Satbir Singh and Rolf D. Reitz Engine Research Center, Department of Mechanical Engineering, University of Wisconsin, Madison

Transcription

1 Comparison of Characteristic Time (), Representative Interactive Flamelet (RIF), and Direct Integration with Detailed Chemistry Combustion Models against Multi-Mode Combustion in a Heavy-Duty, DI Diesel Engine Satbir Singh and Rolf D. Reitz Engine Research Center, Department of Mechanical Engineering, University of Wisconsin, Madison Abstract Three different approaches for modeling diesel engine combustion are compared against cylinder pressure, NO x emissions, and simultaneous optical diagnostic images from a heavy-duty, DI diesel engine. A characteristic time combustion (KIVA-) model, a representative interactive flamelet (KIVA-RIF) model, and direct integration using detailed chemistry (KIVA-) were integrated into the same version of the KIVA-v computer code. In this way, the computer code provides a common platform for comparing various combustion models. Five different engine operating strategies that are representative of several different combustion regimes were explored in the experiments and model simulations. Comparison of simulated cylinder pressure and heat-release rates with the experimental results shows that all the models predict the cylinder pressure and heat release rate reasonably well. All the models predict NO x emissions trends very well but the absolute magnitudes are different from the experimental measurements. A comparison of the model predicted and experimentally observed in-cylinder details is also presented. The KIVA- model better predicted the details of the flame structure and emissions for the LTC conditions. INTRODUCTION One significant benefit of computational models is that they can reduce engine research and development time and resources through their predictive capabilities. They also complement experimental observations by providing information not available from the experiments []. Model predictions are reliable over a limited range of operating conditions, which depends upon the complexity of the model and the experimental data against which it was calibrated. As the engine community is exploring new engine technologies to enhance fuel conversion efficiency and reduce pollutant emissions, computer models are being challenged to perform over a wider range of engine operating conditions. These operating conditions range from conventional, hightemperature, heterogeneous diesel diffusion combustion to more homogeneous low-temperature combustion (LTC). This requires high-fidelity models that can cover an expanded range of combustion environments. In addition to maintaining their good predictive capability, short calculation times are also desirable. Multi-dimensional computational fluid dynamics (CFD) models have contributed to various aspects of engine development in recent years, with practical computational cost [,]. For example, a multi-dimensional computer code KIVA-v coupled with a genetic algorithm was recently used by Wickman et al. [] to optimize the combustion chamber bowl geometry for reduced emissions and improved fuel conversion efficiency. The predictive capability of multidimensional models is limited by the accuracy of their submodels, the most important of which are the turbulence, spray and combustion models. This paper presents a comparison of three previously proposed combustion models with experimental data. These models are the characteristic time combustion (KIVA-), CFD integrated with detailed chemistry (KIVA-), and the representative interactive flamelet (KIVA-RIF) models. In order to test the predictive capability of these models in different types of combustion environments, a set of experiments that belong to five different combustion strategies was designed. All three models are first validated against the measured cylinder pressure, heat release data and NO x emissions and a comparison with in-cylinder images of combustion is also presented. EXPERIMENTAL SETUP ENGINE AND OPERATING CONDITIONS A single-cylinder, direct-injection (DI), -stroke diesel engine based on a Cummins N-series production engine was used in this investigation. The engine has a bore of 9. mm and stroke of. mm. A schematic diagram of the engine is shown in Fig., and the complete description of the engine is available in Refs. [,]. The engine is equipped with a nonproduction, high-pressure, electronically-controlled, commonrail fuel injector. Specifications for the fuel injector are also available in Refs. [,]. The fuel used is EPA certified diesel fuel. Five different operating conditions were selected to be representatives of five modes of diesel combustion, as described in Table. Note that the intake-temperatures and pressures in Table are significantly higher than typical for production engines because of the low compression ratio of the optical engine.

2 Table Test Matrix High-T Short-ID High-T, Long-ID Low-T, Early-Inj. Low-T, Late-Inj. Low-T, Double-Inj. Speed (RPM) IMEP (bar) Injection Pressure (bar) Intake Temp ( C) 9 9 BDC Temp ( C) 9 9 Intake Pressure (kpa) 9 TDC Motored Temp. (K) 9 TDC Motored Density (kg/m) Peak Adiabatic Flame Temp. [K] SOI (ºATDC) , + Injection Quantity (mg), DOI (CAD), O Conc. (Vol %)... The first two conditions in Table are characterized as hightemperature because the charge gas is air (% oxygen by volume), and the resulting stoichiometric flame temperatures are relatively high. For the three remaining conditions, the oxygen volume fraction was reduced to.% via nitrogen dilution. Because of the dilution effect of the added nitrogen, the adiabatic, stoichiometric flame temperatures for the three conditions with.% oxygen are much lower than for the two undiluted conditions, and are therefore characterized as LTC conditions. OPTICAL DIAGNOSTIC TECHNIQUES A number of optical diagnostics were applied to investigate various in-cylinder processes of spray, combustion and emissions for operating conditions described in Table. A brief summary of these diagnostics is given in Table and a detailed description can be found in references given next to each diagnostic. COMPUTATIONAL MODELS The simulations were performed using the KIVA-v release code implemented with three different combustion models: KIVA- [], KIVA- [], and KIVA-RIF [9]. The hybrid Kelvin Helmholtz Rayleigh Taylor (KH-RT) model [] predicted the spray breakup and the RNG- k-ε model developed by Han and Reitz [] predicted the turbulence. The mesh was composed of about, computational cells at bottom dead center (BDC) with. x. x. mm cell size Table Optical Diagnostics Summary Figure. Experimental setup Diagnostic Laser-elastic (Mie) scattering [] OH PLIF [] Broadband (BB) PLIF [] Soot PLII [,,] Chemiluminescence [,] Soot luminosity [,9] Measurement Planar liquid-fuel spray imaging (LL) Planar imaging of OH radicals Planar imaging of unburned fuel and combustion intermediates Planar soot distribution Line-of-sight ignition locations Line-of sight hot soot distribution

3 in the bowl near TDC for adequate resolution. RESULTS AND DISCUSSION HEAT RELEASE RATE, CYLINDER PRESSURE AND NITROGEN OXIDE (NO X ) EMISSIONS Figures (a) to (e) show comparisons of the measured and the model predicted cylinder pressure and AHRR for all five operating conditions. For the KIVA-RIF model, the results are presented for a single flamelet () in the computational domain. For the high-t, short-id condition (Fig. a, all the models predicted the ignition delay reasonably well but the peak cylinder pressure is over predicted. For the high-t, long-id condition (Fig. b), KIVA- and KIVA-RIF predicted a shorter ID, and KIVA- predicted a longer ID than the experiment. The KIVA-RIF model predicted very high peak AHRR as compared to the KIVA- and the KIVA- models. For the three low-t conditions (Figs. c,d, and e), the cool flame (CF) ignition delay is under-predicted by the KIVA- and the KIVA-RIF models. The KIVA- model that uses the simplified SHELL model for predicting auto-ignition, does not predict a distinct CF heat release. For the low-t, early-inj. (Fig. c), and the first combustion event of low-t, double-inj condition (Fig. e), all the models predicted the second stage combustion (SSC) ignition delay accurately, but for the low-t, late-inj. condition (Fig. d), the KIVA-RIF model overpredicted and the KIVA- model slightly underpredicted the ignition delay. The peak AHRR prediction is different for each model. Generally, the KIVA-RIF and the KIVA- models predicted faster rate of combustion and higher peak AHRR. For the KIVA- model, the standard ignition and combustion model constants [] were slightly tuned to get a better prediction of the ID and rate of combustion, but they were kept the same for all of the operating conditions. Figure (f) shows the measured and predicted nitric oxide (NO x ) emissions for all the five operating conditions. The ppm and ppm horizontal lines on Figure (f) identify a break in the vertical scale. The scale for the low NO x levels is set from - ppm and for the high levels, it is set from - ppm. The two high-temperature operating conditions have high NO x emissions, while the three lowtemperature conditions have very low NO x emissions. The KIVA- model over-predicts NO x emissions for the high-t, short-id condition due to the over-prediction of cylinder pressure. Although the KIVA-RIF and the KIVA- models also over-predicted the cylinder pressure, due to their low in-cylinder peak temperature prediction [], the NO x emissions are lower than the KIVA- prediction and compare well with the experimental measurements. Similarly, for the high-t, long-id condition, KIVA- predicted NO x emissions are higher than the KIVA-RIF and the KIVA- predictions, but they match better with the experiment. For the three low-t conditions, the KIVA- model gives a slightly better prediction of NO x emissions. MODEL COMPARISON WITH IN-CYLINDER IMAGES The experimental investigation provided in-cylinder details of spray, auto-ignition sites, OH radicals, and soot distributions in the jet for all five operating conditions. However, in this paper, only a very brief comparison of the model predictions with the experimental observations is made. Additional detailed comparisons can be found in Refs. [,].. High-T, Long-ID Condition Figure shows comparisons of experimentally observed OH PLIF (green) and the line-of-sight soot luminosity (red) with the model predicted OH (green) and soot (red) in a plane along the fuel jet axis for the high-t, long-id condition (Table ). Since the KIVA- model does not predict OH, only the soot distribution is shown. Only one image at the time of peak AHRR is presented from the model predictions and the experimental observations (Figure b). The experimental image in the left column shows a high concentration OH ring at the jet periphery surrounding a high soot concentration region. This distribution of OH and soot is very similar to the conceptual model proposed by Dec []. The KIVA- model that uses fuel as soot pre-cursor, predicts soot throughout the jet. The soot model is activated when the temperature of the computational cell approaches K, and the local soot concentration is proportional to the amount of fuel in the cell. Comparison with the experimental images shows that the KIVA- model predicts soot over a larger region of the jet in disagreement with the experiment. Also, the model predicts soot in the upstream regions of the jet long after the end of combustion. This is also clearly in disagreement with the experimental observations. Both the KIVA- and the KIVA-RIF models predict a diffusion flame surrounding the soot-producing region. The predicted flame structures are somewhat different for the two models. The KIVA-RIF model predicted a broader OH distribution and the flame extends all the way back to the injector while the KIVA- model predicted a thinner diffusion flame that terminates some distance away from the injector in agreement with the experiments. The experimental image shows some weak fluorescence in the near nozzle area but it is believed that this fluorescence is from unburned fuel, and OH is not present near the injector. For the KIVA-RIF model, a large soot producing region overlaps with the OH regions as indicated by the light yellow colored regions of the image. On the other hand for the KIVA- model, the soot and OH regions are spatially displaced, as also seen in the experiments.

4 9 High-T, Short-ID (a) (b) High-T, Long-ID (c) Low-T, Early-Inj. 9 9 Low-T, Late-Inj (d) 9 Low-T, Double-Inj (e) NOx (ppm) ppm ppm High-T Short ID High-T Long ID Low-T Early Inj. (f) Low-T Late Inj. Low-T Double Inj. Figure. Cylinder pressure, AHRR, and NOx emissions for all the five operating conditions Figure. Comparison of the experimentally observed OH PLIF (green) and soot luminosity (red) with the model predicted OH (green) and soot (red). The white dot donates the injector location

5 Figure. Comparison of the experimentally observed liquid fuel (blue) and ignition chemiluminescence (green) with the model predicted liquid fuel (blue) and gas temperature (green) Figure. Comparison of the experimentally observed OH PLIF (green) and soot LII (red) with the model predicted OH (green) and soot (red) CONCLUSIONS. Low-T, Early-Injection Condition Figure shows simultaneous images of the experimentally observed liquid-fuel (blue) and naturally occurring luminous emission from chemiluminescence (green) and the model predicted liquid-fuel (blue color and red particles) and local gas temperature (green) in a plane along the fuel jet axis. The images are presented at the peak of cool flame (CF) heat release (Fig. c) and therefore, the temperature scale for the model images is set from K only. At the peak of the CF heat release, the experimental image shows strong chemiluminescence from mid-stream to downstream regions of the jet. The KIVA- and the KIVA-RIF models also predict CF ignition in similar regions of the jet; however, the actual region of high temperature is different for the two models. The KIVA-RIF model also predicts ignition near the injector tip, in disagreement with the experimental image. The comparison of the model predicted OH and soot distributions with the experimentally observed OH via OH PLIF and soot via soot LII is also given in Figure. The results show that both the KIVA- and the KIVA- RIF models predict the OH distributions very well. Again, in the experimental image, the fluorescence in the upstream regions is believed to be from unburned or partially burned fuel. Similar to the experiments, the KIVA- and KIVA-RIF models predict soot near the bowl edge but again, for the KIVA RIF model, some of the soot is produced in regions that overlap with the OH regions (yellow color). Three different approaches of modeling ignition and combustion in diesel engines were compared against multimode combustion experiments on a heavy-duty DI diesel engine. The models were implemented into the same version of KIVA-v to provide a common platform for comparing previously proposed ignition and combustion models. Five different engine operating strategies that are representative of several different combustion regimes were explored in the experiments and model simulations. The following conclusions can be drawn based on the observations:. All three combustion models give reasonable predictions of the cylinder pressure and heat release rate trends for all five combustion strategies modeled.. The NO x emission predictions by KIVA- follow the cylinder pressure prediction, i.e., overprediction of cylinder pressure results in overprediction of NO x emissions (e.g., the high-t, short-id condition). On the other hand, KIVA-RIF tends to under-predict NO x emissions. This is thought to be due to the use of a global scalar dissipation rate, which introduces excessive averaging into the physical domain. In spite of simplified representation of fuel chemistry, the KIVA- model predicts trends in cylinder pressure and NOx emissions reasonably well.. The KIVA- and the KIVA-RIF predicted OH and soot distributions are similar to the experimental observations, however, the flame structure predicted by the KIVA-RIF model is more smeared out. Also, the KIVA-RIF model predicts

6 ignition close to the nozzle, which was not observed in the experiments. REFERENCES. Miles, P., Choi, D., Megerle, M., RempelEwert, B., Reitz, R. D., Lai, M-C.D., and Sick, V., "The Influence of Swirl Ratio on Turbulent Flow Structure in a Motored HSDI Diesel Engine~ A Combined Experimental and Numerical Study," SAE Paper --, SAE Transactions,, No., pp. 9-9,.. O'Rourke, P. J. and Amsden, A. A., "Three Dimensional Numerical Simulations of the UPS-9 Stratified Charge Engine," SAE Paper 9, SAE Transactions, 9, No., pp. -, 9.. Reitz, R. D. and DiWakar, R., "Effect of Drop Breakup on Fuel Sprays," SAE Paper 9, SAE Transactions, 9, No., pp. -, 9.. Wickman, D. D., Senecal, P. K., and Reitz, R. D., "Diesel Engine Combustion Chamber Geometry Optimization using Genetic Algorithms and Multi-Dimensional Spray and Combustion Modeling," SAE Paper --,.. Dec, J. E., "A Conceptual Model of D.I. Diesel Combustion Based on Laser-Sheet Imaging," SAE Paper 9, SAE Transactions,, No., pp. 9-, 99.. Musculus, M. P. B., "On the Correlation between NOx Emissions and the Diesel Premixed Burn," SAE Paper --, SAE Transactions,, No., pp. 9-,.. Kong, S-C., Han, Z. Y., and Reitz, R. D., "The Development and Application of a Diesel Ignition and Combustion Model for Multidimensional Engine Simulations," SAE Paper 9, SAE Transactions,, No., pp. -, 99.. Kong, S-C. and Reitz, R. D., "Use of Detailed Chemical Kinetics to Study HCCI Engine Combustion with Consideration of Turbulent Mixing Effects. " J. Engng Gas Turbines and Power, (), -, 9. Kong, S. C., Kim, H., Reitz, R. D., and Kim, Y., "Comparisons of Combustion Simulations using a Representative Interactive Flamelet Model and Direct Integration of CFD with Detailed Chemistry," ASME ICES,.. Patterson, M. A. and Reitz, R. D., "Modeling the Effects of Fuel Spray Characteristics on Diesel Engine Combustion and Emissions," SAE Paper 9, SAE Transactions,, No., pp. -, 99.. Han, Z. Y. and Reitz, R. D., "Turbulence Modeling of Internal Combustion Engines using RNG k-e Models," Comb. Sci. Tech.,,, 99.. Espey, C. and Dec, J.E. The effect of TDC Temperature and Density on the Liquid-Phase Penetration in a D.I. Diesel Engine. SAE Paper 9, SAE Transactions, (), -, 99.. Dec, J.E. and Coy, E.B. OH Radical Imaging in a DI Diesel Engine and the Structure of Early Diffusion Flame. SAE Paper 9, SAE Transactions, (), -, 99.. Musculus, M.P.B. Multiple Simultaneous Optical Diagnostic Imaging of Early-Injection Low-Temperature Combustion in a Heavy-Duty Diesel Engine, SAE Paper P-,.. Dec, J.E. Soot Distribution in a D.I. Diesel Engine using -D Imaging of Laser-Induced Incandescence, Elastic Scattering, and Flame Luminosity, SAE Paper 9, SAE Transactions, (), -, 99.. Dec, J. E. and Espey C. Ignition and Early Soot Formation in a D.I. Diesel Engine using Multiple -D Imaging Diagnostics, SAE Paper 9, SAE Transactions, (), 99.. Dec, J.E. and Zion, P.L.K. The Effects of Ignition Timing and Diluent Addition on Late-Combustion Soot Burnout in a DI Diesel Engine based on Simultaneous - D Imaging of OH and Soot SAE Paper --,.. Dec, J.E. and Espey, C. Chemiluminescence Imaging of Autoignition in a DI Diesel Engine, SAE Paper 9, SAE Transactions,, Picket, L. M., Siebers, D.L., and Idicheria, C.A. Relationship between Ignition Processes and the Lift-Off Length of Diesel Fuel Jets, SAE Paper --,.. Singh, S, Reitz, R.D., and Musculus, M.P.B. Comparison of the Characteristic Time (), Representative Interactive Flamelet (RIF), and Direct Integration with Detailed Chemistry Models against Optical Diagnostic Data for Multi-Mode Combustion in a Heavy-Duty DI Diesel Engine, SAE Paper --,.. Singh, S, Reitz, R.D., Musculus, M.P.B., Lachaux, T., Validation of Three Different Engine Combustion Models Against Detailed in-cylinder Optical Diagnostics Data for a Heavy-Duty DI Diesel Engine, Submitted to IJER,.