Development Process of TGDI SI Engine Combustion Simulation Model Using Ethanol–Gasoline Blends as Fuel

Abstract

The Fit for 55 package introduced by the European Union aims to achieve a 55% reduction in greenhouse gas emissions by 2030. In parallel, increasingly stringent exhaust gas regulations have intensified research into alternative fuels. Ethanol presents a promising option due to its compatibility with gasoline, higher octane rating, and lower exhaust emissions compared to conventional gasoline. Additionally, ethanol can be derived from agricultural waste, further enhancing its sustainability. This study examines the impact of two ethanol–gasoline blends (E10, E20) on emissions and performance in a turbocharged gasoline direct injection (TGDI) spark-ignition (SI) engine. The investigation is conducted using three-dimensional computational fluid dynamics (3D CFD) simulations to minimize development time and costs. This paper details the model development process and presents the initial results. The boundary conditions for the simulations are derived from one-dimensional (1D) simulations, which have been validated against experimental data. Subsequently, the simulated performance and emissions results are compared with experimental measurements. The E10 simulations correlated well with experimental measurements, with the largest deviation in cylinder pressure being an RMSE of 1.42. In terms of emissions, HC was underpredicted, while CO was overpredicted compared to the experimental data. For E20, the IMEP was slightly higher at some operating points; however, the deviations were negligible. Regarding emissions, HC and CO emissions were higher with E20, whereas NOx and CO₂ emissions were lower.

1. Introduction

As a consequence of the continuous growth of the automotive industry, exhaust gas emissions have increased, significantly affecting air quality [1]. On a global scale, the transport sector accounts for 26% of greenhouse gas emissions, and this shows a growing trend year by year [2]. In the cities of developing countries, poor air quality causes the death of approximately 800,000 people every year [3].

European legislators have pointed out that laboratory emission tests do not reflect real-world conditions, particularly in the case of NOx. The New European Driving Cycle (NEDC), which has been in use since 1970, did not reflect real driving conditions, a fact that was highlighted by the 2015 diesel scandal [4]. Driving Emissions (RDE) tests were introduced to overcome this, where emissions are measured under real conditions using Portable Emission Measurement Systems (PEMS). However, since these measurements are influenced by several factors such as traffic, weather, and air quality, they were supplemented by the Worldwide Harmonized Light Duty Vehicle Test Cycle (WLTC), a laboratory test that can be carried out under controlled conditions [5,6].

Another consequence of the growing demand for vehicles is the accelerated depletion of fossil fuel reserves [7]. Despite this, according to the World Energy Outlook (WEO), 84% of the global energy demand will still be met by fossil fuels [8]. These two main reasons have motivated researchers to explore various renewable energy sources to meet the world’s continuously increasing energy needs in the interest of sustainability [9].

Several gasoline extenders are under investigation as at least partial substitutes for gasoline. Hydrogen–gasoline solutions [10] combined with engine exhaust gas optimization [11] seem promising, but here the barrier is the hydrogen’s production and logistics. Liquid alternatives are also being researched in depth [12]. For renewable components, aging degradation handling is crucial [13]. Ethanol is an excellent alternative to replace fossil fuels, as it can be produced from renewable sources and is suitable for internal combustion engines. Its oxygen content reduces exhaust gas emissions, while its higher laminar flame speed improves combustion efficiency [14]. The use of alcohol in engines is not a new idea; experiments were already being conducted in the early 1900s, but it has only received significant attention in recent years [15,16]. Globally, the United States and Brazil are the largest ethanol producers, but in many other countries ethanol has been integrated into commercial gasoline with the help of legislation [7,17].

In the last 20 years, many researchers have studied the effects of various ethanol-blended fuels on engine performance and exhaust emissions under different conditions. Most of these studies were experimental, while numerical methods have only begun to spread in the past few years. In terms of results, the research consistently shows that increasing the ethanol content in gasoline slightly increases engine performance [18].

Regarding exhaust emissions, results vary. According to some researchers, NO_x emissions increase at higher engine speeds due to increased fuel burned. Other researchers have reported opposite results, which they explained by the shorter time available for combustion [19]. As for HC and CO emissions, different research results consistently show that increasing ethanol content reduces these values due to its oxygen content and higher burning speed [20,21,22].

Regarding CO₂ emissions, the results are also mixed: D. Y. Dhande et al. [23] investigated the effect of ethanol produced from pomegranate blended with pure gasoline on SI engine performance and emission characteristics. According to the results, CO₂ emissions increased due to the decrease in CO emissions. In contrast, Wang et al. [24] investigated the effects of anhydrous and hydrous ethanol–gasoline blends on the combustion characteristics and emissions of a port fuel injection (PFI) spark-ignition (SI) engine. Their results showed that E10 has lower CO₂ emissions than conventional gasoline. This is because the ethanol blend has lower carbon content, so the reduction in CO does not necessarily coincide with an increase in CO₂ [25].

2. Materials and Methods

2.1. Engine Data

In this study, a 2 L four-cylinder 4-stroke turbocharged spark-ignition engine was investigated.

Table 1. Engine parameters.

Data	Value
Bore [mm]	82.5
Stroke [mm]	92.8
Connecting rod length [mm]	144
Compression ratio [-]	12.2:1
Displacement [cm3]	1984
Injection type	Direct injection

contains the important parameters of the engine.

2.2. Operating Points

As previously discussed, the operation points were selected based on the RDE measurements. The four most common operating points were selected for further investigation, ensuring that each point has a different RPM and engine load.

Table 2. Operating points parameters.

Parameter	OP1	OP2	OP3	OP4
RPM [1/min]	2000	2500	3000	3500
BMEP [bar]	4	16	20	20
Ignition timing [CAD]	−19.14	−2.646	0.251	1.226
Lambda [-]	0.999	0.999	0.997	0.949

summarizes the data of the selected operation points.

2.3. Experimental Apparatus

The engine measurements were performed on a 380 kW AVL DynoSpirit (AVL List GmbH, Graz, Austria) dynamic testbench at the Department of Propulsion Technology of Széchenyi István University. In-cylinder pressure measurements were conducted to enable a detailed process analysis. The data was acquired using the AVL IndiMicro (AVL List GmbH, Graz, Austria) measurement system equipped with Kistler pressure 6056 sensors. Signal amplification was performed with AVL MicroIFEM (AVL List GmbH, Graz, Austria) indicating amplifiers and data processing was carried out using the AVL Indicom software (Version 2.7). Raw exhaust gas samples were taken upstream of the catalytic converter. The exhaust gas composition was analyzed using the AVL SESAM i60 FT SII system, which applies Fourier-transform infrared (FTIR) spectroscopy. The fuel consumption was measured using the AVL FuelExact (AVL List GmbH, Graz, Austria), which is a high-precision mechanical fuel flow measurement system.

2.4. Numerical Methods

The structure of the simulations was divided into three parts. The first simulation model set was built based on 1D simulation data. These simulations were used to determine the gas exchange process and exact oxygen content in the cylinder without spray and combustion to precisely determine the injected mass per cycle, because experimental data was unavailable. The results of the first set of simulations will not be presented in this paper. The second set of models contains all the physical models which are necessary to simulate a real engine cycle. These simulations are calibrated against the experimental data with E10 fuel. The last phase of the simulations run on E20 fuel. They are made from the previous set of simulations, assuming the predictability as no E20 experimental data was available. These simulations have the same circumstances, except the fuel. Naturally, with the increase in ethanol content, the higher laminar flame speed allows for a reduction in spark advance. However, this study does not address modifications to spark timing. Additionally, due to the higher oxygen content, the mixture would become leaner if the amount of injected fuel were not corrected. In this study, the injected fuel quantity was adjusted to ensure that the air–fuel ratio matches that of the E10 cases.

The CONVERGE CFD (Version 3.1) software package was used for pre-processing and running the simulations. This software is employed to calculate three-dimensional turbulent reacting fluid flows in complex geometries including moving elements. Using the cut-cell Cartesian method, this code automatically generates the mesh during run-time. This allows the use of a simple orthogonal mesh which represents the boundary shape precisely. The meshing system contains user-defined functions to change the mesh resolution locally during the simulation. Also, it features Adaptive Mesh Refinement (AMR), which refines the mesh based on the gradient of the variables, such as temperature or velocity [26].

The SAGE detailed chemistry solver was used to model combustion phenomena. This chemistry solver solves detailed chemical kinetics via a reaction mechanism file, which contains chemical kinetics properties of species. A built-in feature called multi-zone was used to accelerate the solving process [27]. A reaction mechanism which was mostly based on [28] with 201 species and 1013 reactions was used for the following simulations.

Two fundamental requirements for combustion are mixing and chemical reactions. In numerical simulations, the mixing process is primarily governed by the turbulence model. Several studies have demonstrated that the RNG k–ε turbulence model is well-suited for simulating combustion in internal combustion engines [29,30,31]. In recent years, large eddy simulation (LES) models have gained increasing popularity due to their ability to resolve more detailed flow structures. However, Reynolds-averaged Navier–Stokes (RANS) models, such as the RNG k–ε, remain widely used due to their significantly lower computational cost [32].

Pressure–velocity coupling was handled using the PISO algorithm, and a pressure-based solver was employed for the simulations. Spatial discretization of the governing equations was performed using a first-order non-conserving upwind numerical scheme to ensure repeatability. Convergence criteria were set to a residual threshold of 10⁻⁵ for all solved variables. A variable time-step approach was used, with a minimum time-step size set to 10⁻⁸ s.

Conventional gasoline combustion involves complex chemistry with reaction mechanisms compromising several thousand chemical species. Computationally determining the reaction kinetics of combustion with that level of complexity is not practical for engineering applications. In CFD simulations, real fuels are represented by surrogate fuels, which are simplified mixtures composed of the main representative species of each major chemical family. Fuel surrogates are designed to mimic real fuels’ physical and chemical properties [33,34]. In this study, a 4-component surrogate called ETRF was used in the simulations, and the chemical composition of surrogates is reported in

Table 3. Composition of fuel surrogates.

Parameter	E10Sur	E20Sur
Isooctane [vol%]	43.8	35.69
n-heptane [vol%]	14.4	13.86
Toluene [vol%]	31.8	31.95
Ethanol [vol%]	10	18.5

.

In the present work, the surrogate formation methodology used was from [35], and is based on solving an overdetermined linear system with the least-squares technique. The number of components in a surrogate determines the number of real fuel properties that can be matched: n−1 properties can be targeted using n-component surrogate. The selection of the target properties depends on the goal of the investigation [36]. The focus of this study is to investigate performance and emissions, and therefore the H/C, O/C ratio and LHV were chosen to match precisely.

Table 4. Fuel properties.

Parameter	E10	E10Sur	E20	E20Sur
RON [-]	96.5	97.2	99	99
MON [-]	85.1	88.5	87.8	88.3
Density [kg/m3]	753.6	756.3	783.1	764.8
H/C ratio [-]	1.906	1.86	1.89	1.89
O/C ratio [-]	0.0307	0.033	0.0614	0.0617
LHV [MJ/kg]	42.04	41.5	40.34	40
Stoich. A/F [-]	14.019	13.87	13.42	13.34

shows the investigated fuels and their surrogates’ main properties.

In this study, the fuel injection is modeled using the method called “blob” developed by Reitz and Diwakar, where diameter of parcels of liquid are equal to the effective nozzle diameter. The primary and secondary breakup of these blobs into droplets is simulated using instability-based models that incorporate both Kelvin–Helmholtz and Rayleigh–Taylor mechanisms [37,38].

3. Results and Discussion

In this section, the results of the 3D simulations are presented. First, the validation of the E10 simulations against test bench measurements based on cylinder pressure and pollutant emissions is discussed. In the second step, the results of the E10 and E20 simulations are compared. Each simulation covers 720 degrees of crankshaft rotation from exhaust valve opening to the next opening.

3.1. Mesh Independence Study

The quality of the mesh greatly influences the simulation results; therefore, its assessment is essential. The in-cylinder simulation includes numerous physical models, and during a single cycle, the gradients of thermodynamic parameters can change significantly. Adaptive meshing is very helpful for handling these variations appropriately. However, to accurately model complex physical processes such as injection, an adequately dense mesh is required. During the mesh sensitivity study, three different settings were tested. The base mesh size and the maximum density of the adaptive mesh were varied, as illustrated in Figure 1.

It can be observed that in the case of the coarsest mesh (base5-amr2), there is a difference of nearly 8 bar in the maximum cylinder pressure, and the pressure rise rate is much lower, indicating that the combustion process is modeled completely inaccurately. In the next iteration (base5-amr3), a slight improvement can be seen; however, the same conclusions as in the previous case still apply. In the third iteration (base4-amr3), a significant change occurred: since cylinder pressure data from test bench measurements was available and the simulated cylinder pressure showed a high degree of agreement with it, there was no justification for using a finer mesh. Therefore, the third iteration was selected for the meshing setup: the base mesh size was 5 mm, and the AMR maximum embedding level was set to 3.

3.2. Validation of the E10 Simulations

The reaction mechanism and fuel surrogate used during the simulations were sufficiently detailed, so there was no need to apply the combustion model’s reaction rate multiplier function or to modify the turbulence parameters. The regions were initialized using experimental and 1D simulation data, which allowed a single cycle to be simulated. This contributed to a reduction in computational time. One cycle simulation took approximately 20 h. To validate the model, simulations were also performed over multiple cycles to examine the variations between them. However, thanks to the use of a first-order non-conservative solver, no significant differences were observed between the cycles. Figure 2 compares the simulation results and the test bench measurements in terms of cylinder pressure. It can be observed that the accuracy of the simulation decreases with increasing engine speed and load, as reflected in the Root Mean Square Error (RMSE). The smallest deviation occurred in the first operating point. In this case, the compression, combustion, and expansion phases matched the measured values from the test bench. The largest deviation was during the compression phase, where the pressure difference reached 2.5 bar. Two major discrepancies were observed in the second operating point—during the compression phase and the combustion process. The simulation showed slightly higher pressure at the end of the compression stroke, suggesting that the amount of air in the cylinder at the intake valve closing moment differed slightly from the actual condition. After the top dead center (TDC), when combustion had already begun, the cylinder pressure was lower in the simulation, indicating that flame propagation was slower than the measurements. The maximum pressure difference in this case was also 2.5 bar, but in the opposite direction. There were some discrepancies during the combustion and expansion phases in the third operating point. Here, the simulation overpredicted the pressure during combustion, while during expansion, the simulated pressure remained consistently lower. This operating point showed the largest pressure deviation—up to 5 bar. The fourth operating point was the least accurate. The reaction mechanism used in the simulations likely becomes less accurate under high-pressure and high-temperature conditions. This point had the highest RMSE, although the maximum pressure deviation was similar to that of the first two operating points. No clear trend was observed in terms of peak cylinder pressure or combustion progress, as the results varied across the operating points. Overall, the simulation results were sufficiently accurate in all operating points compared to the test bench measurements. The models demonstrated predictive capabilities, making them suitable for further investigations.

Figure 3 illustrates the comparison of IMEP values. It is clearly visible that the simulation closely matches the measured data, with the maximum deviation being 4%, occurring at the third operating point. However, the cylinder pressure analysis above indicated that the largest discrepancy appeared at the fourth operating point. This difference arises because the RMSE was calculated only within the evaluation range, whereas the IMEP calculation considers the entire engine cycle. A clear trend can be observed: except for the first case, the simulation slightly underpredicts the values measured on the test bench across all operating points.

The main focus of this study is the investigation of pollutant emissions. As mentioned above, the emission measurements on the test bench were carried out upstream of the catalyst. However, this is not feasible within the simulation model. In the simulation, the most suitable method for evaluating emissions is to analyze the composition of the in-cylinder gases at the moment of exhaust valve opening. Nonetheless, this does not affect the validity of the comparison.

In the following, the HC, CO, CO₂, and NO_x pollutant emissions are compared. Figure 4 illustrates the comparison of HC emissions between the E10 simulations and the test bench measurements. It is clearly visible that the discrepancies are extremely large and not consistent at all. The possible reasons for this are as follows: the simulation overpredicts the combustion efficiency, which results in significantly lower HC emissions in the simulation; however, in this case, the cylinder pressure should also be noticeably higher compared to the test bench measurements, which is not the case. The second possible reason could be the difference in the H/C ratio between the real and the surrogate fuel, but this 2% difference should not cause such a large discrepancy in emissions. Additionally, these large deviations may also result from the reaction mechanism used in the simulations not being sufficiently detailed, which leads to an inadequate modeling of HC emissions. Another possible reason for the discrepancies is that the experminetal and simulation HC calculation methods differ. Additionally, since the amount of injected fuel was determined through simulation, there may also be slight differences in that, which could cause variations in HC emissions However, it was not possible to examine the impact of these methodological differences during the course of the study.

Figure 5 compares CO emissions between the simulation and the test bench measurements. In this case, significant discrepancies are clearly visible, and except for the last operating point, the CO values in the simulation are higher at every other operating point. The reason for this is that CO emissions are closely linked to HC emissions. Since HC emissions are much lower in the simulation, it can be inferred that the differences in HC emissions will manifest oppositely in CO emissions. An exception to this is the high-load OP4 operating point, where this relationship does not hold.

Regarding CO₂ emissions, compared to the previous two components, the magnitude of the discrepancies is much smaller, and the direction of the deviations is consistent across all operating points. The simulation underpredicts the values as shown in Figure 6, which the higher CO emissions in the simulation can explain.

Figure 7 illustrates the comparison of NO_x emissions. In this case, no clear trend can be observed in the pattern of the discrepancies. Based on maximum cylinder pressure, OP1 and OP3 were the closest to the test bench measurements, which is also reflected in the NO_x emissions, as temperature has a strong influence on their formation. The largest deviation in maximum cylinder pressure occurred at OP2, which is similarly evident in the NO_x emissions. Except for OP2, there is a good correlation between the simulation and the measurements.

3.3. Comparison of the E10 and E20 Simulations

The E10 simulations showed good correlation with the test bench measurements, and no tuning of the physical models was necessary. This indicates that the model is likely predictive. In the E20 simulations, only the fuel composition and the injected fuel quantity were modified. Due to the differences in air–fuel ratio (AFR) requirements between the fuels, it was necessary to increase the injected amount to maintain the same value of lambda. The simulation strategy remained the same: boundary and initial conditions were provided by 1D simulation data, and a single cycle was analyzed. Figure 8 shows the comparison of cylinder pressures between the two fuels. It can be observed that no significant pressure differences developed between the fuels. There is no clear trend in the maximum cylinder pressure values; the changes vary in direction regardless of engine speed or load. In OP1 and OP4, the E20 fuel results in slightly higher maximum cylinder pressure, despite E20 having a smaller lower heating value (LHV). This can be attributed to ethanol’s higher combustion efficiency and faster flame speed, which compensate for the lower energy content. The largest pressure difference occurs at OP3, reaching nearly 2 bar.

Figure 9 illustrates the temperature field distribution in a side-view cross-section of the cylinder for both E10 and E20 fuels, at the time of ignition and at various crank angle degrees after ignition, across all four operating points. In OP1, it can be observed that at every examined time point, the flame front is smaller for E20 fuel. At 30° after ignition (AIGN 30), the flame sizes are similar, but the temperature is lower for E20. No significant differences can be seen in flame propagation within the observed crank angle range in the second operating point. However, it is clearly visible that 30° after ignition, the core flame temperature is lower for E20. In the third operating point, E20 shows slightly higher flame propagation speed and core flame temperature, but no substantial differences are observed at this point either. In the fourth operating point, the flame propagation speed is also slightly higher with E20.

Figure 10 illustrates the IMEP values for E10 and E20 across the four investigated operating points. It can be seen that at lower engine speeds and loads, the difference between the fuels is negligible. However, as the load increases, IMEP rises in favor of E20, due to the improved combustion efficiency. The largest deviation occurs at OP4.

To further investigate the combustion process, examining the mass fraction burned (MFB) curve is useful—specifically, the duration between the 10% and 90% points. Figure 11 shows the MFB10–90 duration. A similar trend can be observed with IMEP. At lower speeds and loads, combustion takes longer with E20. However, due to the higher ethanol content, the combustion process becomes shorter at higher speeds and loads.

Despite the fact that the simulation overpredicted CO emissions and underpredicted HC emissions, it is still worthwhile analyzing how the emission components change with different fuels. Figure 12 presents the HC emissions for both fuels across the examined operating points.

Except for the low-load operating point, HC emissions are consistently higher when using E20. This contradicts several studies, but the lower AFR of the E20 fuel can explain it. To maintain the same lambda value, a larger fuel mass must be injected. Even though combustion efficiency improves, it is not enough to offset the increase in fuel quantity. It is also noticeable that the difference in HC emissions increases proportionally with load. CO emissions show a similar trend, despite the fact that E20 has twice the O/C ratio compared to E10. In every operating point, CO levels are higher with E20, as shown in Figure 13. However, no clear relationship can be observed between load and CO emissions in this case.

In terms of CO₂ emissions, illustrated in Figure 14, the E20 fuel consistently results in lower values across all operating points. This is due to its lower carbon content and higher CO formation. The differences are fairly consistent among all operating points.

Figure 15 shows the NO_x emissions comparison. In all cases, emissions are lower when using E20. This is because E20 has a lower LHV, so even though the combustion process is slightly faster, it does not result in significantly higher in-cylinder temperatures. The magnitude of the deviations varies between operating points, with the smallest difference occurring at the low-load operating point.

4. Conclusions

This study successfully developed an accurate and predictive numerical 3D CFD model to simulate the combustion process and pollutant emissions of fuels with different ethanol contents in a turbocharged, direct-injection spark-ignition gasoline engine. The CFD simulations were conducted using the CONVERGE CFD commercial software (Version 3.1).

The operating points investigated were defined based on RDE measurements. The four most common operating conditions were selected for numerical analysis, with each operating point having different engine speeds and loads. The simulations can be divided into three sets. The results of the first set are not presented here, as their sole purpose was to refine the oxygen content in the cylinder using 1D simulation data to determine the injection quantity as accurately as possible. The second set of simulations was validated using E10 fuel against test bench measurements, and assuming the predictive capability of the models, simulation models were then built for E20 fuel. The only differences between the E10 and E20 simulations were the fuel composition and the injected quantity, in order to maintain the same lambda value despite the different AFR.

For combustion modeling, the SAGE detailed chemistry model was used, which included a reaction mechanism with 201 species and 1013 reactions; thanks to the four-component ETRF fuel surrogate, no tuning was necessary. For turbulence modeling, the RNG k-epsilon turbulence model was employed, which also required no tuning. After conducting a mesh sensitivity analysis, the E10 simulations were compared with the test bench measurements.

When comparing cylinder pressures, the smallest deviation in the studied range was 0.35 RMSE, while the most significant deviation reached 1.43. From the perspective of the combustion process, the simulations closely matched the cylinder pressure measurements from the test bench across all operating points. This statement similarly applies to IMEP as well. In terms of pollutant emissions, the results were mixed. The simulations significantly underpredicted HC emissions at all operating points, likely due to the level of detail in the reaction mechanism. Conversely, CO emissions were overpredicted, which is logical given the close relationship between HC and CO. For CO₂ and NO_x emissions, the simulations approximated the test bench values satisfactorily, with the magnitude of NO_x deviation correlating with differences in cylinder pressure during combustion.

Assuming the predictive nature of the model, the E20 simulations were analyzed by adjusting the fuel composition and injection quantity based on the validated E10 simulations. Overall, it was found that at lower speeds and loads, engine performance with E20 was slightly lower than with E10; however, as speed and load increased, this trend reversed. Regarding emissions, HC emissions increased slightly, while CO emissions increased more significantly. This can be explained by the lower AFR of E20, requiring a higher amount of fuel to maintain the same lambda, which was not offset by E20′s higher combustion efficiency. NO_x emissions slightly decreased due to the faster combustion process and lower combustion temperatures. CO₂ emissions also decreased in all operating points, similar to NO_x, thanks to the lower carbon content of E20.