Applied Energy 86 (2009) 630–639

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Performance and exhaust emissions of a gasoline engine with ethanol blended
gasoline fuels using artificial neural network
G. Najafi a,*, B. Ghobadian a, T. Tavakoli a, D.R. Buttsworth b, T.F. Yusaf b, M. Faizollahnejad a
a
b

Tarbiat Modares University, Jalale-E-Aleahmad Highway, Tehran, P.O. Box: 14115-111, Iran
University of Southern Queensland, Toowoomba, 4350 QLD, Australia

a r t i c l e

i n f o

Article history:
Received 19 June 2008
Received in revised form 16 September
2008
Accepted 20 September 2008
Available online 4 November 2008
Keywords:
Artificial neural network
SI engine
Engine performance
Exhaust emissions
Ethanol–gasoline blends

a b s t r a c t
The purpose of this study is to experimentally analyse the performance and the pollutant emissions of a
four-stroke SI engine operating on ethanol–gasoline blends of 0%, 5%, 10%, 15% and 20% with the aid of
artificial neural network (ANN). The properties of bioethanol were measured based on American Society
for Testing and Materials (ASTM) standards. The experimental results revealed that using ethanol–gasoline blended fuels increased the power and torque output of the engine marginally. For ethanol blends it
was found that the brake specific fuel consumption (bsfc) was decreased while the brake thermal efficiency (gb.th.) and the volumetric efficiency (gv) were increased. The concentration of CO and HC emissions in the exhaust pipe were measured and found to be decreased when ethanol blends were
introduced. This was due to the high oxygen percentage in the ethanol. In contrast, the concentration
of CO2 and NOx was found to be increased when ethanol is introduced. An ANN model was developed
to predict a correlation between brake power, torque, brake specific fuel consumption, brake thermal efficiency, volumetric efficiency and emission components using different gasoline–ethanol blends and
speeds as inputs data. About 70% of the total experimental data were used for training purposes, while
the 30% were used for testing. A standard Back-Propagation algorithm for the engine was used in this
model. A multi layer perception network (MLP) was used for nonlinear mapping between the input
and the output parameters. It was observed that the ANN model can predict engine performance and
exhaust emissions with correlation coefficient (R) in the range of 0.97–1. Mean relative errors (MRE) values were in the range of 0.46–5.57%, while root mean square errors (RMSE) were found to be very low.
This study demonstrates that ANN approach can be used to accurately predict the SI engine performance
and emissions.
Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction
Developing renewable energy has become an important part of
worldwide energy policy to reduce greenhouse gas emissions
caused by fossil fuel [1]. Alternative transport fuels such as hydrogen, natural gas and biofuels are seen as an option to help the
transport sector in decreasing its dependency on oil and reducing
its environmental impact [2].
According to [3], using ethanol–gasoline blend fuel in a sparkignition (SI) engine caused a higher engine torque than that of
gasoline fuel. The maximum torque was obtained at 0.9 relative
air–fuel ratio [3]. The effects of ethanol–gasoline blends (E0, E10,
E20, E40 and E60) on engine exhaust emissions and performance
has been investigated by [4]. According to the results of the experiment, engine torque increased. It was also reported that blends
with ethanol allowed the compression ratio to increase without
* Corresponding author. Tel.: +98 261 2203289.
E-mail address: NAGAFY_14@yahoo.com (G. Najafi).
0306-2619/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.apenergy.2008.09.017

any knock [5]. Ref. [6] reported that the most suitable ethanol–gasoline fuel blend in terms of performance and emissions was E50 in
a small gasoline engine with low efficiency. Engine power increased by about 29% running with E50 fuel at high compression
ratio compared to running with E0 fuel. The specific fuel consumption was reduced by approximately 3% [6,7]. With increasing the
ethanol content in gasoline fuel, the heating value of the blended
fuels is decreased, while the octane number of the blended fuels
increases. Ref. [8] reported that blending unleaded gasoline with
ethanol increases the brake power, torque, volumetric and brake
thermal efficiencies and fuel consumption, while it decreases the
brake specific fuel consumption (bsfc; the same conclusions can
be reached from Refs. [9,10].) The 20 vol.% ethanol in fuel blend
gave the best results for all measured parameters at all engine
speeds [8].
Using E40 and E60 blends led to a significant reduction of CO
and HC emissions [4]. For E50, the CO, CO2, and HC emissions were
reduced by 53%, 10% and 12%, respectively [6,7]. NOx emission depends on the engine operating condition rather than the ethanol

throttle position sensor. Najafi et al. 2.1. brake specific fuel consumption.% ethanol in fuel blend gave the best results [13]. As an alternative. It was reported that using ethanol-unleaded gasoline blends as a fuel decreased the coefficient of variation in indicated mean effective pressure. A series of experiments were carried out using gasoline. The effects of using ethanol-unleaded gasoline blends on cyclic variability and emissions in a spark-ignited engine have been investigated by Ref. emission parameters such as CO. The ANN approach was used to predict the performance and exhaust emissions of internal combustion engines [15. The fuel blends were prepared just before starting the experiment to ensure that the fuel mixture was homogenous and to avoid the reaction of ethanol with water. E10. After the engine reached the stabilized working condition. manifold air pressure sensor. therefore. A 190 kW SCHENCK-WT190 eddy–current dynamometer was used in the experiments. The engine was started using gasoline fuel and it was operated until it reached the steady state condition. Ethanol is reported to be an important contributor to decreased engine-out regulated emissions and decreased brake specific energy consumption [10]. the use of ANN has been proposed to determine the engine power. enclosed with forced circulation of a cooling fluid . / Applied Energy 86 (2009) 630–639 631 Nomenclature a ANN A/F bsfc MLP MRE n p R actual output artificial neural network air fuel ratio brake specific fuel consumption (kg/k Wh) multi layer perception mean relative errors number of the points in the data set predicted output correlation coefficient content [11]. four-stroke. The engine speed. All the blends were tested under varying engine speed conditions. Description of the experimental setup and ECU In this study. and load were measured. fuel consumption. emission parameters and the exhaust gas temperature from an online and accurately calibrated exhaust gas analyser DIGAS 4000 were recorded. Testing the engine under all possible operating conditions and fuel cases is both time consuming and expensive. 1.4–45 kg/h by using laminar type flow meter.G. volumetric efficiency and emission components based on different gasoline–ethanol blends and speeds using results of experimental analysis. Artificial neural networks (ANN) are data-processing systems inspired by biological neural system and are used to solve a wide variety of problems in science and engineering. particularly for some areas where the conventional modelling methods fail. The 10 vol. Refs. [28] reported the effects of valve-timing in a spark-ignition engine on the engine performance and fuel economy using ANN [28]. injection timing.2. Above 20% ethanol. brake specific fuel consumption (bsfc). HC. [16– 24] investigated the performance of various thermal systems with the aid of ANN. Ref. An ANN has the ability to relearn to improve its performance if new data are available [14]. The relative air–fuel ratio.3 SOHC. The predictive ability of an ANN results from the training on experimental data and then validation by independent data. The properties of ethanol fuel are given in Table 2. Five separate fuel tanks were fitted to the gasoline engine and these contained gasoline and the bioethanol–gasoline blends. engine could not run smoothly. only experimental results obtained up to this percentage of ethanol will be presented. knock sensor. Testing procedure The performance and emission from the engine running on ethanol (derived from potato waste) and blended with gasoline (E5. These sensors are oxygen sensor. intake air temperature sensor. The ignition system was semi-static distributor less ignition (DLI).7 103/2750 47/5200 6200 Liquid. Fuel consumption rate was measured in RMSE SI SOHC gb.th. A schematic diagram of the experimental setup is shown in Fig. brake thermal efficiency. Pierburg model. and the various bioethanol blends. The effect of ethanol blended gasoline fuels on emissions was investigated in a spark-ignition engine with an electronic fuel injection (EFI) system [10]. four cylinder. E15 and E20) were evaluated and compared with gasoline fuel. the performance and exhaust emissions of an engine can be modelled using ANNs [15]. The engine specification is given in Table 1. brake thermal efficiency and volumetric efficiency were computed.25. A well-trained ANN can be used as a predictive model for a specific application. torque. The addition of ethanol to gasoline fuel enhances the octane number of the blended fuels and changes distillation temperature. while the brake power. NOx and the exhaust gas temperature from an online and accurately calibrated exhaust gas analyser were Table 1 Main characteristics of the test engine Engine type Combustion order Bore  stroke (mm) Displacement volume (cc) Compression ratio Maximum torque (N m/rpm) Maximum power (kW/rpm) Maximum speed (rpm) Cooling system 8 Valves – 4 cylinder-inline-SOHC 1-3-4-2 71  83.26] and the specific fuel consumption and fuel air equivalence ratio of a diesel engine [27]. The engine control unit (ECU) that was used in this engine was a Johnson Controls JCAE S2000. the experiments were performed on a KIA 1. and spark-ignition (SI) gasoline engine. CO2. ignition timing and engine speed by receiving signals from seven sensors. [13]. gi gv k / root mean square error spark ignition single over head camshaft brake thermal efficiency (%) indicated efficiency (%) volumetric efficiency (%) relative air–fuel ratio equivalence fuel–air ratio the range of 0. 2. Experimental work 2. NOx concentrations are increased due to rising of the cylinder temperature with increasing ethanol percentage in the blends [9]. water temperature sensor and engine speed sensor. Air consumption was measured using an air flow meter. Ref. In this study. [12] reported that using E10 blended fuel at a relative air–fuel ratio (k) slightly greater than one can generally reduce pollutant emission components [12]. A multi point fuel injection (MPFI) system with top-feed injectors is used to inject the fuel into the combustion chamber.6 1323 9. ECU function is to control the quantity of fuel.

the bsfc decreases reaching its minimum values.000 108.4 45. the brake thermal efficiency increases reaching its maximum values. This is due to the increase in brake thermal efficiency. 3.4 144. Brake specific fuel consumption The relationship between engine speed and brake specific fuel consumption for different blends and gasoline is shown in Fig. 10% ethanol (E10).0 99.6 gine brake power slightly increased for all engine speeds. 3. Schematic diagram of experimental setup.9 53.3.7 92.16 55. the enTable 3 Properties of different ethanol/gasoline-blended fuels Property item Test fuel E0 Vapour pressure (kPa).9 54.1 38. As shown from Fig. The properties of the five fuels have been summarized in Table 3. reid@37.1 44. 2b shows the influence of different ethanol–gasoline blended fuels on engine torque. 2d. 1. When the ethanol content in the blended fuel is increased.4 71. the indicated efficiency gi increases).6 142. 2d shows the relationship between the volumetric efficiency (gv) and the percentage of ethanol in the fuel blends.1. Ethanol is an . 2a shows the effect of various fuels on engine brake power.3 93. and thus higher power output is obtained [6]. electronic ignition system was used. The relationship between the engine speeds and the brake thermal efficiency is given in Fig. As engine speed increases reaching 3500 rpm. the engine was operated using the new fuel for sufficient time to clean out the remaining fuel from the previous blend. Five test fuels were used in this study: 0% ethanol (E0). 3.) was approximately 35% when 20% ethanol was in the fuel blend.2. the indicated work increases (i. Volumetric efficiency Fig.3 94.9 143.16 55.8 55. The maximum brake thermal efficiency (gb. Experimental results 3. Table 2 The properties of potato’s wastes ethanol Property Method Density (kg/m3) Viscosity (cSt) Calorific value (kJ/kg) Research octane number Pour point (°C) Flash point (°C) Ash content (mass%) ASTM ASTM ASTM ASTM ASTM ASTM ASTM D D D D D D D Ethanol (E100) 4052 88 240 2699 97 93 482 785 1. Brake power Fig.1.1. As the E% increases in the fuel blend.7 40. 2c presents the effect of using ethanol–gasoline blends on brake thermal efficiency. With the increase in ethanol percentage. 3. The increase of ethanol content increases the torque of the engine.1.6 93.8 58. Equivalence fuel–air ratio (/) Equivalence fuel–air ratio (/) is one of the important parameters that effects engine performance parameters. Najafi et al.3 89. 3.1 71.632 G.1.9 184.5 147.. The gain of the engine power can be attributed to the increase of the indicated mean effective pressure for higher ethanol content blends [9].4.6. which raises the octane number) allowed a more advanced timing that results in higher combustion pressure and thus higher torque [7.16 55.8 °C Octane number (research) Gross heat of combustion (MJ/kg) Distillation range (°C) @760 mmHg IBP 10 vol% 50 vol% 90 vol% End point E5 E10 E15 E20 48. the brake thermal efficiency increases as the ethanol percentage increases. the volumetric efficiency increases as the ethanol percentage increases for all engine speeds.1. Fuel properties were determined at the laboratories of Research Institute of Petroleum Industry (RIPI) in Iran.61 40. the quantity EX represents a blend consisting of X% bioethanol by volume. Torque output Fig. As the engine speed increases reaching 3000 rpm. / Applied Energy 86 (2009) 630–639 Fig.7 182. E5 indicates a blend consisting of 5% ethanol in 95% gasoline. As shown in this figure.1. Added ethanol produces lean mixtures that increase the relative air–fuel ratio (k) to a higher value and makes the burning more efficient [11]. the bsfc decreases as the ethanol percentage increases.16 85. 3. 15% ethanol (E15).29].4 40. and 20% ethanol (E20).th.1 27. Before obtaining data from the engine operated with a new blended fuel.0 57.26 55. the density of the mixture and the engine volumetric efficiency increases and this causes the increase of power [8]. this provides fuel–air charge cooling and increases the density of the charge. As shown in this figure. To adjust ignition timing.9 175.1.3 146.2 71. The heat of evaporation of ethanol is higher than that gasoline.1 176.6 <<50 14 0 recorded.51 Test method ASTMD323 ASTMD2699 ASTMD340 ASTMD86 35. All experiments have been carried out at full throttle setting. 5% ethanol (E5).5.87 41. Engine performance 3.g.e. 2c. The improved antiknock behaviour (due to the addition of ethanol. Brake thermal efficiency Fig. 2e. In this paper.12 44.. e.15 42.0 176.

the CO concentration decreases which means the combustion is tuned to be completed. Experimental results of (a) brake power.22 0 1000 2000 3000 4000 5000 6000 Engine speed (rpm) Equivalence fuel-air ratio (φ) e bsfc (kg/kW. / Applied Energy 86 (2009) 630–639 b 50 45 40 35 30 25 20 15 10 5 0 110 E0 105 E0 E5 E10 E15 E20 Torque (Nm) Brake power (kW) a E5 E10 100 E15 95 E20 90 85 80 0 1000 2000 3000 4000 5000 6000 0 1000 Engine speed (rpm) 3000 4000 5000 d 40 E0 35 E5 E10 E15 E20 30 25 20 0 1000 2000 3000 4000 5000 Volumetric efficiency (%) Brake thermal efficiency (%) c 6000 E0 105 E5 E10 100 E15 95 E20 90 85 80 75 0 6000 1000 Engine speed (rpm) 2000 3000 4000 5000 6000 Engine speed (rpm) f 0.55.2. E10.32 0.28 E5 0. while the CO2 concentration of E5.th.8 (%V). CO2 emission depends on relative air–fuel ratio and CO emission concentration [6. volumetric efficiency and brake thermal efficiency are increased when the ethanol amount in the blended fuel is increased. Relative air–fuel ratio (k) approaches 1 as the ethanol content of the blended fuel increases. (b) Torque. 3. E15 and E20 at 3000 rpm was 4.2.93% and 45. increasing ethanol blend ratio leads to lean and complete combustion and therefore the bsfc was slightly decreased shown in Fig. Engine emission studies 3. 24. Najafi et al. The most significant parameter affecting CO concentration is the relative air–fuel ratio (k) [6. E10.56 (%V). The CO concentration in the exhaust gas emission at 3000 rpm for gasoline fuel was 4. The reduction in CO concentration using blended fuels is due to the fact that ethanol (C2H5OH) has less carbon than gasoline (C8H18). respectively. adding ethanol to gasoline leads to leaner operation and improves combustion [9]. (c) gb. 2f).2. CO2 emission Fig. while the CO concentration of E5. respectively in comparison to gasoline.26 E10 E15 0. oxygenated fuel. respectively.38 and 2.0 E15 E20 0.7%. As previously mentioned.12]. E15 and E20 was decreased by 13.69 (%V).4 (%V).05. CO concentrations at 3000 rpm using E5.42%.31%. 27. Engine performance parameters such as brake power. 3b shows the relationship between the CO2 concentrations and engine speeds for different blends percentage.9. The CO2 concentrations at 3000 rpm using E5.24 E20 0.1. 3.2. 2. (e) bsfc and (f) equivalence fuel–air ratio at different fuel blends and engine speeds.11. E10. E15 and E20 was increased by 3.2 E0 1. It can be seen from this figure that when ethanol percentage increases. 3b indicates that CO2 concentration increases as the ethanol percentage increases. CO emission Fig.30 E0 0. 2. Fig. and for this reason.3 and 13. 3a shows the concentrations of CO emission for different engine speeds.2. E15 and E20 at 3000 rpm was 12. (d) gv. torque. The . and consequently combustion becomes complete [9.9 0 1000 2000 3000 4000 5000 6000 Engine speed (rpm) Fig. 13.34 0.633 G. 3. 13.1 E5 E10 1. E10.12].hr) 2000 Engine speed (rpm) 1. This is due to the reduction in the / values.12]. The CO2 concentration in the exhaust gas emission at 3000 rpm for gasoline fuel was 12. Experimental results show that equivalence fuel–air ratio decreases as the percentage of ethanol (in volume) in the blended fuel increases (Fig. Another significant reason of this reduction is that the oxygen content in the blended fuels increases the oxygen-to-fuel ratio in the fuel-rich regions..3 1.

HC emission HC emissions for different speeds are illustrated in Fig. respectively.11. The HC concentration in the exhaust gas emission at 3000 rpm for gasoline fuel was 183 ppm.94%. while the NOx concentration of E5.2. 6. The NOx concentration in the exhaust gas emission at 3000 rpm for gasoline fuel was 876 ppm. E10. the NOx emission is increased. 4. ANN has been applied widely.69% at 3000 rpm.14% and 31. 24.3. particularly by the increase of thermal NO [9. 4. 33. 3c. 137 and 125 ppm.06%.12]. 1319 and 1609 ppm. E15 and E20 was decreased by 16.76% and 10. respectively in comparison to gasoline. (c) HC and (d) NOx at different gasoline–ethanol blends and engine speeds. (b) CO2. while the HC concentration of E5. NOx emission was increased.87%. nonlinear tool and since many phenomena in industry have non-linear characteristics.2.04%.6% and 45.4. therefore. The performance of the ANN-based predictions is evaluated by regression analysis of the network outputs (predicted parameters) and the experimental values [15]. It shows that as the percentage of ethanol in the blends increased.14%.55%. When the combustion process is closer to stoichiometric. Artificial neural networks ANN is an analytical method for simulating system performance. NOx emission Considering the NOx emission. E10. E15 and E20 at 3000 rpm was 1002. respectively. 3.11].12]. The HC concentration at 3000 rpm using E5. 33. 3. flame temperature increases. ‘train’ the ANN so that it can precisely predict the system performance at other conditions. This technique has found application in situations where the simulation of complex systems is required but limited experimental data is available. Experimental results of (a) CO. The method relies on experimental data that is used to Fig. 25.94%. The structure of ANN for gasoline engine with gasoline–ethanol blended fuels. respectively in comparison to gasoline. the CO2 emission increased because of the improved combustion [8. This result indicates that ethanol can significantly reduce HC emissions. respectively in comparison to gasoline. 6. The error identified during the learning process is called the root-mean-squared-error (RMSE) and is defined as follows: 3. The concentration of HC emission decreases with the increase of the relative air–fuel ratio. E15 and E20 was increased by 12. Fig. / Applied Energy 86 (2009) 630–639 a b 8 E0 E5 14 E0 4 E5 E10 2 CO2 (%V) 6 CO (%V) 15 E15 E10 13 E15 E20 12 11 E20 0 10 0 1000 2000 3000 4000 5000 6000 0 1000 Engine speed (rpm) c 2000 3000 4000 5000 400 E0 d 2000 E0 E5 300 E5 E10 E15 E20 200 1500 NOx(ppm) HC (ppm) 6000 Engine speed (rpm) E10 E15 1000 E20 500 100 0 0 0 1000 2000 3000 4000 5000 6000 Engine speed (rpm) 0 1000 2000 3000 4000 5000 6000 Engine speed (rpm) Fig. 139.57%. E10. . The NOx concentrations at 3000 rpm using E5. E15 and E20 at 3000 rpm was 152. the reason for the decrease of HC concentration is similar to that of CO concentration described above [11. 3d shows that the NOx concentration is higher when ethanol percentage increases. 1326. ANN is a powerful.634 G. 3. As a result of the lean burning associated with increasing ethanol percentages. Najafi et al. E10.

brake specific fuel consumption.39  106 8.091  105 2.635 G.981 RMSE=0.hr) 105 110 95 90 R=0.. A standard Back- ð1Þ : a 40 30 20 R=0. and a and p are actual output and predicted output sets. brake thermal efficiency.0137 8.11  104 2.74 (kW) MRE=2. (b) Torque.99929 – 0.99997 0.48 (%) MRE=0.999 RMSE=0. The experimental data set for every output parameter includes 45 values. Approximately 70% of the total experimental data (405 values) was selected at random and was used for training purpose.51  103 6  103 6.26 0. The correlation coefficient can vary between 1 and +1.28  106 0. respectively [15].32 100 d 100 c 25 95 Experimental Torque (Nm) Predicted Volumetric efficiency (%) Predicted Brake thermal efficiency (%) 40 ð2Þ where n is the number of the points in the data set. .48 (%) 85 80 80 40 Experimental Brake thermal efficiency (%) 0. The ANN predictions for the (a) brake power.hr) MRE=0.99  106 4.24 0.30 0. (d) gv and (e) bsfc versus experimental values.th.hr) Fig.75  106 9. but R values closer to +1 50 10 20 30 40 105 100 95 R=0.22 0.26 R=0. which shows the mean ratio between the error and the experimental values. of which 30 values were used for training the network and 15 values were selected randomly to test the performance of the trained network.9993 0.59 (%) MRE=1. is defined as Table 4 Summary of different networks evaluated to yield the criteria of network performance Activation function Training rule Neurons in hidden layer Training error R sig/lin tan/lin sig/lin sig/lin sig/lin sig/lin sig/lin sig/lin sig/lin sig/lin sig/lin trainlm trainlm traingdx trainscg trainrp trainlm trainlm trainlm trainlm trainlm trainlm 20 20 20 20 20 19 18 17 21 22 23 9.995 RMSE=0.85 (%) 0.99997 0. Najafi et al.003 (kg/kW.72  104 9. !12 n  n   1X 100  ðai  pi Þ: n i¼1  ai  85 90 95 100 Experimental Volumetric efficiency (%) e 0.61  104 0.986 RMSE=0. torque.9992 0.46 (%) 90 85 85 50 90 Experimental Brake power (kW) 35 30 20 20 R =0.30 0. Modelling with ANN An ANN model was developed to predict a correlation between brake power.24 0.28 (%) 25 30 35 Predicted bsfc (kg/kW.22 0. while the 30% was reserved for testing.28 0.28 0. The mean relative error.99997 Pn RMSE ¼  pi Þ2 i¼1 ðai MRE ð%Þ ¼ 5. / Applied Energy 86 (2009) 630–639 indicate a stronger positive linear relationship.99995 0.985 RMSE=0.32 (%) 10 0 0 b 110 Predicted Torque (Nm) Predicted Brake power (kW) The correlation coefficient (R) and mean relative error (MRE) are used for characterising the network performance. volumetric efficiency and emission components using different gasoline–ethanol blends and speeds as inputs data. while R values closer to 1 indicate a stronger negative relationship [15].99221 0.49 (Nm) MRE=0.32 Experimental bsfc (kg/kW. (c) gb.99106 0.9995 0. 5.

5. The results showed that the training algorithm of Back Propagation was sufficient for predicting engine torque. This ANN model is limited to the research engine that was used in this project operating on at wide open throttle conditions (the engine specifications are given in Table 1).18 (%V) MRE=4. A multi layer perception network (MLP) was used for nonlinear mapping between the input and the output parameters.987 RMSE=5. This algorithm uses the supervised training technique where the network weights and biases are initialized randomly at the beginning of the training phase. However. It was experimentally demonstrated that adding 20% ethanol to the blends led to an increase in the engine brake power. The backpropagation algorithm was utilized in training of all ANN models. Three criteria R. 22 and 23. torque and brake thermal efficiency.23 (%) Predicted NOx (ppm) Predicted HC (ppm) c 200 150 100 100 150 200 250 Experimental HC (ppm) 11 12 13 14 15 Experimental CO2 (%V) Exxperimental CO (%V) 300 2000 R=0. (c) HC and (d) NOx versus experimental values. Therefore. R values did not increase when neurons in the hidden layer is more than 20 such as 21.973 RMSE=89.985 RMSE=0.54 (%) 14 13 12 11 10 10 0 0 15 8 300 250 d R=0. Results and discussion Adding ethanol to gasoline will lead to a leaner better combustion. The experimental results confirmed that by adding more ethanol. To get the best prediction by the network. The two input variables are engine speed in rpm and the percentage of bioethanol blending with the conventional gasoline fuel. volumetric efficiency and decreases the brake specific fuel consumption. (b) CO2. The ANN predictions for the (a) CO. CO2 emission varies with the A/F ratio and CO concentration. 2000 . There is 1 neural network structure with 2 inputs and 9 outputs.636 G. The number of hidden layers and neurons within each layer can be designed by the complexity of the problem and data set. It is noted that adding ethanol to the blends reduces the HC emission because of oxygen enhancement. brake thermal efficiency. specific fuel consumption and exhaust gas components for different engine speeds and different fuel blends ratios. a network with one hidden layer 6. the input layer consisted of 2 neurons which corresponded to engine speed and levels of biofuel blends and the output layer had 9 neurons. 4. several architectures were evaluated and trained using the experimental data. A linear function was best suited for the output layer. The ANN predictions for the (a) brake power. (b) engine torque. 4 2 2 4 6 R=0. many other networks with several functions and topologies were examined. Unburned HC is a product of incomplete combustion which is related to A/F ratio. and therefore smaller ANNs had to be selected. therefore. The nine outputs for evaluating engine performance are indicated in Fig. The oxygen enrichment generated from ethanol increased the oxygen ratio in the charge and lead to lean combustion. Different training algorithms were tested and Levenberg–Marquardt (trainlm) was selected. A regression analysis between the network response and the corresponding targets was performed to investigate the network response in more detail. the NOx emission increased. brake power. the network was decided to consist of one hidden layer with 20 neurons.85 (ppm) MRE=5. In this study.57 (%) 1500 1000 500 0 0 500 1000 1500 Experimental NOx (ppm) Fig.41 (ppm) MRE=2. When the combustion process is closer to stoichiometric. As a result. An artificial neural network (ANN) was developed based on this experimental work.21 (%) 6 Predicted CO2 (%V) Predicted CO (%V) a and 20 neurons was selected as the optimum ANN. R values in Table 4 represent the correlation coefficient between the outputs and targets.24 (%V) MRE=1. volumetric efficiency. Simulations were performed using MATLAB. The predicted versus experimental values for the experimental parameters are indicated in Fig. The lean combustion improves the completeness of combustion and therefore the CO emission was expected to be decreased. the CO was decreased. Najafi et al. / Applied Energy 86 (2009) 630–639 Propagation algorithm for the engine was used in this model. There were two input and nine output parameters in the experimental tests. flame temperature increases. the CO2 emission increased because of the improved combustion. As seen in Table 4. RMSE and MRE were selected to evaluate the networks to find the optimum solution. (c) brake thermal efficiency. The error minimization process is achieved using gradient descent rule. The activation function for hidden layer was selected to be sigmoid function.994 RMSE=0. 6. (d) volumetric efficiency and (e) brake specific fuel consumption yield a correlation coefficient b 8 R=0. The complexity and size of the network was also an important consideration.

respectively. 6b indicates the predicted versus experimental values for the CO2 emission with correlation coefficient (R) of 0. It is seen that the test pattern consist of the results of 15 tests. 6a.32% (Fig.21%. (c) gb. 0. 5b indicates the predicted versus experimental values for the engine torque with mean relative error (MRE) of 0.49 N m.18%V and mean relative error (MRE) of 4.003 kg/kW h for the (a) brake power. 7.46–5. 0.637 G. 5e).22 0 5 10 15 Test pattern Fig. Fig.74 kW.26 0. root mean square error (RMSE) of 89.23%.985 and 0.57%.. The ANN pre- 40 dictions for the CO yield a correlation coefficient (R) of 0. 6c.85 pm and mean relative error (MRE) of 5. Comparisons of experimental results and the ANN predictions for the (a) brake power. root mean square error (RMSE) of 0. root mean square error (RMSE) of 5. The ANN predictions for the NOx yield a correlation coefficient (R) of 0.973. Mean relative error (MRE) of brake thermal efficiency was 1. respectively. The predicted versus experimental values for the HC emission are indicated in Fig. Mean relative errors (MRE) values were in the range of 0. / Applied Energy 86 (2009) 630–639 a 50 Brake power (kW) (R) of 0. 7 and 8.46%. (d) gv and (e) bsfc for various test patterns. The ANN predictions for the brake specific fuel consumption (bsfc) yield mean relative error (MRE) of 0. The ANN predictions for the volumetric efficiency yield a mean relative error (MRE) of 0.995. The ANN predictions for the performance of the tested SI engine yielded a good statistical performance. (b) Torque. . Fig.28% (Fig. The ANN predictions for the brake power yield a mean relative error (MRE) of 2.986. 6d indicates the predicted versus experimental values for the NOx emission.981.985.30 0. Fig. It was observed that the ANN model can predict engine performance and exhaust emissions with a correlation coefficient (R) in the range of 0.th. 5a). 5c). A comparative presentation of the error during testing by using ANN and experimental results is shown in Figs. Najafi et al.48%. It was found the root mean square error (RMSE) values were 0.59%. Comparisons of the experimental results and the ANN predictions demonstrate that SI engines using ethanol–gasoline blended fuels can be accurately simulated using ANN. while root mean square errors (RMSE) were found to b 110 Experimental Predicted Torque (Nm) 105 30 20 Experimental 10 100 95 90 Predicted 85 0 0 5 10 15 0 Test pattern 10 15 Test pattern c d 40 Experimental Volumetric efficiency (%) Brake thermal efficiency (%) 5 Predicted 35 30 25 20 0 5 10 15 100 Experimental Predicted 95 90 85 80 0 Test pattern bsfc (kg/kW. (d) volumetric efficiency and (e) brake specific fuel consumption.32 Experimental Predicted 0. (c) brake thermal efficiency. 0.85% (Fig.999.41 ppm and mean relative error (MRE) of 2.57%. Fig. 0. The predicted versus experimental values for the CO emission are indicated in Fig.54%.24%V and mean relative error (MRE) of 1. 5d indicates the predicted versus experimental values for the volumetric efficiency. root mean square error (RMSE) of 0.24 0.48% and 0. 0. (b) engine torque.987.994. 0.28 0. The ANN predictions for the HC yield a correlation coefficient (R) of 0.97–1.hr) e 5 10 15 Test pattern 0.

01. Arcakliog˘lu E. [16] Kalogirou SA. Effect of ethanol-unleaded gasoline blends on engine performance and exhaust emissions. Leduc G. (b) CO2. Application of artificial neural-networks for energy systems. Renew Sust Energy Rev 2008. Neural network analysis of fintube refrigerating heat exchanger with limited experimental data. Bansal PK. Lin TH.011.2008.30: 1733–47. 1991. Experimental determination of suitable ethanol–gasoline blend rate at high compression ratio for gasoline engine.26(17–18):2272–8. Prog Energy Combust Sci 2007.36:403–10. in press. Effect of ethanol–gasoline blends on engine performance and exhaust emissions in different compression ratios. Renew Sust Energy Rev. Appl Energy 2000. References [1] Huang Y-H. [8] Al-Hasan M. Palmer RG. Conclusions The present work demonstrates that the use of ethanol–gasoline blended fuels will marginally increase the brake power and decrease the brake specific fuel consumption. Hosoz M. [5] Topgül T.1016/j. Sozen A. Yan XG.31(15):2534–42. Canakci M. [10] He BQ. Energy Conv Manage 2003(44):1547–61. Engine performance and pollutant emission of an SI engine using ethanol–gasoline blended fuels. Krogh A. A study on emission characteristics of an EFI engine with ethanol blended gasoline fuels. Christidis P. The effects of ethanol–unleaded gasoline blends and ignition timing on performance and exhaust emissions. R values in this model are very close to one. Wu J-H. Analysis of the experimental data by the ANN revealed that there is a good correlation between the ANN-predicted results and the experimental data. The ANN results are very good. Kilicaslan I. [12] Wu CW. [11] Hsieh WD. doi:10. Redwood City. Performance and exhaust emissions of a gasoline engine using artificial neural network. Power plant condenser performance forecasting using a non-fully connected ANN. Topgül T. Wu TL. Topgül T. Çinar C. Effects of ethanol–unleaded gasoline blends on cyclic variability and emissions in an SI engine. 7. McClain RL. Wang JX. It was also found that the brake thermal efficiency and volumetric efficiency increase when ethanol–gasoline blends are used. Pelkmans L. Hao JM. [7] Agarwal AK. . The CO2 and NOx concentrations were increased while the concentrations of CO and HC were decreased when ethanol–gasoline blends are used. ANN appears to be a useful method for simulating engine parameters. Renew Energy 2005. / Applied Energy 86 (2009) 630–639 a 8 Experimental b Predicted CO2 (%V) CO (%V) Experimental Predicted 14 6 4 15 2 13 12 11 0 10 0 5 10 15 0 Test pattern c 300 5 Experimental d 2000 15 Experimental Predicted Predicted 1500 NOx (ppm) 250 HC (ppm) 10 Test pattern 200 1000 500 150 0 100 0 5 10 15 Test pattern 0 5 10 15 Test pattern Fig. Energy 2001. Yücesu HS.27: 358–68. ANN provided an accurate and simple approach in the analysis of this complex. Appl Therm Eng 2005. Atmos Environ 2003. [13] Ceviz MA. Biofuel support policies in Europe: lessons learnt for the long way ahead.25: 917–25.26:65–79. (c) HC and (d) NOx for various test patterns. Chen RH.33:233–71. [9] Bayraktar H. Najafi et al. Comparative study of mathematical and experimental analysis of spark ignition engine performance used ethanol–gasoline blend fuel. [3] Yücesu HS. [6] Celik MB. Schade B. Lin TH. [17] Pacheco-Vega A. Therefore. Pu JY. Introduction to the theory of neural computation. Koca A. Renew Energy 2006. Montanes E. Govaerts L et al. Çinar C. [2] Wiesenthal T.67:17–35.37(7):949–57. Neural networks – a new approach to model vapour-compression heat pumps.28:396–404.25:591–9. Comparisons of experimental results and the ANN predictions for the (a) CO. Int J Energy Res 2001. [4] Yücesu HS. multivariate problem.rser.27:46–54. the analysis of the SI engine performance and emissions. Analysis of biodiesel promotion in Taiwan. [15] Sayin C. Atmos Environ 2004. Appl Therm Eng 2006. Experimental and theoretical investigation of using gasoline– ethanol blends in spark-ignition engines. while root mean square errors (RMSE) were found to be very low. Xiao JH. Atmos Environ 2002. There is a good correlation between the simulations from ANN and the measured data. [18] Bechtler H. Ertunc HM. be very low.44:763–70. Therefore ANN proved to be a useful tool for correlation and simulation of engine parameters. Appl Therm Eng 2007. Chen RH. Acknowledgements The authors wish to thank the Iranian Fuel Conservation Organization (IFCO) of NIOC for the research grant provided to complete this project and Mega Motor Company for providing of laboratory facilities. Appl Therm Eng 2008. Kecman V.638 G. The influence of air–fuel ratio on engine performance and pollutant emission of an SI engine using ethanol–gasolineblended fuels. Appl Therm Eng 2007. Int J Heat Mass Transfer 2001. 8. Yang KT.38:7093–100. Sen M.12:1176–86. Browne MW. Yüksel F. [19] Prieto MM. Biofuels (alcohols and biodiesel) applications as fuels for internal combustion engines. [14] Hertz J. NJ: Addison-Wesley Publishing Company. Menendez O. Okur M.

81:187–97.23:937–52. Celikten I. Erduranli P.80(4):565–74. Performance comparison of CFCs with their substitutes using artificial neural network. Appl Therm Eng 2003. / Applied Energy 86 (2009) 630–639 [20] Chouai A. Vaitilingom G.47:1574–87.28:1113–25. [24] Hosoz M. Berge JC. A diesel engine’s performance and exhaust emissions.80:11–22. Arcaklioglu E. Appl Therm Eng 2006. Modelling of thermodynamic properties using neural networks – application to refrigerants. Appl Energy 2005. [26] Arcaklioglu E.26:627–35. Appl Energy 2006.G. [21] Sozen A. Sekmen Y. Fuel 2001. [28] Golcu M. [23] Ertunc HM. Richon D. Ozalp M. Erdil A. Artificial neural network analysis of a refrigeration system with an evaporative condenser. Performance maps of a diesel engine. Artificial neural network based modelling of variable valve-timing in a spark ignition engine. Fluid Phase Equilibr 2002. Hosoz M. Energy Conv Manage 2006. Interest of combining an additive with diesel ethanol blends for use in diesel engines. Appl Energy 2005. Caro PS. [27] Celik V.83:594–605. Appl Energy 2005. [29] Mouloungui Z. . Salman S. 639 [25] Canakci M.81:247–59. Ertunc HM. A new approach to thermodynamic analysis of ejector-absorption cycle: artificial neural networks.199:53–62. Artificial neural network analysis of an automobile air conditioning system. Laugeier S. [22] Arcaklioglu E. Arcaklioglu E. Int J Energy Res 2004. Arcaklioglu E. Najafi et al. Performance and exhaust emissions of a biodiesel engine.