Two-stage ignition behavior and octane sensitivity of toluene reference fuels as gasoline surrogate

Combustion and Flame 210 (2019) 100–113

Contents lists available at ScienceDirect

Combustion and Flame journal homepage: www.elsevier.com/locate/combustflame

Two-stage ignition behavior and octane sensitivity of toluene reference fuels as gasoline surrogate Doohyun Kim a,∗, Charles K. Westbrook b, Angela Violi c a

Department of Mechanical & System Design Engineering, Hongik University, 94 Wausan-ro, Mapo-gu, Seoul 04066, Republic of Korea Lawrence Livermore National Laboratory, P.O. Box 808, Livermore, CA 94550, United States c Departments of Mechanical Engineering, Chemical Engineering, Biophysics Program, University of Michigan, 2350 Hayward St., Ann Arbor, MI 48109, United States b

a r t i c l e

i n f o

Article history: Received 12 May 2019 Revised 14 August 2019 Accepted 14 August 2019 Available online 28 August 2019 Keywords: Octane sensitivity Two-stage ignition Knocking Toluene reference fuel

a b s t r a c t Current approaches to improve the efficiency of Spark-Ignition (SI) gasoline engines have been focusing on turbocharging, increasing the compression ratio, and pursuing advanced low-temperature combustion concepts. In order to maximize these strategies, it is important to optimize the knock resistance of the fuel, and therefore knowledge of the sensitivity of the ignition process under a wide range of engine operating conditions is required. Octane sensitivity (OS), which is defined as the difference between Research Octane Number (RON) and Motored Octane Number (MON), has been introduced to represent how fuel’s ignition reactivity changes relative to the primary reference fuels (n-heptane/iso-octane) within RON/MON conditions. Previous works have indicated that OS is intimately related to low temperature reactivity of the fuel, which can be revealed as two-stage heat release characteristics during an ignition event. Prompted by these findings, in this paper, we investigate the relationship between two-stage ignition behavior and OS, using chemical kinetic simulations of 24 Toluene Reference Fuels (TRFs)/ethanol blends. TRFs are ternary mixtures of n-heptane/iso-octane/toluene, which is capable of capturing aromatic content and positive values of OS of real gasoline fuels. Simulation results show that fuels with weak or no two-stage ignition behavior tend to have high OS, due to their lack of Negative Temperature Coefficient (NTC) effect and high sensitivity in ignition delay time. Leveraging such observations, we develop a correlation between two-stage behavior and OS as an OS prediction method. Two metrics that represent the strength of the two-stage ignition behavior are proposed and used as OS predictors, which are Low Temperature Heat Release percentage (LTHR%) and Heat Release Rate at the end of first stage (HRRinf ) calculated from a simple kinetic simulation. Regression analysis shows a clear trend between decreases in the proposed two-stage behavior metrics and increases in the value of OS of the fuel. We also test the new metric (LTHR%) using simulation results of 0-D reactors with imposed pressure time histories obtained from engine experiments, as well as using different TRF kinetic mechanisms. The results demonstrate the effectiveness of the metric as a representation of the two-stage ignition behavior in practical combustion systems, highlighting the importance of the proposed relationship, and its potential as a simple and effective OS predictor. © 2019 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

1. Introduction Spark ignition (SI) engines have been the primary technology for light duty passenger vehicles for more than a century. The combustion efficiency of SI engines can usually be improved by operating them at increased compression ratios so that the combustion takes place at relatively high pressures. These conditions, however, have been found to lead directly to knocking operation, caused by premature autoignition of the unburned “end gases”, the last ∗

Corresponding author. E-mail address: [email protected] (D. Kim).

portion of the fuel/air mixture to be consumed by the propagating flame, before the flame front consumes those end gases. In addition to unpleasant engine noise, knock can lead to material damage of various engine components inside the combustion chamber and ultimately to engine failure [1]. Minimizing knock in SI engines is controlled by limiting the compression ratio or retarding the spark timing, usually at the expense of thermal efficiency. Also, such control is often difficult because each of the hundreds or thousands chemical species included in the gasoline fuel has a relatively different tendency to knock [2]. The knocking tendency of individual fuel components is usually quantified by using its octane number (ON). By definition, ON

https://doi.org/10.1016/j.combustflame.2019.08.019 0010-2180/© 2019 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

101

Table 1 Operating conditions for RON/MON measurements [3]. RON Measuring device Engine speed Intake temperature Spark advance a

MON

CFR octane rating engine, variable compression ratio 600 rpm 900 rpm 325 K (air) 422 K (charge) 19°–26° BTDC 13° BTDCa

Before top dead center.

compares the knocking behavior of a given test fuel with that of a binary mixture of n-heptane and iso-octane, the “primary reference fuels” or PRFs, under highly constrained engine test conditions, in a particular type of Cooperative Fuels Research (CFR) test engine [2]. The PRF n-heptane has a defined ON of zero (0), and iso-octane has a defined ON of 100, and the ON of any mixture of these PRFs defines the ON equal to the percent of iso-octane in the mixture. The ON of the fuel is therefore assigned to be equal to the ON of the PRF mixture that ignites at the same conditions in the CFR engine as the test fuel. Fuels with higher ON are generally more resistant to knock and slower to autoignite under elevated temperatures and pressures than fuels with lower ON values. The ON of a mixture of fuel components is measured similarly to the single-component fuel, using the fuel mixture as the test fuel. Two common octane numbers, Research Octane Number (RON) and Motor Octane Number (MON), are used to describe the knock resistance of fuels. While using the same CFR engine and binary mixtures of n-heptane/iso-octane as the reference fuels, engine operating conditions for RON and MON are different [3] as summarized in Table 1. Both RON and MON are representations of ignition characteristics relative to the PRFs, but RON is measured at slower engine speed and lower initial temperatures than MON, so RON compares fuel reactivities over a lower range of operating temperatures than does MON, while MON emphasizes higher temperature fuel ignition. Together, the two indices provide a complex metric of relative ignitability of SI engine fuels. A test fuel’s RON and MON are equal only when the ignition delay time of the test fuel is achieved with the same blending ratio of n-heptane/iso-octane in both RON and MON measurements. However, the ignition behavior of real transportation fuels is typically different from that of PRFs within nominal RON/MON test conditions, and the values of RON and MON are usually not equal. Thus, to further characterize the knock resistance of fuels in SI engines, Octane Sensitivity (OS) has been introduced, which is defined as the difference between RON and MON, i.e., OS = RON − MON. OS has been recently drawing attention as a fuel property that should be optimized to increase fuel efficiency. Particularly, it has been reported that high OS fuels have better knock resistance compared to other fuels with similar RON and lower OS when higher pressure/temperature operation is expected [2,4]. This result implies that high OS fuels have better knock resistance for the operation of modern SI engines with higher compression ratio and turbocharging. The other application where high OS fuel is desirable is in future engine operating concepts that employ dual modes of conventional SI and advanced low temperature combustion strategy. Noting that normal SI operation favors highly knock resistant fuel but HCCI mode requires good ignition quality (or poor knock resistance) [4,5], fuels with high OS, or high ignition sensitivity, can be very attractive for the SI/HCCI dual mode operation. Due to the empirical nature of RON/MON measurements and the definition of OS, a fundamental understanding of the physics and chemistry behind OS is still lacking. Previous works have indicated that the low temperature reactivity of a fuel, particularly its Negative Temperature Coefficient (NTC) behavior, is related to its OS value. The early investigation by Leppard [6] reported experiments using a modified CFR engine with paraffin, olefin, and

Fig. 1. Example of the two-stage ignition behavior depicted as calculated heat release rate of stoichiometric fuel/air mixtures of n-heptane and iso-octane at 25 atm, 775 K. Time axis is normalized by ignition delay time of the respective fuel.

aromatic fuels and showed that fuels with considerable NTC behavior had low values of OS and fuels without significant NTC behavior usually had high values of OS. By comparing ignition characteristics predicted by a detailed kinetic model, Mehl et al. [7] showed that the higher value of OS for 1-pentene compared with n-pentane was due to its weaker NTC effect. Later, Mehl et al. [8] utilized calculated ignition delay times from constant volume simulations using detailed kinetics to predict RON/MON of gasoline surrogate mixtures. Work by Szybist and Splitter [4] and numerical investigations by Westbrook et al. [2] and Singh et al. [9] also indicated that significant low temperature reactivity is strongly correlated with very small values of OS. While most small aromatic fuels have very weak low temperature reactivity and high OS in general, some aromatic molecules with adjacent methyl pairs attached to a benzene ring, such as 1,2-dimethylbenzene (o-xylene) and 1,2,4-/1,2,3-trimethylbenzene, possess noticeably stronger reactivity in low temperature regimes when compared to their structural isomers without adjacent methyl pairs. It has been shown that benzylperoxy radical → benzylhydroperoxide isomerization reactions (RO2 => QOOH), which promote chain branching in the low temperature regime, are possible for those molecules due to the proximity of two methyl groups, explaining their alkane-like NTC behavior and lower OS values compared with their isomers [10–13]. In a very recent paper, Tao et al. [14] studied the interconnected topics of octane numbers and influences of low temperature heat release on OS using chemical kinetic modeling. Under both SI and advanced compression ignition engine conditions, it also showed that smaller OS results in earlier low temperature heat release and shorter first stage ignition delay time. While NTC behavior is a low/intermediate temperature phenomenon expressed as total ignition delay time at multiple initial temperature points, two-stage behavior is another ignition phenomenon also caused by low/intermediate temperature chemistry [15,16] and can be observed in a single ignition event. Thus, it is reasonable to hypothesize that two-stage ignition behavior is also somehow related to OS similar to the NTC behavior. An example of a two-stage behavior is illustrated in Fig. 1 that shows the calculated heat release rates of stoichiometric mixtures of n-heptane/air and iso-octane/air with initial condition of 25 atm and 775 K, as functions of time, each normalized by its total ignition delay time. During the earlier part of the ignition process, the heat release rate increases, then peaks at a normalized time of about 0.85 for n-heptane and about 0.4 for iso-octane, and then decreases. This “first stage ignition” is the low temperature heat release phase. After a period of time, a second sharp increase in the heat release

102

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

rate occurs and full ignition is achieved due to the onset of high temperature oxidation chemistry. This well-known two-stage behavior begins with low temperature reactivity of both fuels, caused by the competition among low and intermediate temperature reaction pathways and their associated variabilities in chain branching vs chain propagation/termination and their differences in temperature dependency [17]. Not shown by the curves in Fig. 1, the actual ignition delay times for the two PRF fuels are quite different; nheptane ignites at 0.8 ms and iso-octane at 12 ms. In spite of these different ignition delay times, both fuels display two peaks in the heat release rate and a two-stage ignition. Prompted by these results, in this paper we investigate the chemical nature of octane sensitivity, focusing on the relationship between two-stage ignition behavior and the sensitivity of the ignition process to the temperature/pressure change within RON/MON measurement conditions. For this numerical study, we used 24 mixtures of n-heptane/iso-octane/toluene/ethanol with OS values ranging from 0 to 16, for which extensive experimental characterizations and kinetic mechanism development have been conducted [18–21]. While binary mixtures of n-heptane/iso-octane (PRF) may be the simplest option as a gasoline surrogate, PRF mixtures cannot capture the significant aromatic content and positive OS values of many real gasoline fuels. In fact, surrogate fuels consisting only of PRF components have, by definition, zero OS. Instead, ternary mixtures of n-heptane/iso-octane/toluene called Toluene Reference Fuel (TRF) can provide the opportunity of emulating aromatic content, RON/MON differences, and OS. Another common fuel component, ethanol, provides similar opportunities to vary RON/MON and OS when combined with PRF, TRF mixtures, or with other surrogate gasoline mixtures. The results below include the development of a metric that represents the strength of the two-stage behavior (low temperature heat release percentage), which can be used to predict OS of TRF/ethanol blends. This two-stage behavior metric is validated in various conditions using different kinetic mechanisms, providing insights on future kinetics development. This new prediction method can influence future surrogate fuel development to include the variability in ignition sensitivity of real fuels.

2. Sensitivity in ignition process 2.1. Simulation methodology Ignition characteristics of TRFs and their mixtures with ethanol were investigated using the 0-D homogeneous reactor model in the Chemkin package [22] and the detailed kinetic mechanism from ANSYS Model Fuel Library (MFL) [23] that includes 2631 species and 11,057 reactions and describes the oxidation process of various hydrocarbon classes pertinent to gasoline application (n-alkanes, iso-alkanes, alkenes, aromatics, alcohols). The reactor is a single zone, constant-volume, adiabatic system, with initial temperature/pressure of stoichiometric air/fuel mixture assigned as inputs. For this type of simulation, only the oxidation chemistry and the initial conditions of the air/fuel mixture determine the time histories of pressure and temperature during the whole ignition process. Note that this approach is much simpler than previous kinetic approaches to simulations of knock and octane sensitivity [2] in which the model is required to follow a pre-determined pressure/time history of a fuel/air mixture taken from real CFR engines following ASTM engine cycles [3]. Since the main factor that determines RON/MON/OS of a fuel is its chemical ignition characteristics rather than any engine-related dynamics, 0-D constant-volume, adiabatic reactor simulations should be sufficient for characterizing the ignition behavior of the fuel and the resulting octane sensitivity, although the exact trajectories of temperature and pressure of actual RON/MON testing using the CFR engine cannot be followed. Following Foong et al. [20], four TRF mixtures were investigated with an identical RON value of 91 but with varying OS from 0 to 7.6. For pure TRF mixtures, OS variation was implemented by varying toluene content from 0, 15, 30 and 45 vol%. In addition, TRF/ethanol blends containing as much as 80% ethanol were also studied, which further increase the mixture OS to values up to 16. The compositions of the 24 TRF blends with ethanol and the experimental values for RON, MON, and OS are summarized in Table 2. Beside their relevance as gasoline surrogates, these TRF/ethanol blends are appropriate for the current study due to their substantial range in OS from 0 to 16. Simulation results of all of these

Table 2 Compositions, RON, MON, and OS of TRFs and their blends with ethanol [20]. IDRON and IDMON are computed ignition delay times in 0-D homogeneous reactor simulations with assigned initial conditions (775 K, 25 atm for IDRON , 875 K, 25 atm for IDMON ) using MFL mechanism. SID is relative difference between IDRON and IDRON in percentage. Fuel name

TRF0-E0 TRF0-E10 TRF0-E20 TRF0-E40 TRF0-E60 TRF0-E80 TRF15-E0 TRF15-E10 TRF15-E20 TRF15-E40 TRF15-E60 TRF15-E80 TRF30-E0 TRF30-E10 TRF30-E20 TRF30-E40 TRF30-E60 TRF30-E80 TRF45-E0 TRF45-E10 TRF45-E20 TRF45-E40 TRF45-E60 TRF45-E80

Composition (volume fraction)

RON

toluene

iso-octane

n-heptane

ethanol

0.000 0.000 0.000 0.000 0.000 0.000 0.150 0.135 0.120 0.090 0.060 0.030 0.298 0.268 0.238 0.179 0.119 0.060 0.450 0.405 0.360 0.270 0.180 0.090

0.910 0.819 0.728 0.546 0.364 0.182 0.726 0.653 0.581 0.436 0.290 0.145 0.532 0.479 0.426 0.319 0.213 0.106 0.347 0.312 0.278 0.208 0.139 0.069

0.090 0.081 0.072 0.054 0.036 0.018 0.124 0.112 0.099 0.074 0.050 0.025 0.170 0.153 0.136 0.102 0.068 0.034 0.203 0.183 0.162 0.122 0.081 0.041

0.000 0.100 0.200 0.400 0.600 0.800 0.000 0.100 0.200 0.400 0.600 0.800 0.000 0.100 0.200 0.400 0.600 0.800 0.000 0.100 0.200 0.400 0.600 0.800

91 98.7 103.8 108 108.4 108.4 91 97.8 102.6 107.1 107.7 107.8 91.3 97 101.4 106 107.1 107.5 91.1 96 100.2 104.6 106.3 107.1

MON

91 94.3 95.3 94.5 93.4 92.2 88.4 91.7 93.2 93.6 92.6 91.7 86.1 89.4 91.1 92.1 92 91.4 83.5 87.2 89.1 90.9 91.2 91.1

OS

0 4.4 8.5 13.5 15 16.2 2.6 6.1 9.4 13.5 15.1 16.1 5.2 7.6 10.3 13.9 15.1 16.1 7.6 8.8 11.1 13.7 15.1 16

IDRON

IDMON

SID

(ms)

(ms)

(%)

6.7 9.2 12.6 22.0 34.7 51.2 6.9 9.4 12.7 21.6 33.0 47.0 7.7 10.4 13.7 22.5 33.6 47.0 9.9 12.8 16.3 25.2 35.8 48.2

6.5 6.6 6.6 6.7 6.8 7.1 6.5 6.5 6.4 6.4 6.3 6.4 6.7 6.6 6.5 6.4 6.3 6.3 7.6 7.2 7.0 6.6 6.4 6.3

2.0 28.5 47.4 69.5 80.3 86.2 5.5 31.2 49.4 70.5 80.8 86.3 13.4 36.2 52.4 71.7 81.3 86.5 23.5 43.3 57.2 73.7 82.2 86.9

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

103

TRF/ethanol blends depend primarily on the kinetic reaction pathways and rate parameters of n-heptane, iso-octane, toluene, and ethanol, all of which have been studied and validated extensively. Computed predictions of OS behavior using other, widely varying neat fuels and mixtures from different hydrocarbon classes for which reliable reaction mechanisms have been developed, such as those included in the previous, more complex modeling approach [2], should be feasible and will be pursued in future studies. For each of the present constant-volume, adiabatic simulations, the equivalence ratios of the air/fuel mixtures were set to 1. Ignition delay time was determined when the computed temperature increased by 400 K from the initial temperature. Computed values for Ignition Delay (ID) times are shown in Table 2, which will be discussed in the following sections. 2.2. Results As discussed above, OS represents the sensitivity of the test fuel’s ignition behavior relative to PRFs within RON/MON test conditions. The key to the present approach using simplified 0-D reactor is the selection of the initial conditions for the constant-volume ignition calculations that can properly represent thermodynamic conditions of RON/MON testing. For our study, the initial condition of 775 K, 25 atm was chosen as the RON case, and 875 K, 25 atm for the MON case. This choice was motivated by the need to approximate the thermodynamic conditions of actual RON/MON tests after compression based on the pressure and temperature histories of CFR engine experiments presented in Szybist and Splitter [4] and Mehl et al. [7]. To quantify the sensitivity of the ignition process within RON/MON conditions, comparisons were made between the RON case simulation (775 K, 25 atm) and the MON case simulation (875 K, 25 atm) for each fuel tested in Table 2. As a measure of the sensitivity, the relative change (or %difference) in computed ignition delay time from the RON case to the MON case is defined as ignition delay sensitivity (SID ) and is calculated as follows:

SID = ( (IDRON − IDMON )/IDRON ) × 100 where IDRON is ignition delay time at 25 atm, 775 K, and IDMON is ignition delay time at 25 atm, 875 K. Using the second mixture in Table 2 for TRF0-E10 as an example, IDRON = 9.2 ms and IDMON = 6.6 ms, so the value of SID = (9.2–6.6)/9.2 × 100, is equal to 28.5%. The value of OS for this mixture shown in Table 2 is 4.4, measured in engine experiments by Foong et al. [18]. By definition, a positive SID value indicates a shorter ignition delay time for the MON case. Also, SID is a normalized value relative to the ignition delay time of the RON case (775 K, 25 atm). Ignition delay sensitivity (SID ) of the 24 fuels reported in Table 2 is plotted in Fig. 2 as percentage difference in computed ignition delay times versus octane sensitivity. While pure TRFs without ethanol are placed a little off line (4 lowest SID points) compared to the rest of the data points, the simulation results show a linear trend in general, with high OS fuels presenting a larger difference between ignition delay times computed under RON conditions than MON conditions. In other words, the higher the value of OS, the higher is the sensitivity of the ignition delay time to the change of conditions within the thermodynamic regime relevant to RON/MON measurement. In addition, results reported in Fig. 2 confirm the ability of the reaction mechanism used in this study to provide accurate values for the ignition delay times since the values of RON, MON, and OS in Table 2 are measured in CFR engine experiments [20]. Figure 3 shows the calculated ignition delay times of two families of fuels taken from Table 2 at 25 atm over a range of initial temperatures, together with vertical lines indicating those temperatures corresponding to the RON and MON cases of constant-volume simulations. These computed results show that

Fig. 2. Ignition delay sensitivity (SID ) of RON-representative (775 K, 25 atm) and MON-representative (875 K, 25 atm) cases. X-axis is measured OS from Foong et al. [20].

strong NTC behavior of low OS fuels can produce small differences in ignition delay times between the temperatures used to measure RON and MON. Fuels with lower OS values reported in Fig. 3(a) exhibit smaller ignition delay time differences between the RON case (775 K) and MON case (875 K) due to strong NTC behavior between those two temperature conditions. The computed ignition delay time actually increases (i.e., slower ignitions) when the initial temperature increases from 775 K to about 850 K for TRF0-E0 (OS=0) and TRF15-E0 (OS=2.6), illustrating the well-known NTC behavior of the overall autoignition reaction. For fuels with higher OS values, such as TRF30-E0 (OS=5.2) and TRF45-E0 (OS=7.6), only small increases or monotonic decreases in ignition delay time as function of the temperature are observed. Such behavior is more pronounced in Fig. 3(b) that shows the ignition delay trends of TRF0/ethanol blends. Indeed, fuels with higher OS show little or no NTC behavior. These results demonstrate that non-monotonic changes in ignition delay time, including NTC behavior, lead to net differences in total ignition delay time that are smaller for fuels with strong NTC behavior than for fuels with negligible NTC behavior. In addition, the relationship shown in Fig. 2 indicates that the strength of the NTC characteristics and OS are closely related, which agrees with previous analyses reported in the literature [4,7]. This overall discussion revolves around the curvature in plots of the ignition delay times for different fuel mixtures. When such a plot has significant curvature, that shape brings the lower temperature part of the curve lower (i.e., to shorter ignition delay times), therefore nearer to the ignition delay time at the higher temperature and therefore with considerably smaller OS. All of this discussion is really a result of the initial temperature required for RON simulations at which the differences between fuels with and without NTC kinetics is greatest and the fact that the MON test covers temperatures above most NTC kinetics behavior. Another way to understand this point is to examine the trends in RON and MON for the four “families” of fuel mixtures in Table 2, specifically the TRF0, TRF15, TRF30, and TRF45 mixtures. Within each family, the RON values (which are ASTM test results) vary from 91 to about 108, a range of 17 for each family, while the MON values vary by a much smaller range of approximately 5–8, while the values of OS vary by 10–16 as the amount of ethanol increases from zero to 80%. Thus, the major contributions to the increased OS is due to differences primarily in RON, in agreement with the graphical analysis of the ID curvature. Figure 4 shows the calculated heat release rate profiles corresponding to the two families of fuels in Fig. 3 at the initial

104

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

Fig. 3. Calculated ignition delay times of stoichiometric air/fuel mixtures of (a) TRF0-E0–TRF45-E0 (OS = 0–7.6) and (b) TRF0-E10–TRF0-E60 (OS = 4.4–15) at 25 atm.

Fig. 4. Calculated heat release rate profiles of (a) TRF0-E0–TRF45-E0 and (b) TRF0-E10–TRF0-E60 from the RON case (775 K) and MON case (875 K). Time axis is normalized by ignition delay time of the RON case of each fuel.

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

temperatures used to characterize the RON and MON cases. Note that Fig. 4(a) shows pure TRFs without ethanol (all with 91 RON, zero to 45% toluene, and varying OS from 0 to 7.6), and Fig. 4(b) shows TRF0/ethanol blends (varying RON from 98.7 to 108.4, with 10% to 60% ethanol, and varying OS from 4.4 to 15). These are the points in Fig. 3(a) and (b) where the ignition delay curves intersect the vertical lines denoting initial temperatures in the RON (775 K) and MON (875 K) cases. The values on the x-axis for both RON and MON cases are normalized by the ignition delay time of the RON case to highlight their relative differences. In Fig. 4(a), it is clear that TRF0-E0 shows very similar total ignition delay times at 775 K and 875 K, despite significantly different heat release rate profiles. On the RON-representative condition (775 K), TRF0-E0 shows clear two-stage behavior, while a monotonic increase in heat release rate is observed for the MON-representative condition (875 K) over the whole ignition delay period. Fuels with higher values of OS show a bigger difference in total ignition delay time, as reported in Fig. 4(a). Going from TRF0-E0 to TRF45-E0 by adding toluene, the two-stage behavior for the RON case (initial temperature of 775 K) represented by the reduction in heat release rate after the first heat release peak (i.e., the NTC curvature in the heat release curve) gets weaker as OS increases, which is due to the fact that neat toluene fuel being added does not produce any NTC behavior and ignites in a single stage. This trend is even clearer in Fig. 4(b) with TRF0/ethanol blends. Ethanol, like toluene, is a single stage ignition fuel by itself, unlike large alkanes, and increasing the fraction of ethanol leads steadily to weaker two-stage ignition behavior. When ethanol content is higher than 40%, even the RON case at 775 K becomes a single-stage ignition process, indicated by monotonic increase in heat release rate. Consistent with numerous previous studies [2,20], the current simulations show that ethanol blends with PRF fuels are particularly effective at producing high OS mixtures that have great sensitivity in ignition delay time within the RON/MON regime. Despite the empirical definitions of RON and MON relative to the reactivity of PRF, the above simulations show that OS can be simply regarded as the sensitivity of the total ignition delay time under engine-relevant conditions. Thus, the relationship between OS and the ignition delay time sensitivity reported in Fig. 2 may be utilized to develop an OS prediction method. However, the such method requires an arbitrary selection of the initial conditions, which inherently differs from those of actual RON/MON measurement devices where temperature and pressure varies because of the motion of the piston. Alternatively, Fig. 4 implies that the strength of the two-stage ignition behavior depicted in the heat release rate profiles may also be a suitable OS predictor. Indeed, in the next section, the characteristic heat release rate profiles will be parameterized to derive a metric that represents the strength of the two-stage ignition behavior and can be used as a potential OS predictor. 3. OS prediction method for TRF/ethanol mixtures 3.1. Metrics for the two-stage ignition behavior Low Temperature Heat Release percentage (LTHR%), defined as (total heat release during the first stage of the overall ignition) / (total heat release until full ignition), has been used to quantify the strength of two-stage ignition behavior [24], but there is no formal definition to identify the end of the first stage heat release phase. One option is to employ the times at which the local maximum or local minimum of the heat release rate occurs during the ignition process, as marked for the TRF0-E0 in Fig. 5. However, when the heat release profile monotonically increases until ignition occurs, as shown as the lower curve in Fig. 5 for TRF0-E80, a local maximum or local minimum does not exist, and the LTHR% cannot be

105

Fig. 5. Computed heat release profiles for a two-stage (TRF0-E0) and a single-stage (TRF0-E80) ignition process at 800 K, 20 atm. Time axis is normalized by ignition delay time of the respective fuel.

defined in this way. Considering that high OS fuels are expected to go through single-stage ignitions, a more robust definition is necessary to incorporate such cases. Therefore, in this study, we have observed that when there is no local minimum or maximum in the heat release curve, there is still an inflection point in that curve, and we have used this inflection point of the log scale heat release rate profile as the definition for the end of the first stage heat release phase. In this way, LTHR% can be defined for both two-stage and single-stage cases as illustrated by the triangles on both curves in Fig. 5. In addition to LTHR%, the value of the heat release rate at the inflection point (the end of the first stage) is also tested as a secondary metric, which may roughly represent the low temperature reactivity of the fuel. However, LTHR% may be a more comprehensive metric to describe the strength of the two-stage ignition, since LTHR% takes account of the low temperature reactivity and the timing of when the heat release rate slows down, while the heat release rate at the inflection point describes only the strength of the initial reactivity. In this study, detection of the inflection point and the subsequent calculation of LTHR% were not manually executed, but rather using scripts that batch-processes heat release rates of the fuel mixtures tested. The inflection point was defined by the local minimum of the first derivative of the log scaled heat release rate profile. 3.2. OS vs two-stage behavior metrics In order to determine an appropriate initial condition for the homogeneous constant-volume simulations to explore the relationship between OS and the two-stage ignition behavior metrics, we investigated the relationship between OS and LTHR% and OS and HRRinf (Heat Release Rate at the inflection point). Initial conditions of 10–30 atm and 775–900 K were explored, which are approximate ranges for thermodynamic conditions of actual RON/MON measurements [4,7] covering the NTC regime. To understand the effect of temperature, a temperature sweep at a common initial pressure of 20 atm was performed. Figure 6 shows computed results at initial temperatures of 775 K, 825 K, and 875 K at 20 atm. No clear and strong trends could be identified at 875 K and 825 K. At 775 K, both OS–LTHR% and OS–HRRinf show clear inversely proportional trends. Also, the distributions in magnitude of the metrics are wider, which may provide better resolution for the OS prediction method. Fig. 4 in the previous

106

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

Fig. 6. (a) OS vs LTHR% and (b) OS vs heat release rate at the inflection point at 20 atm. Temperature was varied between 775 K and 875 K.

Fig. 7. (a) OS vs LTHR% and (b) OS vs heat release rate at the inflection point at 800 K from 10 atm to 30 atm.

section showed that when the initial temperature is higher within the tested conditions, the two-stage behavior gets weaker, and none of the curves in Fig. 4 with initial temperatures of 875 K showed any two-stage ignition behavior. Two-stage behavior is caused by a transition of preferred kinetic reaction pathways as the charge temperature increase from low to high temperature regime during the ignition delay period, specifically through the competition between QOOH → Q or carbonyl or cyclic ether vs QOOH → O2QOOH → ketohydroperoxide for alkanes [17]. Thus, it is clear that the lowest initial temperature in the NTC regime is necessary to parameterize the strength of two-stage behavior. As the second step, we analyzed the effect of pressure on the metrics. Initial pressures of 10, 20, and 30 atm were studied at a common initial temperature of 800 K, as shown in Fig. 7. At the lowest pressure tested of 10 atm, computed LTHR% values are similar for all fuels regardless of their OS values, and a slightly decreasing trend is observed for the heat release rate at the inflection point. As pressure increases, we observed a well-defined inversely proportional trend between the two-stage metrics and fuel OS. This result agrees with experimental studies that observed that the NTC regime of typical transportation fuels shifts toward higher temperature ranges as the pressure increases [15,25]. Various detailed chemical kinetic mechanisms currently available, including the MFL mechanism used in this study, also capture this pressure effect on the NTC temperature range [25,26]. As an ex-

Fig. 8. Calculated ignition delay time of stoichiometric TRF0-E0/air mixture from 10 to 30 atm.

ample, Fig. 8 shows ignition delays of stoichiometric TRF0-E0/air mixture calculated by the MFL mechanism and a homogeneous constant-volume reactor model for initial pressures from 10 to

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

107

Fig. 9. (a) OS vs LTHR% and (b) OS vs heat release rate at the inflection point at 25 atm, 775 K for 24 TRF/ethanol mixtures. The solid lines are the regression equations generated by curve-fitting the calculated two-stage behavior metrics to the measured OS.

30 atm. The results show that as pressure increases, the beginning and end points of the NTC behavior moves toward higher temperature, which implies that the transition from the low/intermediate temperature regime to the high temperature chemistry regime occurs at higher temperature. In Fig. 8, the location of 800 K as well as the initiation of the high temperature oxidation chemistry denoted by the roll off points of the Arrhenius plot are shown. At 10 atm, 800 K is very close to the roll off point of the ignition delay curve, which indicates that the high temperature chemistry is quick to start when simulation starts at 10 atm, 800 K. As pressure goes up, the distance from 800 K to the roll off point increases, making it a more favorable condition to observe the effects of the low/intermediate to high temperature regime during the ignition process, which causes two-stage ignition behavior. The above analysis shows that within the initial temperature/pressure conditions tested, the proposed two-stage behavior metrics based on LTHR% and the heat release rate at the inflection point, show well-defined correlations with OS when temperature is low and pressure is high. Considering such observations as well as the realistic engine operating conditions, we chose 25 atm and temperature of 775 K as the nominal initial conditions for generating OS predictions. Figure 9 reports OS vs the two-stage behavior metrics at 25 atm, 775 K. Using the two-stage behavior metrics as the OS predictors, regression equations were derived by curvefitting the calculated two-stage behavior metrics to the measured OS values. For simplicity, a second order polynomial was used for regression. The regression equations and the goodness of fit represented by R2 are as follows:

blends are required to increase the number of training sets. More specifically, experimental and computational analyses of different TRF/ethanol mixtures with identical RON/MON would be a great contribution to this topic. However, clear inverse trends and high R2 values demonstrate that the proposed two-stage behavior metrics derived from a simple homogeneous reactor simulation seem very promising as an OS predictor. 4. Applicability of the two-stage behavior metric 4.1. Pressure-imposed 0-D simulations To highlight that the LTHR% results investigated in the previous section are fundamentally sound OS predictors and, consequently, are widely applicable to different types of simulations, the methodology was applied to homogeneous reactor simulations with imposed pressure time histories from RON/MON-like engine experiments presented in Westbrook et al. [2]. Note that the simulations in previous sections were carried out with only initial temperature and pressure, but no pressure time history was imposed. The cylinder pressures from the RON/MON-like engine experiments are shown in Fig. 10. Similar to the standard RON and MON test

OS = −0.04032(LTHR% )2 + 0.2314(LTHR% ) + 15.46, R2 = 0.9806

OS = 2.831 × 10−19 (HRRinf )2 − 4.402 × 10−9 (HRRinf ) + 16.42, R2 = 0.9802 where HRRinf stands for the heat release rate at the inflection point. It should be noted that the above regression equations should be regarded as preliminary OS prediction methods for TRF/ethanol blends and cannot be generalized to other types of fuels. To improve the robustness of the regression equation, more experimental RON/MON measurements for additional TRF/ethanol

Fig. 10. Pressures of RON/MON-like engine experiments used for constant volume simulations.

108

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

Fig. 11. Octane sensitivity vs LTHR% from (a) RON-like and (b) MON-like simulations.

procedures, RON-like experiments were performed at 600 RPM, and MON-like experiments at 900 RPM. More details on the experiments are available in [2]. The constant-volume simulations of imposed pressures from RON/MON-like cycles were initialized at 502 K, 3.06 atm for the RON-like case, which corresponds to 65.3° BTDC of the RON-like engine experiment, and 501 K, 2.07 atm for the MON-like case, which corresponds to 80.6° BTDC of the MONlike experiment. While pressure profiles are given as input, the temperature profiles inside the constant-volume reactor are calculated based on the pressure change and the heat release from the oxidation chemistry. Westbrook et al. [2] showed that this simulation method is effective to reproduce the actual RON/MON measurement conditions with 0-D reactor simulations, and is able to predict RON of a test fuel using the calculated ignition timing. The present MFL mechanism was used for comparison with those simulations for the same 24 TRF/ethanol fuel blends, and the same method for detecting the inflection point (the local minimum of the first derivative of log scaled heat release rate profile) was used as described above. Figure 11 shows OS vs LTHR% from RON-like and MON-like simulations with imposed pressure time histories. The inversely proportional LTHR%-OS trend is similar to the one reported in Fig. 9(a) for RON-like simulations. However, MON-like case shows some inverse LTHR%-OS trend with significant scatter, and LTHR% values are very small, mostly less than 1%. Such weak LTHR%-OS relationship for MON-like simulation is due to its temperature conditions. Figure 12 shows heat release rates, temperatures, and pressures of RON/MON-like simulations of TRF0-E0. For the MON-like case, the charge temperature and pressure reach around 850 K and 20 atm during the initial phase of the heat release (0.013–0.015 s). As indicated in the previous section, such high temperature is not suitable for observing the two-stage ignition behavior due to the quick initiation of the high temperature chemistry. Heat release profile in Fig. 12 also shows that even TRF0-E0 that has OS of 0 and the strongest two-stage behavior among all the fuels tested, results in monotonic increase in heat release rate in MON-like simulations. This result implies that not only for simulations but for experiments, thermodynamic conditions relevant to MON experiments are not appropriate to investigate the two-stage ignition behavior and its relationship with OS, since the contribution of low/intermediate temperature chemistry to the ignition process is considerably small. On the other hand, apparent two-stage ignition behavior is revealed in the RON-like simulations. Since the RON-like pre-ignition charge temperature

Fig. 12. Time history of heat release rates, temperatures, and pressures of RON/MON-like simulations of TRF0-E0. Markers in the heat release rate profiles indicate the inflection points.

before combustion is low enough, the ignition process should go through all low/intermediate/high temperature chemistry regimes before the ignition is achieved, which causes the distinctive two-stage behavior to occur as extensively discussed in previous sections. The clear decreasing trend of the LTHR% from the RON-like simulations shown in Fig. 11(a) is encouraging, as it confirms that

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

109

Fig. 13. LTHR% vs OS at 775 K, 25 atm and SID vs OS of RON-representative (775 K, 25 atm) and MON-representative (875 K, 25 atm) cases using (a) (e) MFL mechanism [23], (b) (f) KAUST mechanism [27], (c) (g) Westbrook mechanism [2], and (d) (h) CoOptima mechanism [28]. The lowest OS point in each line indicates pure TRFs without ethanol.

the idea of correlating the two-stage behavior metric to OS is fundamentally sound and thus applicable not only to the simple simulations with assigned initial conditions (in Sections 2 and 3), but also to more engine-like simulations with varying pressure/temperature. Moreover, similar results for LTHR%-OS from initial condition simulations and imposed pressure history simulations imply that analyzing the chemical ignition of a fuel using 0-D simulations with appropriate initial conditions is sufficient to reveal the chemical nature of octane sensitivity. 4.2. 0-D simulations using different kinetic models To add generality to the analysis of the two-stage ignition behavior and its metric (LTHR%), we tested different chemical mechanisms available in the literature, using the homogeneous, constant-volume, adiabatic reactor simulations with the initial pressure/temperature conditions described in Sections 2 and 3. The three mechanisms considered are: a gasoline surrogate kinetic mechanism developed by research scientists at KAUST [27] that includes 2406 species and 9633 reactions; a mechanism developed by Westbrook et al. [2] that includes 2678 species and 11,565 reactions; and a detailed mechanism developed by the CoOptima program [28] that consists of 2108 species and 9046 reactions. All three detailed mechanisms as well as the MFL mechanism used in the previous sections have been recently published (2016 or later), developed for gasoline surrogate applications including TRFs, and extensively validated against experimental data. These mechanisms can be regarded as the state-of-theart kinetic models for TRFs. Details on their pathways and reaction rate parameters are extensively described in their respective publications. The intention is to examine the applicability of the proposed OS prediction method when different mechanisms are utilized.

Figure 13 shows the two-stage behavior metric (LTHR%) vs OS at 775 K, 25 atm and %differences in ignition delay time of RON vs MON representative 0-D simulations (775 K, 25 atm vs 875 K, 25 atm) using the three mechanisms together with the results obtained using the MFL mechanism, as reported in Figs. 9 and 2. First to note is that the effect of ethanol addition to the base TRFs is consistently captured by all four mechanisms: as the content of ethanol increases, and therefore higher values of OS, the LTHR% decreases and ignition delay time differences increase. Pure ethanol is a single-stage ignition fuel with very high OS value, and it suppresses the two-stage ignition behavior of n-heptane/isooctane very effectively, making the ignition delay time of the mixture more sensitive to the condition change. Such ethanol effect is commonly captured by all mechanisms. However, the results obtained from the KAUST and Westbrook mechanisms for the %difference in ID show a different shape of the curves as compared to the linear trends obtained using the MFL and CoOptima mechanisms. Since a very recent study on RON and MON predictions for PRF/ethanol mixtures by Westbrook et al. [29] showed the same curves for the RON vs. ethanol fraction as in Fig. 13 with good agreement with experimental results from Foong et al. [20], the differences in the PRF and ethanol submodels between the present mechanisms are likely responsible for these differences. The impact of ethanol chemistry on the overall TRF/ethanol mixture seems more preeminent in the KAUST and Westbrook mechanisms when compared to the MFL and CoOptima models, as indicated by very similar values of LTHR% and ignition delay differences for mixtures with ethanol content 40% or higher (E40, E60, E80) which have OS of 13–16. While the effect of ethanol addition to the same base TRFs is relatively consistent for all four mechanisms, there are significantly greater scatter among different TRFs for mechanisms other than MFL. Figure 13(a) and (e) shows that both LTHR% and ignition de-

110

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

Fig. 14. LTHR% vs OS at 775 K, 25 atm and SID vs OS of RON-representative (775 K, 25 atm) and MON-representative (875 K, 25 atm) cases for pure n-heptane and pure iso-octane doped with toluene calculated by (a) MFL, (b) KAUST, (c) Westbrook, and (d) CoOptima mechanisms.

lay differences of all four TRF cases collapse into one trend line for the MFL mechanism. However, all the other mechanisms generate separated trend lines except for high OS mixtures with high ethanol content where ethanol chemistry is dominating regardless of base TRF composition. Even for pure TRFs without ethanol (lowest OS points in each line in Fig. 13), all mechanisms exhibit quite different behaviors and values for the two-stage behavior metric (LTHR%) and ignition delay time difference. The MFL mechanism gives positive values of %difference in ignition delay time for all pure TRFs, which indicates that ignition delay time of the 875 K, 25 atm cases are always shorter than the 775 K, 25 atm case, as illustrated in Fig. 4(a) with heat release profiles. Recall that, by definition, positive value of %difference in ignition delay indicates the MON representative case (875 K, 25 atm) ignites earlier than the RON representative case (775 K, 25 atm), while negative value indicates later ignition of the MON representative case. However, the other three mechanisms generate negative values except for TRF45-E0 with Westbrook mechanism, meaning that the 875 K, 25 atm cases ignite later than the 775 K, 25 atm cases, which implies that the NTC behavior predicted by these mechanisms is stronger than that from MFL mechanism within the conditions tested in this study. In the meantime, MFL, KAUST, and Westbrook mechanisms commonly show the increasing trend in the ignition delay time difference as the toluene content increase in pure TRFs, but LTHR% trends are completely opposite – MFL mechanism predicts decreasing trend, while KAUST and Westbrook mechanisms predict increasing trend. Note pure TRFs are the lowest OS point in each line in Fig. 13. On the contrary, CoOptima mechanism generates relatively constant values for both LTHR% and ignition delay time difference for all four pure TRFs. These results demonstrate that the heat release characteristics during the ignition process of TRF components (n-heptane, iso-octane, toluene) and how they interact with each other when blended, are predicted differently by these

mechanisms. Figures S1–S3 in the Supplementary material report the heat release rates calculated with the KAUST, Westbrook, and CoOptima mechanisms. The mechanisms show different heat release profiles, timing of the first-stage heat release relative to the ignition point, and different responses to the fuel change from TRF0-E0 to TRF45-E0. One aspect that should be considered for the analysis of Fig. 13 is the change in base TRF compositions, particularly nheptane content, which is the most reactive fuel and has the strongest two-stage behavior among the TRF components. As noted earlier, all base TRFs tested in this study have same RON at 91 with varying toluene content from 0 to 45 vol%. Since toluene has significantly higher RON than 91, n-heptane (the component with lowest RON) content increases from 9% to 20% as toluene content increases from 0% to 45% to maintain RON at 91, as noted in Table 2. Thus, the analysis of four pure TRFs does not solely exhibit the effect of toluene addition and the consequent increase of OS, but also it demonstrates the complicated effect of change in n-heptane/isooctane ratio. In order to isolate the effect of toluene addition, homogenous reactor simulations were additionally performed at RON/MON representative conditions using pure n-heptane or pure iso-octane as base fuels, doped with toluene up to 40 vol%. OS of these binary mixtures were predicted using the correlation presented in Morgan et al. [30]. Figure 14 shows the LTHR% and %differences in ignition delay time for these fuel mixtures using the four mechanisms. All mechanisms predict a clear decreasing trend for both n-heptane and iso-octane blends with increasing toluene content for LTHR%. Encouragingly, the LTHR% values calculated by all mechanisms are relatively similar among n-heptane or iso-octane blends, showing the potentials of LTHR%, a two-stage behavior metric, as an OS predictor for those binary blends. Meanwhile, LTHR% trends also reveal the inherent limitations of the OS analysis when PRF blending ratio varies. n-heptane has stronger low temperature reactivity,

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

111

Fig. 15. Comparisons of calculated ignition delay times against experimental measurements from shock tube [31–37] for pure n-heptane, two n-heptane/toluene binary blends, and pure toluene. Stoichiometric fuel/air mixtures at 40 atm are shown.

NTC behavior, and two-stage ignition characteristics compared to iso-octane. However, due to the definition of OS, both molecules have OS value of zero, which does not correspond to the general notion that OS of fuels are somehow related to the two-stage ignition behavior. It is clear in Fig. 14(a)–(d) that pure n-heptane and iso-octane have identical values of OS but LTHR% values are 3–5 times higher for n-heptane. Similarly, although all PRF mixtures have OS equal to zero, the two-stage behavior is stronger as the content of n-heptane increases. Thus, one of the reasons for different LTHR% trends among the four TRFs (TRF0–TRF45) reported in Fig. 13 might be the change in n-heptane/iso-octane ratio. And it also indicates that investigating the OS – two-stage behavior relationship using fuel mixtures with different n-heptane/iso-octane ratios could lead to potential issues. Figure 14 also shows that the behavior of %difference in ignition delay time predicted by the four mechanisms is substantially different, particularly for n-heptane/toluene blends. Differently from iso-octane/toluene blends that show clear trends and similar values, n-heptane/toluene blends show strikingly different behaviors. MFL mechanism starts at about SID value of 20% with pure n-heptane, and as toluene is added, %difference in ignition delay time considerably decreases and becomes negative, as shown in Fig. 14(e). The KAUST and Westbrook mechanisms predict close to 0% for pure n-heptane, and with increasing toluene content there is slight decrease for the KAUST predictions in Fig. 14(f), and nearly constant for Westbrook Fig. 14(g). Predictions from the CoOptima mechanism Fig. 14(f) are drastically different from the other three mechanisms, as can be observed from the highly negative value (∼60%) for pure n-heptane and increasing trend as the concentration of toluene increases. These results might be another

key reason for having different trends for pure TRFs shown in Fig. 13. As the calculated ignition characteristics of n-heptane/toluene blends have large discrepancies, we compared the computed results with ignition delays measured in shock tubes. It should be noted that for more direct comparisons of the two-stage characteristics, heat release profiles of individual ignition events from experiments may be ideal. However, obtaining meaningful resolutions for the magnitude of heat release rate during the first stage is experimentally challenging, particularly for fully-igniting cases. Although the total ignition delay times may not be the ideal dataset to examine the two-stage ignition behavior, we can assess the overall agreement in ignition delay predictions, as well as the NTC behavior, which is known to be closely related to the twostage ignition characteristics. Ignition delay measurements of pure n-heptane, pure toluene, and two binary n-heptane/toluene mixtures from shock tube near 40 atm were taken from [31–37] and scaled to 40 atm using a pressure exponent of −1. In general, Fig. 15 shows that all mechanisms follow the experimental ignition delay trends of all four fuels, while noticeable differences exist among mechanisms. For pure n-heptane, all four mechanisms are within the experimental variability among different datasets in high temperature and NTC regime, while CoOptima mechanism shows better agreement in low temperature regime. KAUST and Westbrook mechanisms predict narrower NTC regime compared to MFL and CoOptima mechanisms. While MFL mechanism predicts the shortest ignition delay times in NTC regime for pure n-heptane and n-heptane/toluene 90/10 vol% blend, it predicts the longest ignition delay time for the 35/65 vol% blend and considerably weaker NTC behavior represented by the

112

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113

slope of the Arrhenius plot. It implies that suppression of NTC behavior by toluene is stronger in MFL mechanism than other three mechanisms. For pure toluene, all mechanisms over-predict ignition delay times in the temperature range 80 0 K–10 0 0 K where measurements are available, while MFL shows smallest deviation. Through these comparisons in Fig. 15, it is very difficult to conclude that any one of the drastically different trends for n-heptane/toluene binary blends observed in Fig. 14(e)–(h) is more physically valid. Thus, it is suggested that further efforts are required in the mechanism development community to reach consensus in terms of ignition behavior of pure n-heptane and toluene as well as the blending effect of these two TRF components, with additional experiments and theory analyses needed. Above results that investigated 24 TRF/ethanol blends clearly show that both LTHR% and %difference in ignition delay time calculated by different mechanisms can capture very similar correlations with OS, highlighting their potentials as OS predictors. Also, the major hypothesis of this paper, which is the close correlation between OS and the strength of the two-stage ignition behavior, can be consistently observed from simple 0-D simulations using various current generation TRF mechanisms. While the trends among four different pure TRFs were not consistent due to the limitations of OS definition and substantial differences in chemistry within the mechanisms, the effect of OS increase by ethanol or toluene addition suppressing the two-stage behavior was consistent throughout the four mechanisms, particularly when base n-heptane/iso-octane blending ratio is fixed. 5. Conclusions In this study, chemical kinetic simulations of 24 TRF/ethanol blends were performed to investigate the chemical nature of octane sensitivity. Homogeneous, constant-volume, adiabatic reactor simulations using a detailed chemical mechanism (MFL) showed that higher OS fuels have higher sensitivity in calculated ignition delay time when transitioning from RON-representative to MON-representative conditions. In addition, analysis of heat release rate profiles from simulations indicated that higher OS fuels have weaker two-stage ignition behavior. To be used as an OS predictor, a simple metric to quantify the strength of the two-stage ignition behavior was proposed, which can be derived from simulation results. A clear inverse trend of OS-LTHR% was observed at higher pressure and lower temperature conditions within the NTC regime. Using ignition calculations of 24 TRF/ethanol mixtures with the initial condition at 775 K and 25 atm, a preliminary correlation was developed as a tool to predict OS of TRFs and their blends with ethanol. For further validation of the proposed method, its applicability was tested by using simulation results of constant-volume simulations with pressure time history imposed throughout the simulation. Despite some drastically different simulations, the results indicated a clear decreasing OS-LTHR% relationship similar to the cases where only initial temperature/pressure is assigned. Moreover, LTHR%–OS relationship was calculated by three different TRF mechanisms. While some discrepancy existed, the general idea that higher OS fuels have weaker two-stage ignition behavior could be reproduced from different mechanisms. The results of this paper demonstrate a great potential of the proposed LTHR% as an OS predictor. At the same time, it indicates further refinement in kinetic mechanisms for TRF components and the need for additional experiments. Acknowledgment Authors would like to thank Dr. S. Goldsborough and Dr. D. Kang at Argonne National Laboratory for discussions on experi-

mental aspects of the two-stage ignition behavior. Authors would also like to thank Dr. W. Pitz at Lawrence Livermore National Laboratory for providing the CoOptima mechanism before public release, and Dr. C. Naik at Ansys Inc. and Prof. M. Sarathy at KAUST for technical assistance with kinetic simulations. This work was supported by 2019 Hongik University Research Fund. This work was also supported in part by the US Army Research W911NF-14-1-0359. The portion of this work performed at LLNL was supported by the U.S. Department of Energy (DOE), Office of Energy Efficiency and Renewable Energy (EERE), Vehicle Technology Office (VTO) under contract no. DE-AC52-07NA27344.

Supplementary material Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.combustflame.2019.08. 019.

References [1] J.B. Heywood, Internal Combustion Engine Fundamentals, McGraw-Hill, New York, 1988. [2] C.K. Westbrook, M. Mehl, W.J. Pitz, M. Sjöberg, Chemical kinetics of octane sensitivity in a spark-ignition engine, Combust. Flame 175 (2017) 2–15. [3] Knocking Characteristics of Pure Hydrocarbons, ASTM International, 1958. [4] J.P. Szybist, D.A. Splitter, Pressure and temperature effects on fuels with varying octane sensitivity at high load in Si engines, Combust. Flame 177 (2017) 49–66. [5] J. Lacey, K. Kameshwaran, S. Sathasivam, Z. Filipi, W. Cannella, P.A. Fuentes-Afflick, Effects of refinery stream gasoline property variation on the auto-ignition quality of a fuel and homogeneous charge compression ignition combustion, Int. J. Engine Res. 18 (2017) 226–239. [6] W.R. Leppard, The chemical origin of fuel octane sensitivity, SAE Technical Paper 902137 (1990). [7] M. Mehl, T. Faravelli, F. Giavazzi, E. Ranzi, P. Scorletti, A. Tardani, D. Terna, Detailed chemistry promotes understanding of octane numbers and gasoline sensitivity, Energy Fuels 20 (2006) 2391–2398. [8] M. Mehl, J.Y. Chen, W.J. Pitz, S.M. Sarathy, C.K. Westbrook, An approach for formulating surrogates for gasoline with application toward a reduced surrogate mechanism for cfd engine modeling, Energy Fuels 25 (2011) 5215–5223. [9] E. Singh, J. Badra, M. Mehl, S.M. Sarathy, Chemical kinetic insights into the octane number and octane sensitivity of gasoline surrogate mixtures, Energy Fuels 31 (2017) 1945–1960. [10] A. Roubaud, R. Minetti, L.R. Sochet, Oxidation and combustion of low alkylbenzenes at high pressure: comparative reactivity and auto-ignition, Combust. Flame 121 (20 0 0) 535–541. [11] A. Roubaud, O. Lemaire, R. Minetti, L.R. Sochet, High pressure auto-ignition and oxidation mechanisms of o-xylene, o-ethyltoluene, and n-butylbenzene between 600 and 900K, Combust. Flame 123 (2000) 561–571. [12] G. da Silva, J.W. Bozzelli, On the reactivity of methylbenzenes, Combust. Flame 157 (2010) 2175–2183. [13] D. Kang, D. Kim, K.H. Yoo, A. Violi, A. Boehman, The effect of molecular structures of alkylbenzenes on ignition characteristics of binary n-heptane blends, Proc. Combust. Inst. 37 (2019) 4681–4689. [14] M. Tao, P. Zhao, J.P. Szybist, P. Lynch, H. Ge, Insights into engine autoignition: combining engine thermodynamic trajectory and fuel ignition delay iso-contour, Combust. Flame 200 (2019) 207–218. [15] G. Kukkadapu, K. Kumar, C.-.J. Sung, M. Mehl, W.J. Pitz, Experimental and surrogate modeling study of gasoline ignition in a rapid compression machine, Combust. Flame 159 (2012) 3066–3078. [16] C.K. Westbrook, W.J. Pitz, O. Herbinet, H.J. Curran, E.J. Silke, A comprehensive detailed chemical kinetic reaction mechanism for combustion of n-alkane hydrocarbons from n-octane to n-hexadecane, Combust. Flame 156 (2009) 181–199. [17] H.J. Curran, P. Gaffuri, W.J. Pitz, C.K. Westbrook, A comprehensive modeling study of n-heptane oxidation, Combust. Flame 114 (1998) 149–177. [18] M. Mehl, W.J. Pitz, C.K. Westbrook, H.J. Curran, Kinetic modeling of gasoline surrogate components and mixtures under engine conditions, Proc. Combust. Inst. 33 (2011) 193–200. [19] C.K. Westbrook, W.J. Pitz, M. Mehl, H.J. Curran, Detailed chemical kinetic reaction mechanisms for primary reference fuels for diesel cetane number and spark-ignition octane number, Proc. Combust. Inst. 33 (2011) 185–192. [20] T.M. Foong, K.J. Morganti, M.J. Brear, G. da Silva, Y. Yang, F.L. Dryer, The octane numbers of ethanol blended with gasoline and its surrogates, Fuel 115 (2014) 727–739. [21] L. Cai, H. Pitsch, Optimized chemical mechanism for combustion of gasoline surrogate fuels, Combust. Flame 162 (2015) 1623–1637. [22] CHEMKIN version 18.0, ANSYS, 2016.

D. Kim, C.K. Westbrook and A. Violi / Combustion and Flame 210 (2019) 100–113 [23] Model fuel library, ANSYS, 2016. [24] D. Kang, V. Kalaskar, D. Kim, J. Martz, A. Violi, A. Boehman, Experimental study of autoignition characteristics of Jet-A surrogates and their validation in a motored engine and a constant-volume combustion chamber, Fuel 184 (2016) 565–580. [25] H. Wang, M.A. Oehlschlaeger, Autoignition studies of conventional and Fischer–Tropsch jet fuels, Fuel 98 (2012) 249–258. [26] D. Kim, J. Martz, A. Violi, A surrogate for emulating the physical and chemical properties of conventional jet fuel, Combust. Flame 161 (2014) 1489–1498. [27] S.M. Sarathy, G. Kukkadapu, M. Mehl, T. Javed, A. Ahmed, N. Naser, A. Tekawade, G. Kosiba, M. AlAbbad, E. Singh, S. Park, M.A. Rashidi, S.H. Chung, W.L. Roberts, M.A. Oehlschlaeger, C.-.J. Sung, A. Farooq, Compositional effects on the ignition of face gasolines, Combust. Flame 169 (2016) 171–193. [28] M. Mehl, S. Wagnon, K. Tsang, G. Kukkadapu, W.J. Pitz, C.K. Westbrook, Y. Tsang, H.J. Curran, N. Atef, M.A. Rachidi, S.M. Sarathy, A. Ahmed, A comprehensive detailed kinetic mechanism for the simulation of transportation fuels, 10th US National Combustion Meeting (2017). [29] C.K. Westbrook, M. Sjöberg, N.P. Cernansky, A new chemical kinetic method of determining RON and MON values for single component and multicomponent mixtures of engine fuels, Combust. Flame 195 (2018) 50–62. [30] N. Morgan, A. Smallbone, A. Bhave, M. Kraft, R. Cracknell, G. Kalghatgi, Mapping surrogate gasoline compositions into RON/MON space, Combust. Flame 157 (2010) 1122–1131.

113

[31] H.K. Ciezki, G. Adomeit, Shock-tube investigation of self-ignition of n-heptane-air mixtures under engine relevant conditions, Combust. Flame 93 (1993) 421–433. [32] H.-P.S. Shen, J. Steinberg, J. Vanderover, M.A. Oehlschlaeger, A shock tube study of the ignition of n-heptane, n-decane, n-dodecane, and n-tetradecane at elevated pressures, Energy Fuels 23 (2009) 2482–2489. [33] B.M. Gauthier, D.F. Davidson, R.K. Hanson, Shock tube determination of ignition delay times in full-blend and surrogate fuel mixtures, Combust. Flame 139 (20 04) 30 0–311. [34] M. Hartmann, I. Gushterova, M. Fikri, C. Schulz, R. Schießl, U. Maas, Auto-ignition of toluene-doped n-heptane and iso-octane/air mixtures: high-pressure shock-tube experiments and kinetics modeling, Combust. Flame 158 (2011) 172–178. [35] H.-P.S. Shen, J. Vanderover, M.A. Oehlschlaeger, A shock tube study of the auto-ignition of toluene/air mixtures at high pressures, Proc. Combust. Inst. 32 (2009) 165–172. [36] D.F. Davidson, B.M. Gauthier, R.K. Hanson, Shock tube ignition measurements of iso-octane/air and toluene/air at high pressures, Proc. Combust. Inst. 30 (2005) 1175–1182. [37] J. Herzler, M. Fikri, K. Hitzbleck, R. Starke, C. Schulz, P. Roth, G.T. Kalghatgi, Shock-tube study of the autoignition of n-heptane/toluene/air mixtures at intermediate temperatures and high pressures, Combust. Flame 149 (2007) 25–31.