Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

1 s2.0-s0016236111006958-main

392 views

Published on

  • Be the first to comment

  • Be the first to like this

1 s2.0-s0016236111006958-main

  1. 1. A parametric study for specific fuel consumption of an intercooled diesel engine using a neural network Abdullah Uzun ⇑ Sakarya Vocational School, Automotive Programming, Sakarya University, 54187 Sakarya, Turkey a r t i c l e i n f o Article history: Received 23 June 2010 Received in revised form 31 October 2011 Accepted 2 November 2011 Available online 15 November 2011 Keywords: Neural networks Intercooling Specific fuel consumption Scaled conjugate gradient algorithm Diesel engine a b s t r a c t Turbocharging is a process wherein the amount of oxygen used in a combustion reaction is increased to raise output and decrease specific fuel consumption. On account of this, fuel economy and thermal effi- ciency are more important for all engines. The use of an intercooler reduces the temperature of intake air to the engine, and this cooler and denser air increases thermal and volumetric efficiency. Most research projects on engineering problems usually take the form of experimental studies. However, experimental research is relatively expensive and time consuming. In recent years, Neural Networks (NNs) have increasingly been used in a diverse range of engineering applications. In this study, various parametric studies are executed to investigate the interrelationship between a single variable and two steadies and two constant parameters on the brake specific fuel consumption (BSFC, g/kW h). The variables selected are engine speed, load and Crankshaft Angel (CA). The data used in the present study were obtained from previous experimental research by the author. These data were used to enhance, train and test a NN model using a MATLAB-based program. The results of the NN based model were found to be convincing and were consistent with the experimental results. The trained NN based model was then used to perform the parametric studies. The performance of the NN based model and the results of parametric studies are presented in graphical form and evaluated. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction Diesel engines have a number of advantages compared with other types of internal combustion engines, including greater reli- ability, longer operating life, lower operating costs and increased en- gine efficiency. They are able to operate at higher compression ratios than gasoline engines, because the fuel is mixed with the air at the start of the combustion process. This higher compression ratio leads to better fuel efficiency. Since diesel engines are more efficient and robust than gasoline engines, they are widely used. However, owing to economicconsiderations,greater fueleconomyisstill pursued [1]. The emphasis on fuel conservation in the 1970s stimulated the use of turbocharging to improve the fuel economy of those diesel engines that run at or near full load for long periods of time. This category includes marine, stationary and heavy-road vehicle en- gines. By improving structural design, many engines used in these applications are supercharged to high mean effective pressure, of- ten accompanied by reduced piston speed, with resulting im- proved mechanical efficiency [2]. The combustion process in a diesel engine is an extremely complex process. Within a diesel engine, this process depends primarily on the fuel–air ratio. Intake air mass flow intake air flow and its temperature is are important parts of this process in terms of effective engine efficiency [3]. In turbocharged engines, especially at high engine speeds and loads at the compressor output, temperatures rise and the density and, consequently, the amount of air induced into the engine de- crease. To avoid this decrease in engine induction air and engine power, the compressor output air temperature must be cooled. An intercooler cools the intake air charge with air or water coolant from the turbocharger compressor before it enters the en- gine. In vehicles, a charge air cooler system is mounted in front of the radiator. In air-to-water charge cooling, the compressed air is cooled in a heat exchanger, which is either separate from or built into the intake manifold. In air-to air charge coolers, the lower- temperature ambient air is utilized directly for cooling rather than using engine coolant as the intermediate system. During move for vehicles, the coolant must be at the atmospheric and space temperature [4]. Optimum diesel engine performance is related to the engine design, operating parameters and fuel properties. [5]. Brake specific fuel consumption (BSFC) is a measure of engine and fuel efficiency within a shaft reciprocating engine. BSFC is the rate of fuel consumption divided by the power produced. BSFC allows the fuel efficiency of different reciprocating engines to be directly 0016-2361/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.fuel.2011.11.004 ⇑ Tel.: +90 264 295 74 60; fax: +90 264 278 65 18. E-mail address: abdullahuzunoglu@gmail.com Fuel 93 (2012) 189–199 Contents lists available at SciVerse ScienceDirect Fuel journal homepage: www.elsevier.com/locate/fuel
  2. 2. compared. The influence of intercooling on diesel engine perfor- mance parameters has been examined using numerical and exper- imental methods [6–9]. In the present study, a four-stroke, direct-injection, six cylinder, turbocharged, intercooled diesel engine was selected as a test en- gine. First, the engine was tested equipped with a water-cooler intercooler, at different speeds and load conditions. Then, this en- gine was again tested without an intercooler under the same con- ditions. The fuel consumption was measured by operating the test engine at different loads and speeds with and without an inter- cooler. In these tests, it is more economical than trying to find work ranges which were targeted [10]. In many engineering applica- tions, it can be difficult to obtain experimental data for various parameters. Neural Networks (NNs) have increasingly been used in a variety of engineering applications. There are many NN studies within the literature on diesel en- gine performance [11–13]. All diesel engine performance data are difficult to measure experimentally. Test measuring ranges are 1600–2000–2400 rpm and 18–20–22 CA, 400–450–500 N, for all number of engine speeds for each crankshafts angle values, test loads. Similarly, performance data have been developed using NNs. Previous studies have used empirical data to train NNs. After the training period, the network is used to estimate the values of parameters that have not been presented to the system. Table 1 Specification of the test engine [14]. Engine type FORD 6.0 LT T/C intercooling, DI Stroke 4 Cylinder 6 Cylinder diameter (mm) 104.77 Piston stroke (mm) 114.9 Compression ratio 16.5/1 Max. engine power (kW) 136 (2400 rpm) Max speed (rpm) 2750–2780 Idle speed (rpm) 665–685 Engine volume (lt) 5.947 Injection timing (°CA) 20 Ignition order 1–5-3–6-2–4 Engine weight (kg) 500 Fig. 1. Testing scheme. Fig. 2. Test engine set. Table 2 Range of parameters in the database and normalization values. Min Max Normalization values Input parameters CA 18 24 24 Load(F) 300 550 550 Speed (rpm) 1000 2400 2400 Output parameters BSFC w. intercooler 120,58 188,37 250 BSFC w.out intercooler 129,49 207,56 250 190 A. Uzun / Fuel 93 (2012) 189–199
  3. 3. The main purpose of the present study is to conduct parametric studies to investigate the effect of CA, engine speed and loads on diesel engine performance in terms of brake specific fuel consump- tion (BSFC, g/kW h), both with and without intercooling. Firstly, a series of experimental studies were conducted. The results of these tests were used to train the NN based model. The results of the NN based model used in the present study are convincing and are con- sistent with the experimental results. NN based model outputs are compared with experimental results. These results indicated that the NN based model was highly successful in determining the BSFC of the diesel test engine. The well-trained NN based model was then used to perform parametric studies. The performance of the NN based model and the results of parametric studies are pre- sented in graphical form and evaluated. BSFC1 biasbias CA F rpm BSFC2 hidden layer Fig. 3. Architecture of the NN model. 0 0,2 0,4 0,6 0,8 1 0 0,2 0,4 0,6 0,8 1 Experimental NN (a) training set 0 0,2 0,4 0,6 0,8 1 0 0,2 0,4 0,6 0,8 1 Experimental NN (b) testing set Fig. 4. The performance of NN based model for diesel engine with intercooling. 0 0,2 0,4 0,6 0,8 1 Experimental NN (a) training set 0 0,2 0,4 0,6 0,8 1 Experimental NN (b) testing set 0 0,2 0,4 0,6 0,8 1 0 0,2 0,4 0,6 0,8 1 Fig. 5. The performance of NN based model for diesel engine without intercooling. Table 3 General properties of Model A. CA F rpm A1 18 400 Var. 20 22 A2 18 450 Var. 20 22 A3 18 500 Var. 20 22 A4 18 400 Var. 450 500 A5 20 400 Var. 450 500 A6 22 400 Var. 450 500 A. Uzun / Fuel 93 (2012) 189–199 191
  4. 4. 2. Details of the experimental setup The experimental set up is a six-cylinder, four-stroke, diesel en- gine with an intercooler; and turbocharger, which was used with a hydraulic dynamometer. The specifications of the diesel engine are given in Table 1. All of the experiments [10] were performed for various speeds and loads, both with and without an intercooler. A heat exchanger is used to cool the engine water. Tank capacity is 300 L. Intake and outlet air temperature of the intercooler are measured. Intercooler cooling water flow rate is measured as 8.5 L/min. During the test, values were measured and recorded for engine speed, load, intake manifold temperature, exhaust gas temperature, cooling water in- let and outlet temperatures, intercooler input and output temper- atures, oil temperature, test cabin temperature, oil pressure, media pressure, ambient temperature, intake manifold pressure, air in- take pressure difference (in oblique manometer), cooling water pressure difference (U-manometer) and fuel consumption. At the end of each experimental period, the injection advance of the en- gine was changed. It seems to repeat information that has already been given in the earlier text. Intercooler and without intercooler different loads, inject angle and speed into tables and graphs in the characteristic values are presented as a comparison. The values of some test variables were calculated using empirical measure- ments, namely specific fuel consumption, effective engine efficiency, excess air coefficient. The engine installation and testing scheme are given in Fig. 1 and the test engine is shown in Fig. 2. The accuracy in measurements is about 3% in the experiment. 3. Neural Networks (NN) Recently, NNs have been successfully employed as a computa- tional tool to solve complex problems in various engineering fields. NNs provide an alternative computational model inspired by a neu- rological model of the human brain. NNs can simulate operational features of the human brain. The NN architecture is composed of one input layer, one output layer and one or more hidden layers. In the NN architecture, the layers are composed of neurons, which are the fundamental processing element of the NN [15,16]. Neurons in each layer are totally connected to other neurons in following layers. These NN architectures are commonly referred to as a fully interconnected feed forward multilayer perception. There are two biases in the NN. One of them is connected to neurons in the hidden layer and the other one is connected to neurons in the output layer. The numbers of neurons in the input and output layers are designated according to the problem [15]. The back-propagation (BP) algorithm is a widely used training algorithm for multi-layered feed forward networks. The BP algo- rithm basically consists of two phases. The first is the forward phase, where the activations are propagated from the input to the output layer. In the first phase (the forward phase), the activations are propagated from the input to the output layer. In the next phase (the backward phase), the error between the observed actual value and the desired nominal value in the output layer is propagated backwards in order to modify the weights and bias values NN. All the data used in the NN must be initialized before the train- ing procedure. In the forward phase, the weighted sum of input components is calculated as netj ¼ Xn i¼1 wijxi þ biasj ð1Þ where netj is the weighted sum of the jth neuron for the input re- ceived from the preceding layer with n neurons, wij is the weight be- tween the jth neuron and the ith neuron in the preceding layer, xi is the output of the ith neuron in the preceding layer. The output of the jth neuron outj is calculated with a sigmoid function, as follows: outj ¼ fðnetjÞ ¼ 1 1 þ eÀðnetjÞ ð2Þ During the training process, the weights are modified to capture the relationship between the input and output patterns. The training of the network is carried out by adjusting the weights. The objective of the training procedure is to find the optimal set of weightings. The best-known training algorithm for the multi-layer feed-for- ward neural network is the back-propagation algorithm. Two back- propagation training algorithms, called gradient descent and gradi- ent descent with momentum, run very slowly. Therefore, several adaptive training algorithms for NN have recently been developed, such as Conjugate Gradient Algorithm (CG) and Scaled Conjugate Gradient Algorithm (SCG). The present study is used SCG as the optimization algorithm. Details of the SCG algorithm can be found in the literature [17]. The output of the NN model is compared with the experimental results to obtain error values. The sum of the squares error (SSE) is used as the performance function for feed forward networks. The training process continues until the optimal sum of the squares er- ror is determined. The SSE is defined as; SSE ¼ Xm i¼1 ðTi À outiÞ2 ð3Þ Table 4 General properties of Model B. CA F rpm B1 18 Var. 1600 20 22 B2 18 Var. 2000 20 22 B3 18 Var. 2400 20 22 B4 18 Var. 1600 2000 2400 B5 20 Var. 1600 2000 2400 B6 22 Var. 1600 2000 2400 Table 5 General properties of Model C. CA F rpm C1 Var. 400 1600 2000 2400 C2 Var. 450 1600 2000 2400 C3 Var. 500 1600 2000 2400 C4 Var. 400 1600 450 500 C5 Var. 400 2000 450 500 C6 Var. 400 2400 450 500 192 A. Uzun / Fuel 93 (2012) 189–199
  5. 5. where Ti and outi are the target outputs and the actual output of neural network values, respectively, for ith output neuron; and m is the number of neurons in the output layer. 4. Numerical study A parametric study is carried out to investigate the effect of CA, engine speed and loads on diesel engine performance in terms of brake specific fuel consumption (BSFC, g/kW h), both with and without intercooling. A NN based model was used to perform the parametric studies. Firstly, experimental studies were conducted. Then, the data obtained from the experimental results were used as initial inputs to the NN based model. The data are divided into two parts, training and testing sets, for both models. Sixty data-sets were selected as the training set and used to train the NN based model. A further 34 data sets, which are not used in the training process, are selected as the testing set and used to validate the generalization capability of the NN based model. Inputs and outputs are normalized in the (0–1) range by using simple normalization methods and values are given in Ta- ble 2. The maximum and minimum values of inputs and outputs are also given in Table 2. In the NN model, the numbers of neurons in the input and output layers are based on the geometry of the problem. How- ever, there is no general rule for determining the number of neu- rons in a hidden layer and the number of hidden layers [15]. 130 140 150 160 170 180 190 200 1000 1200 1400 1600 1800 2000 2200 2400 speed (rpm) BSFC(g/kWh) ca=18 ca=20 ca=22 F=400N )(a1 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) ca=18 ca=20 ca=22 F=400N )(b1 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) ca=18 ca=20 ca=22 F=450N )(a2 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) ca=18 ca=20 ca=22 F=450N )(b2 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) ca=18 ca=20 ca=22 F=500N )(a3 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) ca=18 ca=20 ca=22 F=500N )(b3 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 Fig. 6. Interrelationship between the BSFC and speed (According to F): (a) with intercooling (b) without intercooling. A. Uzun / Fuel 93 (2012) 189–199 193
  6. 6. Therefore, the number of hidden layers and the number of neu- rons in each layer are selected by trial and error. In order to determine the most appropriate NN model, many different NN models with neurons in hidden layers were trained and tested for 2500 epochs. The sum of the squares errors (SSE) is used to establish the most appropriate NN model. The required sum of the squares error (SSE) was selected as 10À5 . The most appropri- ate NN models corresponding to the performance of both training and testing sets in terms of SSE are chosen. Consequently, the NN model selected to define specific fuel consumption of an inter- cooled diesel engine had three neurons in the input layer, five neurons in the hidden layer and two neurons in the output layer (Fig. 3). A MATLAB program with a graphical user interface (GUI) was developed to train and test the NN model [10]. In the NN model, the type of back-propagation used was the scaled conjugate gradi- ent algorithm (SCGA). The activation function was a sigmoidal function, whose epochs (learning cycle) are 40,000. The performance of the NN based model of the intercooled die- sel engine shows that the correlations between experimental re- sults and NN based model results are consistent for training and testing sets, as shown in Fig. 4. The performance of the NN based model for the diesel engine without intercooling are shown in Fig. 5. Fig. 5 shows that the cor- relations between targets and outputs are reliable for training and testing sets. 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) F=400 F=450 F=500 CA=18 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) F=400 F=450 F=500 CA=18 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) F=400 F=450 F=500 CA=20 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) F=400 F=450 F=500 CA=20 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) F=400 F=450 F=500 CA=22 130 140 150 160 170 180 190 200 speed (rpm) BSFC(g/kWh) F=400 F=450 F=500 CA=22 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 )(a1 )(b1 )(a2 )(b2 )(a3 )(b3 Fig. 7. Interrelationship between the BSFC and speed (According to CA): (a) with intercooling (b) without intercooling. 194 A. Uzun / Fuel 93 (2012) 189–199
  7. 7. The results of Fig. 4 indicate that the NN based models are suc- cessful in learning the relationship between the input parameters and outputs. The results of Fig. 5 show that the NN based models are capable of generalizing between input and output variables. As can be seen from the Figs. 4 and 5, NN based model results agree well with the experimental results. 5. Parametric study By using a well-trained NN based model, various parametric studies are conducted to investigate the BSFC at different positions. The three main models are formed and the interrelationship be- tween the one variable and two steadies on the BSFC are studied for each model. The selected variables are engine speed, load and CA in Models A, B and C, respectively (Table 3–5). 5.1. Model A In Model A, a parametric study is carried out to reveal the behavior of BSFC with respect to engine speed for various load (F) and CA values in a diesel engine, both with and without interco- oling. The interrelationship between the one variable and two steadies for Model A are shown six positions in Table 3. The results are presented graphically in Fig. 6 and 7, indicating the relationship between BSFC and engine speed. It can be seen in Figs. 6 and 7 that, with variable engine speed (rpm), BSFC decreases in the case of CA = 20. Lower fuel 130 140 150 160 170 180 190 200 210 300 325 350 375 400 425 450 475 500 525 550 Load (N) BSFC(g/kWh) ca=18 ca=20 ca=22 speed=1600 rpm 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) ca=18 ca=20 ca=22 speed=1600 rpm 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) ca=18 ca=20 ca=22 speed=2000 rpm 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) ca=18 ca=20 ca=22 speed=2000 rpm 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) ca=18 ca=20 ca=22 speed=2400 rpm 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) ca=18 ca=20 ca=22 speed=2400 rpm 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 )(a1 )(b1 )(a2 )(b2 )(a3 )(b3 Fig. 8. Interrelationship between the BSFC and load (According to speed): (a) with intercooling (b) without intercooling. A. Uzun / Fuel 93 (2012) 189–199 195
  8. 8. consumption occurs with low load (400 N) and 18 CA; and with middle and high loads (450 and 500 N) and 20 CA at approximately 2000 rpm. The results of this section, shown in Figs. 6a and b, also suggest that the trend of increased fuel consumption with higher engine speed is less pronounced in used intercooled engine than without used intercooling. According to Fig. 6, the best operating ranges are in Fig. 6 a3 and b3. In case of F = 500 N, conversely the poor working conditions seen in Fig. 6a1 and b2 in case of F = 400 N. As can be seen Fig. 6, the best work range conditions clearly correspond to the case of CA = 20. In the case of variable CA, the ideal working range can be observed as 18 CA–400 N, 20 CA–450 N and 22 CA–500 N. Also, the intercooled diesel engine achieved lower fuel consumption than the non-intercooled engine, as might be expected (Fig. 7a and b). As shown in Fig. 7a1 in the case of CA = 18, BSFC decreases slightly until nearly 5–8%. Since the starter value of BSFC is very high, this case does not represent an ideal working range. An ideal working range can be seen in Fig. 8a2 in the case of CA = 22, in which the starter value of BSFC and the ratio of increase in BSFC are very low. 5.2. Model B In Model B, the parametric study is carried out to reveal the behavior of BSFC with respect to loads (F) for various engine speeds (rpm) and CA values, both with and without intercooling. The 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) speed=1600 speed=2000 speed=2400 CA=18 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) speed=1600 speed=2000 speed=2400 CA=18 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) speed=1600 speed=2000 speed=2400 CA=20 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) speed=1600 speed=2000 speed=2400 CA=20 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) speed=1600 speed=2000 speed=2400 CA=22 130 140 150 160 170 180 190 200 210 Load (N) BSFC(g/kWh) speed=1600 speed=2000 speed=2400 CA=22 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 300 325 350 375 400 425 450 475 500 525 550 )(a1 )(b1 )(a2 )(b2 )(a3 )(b3 Fig. 9. Interrelationship between the BSFC and load (According to CA): (a) with intercooling (b) without intercooling. 196 A. Uzun / Fuel 93 (2012) 189–199
  9. 9. relationship between the one variable and two steadies for Model B are shown six positions in Table 4. The results are presented graphically in Figs. 8 and 9, indicating the relationship between BSFC and engine loads. It can be seen in Fig. 9 that, with variable engine load, all of the BSFC values are close to the case of low load. BSFC generally decreased with increased engine load. However, there are sudden changes in all BSFC values at loads of around 400–450 N. According to Fig. 9, the best operating range occurs in Fig. 9a1 at an engine speed of 1600 rpm. As can be seen from Fig. 8, BSFC with interco- oling diesel engine is less than without intercooling. As shown in Fig. 9, there is no significant difference between BSFC at all engine loads, engine speeds; or with and without intercooling. Fig. 9a1–3 indicates that all BSFC values are approxi- mately equal. However, it can be seen that when BSFC decrease in starter values, at around F = 400–450 N, BSFC increases until around F = 450–500 N. As explained above, the best work range in this model is found in Fig. 10a2. In Fig. 9, at middle load range (400–450 N), the fuel consumption appears to be lowest at CA 20, which is clearly much lower than the result sin both CA 18 and CA 22 (Fig. 9a and b). 5.3. Model C In Model C, the parametric study examines the relationship between BSFC and CA. The trend of BSFC with respect to CA for 130 140 150 160 170 180 190 200 210 18 19 20 21 22 23 24 CA BSFC(g/kWh) speed=1600 speed=2000 speed=2400 F=400 N 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) speed=1600 speed=2000 speed=2400 F=400 N 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) speed=1600 speed=2000 speed=2400 F=450 N 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) speed=1600 speed=2000 speed=2400 F=450 N 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) speed=1600 speed=2000 speed=2400 F=500 N 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) speed=1600 speed=2000 speed=2400 F=500 N 18 19 20 21 22 23 24 18 19 20 21 22 23 24 18 19 20 21 22 23 24 18 19 20 21 22 23 24 18 19 20 21 22 23 24 )(a1 )(b1 )(a2 )(b2 )(a3 )(b3 Fig. 10. Interrelationship between the BSFC and CA (According to F): (a) with intercooling (b) without intercooling. A. Uzun / Fuel 93 (2012) 189–199 197
  10. 10. variable F and engine speed values is investigated for a diesel en- gine both with and without intercooling; the results are shown in Figs. 10 and 11. As clearly indicated in Fig. 10a and b, The fuel consumption made at CA 20 and 450–500 N for all three load lev- els, while the CA is kept variable. The interrelationship between the one variable and two steadies for Model C are shown in six positions in Table 5. The results are presented graphically in Figs. 10 and 11, to indicate the relationship between BSFC and the CA. As shown Fig. 10, BSFC decreases slightly in starter values until around 19 CA. BSFC tends to increase for all engine speeds after 19 CA. As shown in Fig. 10, there is no significant difference in BSFC at all CA values and engine speeds. As can be seen from Fig. 10, in all cases, the engine with intercooling consumes less fuel than the en- gine without intercooling. The results shown in Fig. 10 suggest the ideal BSFC level is around 18–20 CA and 450 N for the lower engine speeds. Additionally, it is observed that increasing loads and en- gine speeds lead to increasing BSFC. The results shown in Fig. 10a and b suggest the same indications made before for the ef- fect of intercooling on engine performance. Similar trends for all BSFC values are observed in Fig. 11 for die- sel engines with intercooling and without intercooling. As shown in Fig. 11, there is no considerable statistically significant differ- ence between BSFC trends in all CA and engine loads for cases with intercooling and without intercooling. As explained above, the best work range is found in Fig. 11a1. In Fig. 11, at middle load range (450–500 N), fuel consumption is lowest at F = 450 N, which is clearly much lower than the results for both F = 400 N and F = 500 N (Fig. 11a2–1a3). 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) F=400 N F=450 N F=500 N speed=1600 rpm 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) F=400 N F=450 N F=500 N speed=1600 rpm 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) F=400 N F=450 N F=500 N speed=2000 rpm 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) F=400 N F=450 N F=500 N speed=2000 rpm 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) F=400 N F=450 N F=500 N speed=2400 rpm 130 140 150 160 170 180 190 200 210 CA BSFC(g/kWh) F=400 N F=450 N F=500 N speed=2400 rpm 18 19 20 21 22 23 24 18 19 20 21 22 23 24 18 19 20 21 22 23 24 18 19 20 21 22 23 24 18 19 20 21 22 23 24 18 19 20 21 22 23 24 )(a1 )(b1 )(a2 )(b2 )(a3 )(b3 Fig. 11. Interrelationship between the BSFC and CA (According to speed): (a) with intercooling (b) without intercooling. 198 A. Uzun / Fuel 93 (2012) 189–199
  11. 11. 6. Conclusion The present study used parametric analysis to examine the ef- fect of CA, engine speed and engine loads on the brake specific fuel consumption (BSFC) of a diesel engine, both with and without intercooling. The parametric analysis used a novel NN approach. Firstly, the NN based model was trained and tested with data obtained experimentally. Following training, the outputs produced by the NN based model were compared with the experimental re- sults and found to be similar. Then, the well-trained NN based model was used to perform the parametric studies. In the paramet- ric studies, three models were used to show the relationship be- tween the one variable and two steadies on BSFC. The selected variables were engine speed, engine load and CA in Models A, B and C, respectively. In Model A, the parametric study examined BSFC with respect to engine speed for various load (F) and CA values, in cases with and without intercooling. The most appropriate operating ranges correspond to F = 500 N and the worst operating conditions corre- spond to F = 400 N. Model B examined BSFC with respect to loads (F) for various en- gine speed (rpm) and CA values in cases with and without interco- oling. There was no statistically significant difference between BSFC at all engine loads, engine speeds and in cases with intercool- ing or without intercooling. The results show that, with variable engine load, all BSFC values are close to the case of low load. BSFC generally decreased with increased engine load. However, there are sudden changes in all BSFC values at loads of around 400–450 N. Model C examined BSFC with respect to CA for various engine speeds (rpm) and load (F) values, for cases with and without int- ercooling. BSFC decreased slightly in starter values until around 19 CA. BSFC tends to increase for all engine speeds after 19 CA. The results suggest the ideal BSFC level at around 18–20 CA and 450 N for the lower engine speeds. It was observed that increasing loads and engine speeds lead to increasing BSFC. There is no statis- tically significant difference between BSFC trends in all CA and en- gine loads for cases with used intercooling and without used intercooling. It can be concluded that 20 CA, 400–450 N and medium engine speed (2000 rpm) produce an optimal operating range for higher or lower specific fuel consumption in the tested diesel engine with and without used intercooling. All of the results suggest that the diesel engine without intercooling consumes more fuel compared to the same engine with intercooling. The results are reasonable and found to be consistent with the literature. The study findings may make an important contribution for researchers in similar fields, since the use of NN models is quicker, more convenient and cost-effective than fully experimental stud- ies. The reasons of BSFC increasing at intercooling diesel engine can be explained as follows: The result indicating that BSFC in- creased in a diesel engine with intercooling can be explained as fol- lows: CA, engine speed and load must be optimized, and the best combustion process like that, air intake mass flow, should be cooled by the intercooler. References [1] Celik V, Arcaklioglu E. Performance maps of the diesel engine. Appl Energy 2005;81:247–59. [2] Stone R. Motor vehicle fuel economy. USA: Brunel University; 1992. [3] Brady RN. Modern diesel technology. New Jersey, Columbus (OH): Prentice Hall; 1996. [4] Jayashankara B, Ganesan V. Effect of fuel injection timing and intake pressure on the performance of a DI diesel engine – a parametric study using CFD. Energy Convers Manage 2010;51:1835–48. [5] Icingur Y, Altiparmak D. Effect of fuel cetane number and injection pressure on a diesel-engine’s performance and emissions. Energy Convers Manage 2003;44:389–97. [6] Mostafavi M, Agnew B. Thermodynamic analysis of combined diesel engine and absorption refrigeration unit-turbocharged engine with intercooling. Appl Thermal Eng 1996;16:733–40. [7] Mostafavi M, Agnew B. Thermodynamic analysis of combined diesel engine and absorption refrigeration unit-supercharged engine with intercooling. Appl Thermal Eng 1996;16:921–30. [8] Mostafavi M, Agnew B. Thermodynamic analysis of combined diesel engine and absorption refrigeration unit-naturally aspirated engine with precooling. Appl Thermal Eng 1997;17:593–9. [9] Parlak A, Islamoglu Y, Yasar H, Egrisogut A. Application of artificial neural network to predict specific fuel consumption and exhaust temperature for a diesel engine. Appl Thermal Eng 2006;26:823–4. [10] Uzun A. Effects of intercooling on performance of a turbocharged diesel engine, PhD thesis, Sakarya university. Science; 1998. in Turkish. [11] Arcaklıog˘lu E, Çelikten I. A diesel engine’s performance and exhaust emissions. Appl Energy 2005;80:11–2. [12] Arcaklıog˘lu E, Çavusßog˘lu A, Erisßen A. Thermodynamic analyses of refrigerant mixtures using artifical neural-networks. Appl Energy 2004;77:273–86. [13] Taylor CF. The internal combustion engine in theory and practice. MIT press; 1985. [14] Otosan Ford Cargo Maintenance book; 1992. [15] Naci Caglar. Neural network based approach for determining the shear strength of circular reinforced concrete columns. Constr Build Mater 2009;23:3225–32. [16] Murat Pala. A new formulation for distortional buckling stress in cold-formed steel members. J Constr Steel Res 2006;62:716–22. [17] Moller AF. A scaled conjugate gradient algorithm for fast supervised learning. Neural Networks 1993;6:525–33. A. Uzun / Fuel 93 (2012) 189–199 199

×