each surrogate component. The same spray and turbulence model constants were used for all the cases simulated. Equations for all chemical species included in the detailed kinetics mechanism are directly solved in the CFD calculations through an operator-splitting method. Several advanced chemistry-solution techniques in ANSYS Forte CFD are used to provide maximum simulation efficiency. This includes the built-in sparse-matrix solver, which offers near-linear scaling between CPU time and the number of species, with zero loss of accuracy. The dynamic cell clustering (DCC) method [8] was also used to minimize the chemistry-related computational time. The accuracy of the DCC method in ANSYS Forte CFD has been previously reported [6]. Six cases have been modeled varying the amount of fuel, engine speed and EGR rate, as shown in Table 3. Typical simulation turnaround time was about 12-15 hours on 16 cores. The use of particle tracking capability based on the Method of Moments did not significantly increase the simulation time. Table 3: Suite of BMW diesel engine cases modeled Injected fuel External EGR Engine speed # of pilot (mg/cycle) (%) (rpm) injections Case 1 19.5 0 1800 2 Case 2 20.3 30 1800 2 Case 3 20.7 33 1800 2 Case 4 43.9 0 2500 1 Case 5 45.4 19 2500 1 Case 6 68.6 0 4400 0 Figure 1 shows the model predicting the trends well for NO x and soot across the six cases modeled. While soot magnitude is over-predicted by a factor of 2x, the trends are predicted well. Figure 1: Predicted emissions compared with measured values for the BMW diesel engine IFPEN diesel engine The IFPEN diesel engine modeling is a continuation of work done earlier by Naik et al., and the details can be found in reference [9]. Several cases were modeled for the IFPEN engine. Here, for brevity we focus on the representative Reference condition, which has 2.5 bar IMEP, 35% EGR, and start of injection at 353 CA degrees. Figure 2 shows the planar-averaged soot volume fraction; the planar-averaging for the model results have been done consistent with the experiments [9]. The predictions for soot as a function of crank angles agree very well with the engine data.
Figure 2. Predicted planar-averaged soot volume fraction for the Reference condition, in comparison with the IFPEN diesel engine data. Figure 3 shows the spatially-resolved values of soot, at various crank angles. The location of the cut plane is not shown here for the sake of brevity; details can be found in [9]. The agreement between the model and measurements for the magnitude and location of the soot cloud is very good. The legend is slightly different for the model values, with a peak of 2 ppm versus 3 ppm for data; this is well within the uncertainty shown for the measurements in Figure 2. Figure 3: Comparison of predicted planar soot volume fraction to measured values for reference conditions. Left panels show experimental data and right panels show the ANSYS Forte CFD results. More analysis of the soot evolution is shown below in Figure 4 and Figure 5. In Figure 4, soot total number density iso-surface of 10 11 particles/cm 3 is plotted for various crank angles. This figure shows that the soot is mainly formed in the engine bowl, and the effect of swirl is seen in the soot structure. Figure 5 shows the distribution of particles in various bins of average particle diameter. A large number of particles (~ 10 13 particles) of less than 1 nm average diameter are initially formed at 364 CA degrees through nucleation of soot precursors. Soot particle coagulation, growth and oxidation result in the soot evolution after this. With time, the small particles coagulate and grow to
larger particles, progressing to 1-2 nm range, then to 2-4 nm range, and so on. Particle oxidation results in a decrease of the number of particles and volume fraction after ~375 CA degrees, as seen in Figure 3 through Figure 5. Figure 4: Predicted soot particle number density iso-surface of 10 11 particles/cm 3 for the IFPEN Reference case Figure 5: Distribution of particles in various bins of average particle diameter for the IFPEN Reference case IFPEN GDI engine The IFPEN GDI modeling is a continuation of the work reported previously by Liang et al. [10], with the current work using updated chemistry. The GDI engine simulated is a 4-stroke 4-valve spray-guided IFPEN engine with optical access [11, 12]. The IFPEN GDI engine was operated at low-load conditions (4.3 bar IMEP). The overall equivalence ratio was 0.4. Start of injection was at 340 CA degrees and injection duration was 4 CA degrees. Simulation using ANSYS Forte CFD was carried out from before exhaust valve closing (EVC) to exhaust valve opening (EVO), including valve motion. Automatic mesh generation of a Cartesian-mesh with local mesh refinement provides a dynamic mesh with the moving boundaries, using ANSYS Forte CFD s Immersed Boundary method. The Linearized Instability Sheet Breakup (LISA) model was used to model the hollow-cone spray. The G- equation model was used to calculate the turbulent (premixed or partially premixed) flame propagation in the cylinder. Soot was found to be relatively insensitive to the G-equation model, and most soot emissions were found to be produced in the post-flame where we used detailed chemistry. Figure 6 shows the pressure/hrr predictions and soot predictions, respectively, for one of the GDI cases. Pressure and heat release rate have been predicted well within the experimental uncertainty. For soot, while the model over-
predicts soot volume-fraction magnitudes by a factor of ~3x, it captures the soot trend as a function of crank angle well. The soot measurements reported were from two cycles under average operating conditions, measured in a planar laser sheet. The scatter in the data is partly due to cycle to cycle variation [11]. Figure 6: Computed and measured in-cylinder pressure, heat release rate and soot traces for the IFPEN GDI engine Summary and Conclusions This paper presents a synopsis of recent progress made in our group towards modeling soot particle numbers in diesel and GDI engines. Towards the goal of more accurate soot simulations, realistic fuel surrogates and detailed chemistry have been used. Multicomponent fuel surrogates have been designed to represent real fuels for both gasoline and diesel. In addition to detailed gas-phase kinetic mechanisms, a 29-reaction soot surface chemistry has been used in the simulations. Statistics on soot particle size distributions have been tracked using the method of moments in ANSYS Forte CFD. The use of particle tracking capability based on the Method of Moments did not significantly increase the simulation time. Some results are presented for two diesel engines operating over a wide range of conditions, and a GDI engine. For some of the cases, the focus has been on predicting soot data at the exhaust, while for other cases we compare to time-resolved data for soot evolution as a function of crank angles. The comparisons with experimental data include spatially-resolved images of soot. The same source of gas-phase kinetics (the Model Fuel Library) was used for all simulations. The same soot surface mechanism was used for gasoline and diesel simulations. Simulations predict soot trends well over the wide range of conditions. References 1. ANSYS Forte CFD 40142: Reaction Design, San Diego, 2014. 2. Reaction Workbench 15131: Reaction Design, San Diego, 2013. 3. Model Fuels Library: Reaction Design, San Diego, 2013. 4. K. Puduppakkam, C. Naik, E. Meeks, C. Krenn, R. Kroiss, J. Gelbmann, and G. Pessl, SAE Technical paper 2014-01-2570, 2014. 5. C. V. Naik, K. V. Puduppakkam, C. Wang, J. Kottalam, L. Liang, D. Hodgson, and E. Meeks, SAE Technical paper 2010-01-0541, 2010. 6. L. Liang, C. V. Naik, K. V. Puduppakkam, C. Wang, A. Modak, E. Meeks, H.-W. Ge, R. D. Reitz, and C. J. Rutland, SAE Technical Paper 2010-01-0178, 2010. 7. C. V. Naik, K. V. Puduppakkam, and E. Meeks, Proceedings of ASME Turbo Expo, GT2014-26259, 2014. 8. L. Liang, J. G. Stevens, and J. T. Farrell, Combustion Science and Technology, 181: 1345-1371, 2009. 9. C. V. Naik, K. Puduppakkam, and E. Meeks, SAE International Journal of Engines, 6: 1190-1201, 2013. 10. L. Liang, C. V. Naik, K. V. Puduppakkam, A. U. Modak, and E. Meeks. Detroit, Michigan, 2014. 11. C. V. Naik, L. Liang, K. V. Puduppakkam, and E. Meeks, SAE Technical Paper 2014-01-1135, 2014. 12. C. Lacour, L. d. Francqueville, and V. Ricordeau, Congrès Francophone de Techniques Laser, CFTL, Vandoeuvre-lès-Nancy, France, 2010.