Investigation of the Effect of Injection and Control Strategies on

Investigation of the Effect of Injection and Control Strategies on Combustion Instability in Reactivity Controlled Compression Ignition (RCCI) Engines D. T. Klos and S. L. Kokjohn Department of Mechanical Engineering University of Wisconsin-Madison Madison, WI 53706 USA Abstract. This paper uses detailed CFD modeling with the KIVA-CHEMKIN code to investigate the influence of injection timing, combustion phasing and operating conditions on combustion instability. Using detailed computational fluid dynamics (CFD) simulations, a large design of experiments (DOE) is performed with small perturbations in the intake and fueling conditions. A response surface model (RSM) is then fit to the DOE results to predict cycle-to-cycle combustion instability. Injection timing had significant tradeoffs between engine efficiency, emissions and combustion instability. Near TDC injection timing can significantly reduce combustion instability, but the emissions and efficiency drop to close to conventional diesel combustion (CDC) levels. The fuel split between the two DI injections has very little effect on combustion instability. Increasing EGR rate, while making adjustments to maintain combustion phasing, can significantly reduce PPRR variation until the engine is on the verge of misfiring. Combustion phasing has a very large impact on combustion instability. More advanced phasing is much more stable, but produces high peak pressure rise rates, higher NOx levels, and can be less efficient due to increased heat transfer losses. The results of this study identify operating parameters that can significantly improve the combustion stability of dual-fuel RCCI engines.

Introduction In the search for more efficient, cleaner engines, premixed compression ignition (PCI) engines have emerged with the potential to simultaneously increase engine efficiency while reducing NOx and soot emissions. Although promising, PCI engines have not been widely implemented due to difficulties controlling combustion phasing, combustion duration, and cycle-to-cycle variation. Recent work has shown that blending two fuels with different autoignition characteristics inside of the combustion chamber can provide control over both the combustion phasing and heat release rate. In this dual-fuel combustion process, the combustion phasing and duration are controlled by the autoignition characteristics, or reactivity of the charge; thus, it was named Reactivity Controlled Compression Ignition (RCCI) combustion. Providing control over combustion phasing and duration represented a significant advance in PCI combustion; however, research has shown that even when the mean combustion phasing is well-controlled, the cycle-tocycle fluctuations in the combustion phasing are still significant. This paper investigates how combustion instability is affected by injection timing, combustion phasing and operating conditions. Using detailed computational fluid dynamics (CFD) simulations, a large design of experiments (DOE) is performed with small perturbations in the intake and fueling conditions. A response surface model (RSM) is then fit to the DOE results to predict cycle-to-cycle combustion instability. Injection timing, combustion phasing and operating conditions are shown to all have significant effects on combustion instability. By having a better understanding of the tradeoffs between operating efficiency

and combustion instability, production LTC engines become a more realistic possibility. The results of this study identify operating parameters that can significantly improve the combustion stability of dual-fuel RCCI engines.

Engine and Simulation The engine simulated was a Caterpillar 3401E Single Cylinder Oil Test Engine (SCOTE). The engine is equipped with a conventional common-rail injector for delivery of diesel fuel and a port-fuel-injection (PFI) system for delivery of gasoline when operating in the RCCI combustion mode. The engine and injector specifications are summarized in Table 1. Table 1: Engine and injector geometry. Caterpillar 3401E SCOTE Displacement [L] 2.44 Bore x Stroke [mm] 137.2 x 165.1 Connecting Rod Length [mm] 261.6 Compression Ratio 16.1:1 Swirl Ratio 0.7 Common Rail Diesel Fuel Injector Number of Holes 6 Hole Diameter [μm] 250 Included Spray Angle [°] 145 Computations were performed using the KIVA-3v release 2 code [1] with improvements to many physical and chemistry models developed at the Engine Research Center (ERC) [2-4]. The KIVA-3v code is coupled with the CHEMKIN II solver for detailed chemistry calculations. In

1

this study, the chemistry of gasoline and diesel fuel are represented by that of iso-octane and n-heptane, respectively. A reduced reaction mechanism made up of 45 species and 142 reactions [3] describes the combined oxidation of nheptane and iso-octane. To reduce computation time, only a 60-degree sector mesh representing a single hole of the six hole injector was simulated. Previous work (e.g., [5, 9]) has shown excellent agreement between RCCI measurements and simulations using this modeling approach. Table 2 shows the operating conditions and the ranges for a 5 factor, 3 level, 35 full factorial DOE. The operating conditions are based on the experiments of Hanson et al. [5] at a load of 9 bar IMEP and a speed of 1300 rev/min. The baseline RCCI case used 42% EGR and 86% premixed gasoline (by mass). Table 2: Swept variables Baseline IVC Temperature [K] 351 Head/Liner/Piston Temp. [K] 550 Gasoline Mass [mg/cycle] 82.88 Diesel Fuel Mass [mg/cycle] 13.49 EGR % 41.74

+/2 25 4.14 1 2

Figure 1: RSM predicted output to CFD output.

The range of each variable corresponds to levels greater than the normal fluctuation that is expected during an engine running at steady-state conditions. Thirty operating conditions were simulated. Each operating condition consisted of a 5 factor, 3 level full factorial DOE with 35, (i.e., 243) KIVA runs for a total of 7290 KIVA runs. Each run took ~15 hours to complete. Normally, running this number of simulations would not be practical, but by using the Universities high throughput computing system, Condor, all of the simulations were completed in just over one week.

accurately over a wide range. CO and HC are predicted accurately for inputs that result in relatively low outputs of CO and HC; however, inputs that result in the highest CO and HC levels showed poorer prediction by the RSM. Cases showing high CO and HC are approaching misfire and evidently exhibit a very different response than the cases under normal operating conditions. Even with the deviation at the near misfire cases, the maximum error in the CO prediction is 11 g/kgf (i.e., only 0.25% of the input fuel energy). The maximum error in the HC prediction is 4 g/kgf (i.e., only 0.4% of the input fuel energy). At most conditions the error in CO and HC is below 0.1% of the input fuel energy.

Fitting a Response Surface Model

Reproducing cycle-to-cycle variations

Once the simulations were finished and post processed, they were used to build a Response Surface Model (RSM). A full quadratic RSM, which includes main effects, interaction effects, and quadratic terms, was fit to the data. Eq. 1 shows the response surface model for an output, 𝑦, with 𝑛𝑥 inputs, 𝑥. 𝑚𝑖𝑗 is the coefficient for the main effect term for 𝑖 𝑡ℎ output and 𝑗𝑡ℎ input, 𝐼𝑖,𝑗,𝑘 is the coefficient on the interaction term, and 𝑞𝑖,𝑗 is the coefficient on the quadratic term.

With reasonable agreement between the RSM and the CFD predictions, the RSM was then exercised to reproduce the cycle-to-cycle instability observed in the experiments of Splitter [8]. To reproduce cycle-to-cycle variations, the inputs to the RSM were sampled from a normal distribution with the mean set to the baseline value of Table 2 and an estimated standard deviation. The standard deviation (σ) of each variable was chosen as an estimate of actual variation that would be expected on a running engine and are given in Table 3. The variation of temperature, pressure, and EGR percent were taken from a combination of previous experience and the literature [6, 7]. Figure 2 shows the GIE, CA50, IMEP, and peak PRR for 200 measured cycles and 200 cycles using the RSM with estimated standard deviations in the operating conditions. Table 4 shows the mean and standard deviation of several measured and RSM predicted outputs. Comparing the RSM predicted cycle-to-cycle variation to that of the experiments, it is clear that, with small perturbations in the operating parameters, the variability

nx

nx 1

yi   mi , j x j   j 1

nx

I

j 1 k  j 1

nx

i , j ,k

x j xk   qi , j x j x j

(1)

j 1

The input variables, x’s, of the RSM are: 1) IVC temperature, 2) Gasoline fuel mass, 3) Diesel fuel mass, 4) EGR Rate, and 5) Combustion chamber surface temperature. The output variables, y’s, of the RSM are the common emissions and performance metrics (e.g., IMEP, CA50, GIE, PPRR, etc…). Figure 1 shows the CFD simulated output versus the RSM’s predicted output; a straight line would indicate a perfect match. Nearly all outputs are predicted very

2

observed in the engine experiments can be adequately reproduced.

Table 4: Comparison of measured and RSM predicted results for RCCI combustion using the standard deviation values shown in Table 3. Exp. RSM Mean Sigma Mean Sigma IMEP [bar] 9.25 0.21 9.03 0.21 CA50 [°ATDC] 0.909 0.35 1.13 0.37 PPRR 13.0 1.32 19.4 1.15 [bar/deg] GIE [%] 52.7 0.11 1 GIE_Exp [%] 53.7 1.22 53.3 1.23 EINOx [g/kgf] 0.422 0.218 0.067 EISoot [g/kgf] 0.138 0.001 2.0e-4 EIUHC [g/kgf] 12.86 11.27 0.52 EICO [g/kgf] 19.48 5.12 0.321 CA10 [°ATDC] -3.96 0.38 -2.14 0.37 CA90 [°ATDC] 4.72 0.73 4.12 0.45

Table 3: Input parameters mean and standard deviation values. Standard Mean Deviation IVC Temperature [K] 365 0.4 EGR Rate 40.1 0.4 Gasoline Mass [mg/cycle] 83 2.4 Diesel Mass [mg/cycle] 14 0.4 Combustion Chamber 500 2 Temperature [K] RSM

Experimental

Figure 2: Comparison of measured and RSM predicted results for RCCI combustion using the standard deviation values shown in Table 3.

Influence of Injection Strategy and EGR on Stability In order for LTC strategies to be widely used in production applications, cycle-to-cycle variability must be reduced. RCCI has a distinct advantage over HCCI in controllability, but further improvements are needed to improve the combustion stability. Using the RSM technique described previously, the different injection and control

strategies can be compared by their predicted variation. Two different injection strategies were evaluated: single and double injection. The injection timing for the single injection was swept from -70 to -10 degrees before TDC in steps of 10° CA. To isolate the influence of injection timing, the premixed fuel percent was varied for each injection timing to maintain CA50 at 1.25° ATDC. For the double injection strategy, the first injection was held constant at -58° ATDC and the second injection was swept from -47° to -7° ATDC. Similarly, the premixed fuel percent was varied for each injection timing to maintain a CA50 of 1.25 deg. For this sweep, 66% of the diesel fuel was injected in the first injection and 34% in the second injection. Figure 3 shows the predicted variation in IMEP, CA50, GIE and PPRR as a function of injection timing for the single and double injection strategies. Figure 4 shows the average emissions and performance (predicted by the RSM) as a function of injection timing for each injection strategy. As can be seen in Figure 3, injection timing has very little effect on IMEP variation for either injection strategy. For a very advanced or retarded single injection, however, the average IMEP decreases slightly. The single injection case with an SOI timing at -10° ATDC has limited mixing time and results in near stoichiometric regions that have high flame temperatures, increasing NOx emissions. This case also shows a significant increase in CO and UHC emissions. This increase is due to the low reactivity of the charge outside of the diesel jet, primarily in the outer region of the combustion chamber. When the diesel injection timing is near TDC, the diesel fuel is confined to the piston bowl region and the fuel in the outer region of the combustion chamber is primarily gasoline. Without the first diesel injection, the ignition delay of the gasoline in the outer region of the chamber is too long

1

The experimentally observed gross indicated efficiency (GIE_Exp) is the GIE calculated using the commanded fueling amount opposed to the exact cycle’s fuel quantity which can only be known in simulation.

3

to achieve complete combustion. Notice that the case with the split injection has nearly constant CO and UHC levels. The increased CO and UHC of the single injection case with a near TDC injection timing explains the drop in IMEP. The very advanced injection timing of -70 case also shows a drop in IMEP. CO and UHC are very low for this case and do not explain the drop in IMEP. Figure 5 shows the average heat loss percent as a function of injection timing. The heat lost to the liner increases significantly for the -70 injection case and explains the drop in IMEP for that case. The injection event is so early that the squish area has a much higher equivalence ratio than the piston bowl region and produces a high temperature zone in the squish region. However, since the surface area to volume ratio is very high in the squish area, heat transfer losses increase, resulting in lower IMEP. CA50 variation is greatly affected by injection timing. Injecting very late can significantly reduce CA50 variation because ignition occurs in near stoichiometric and rich regions. As the equivalence ratio is increased from very lean to near stoichiometric, the sensitivity of ignition delay to equivalence ratio decreases. To illustrate this, Figure 6 shows constantvolume ignition delay calculations at a range of equivalence ratios and PRF numbers. As the equivalence ratio is increased to unity and beyond, the ignition delay remains nearly constant. Although the late single injection case

shows improved stability, NOx, CO and UHC emissions increase significantly. In a double injection strategy, NOx increases to 4.5 g/kgf (~1 g/kW-hr) while CO and UHC are nearly constant. This benefit in CA50 variation is not linear. Retarding the injection timing from near -40° to -20°, CA50 variation increases slightly until the ignition source is rich enough to minimize the sensitivity to equivalence ratio. Excluding the very late injection conditions where NOx emissions are high, the minimum CA50 variation is around 50° ATDC for single injection and -37° ATDC for double injection. As the injection timing is retarded, the PPRR variation linearly decreases. This reduction, however, does not follow the reduction in average PPRR as seen in Figure 4. The cycle average PPRR is at a maximum at -50° ATDC for single injection and decreases slightly with more advanced injection timings and also with more retarded injection timings. The double injection variation trends are similar to the single injection trends, but much slower since only 33% of the DI fuel injection is changing. The effect of fuel mass split between the double injections was then tested. The injection timing was held at 58° and -37° ATDC as in the baseline case and the amount of fuel in the first injection was swept from 16.6% to 100% (i.e., a single injection). Similar to the other cases, the premixed fraction was adjusted to hold CA50 constant at 1.25° ATDC. Figure 7 shows the RSM predicted variation in IMEP, CA50,

Single Diesel Injection

Single Diesel Injection

Split Diesel Injection Split Diesel Injection

Figure 3: Predicted variation of IMEP, CA50, GIE and Peak PRR as a function of injection timing for a single injection strategy (top) and a split injection strategy (bottom).

Figure 4: Predicted average values as a function of injection timing for a single injection strategy (top) and a split injection strategy (bottom).

4

Figure 5: Average heat loss percent at different injection timings for a single and double injection strategies.

Figure 7: Predicted Variation of IMEP, CA50 and PPRR at different fuel mass splits for double injection.

9 PRF 0 PRF 25 PRF 50 PRF 64 PRF 80

8 Ignition Delay [ms]

7 6 5 4 3 2 1 0 0

0.5

1

1.5

2

2.5

Figure 8: Predicted cycle average PPRR at different fuel mass splits.

3

Equivalence Ratio [-]

Figure 6: Constant-volume ignition delay calculations illustrating the effect of equivalence ratio and PRF number on ignition delay [10]. and PPRR as a function of diesel fuel mass splits for the double injection strategy. As can be seen in Figure 7, the fuel split has little effect on IMEP or CA50 variation. However, PPRR variation slightly increases with more fuel in the first injection. This trend approximately matches a sweep from a single injection at -37° ATDC to one at -58° ATDC as shown in Figure 3. This shows that the PPRR variation is primarily a function of the mass average injection timing. This trend is not shown in the average PPRR seen in Figure 8. Average PPRR increases until a mass fraction of 0.66, the base case. Above 0.66 the squish region is generally richer than the piston bowl which accounts for the reduction in average PPRR. The effect of EGR on variation was investigated. The amount of EGR was swept from 11% to 61% and the premixed fuel ratio was adjusted to maintain a CA50 of 1.25° ATDC. Figure 9 shows the RSM predicted variation in IMEP, CA50, and PPRR as a function of EGR. IMEP and CA50 variation are not affected by EGR until greater than 60% EGR. At these high EGR levels the engine is on the verge of misfire and small perturbations to the operating parameters

can wildly change the combustion event. This can be seen by the significant increase in CA50 and GIE variation. IMEP variation at 61% EGR, however, significantly decreases. To explain the unintuitive reduction in IMEP variation with increasing EGR, Figure 12 shows the IMEP, CA50, and GIE calculated from the CFD vs. the total fuel mass for several selected cases at 51% and 61% EGR. For both levels of EGR IMEP increases with fuel quantity; however, the case with 61% EGR increases at a slower rate than the case with 51% EGR. CA50 retards much more in the 61% EGR case than the 51% EGR case and reduces GIE. In the 51% EGR case the change in CA50 is not enough to significantly change the GIE. The reduction of GIE with increasing fuel partially offsets the added fuel energy and these competing trends reduce the IMEP variation as observed in Figure 9. The absolute value of PPRR variation along with the average PPRR continually decreases with higher levels of EGR, as shown in Figure 10. PPRR COV, shown in Figure 11, however decreases to a minimum at about 50% EGR. Next, the injection timings, fuel splits, and EGR rate were held constant at the baseline conditions. The CA50 was then swept from 2° to 12° ATDC by changing the premixed fuel percent. Figure 13 shows the RSM predicted variation in

5

Figure 9: Predicted Variation of IMEP, CA50, GIE and PPRR at different levels of EGR%.

Figure 10: Predicted Average values of IMEP, Peak PRR, EICO and EINOx for different levels of EGR%.

Figure 12: CFD calculated IMEP, CA50 and GIE as a function of total fuel mass for 51% and 61% EGR.

Figure 11: PPRR COV at different levels of EGR IMEP, CA50, GIE, and PPRR as a function of CA50. As expected, IMEP and GIE variation increases with increased combustion phasing retard. CA50 variation increases until a maximum is reached at approximately 8° ATDC and then reduces. After a CA50 of 8° ATDC the number of partial misfires increases significantly, as can be seen by the large decrease in IMEP and increase in CO shown in Figure 14. For these calculations CA50 is calculated from the

normalized AHRR curve and therefore partial or near full misfires still have a calculable CA50. In reality some of these cases would not have even released half the fuel energy and would therefore not have a CA50 (if calculated based on the total fuel energy). Further work will develop an improved methodology to quantify combustion phasing stability capable of considering misfires and their impact on cycle-tocycle instability. PPRR variation slightly increases up to a CA50 of 6° ATDC and then falls off. At earlier combustion phasing, the combustion is much faster and consistent, as can be seen by the very high PPRR and lower CA50 variation. As the combustion phasing is retarded, the CA50 variation increases faster than the PPPR decreases, resulting in higher PPRR variation. After a CA50 of 6° ATDC, the average PPRR drops significantly and, although instability is high, the PPRR variation is lower. The PPRR COV, however, shown in Figure 15 continually increases with retarded combustion phasing.

6

Figure 13: Predicted Variation of IMEP, CA50, Peak PRR and GIE at different combustion phases.

used to simulate hundreds of cycles and reproduce the cycleto-cycle variation of a running engine. The effect of injection timing, combustion phasing and operating conditions on combustion instability were also investigated. Injection timing had significant tradeoffs between engine efficiency, emissions and combustion instability. Near TDC injection timing can significantly reduce combustion instability, but the emissions and efficiency drop to close to conventional diesel combustion (CDC) levels. The fuel split between the two DI injections has very little effect on combustion instability. Increasing EGR rate, while making adjustments to maintain combustion phasing, can significantly reduce PPRR variation until the engine is on the verge of misfiring. Combustion phasing has a very large impact on combustion instability. More advanced phasing is much more stable, but produces high peak pressure rise rates, more NOx and can be less efficient due to increased heat transfer losses. The results of this study identify operating parameters that can significantly improve the combustion stability of dual-fuel RCCI engines.

References 1.

Figure 14: Predicted average value of IMEP, Peak PRR, and EICO at different combustion phases.

Figure 15: PPRR COV at different combustion phases.

Conclusions In order to make RCCI and other advanced combustion strategies a reality in production applications, cycle-to-cycle variations must be thoroughly understood. To determine the main sources of variation, a full factorial DOE using detailed CFD modeling was performed and a RSM was fit to the data. The RSM was able to accurately predict several outputs such as IMEP, CA50, PPRR, NOx, Soot and UHC. It was then

Amsden, A. A. KIVA-3V, Release 2, Improvements to KIVA3V. LA-UR-99-915, 1999, LA-UR-99-915. 2. Beale, J. C. and R. D. Reitz. Modeling Spray Atomization With the Kelvin-Helmholtz/Rayleigh-Taylor Hybrid Model. Atomization and Sprays, 1999, 9, 623-650. 3. Ra, Y. and R. D. Reitz. A Reduced Chemical Kinetic Model for IC Engine Combustion Simulations with Primary Reference Fuels. Combustion and Flame, 2008, 155, 713-738. 4. Abani, N., A. Munnannur, and R. D. Reitz. Reduction of Numerical Parameter Dependencies in Diesel Spray Models. Journal of Engineering for Gas Turbines and Power, 2008, 130. 5. Hanson, R. M., S. L. Kokjohn, D. A. Splitter, and R. D. Reitz. An Experimental Investigation of Fuel Reactivity Controlled PCCI Combustion in a Heavy-Duty Engine. SAE 2010-010864, 2010. 6. Zhong L., Singh I.P., Han J., Lai M-C and Henein N.A. Effect of Cycle-to-Cycle Variation in the Injection Pressure in a Common Rail Diesel Injection System on Engine Performance, SAE paper No. 2003-01-0699 7. Denis Gill, Josef Erhart and Ludwig Burgler, Characterizing Injection System Behaviour under Real Operating Conditions. AVL document library 8. Splitter, Derek. High Efficiency RCCI Combustion. PhD Thesis in Mechanical Engineering, University of WisconsinMadison, 2012 9. Dempsey, Adam. Dual-Fuel Reactivity Controlled Compression Ignition (RCCI) with Alternative Fuels. PhD Thesis in Mechanical Engineering, University of WisconsinMadison, 2013 10. Kokjohn, S., Reitz, R., Splitter, D., and Musculus, M., "Investigation of Fuel Reactivity Stratification for Controlling PCI Heat-Release Rates Using High-Speed Chemiluminescence Imaging and Fuel Tracer Fluorescence," SAE Int. J. Engines 5(2):248-269, 2012, doi:10.4271/2012-01-0375.

7