Fundamental Mechanism Analysis on the Underlying Processes of LSPI Using Experimental and Modeling Approaches

Abstract

In view of growing global mobility requirements, energy security and climate change remain prominent issues. Consequently, there are strong drivers for further improved efficiency of vehicle powertrains, especially for internal combustion engines, which represent today’s mainstream technology.

1 Introduction

In view of growing global mobility requirements, energy security and climate change remain prominent issues. Consequently, there are strong drivers for further improved efficiency of vehicle powertrains, especially for internal combustion engines, which represent today’s mainstream technology.

The three main approaches to achieving thermal efficiency improvement of SI engines are: 1. High compression ratio, 2. High dilution, 3. Loss reduction. As the dominant technology via a loss reduction approach, downsizing technology has been the focus of recent attention and widely applied by many automotive companies.

For such downsized engines, operating with higher boost pressures at lower engine speeds is an essential technical challenge to achieving both fuel economy and driving performance like torque response. However, under these conditions the likelihood of an abnormal combustion phenomenon (LSPI, low speed pre-ignition) is increased, and combating this issue is critical to developing such engines with improved performance.

Figure 1 shows a typical pressure profile of LSPI (suddenly occurred 1st cycle). LSPI can be manifested as heat release before spark ignition timing and subsequent extreme pressure oscillations. It is accepted that LSPI does not arise directly from fuel autoignition, but from hot spots containing material with particularly short autoignition delay times.

Fig. 1.

Pressure profile of LSPI (1st cycle) and normal cycle

Previously, various parameters including engine conditions, lubricant composition and fuel properties have been reported to affect LSPI occurrence frequency. These observations have led to various hypotheses for pre-ignition causes and mechanistic pathways, which have been validated directly or indirectly using various experimental approaches including ignitability tests and endoscopic visualization in engines.

Figure 2 summarizes hypothesized mechanistic pathways for LSPI as reported previously. Much evidence indicates that the primary steps in LSPI are influenced by the extent of fuel adhesion on the liner wall and the properties of the lubricant oil that it interacts with on the liner wall [1,2,3]. Although many pathways to LSPI have been reported [4,5,6,7,8,9,10,11,12], common to all is the involvement of lubricant oil derived contents as pre-ignition sources.

Fig. 2.

Summary of proposed mechanistic pathways for LSPI occurrence

In this work, we follow the well-known mechanism hypothesis which assume lubricant oil containing droplets from piston crevice as a candidate of pre-ignition source. The mechanism hypothesis examined is shown in Fig. 3: Droplet of lubricant oil and fuel mixture, caused by adhesion of fuel spray on liner wall, will fly and pre-ignite before spark ignition.

Fig. 3.

Mechanism hypothesis of droplet (from piston crevice) triggered LSPI

2 Endoscopic Visualization of Droplet Triggered LSPI

2.1 Experimental Methodology: Single Cylinder Engine with Endoscope

To confirm the droplet triggered LSPI phenomenon, endoscopic visualization of LSPI was conducted on a single cylinder engine. Table 1 shows the specification of the single cylinder engine for this work. Experimental condition is shown in Table 2. Single cylinder engine based on downsized SI engine with gasoline direct injection (prototype) was used for this study. Using an external supercharger and intake air heater, boost pressure and intake air temperature was kept constant. Steady operation at low engine speed and highly boosted conditions was performed.

Table 1. Specification of single cylinder engine

Table 2. Experimental setting and condition (SCE)

Figure 4 shows the abstract of the visualization system with endoscope. An endoscope with high pressure resistance (“Bore scope” by NAC Image Technology, φ8.5 mm, pressure resistance up to 30 MPa) was mounted to the side part of combustion chamber. In-cylinder images were visualized and recorded with using relay lens and high speed camera (NAC Image Technology, MEMRECAM GX-8/K4). Flame speed was set to 10000 fps for all experiment. As well, this visualization system can observe almost all area of in-cylinder with changing front lens (straight and lateral view) and mounting angle of the endoscope.

Fig. 4.

Endoscopic visualization system apparatus

2.2 Visualized Droplet-Triggered LSPI (1st Cycle)

Figure 5 shows the 1st abnormal cycle of LSPI, visualized by the endoscope on the single cylinder engine. Start of injection was set to 280° BTDC.

Fig. 5.

Visualized in-cylinder phenomenon of LSPI (1st cycle)

At the last of compression stroke, objects with luminous flame flying from exhaust side to chamber center generated pre-ignition (Fig. 5(a)). Then it was observed that blue flame propagated from these luminous objects at exhaust side to chamber center direction (Fig. 5(b)). After a discharge ignition by spark plug (Fig. 5(c)), pressure oscillation which is expected to be a result of end-gas auto-ignition, was observed (Fig. 5(d)). After that, much amount of luminous flame was observed at whole viewing field (Fig. 5(e)). It is probably a result of rich fuel/air mixture generation with boundary layer destruction by the strong pressure oscillation and its rich combustion.

As a result of visualizing 1st abnormal cycle of LSPI many times, including different view angles, many of LSPI is triggered by flame propagation pre-ignited by flying objects with luminous flame like Fig. 5.

Figure 6 shows the enlarged view of flying object and its behavior on the same 1st abnormal cycle with Fig. 5. Firstly, a luminous object with relatively big diameter was emerged (Fig. 6(a)). Then, the number of luminous flame kernels was increased and simultaneously sizes of these kernels became smaller (Fig. 6(b), (c)) with flying distance (Fig. 6(d)). From these behaviors, we expected that these objects were droplets flying with making breakup and vaporization.

Fig. 6.

Enlarged view of flying objects (flame kernels) triggering LSPI (1st cycle)

As a contrasting reference, Fig. 7 shows the visualized subsequent cycle of LSPI. The cycle just after the 1st abnormal combustion cycle same with Figs. 5 and 6 was visualized. Although fuel supply was cut off for this cycle with trigger controlled system in this case, a luminous flying particle was observed at compression stroke. The number of particle was only one and never increased. Also particle size was kept almost same (From image, its size was estimated to be approximately 0.4–0.7 mm). We can also easily assume that this particle is flying with riding on tumble flow from its trajectory (Fig. 7(a)–(f)). From these behaviors, this particle is expected to be solid, as already visualized in previous studies [9,10,11].

Fig. 7.

Behavior of flying particle at subsequent cycle of LSPI

From the observed behavior and comparison with sold-like particle in the subsequent cycle, we understood that the luminous objects at 1st LSPI cycle were flying droplets in liquid phase.

To supplement the understanding of LSPI phenomenon in 1st cycle, visualization in different viewing field, exhaust side, was also conducted. Figure 8 shows the result.

Fig. 8.

In-cylinder phenomenon of LSPI (1st cycle) visualized in exhaust side

Same to the result in Fig. 5, objects with luminous flame flying from exhaust side (Fig. 8(a)) and flame propagation from these objects (Fig. 8(b)) were again observed. The combustion chamber became suddenly bright due to the end-gas autoignition (Fig. 8(c)). Soon after the flashlight by the autoignition, particles peeled off from piston surface near the edge (Fig. 8(d)) and enormous scattering of these particles (Fig. 8(e)) were observed.

To summarize these visualization results, we could thought that there are LSPI caused by flying droplets. Also, it can be understood that scattering of solid particles, which have been reported as a driver of subsequent LSPI cycles, are triggered by pressure oscillation of 1st LSPI cycle.

For further understanding on LSPI based on this “droplet triggered” mechanism hypothesis, we need to investigate two processes: The physical process of droplet flying and the chemical process of metallic additive’s effect on droplet ignition.

3 Analysis and Modeling on the Process of Droplet Flying

3.1 Experimental Methodology: Visualization Engine with Oil Lubrication

From understanding on the possibility of droplet triggered LSPI, analysis on detailed process and mechanism of droplet flying by using the visualization engine with oil lubrication was conducted.

Table 3 shows the specification of the visualization engine for this work. Experimental condition is shown in Table 4. Using same injection system with the single cylinder engine, fuel amount equivalent to the condition of highly boosted condition was injected. Then steady motoring operation was performed.

Table 3. Specification of visualization engine

Table 4. Experimental conditions (Visualization)

Figure 9 shows the apparatus of this visualization engine. A quart glass cylinder with dimension controlled to be same with production engine, a production piston, and production piston rings lubricated by oil were used for this visualization engine. With exposing light to the visualization area by metal halide lamp, behavior of oil and fuel at oil-lubricated sliding surface and its neighborhood (including piston crevice) can be visualized and recorded by high speed camera.

Fig. 9.

System of visualization engine with oil lubrication

3.2 Visualized Droplet Flying Process

Figure 10 shows an example of visualized images by the visualization engine. In this case, start of injection was set to 240° BTDC. Series of phenomenon from intake stroke to compression stroke, written as below, was well observed.

Fig. 10.

Example of visualized intake & compression stroke

• Adhesion of fuel spray on liner wall (Fig. 10(a))

• Scratch up of liner-adhered fuel by piston (Fig. 10(b))

• Droplets flying at the end of comp. stroke (Fig. 10(c))

Figure 11 shows such typical behavior of droplets flying from piston land. These images were obtained by image procession of subtracting motoring images (background images without any fuel) from main images visualized at each crank angle. Obtained images are useful to understand the phenomenon of droplets flying in detail.

Fig. 11.

Visualized droplet flying process

Lubricant oil flying by piston inertia has been already reported by many literatures [13], mainly from the viewpoint of oil consumption. However, it was confirmed from this visualization that continuous operation at liner wetting condition and accumulation of fuel at piston crevice can easily leads to droplets flying, even at low engine speed (low piston inertia) which is focused on LSPI issues.

Behavior of droplets flying was almost same with the result of numerical analysis using VOF method by Moriyoshi et al. [14]. Mixture of oil and fuel go up on the piston land with generating liquid column and finally get on the piston crown surface and fly into combustion chamber. Such behavior is well explained by Rayleigh Taylor instability. Density difference between oil/fuel mixture (liquid phase) and fuel/air mixture in combustion chamber (gas phase) will develop disturbance and generate waves. Finally the heights of waves will exceed the piston land height then oil/fuel mixture get on piston crown surface and finally fly by piston inertia.

To investigate the effect of fuel adhesion or operating time on droplet flying, amount of flying droplets was calculated by image procession. Using the subtracted images, we defined control area as upper side of piston crown surface line. Then number of pixels brighter than criteria was counted up and used as an indicator for the amount of droplets.

Figure 12 shows the effect of injection timing and operation time on the amount of flying droplets at TDC. With more retarded injection timing, more amount of droplets was observed from earlier timing, due to more amount of fuel adhesion on liner wall. At the same time, after long enough operation time, amount of flying droplets looks saturated to some constant level. This is probably because of outflow from piston crevice. For example, evaporation from re-coated film and flow to the 2nd land across the ring gap/back clearance (Fig. 12(d)) are expected. Thus this system is thought to be a lag system with saturation behavior.

Fig. 12.

Effect of injection timing on the amount of flying droplets

Figure 13 shows the amount of droplets flying at each crank angle with same injection timing (240° BTDC) and different operation time. With longer operation time, droplets flying at earlier timing (around −30° ATDC) was observed. It is because long operation time leads to more accumulation of fuel and oil/fuel mixture will have lower viscosity and surface tension.

Fig. 13.

Effect of operation time on the amount of droplets (SOI = 240° BTDC)

3.3 Modeling of Droplet Flying Process and Discussion

Experimentally visualized droplet flying process seems to be easily explained by Rayleigh-Taylor instability. We tried to model the phenomenon in simple way.

Table 5 shows the simplified process from wave generation to droplet flying based on Rayleigh-Taylor instability.

Table 5. Assumed droplet flying process based on Rayleigh-Tayloer instability

To estimate accurately the amount of oil/fuel mixture in piston top land, consideration of evaporation rate profile after fuel adhesion as well as fuel impingement and adhesion itself is necessary. However, for simple modeling, the amount of mixture expected to be in crevice at the end of compression stroke was assumed to be already in crevice at the calculation start timing. The amount of mixture was estimated from zero dimensional wetting simulation and experimental results.

Rayleigh-Tayler instability is a phenomenon that two types of fluids with different density will cause the instable boundary surface due to the effect of inertia. For the modeling of wave generation based on the instability, wave length and amplitude need to be estimated. In the simplified model, it was assumed that, after the amplitude exceed 0.4 times of wave length, wave growth transition to non-linear zone and wave length fixed. Model equations of Rayleigh-Taylor instability are shown from (1) to (3) [14,15,16]. Growth speed of generated waves are modeled as a function of kinetic viscosity, the amount of mixture and piston speed, based on the Eq. (4) proposed by Thirouard et al. [17, 18]. For the physical properties of mixture, iso-octane was used with assumption of enough local dilution of lubricant oil by fuel and saturation of the mixture properties [5].

Model equations of Rayleigh-Taylor instability:

$$ \gamma^{2}_{peak} = - \frac{{kAta_{pst} }}{2}\left( {1 - \frac{{k^{2} }}{{k_{c}^{2} }}} \right), k = k_{peak} $$

(1)

$$ At = {{\left( {\rho_{liq} - \rho_{gas} } \right)} \mathord{\left/ {\vphantom {{\left( {\rho_{liq} - \rho_{gas} } \right)} {\left( {\rho_{liq} + \rho_{gas} } \right)}}} \right. \kern-0pt} {\left( {\rho_{liq} + \rho_{gas} } \right)}} $$

(2)

$$ k_{c} = \left[ {\frac{{ - \left( {\rho_{liq} - \rho_{gas} } \right)a_{pst} }}{{Ts_{liq} }}} \right]^{1/2} $$

(3)

Model equation to estimate wave growth speed (proposed by Thirouard et al.):

$$ H_{top} = H_{0} + 1.188\left( {\frac{{\left| {u_{p} - u_{{p\left( {max} \right)}} } \right|\sqrt {V_{liq} } }}{2\nu }} \right)^{0.394} \sqrt {V_{liq} } $$

(4)

Figure 14 shows the example of estimated wave height which is a representative parameter to judge the droplet flying. SOI = 300° BTDC, advanced injection timing with less liner wetting, didn’t show the wave height higher than top land height. On the contrary, SOI = 200° BTDC, retarded injection timing with more liner wetting, showed higher wave and early droplet flying around −30° ATDC. These result qualitatively corresponds to the experimental results shown in Fig. 12 or Fig. 13. With SOI = 300° BTDC, no droplet flying was observed even with quite long operation time (Fig. 12). With retarded injection timing, the timing of droplet flying converged to around −30° ATDC. Estimated wave length (approximately 3.0 mm) and experimentally observed gap between liquid columns (approximately 2.0 mm) were also comparable.

Fig. 14.

Example of wave height estimation by RT instability model (1200 rpm)

These calculation is just rough estimation and further prediction accuracy on the amount and properties of mixture are required for the quantitative estimation. However, droplet flying mechanism based on Rayleigh-Taylor instability seems to be relevant.

4 Analysis and Modeling on the Process of Droplet Ignition

4.1 Experimental Methodology: Rapid Compression and Expansion Machine

From understanding on the possibility of droplet triggered LSPI, we focused on the chemical process of metallic additive’s effect on droplet ignition.

To observe the ignition of fuel/oil droplets and subsequent flame propagation in the fuel/air mixture, a rapid compression and expansion machine (RCEM) are selected.

Figure 15 shows a schematic of the RCEM. This RCEM (Bore/Stroke = 125/140 mm) can simulate the piston motion of a reciprocating engine’s compression and expansion strokes with its cam profile. The descending cam (total weight = 180 kg) moves the piston along the cam profile one time. Compression ratio settings can be changed easily by changing the cylinder head position.

Fig. 15.

Experimental schematic of RCEM

A transparent cylinder head made from polished acrylic was used to facilitate visualization of in-cylinder phenomena. The ignition kernels of fuel/oil droplets and subsequent flame propagation were observed using images obtained with the high-speed camera. Both the visualized images and pressure measurements allow quantitative evaluation of ignition delay times.

For the injection of the oil/fuel mixture, we prepared a special six-hole, solenoid-driven DI injector for gasoline engines. To avoid wetting of the spray stream on the piston crown surface, the injector was mounted at an inclined angle. Targets of the spray streams were also optimized. One stream for observation was designed to penetrate freely in the visualization area without wetting any wall, while the other five streams penetrated toward the outside of the visualization area. Figure 16 shows a schematic of the cylinder head, injector mounting design, and geometrical relation between the injector streams and the visualization area.

Fig. 16.

Cylinder head design and geometrical layout of spray direction/visualization area

4.2 Analysis on the Droplet Ignition Phenomenon

Table 6 shows the experimental conditions with combinations of droplet composition with metallic additive variation and ambient mixture conditions with RON variation. Experiments were conducted 5 times per each condition.

Table 6. Experimental conditions of RCEM: ambient conditions and droplet composition

Figure 17 shows the obtained ignition delay times and pressure profiles of fuel droplets for various amounts of calcium additives. It definitely showed the enhanced droplet ignition and subsequent combustion due to calcium additive in droplet. Figures 18 shows the visualized enhancement of droplet ignition and combustion due to the different amount of calcium additives in droplets. Wider spatial spreading of the initial flame was clearly observed with higher amounts of calcium additives.

Fig. 17.

Ignition delay times and pressure profiles with Ca additive variation

Fig. 18.

Visualized combustion enhancement by Ca additive in droplet

Figure 19 shows the obtained ignition delay times and pressure profiles for various RON values of surrounding mixture. It definitely showed the enhanced droplet ignition and subsequent combustion due to lower RON value of surrounding mixture. Figure 20 shows the visualized enhancement of droplet ignition and combustion due to the lower RON value of mixture. Similar to calcium additive effect, wider spatial spreading of the initial flame was clearly observed with lower RON mixture.

Fig. 19.

Ignition delay times and pressure profiles with RON variation mixture

Fig. 20.

Visualized combustion enhancement by low RON mixture (with Ca in droplet)

Figure 21 shows the obtained ignition delay times and pressure profiles of fuel drop-lets for various amounts of zinc additives. It definitely showed the inhibited droplet ignition and subsequent combustion due to zinc additive in droplet. Figure 22 shows the visualized inhibition of droplet ignition and combustion due to the different amount of zinc additives in droplets. Zinc additive clearly show the opposite sensitivity to the calcium additive and to lower RON mixture.

Fig. 21.

Ignition delay times and pressure profiles with Zinc additive variation

Fig. 22.

Visualized combustion inhibition by Zinc additive in droplet

4.3 Modeling of Droplet Ignition

Based on the observation of the ignition delay times and the visualized images in these RCEM experiments, the gas phase reaction with the surrounding premixed mixture seems to be important for droplet ignition and subsequent combustion in the cylinder. It is also expected that calcium additive have a promotion effect on gas-phase reaction and autoignition. On the contrary, zinc additive is expected to have an inhibition effect on gas phase reaction.

Historically, the catalytic promotion effects of CaCO₃ and CaO on hydrogen abstraction reactions have been well known in the research fields of thermal decomposition or gasification of coal [20, 21]. The hypothesized reactions (5) and (6) are shown below. If such reaction are applicable, the observed promotion effects of calcium additives can be explained.

$$ RH + \left( {CaO} \right) = R + H + (CaO) $$

(5)

$$ RH + \left( {CaCO_{3} } \right) = R + H + \left( {CaCO_{3} } \right) $$

(6)

It is well known that ZnDTP functions to prevent oxidation of the lubricant oil by changing radicals or peroxides into stable substances, as shown in (7) [22]. If such a reaction is effective and can be applied in the gas-phase reaction, the observed inhibition effects of zinc additives can be explained.

$$ QOOH + ZnDTP = QOH + ZnO + DTP $$

(7)

We tried to model and simulate the chemical effects of these metallic additive with assuming such specific reactions. For the CFD calculation, KIVA-4 coupled with CHEMKIN was used. Figure 23 shows the computing mesh to simulate RCEM and Table 7 shows the sub-models used for this study.

Fig. 23.

Computing mesh for this study

Table 7. Sub-models used in the CFD code

For the reaction scheme, skeletal PRF scheme developed by Tsurushima [19] was selected as a base scheme. Then promotion effect of calcium additive was simulated by enhancing hydrogen abstraction reaction and inhibition effect of zinc additive was simulated by inhibiting decomposition reaction of peroxides as shown in Fig. 24. To simulate ignition behavior of RCEM experiments, activation energy E₁ for hydrogen abstraction reactions and activation energy E₂ for peroxide decomposition reaction are changed. To separate the droplet-effected reaction and bulk mixture reaction, related hydrocarbon species are treated as isomers. Initial conditions of calculation was summarized in Table 8.

Fig. 24.

Modified reaction scheme used for CFD calculation

Table 8. Calculation condition to simulate experimental results

Figure 25 shows the results of calculation for both calcium and zinc additive effect simulating Test 1 (Fig. 17/18) and Test 3 (Fig. 21/22). With changing activation energy values, ignition timing was reproduced well. The relevance of absolute values of activation energy delta will need further discussion. However, required delta of these activation energies indicate that the effect of these metallic additive on droplet ignition in fuel/air mixture is really drastic.

Fig. 25.

Simulated pressure profiles and ignition delay times with E₁ and E₂ optimization

HCHO concentration in Fig. 26 and temperature distribution in Fig. 27 shows the effects of activation energy change on gas phase reaction process. Reduced E₁ for simulating enhancement by calcium additive shows early LTO represented by HCHO consumption and rapid temperature rise after LTO. Reduced E2 for simulating inhibition by zinc additive shows retarded HCHO consumption and temperature rising.

Fig. 26.

Observation of LTO enhancement and inhibition via HCHO concentration

Fig. 27.

Simulated temperature distribution in the cylinder

5 Summary

From the endoscopic visualization in single cylinder engine, we could assume that there are LSPI caused by flying droplets. Also, it can be understood that scattering of solid particles, which have been reported as a driver of subsequent LSPI cycles, are triggered by pressure oscillation of 1st LSPI cycle.

From the oil lubricated visualization engine experiment, it is confirmed that liner adhered fuel will accumulate in piston crevice then oil/fuel mixture can easily fly into chamber even at low engine speed condition.

Although further improvement on prediction of the amount and properties of mixture are required, droplet flying process and phenomenon can be well explained by Rayleigh-Taylor instability.

From the RCEM experiments on droplet ignition in air/fuel mixture, the gas phase reaction with the surrounding premixed mixture seems to be important for droplet ignition and subsequent combustion in the cylinder.

It is expected that calcium additive have a promotion effect on gas-phase reaction and autoignition. On the contrary, zinc additive is expected to have an inhibition effect on gas phase reaction. These chemical effects were tentatively modeled by enhancement of hydrogen abstraction reaction and inhibition of peroxide decomposition reaction.