Artificial Neural Network Based Modeling of Injection Pressure in Diesel Engines

Transcription

1 Artificial Neural Network Based Modeling of Injection Pressure in Diesel Engines MALI AKCAYOL, CAN CINAR, HIBRAHIM BULBUL, ALI KILICARSALAN 4 Deartment of Comuter Engineering, Gazi University, Maltee, 06570, Ankara, TURKEY Deartment of Mechanical Education, Gazi University, Besevler, 06500, Ankara, TURKEY Deartment of Comuter Education, Gazi University, Besevler, 06500, Ankara, TURKEY 4 Deartment of Mechanical Engineering, Gazi University, 900, Corum, TURKEY Abstract: - Injection ressure in diesel engines has an imortant effect on the engine erformance and soot formation During erforming the work, the measurement of the torque, ower, secific fuel consumtion (SFC) and soot formation values in the diesel engines is a time consuming work and it also requires secific tools, an exert In addition to the difficulties mentioned earlier, some of the oerating oints can be only investigated and evaluated because of difficulties of measuring the arameters at the oerating conditions In this study, to overcome these difficulties, an artificial neural network (ANN) is used for rediction of erformance and soot formation in diesel engines The training data for ANN is obtained from measurements In comarison of erformance analysis of ANN, the deviation coefficients of torque, ower, SFC, and soot formation for the test ressure conditions are less than 66,, 89, and 47, resectively The statistical coefficient of multile determinations for the investigated cases is about 0994 to 0998 The degree of accuracy is accetable in redicting the arameters of the system So, it can be concluded that ANN rovides a feasible method in redicting the system arameters Key-Words: - Artificial neural network, injection ressure, diesel engine Introduction Diesel engines have been enetrating a number of markets all around the world because of their good fuel economy and high reliability The diesel engines are widely used in heavy-duty engine alications such as bus, truck, ower generation They are referred over sark ignition engines because they can achieve greater efficiencies and higher indicated mean effective ressures due to the higher comression ratios where they oerate [,] Diesel engines roduce lower amounts of HC (Hydrocarbon), and CO (Carbonmonoxide) emissions than the sark ignition engines because of more comlete combustion of the air-fuel mixture Soot or articulate emissions occur when there is insufficient air to comletely burn the fuel [] And it is well established that these emissions from diesel engines may have a harmful effect on human health [4] There are several factors that the engine designer considers to rovide both low emission levels and high erformance with good fuel economy Some of these factors are the shae of the combustion chamber, the injection rate and nozzle sray attern, injection timing, and injection ressure [] In recent years, a number of studies have been conducted on injection ressure to increase engine erformance and to decrease exhaust emissions in diesel engines [5] In these studies, some of the oerating oints of the system have been investigated For this tye of works,

2 exerts and secial equiments are needed It also requires too much time and high cost [6] In the last decade, ANNs have been widely used for many different industrial areas such as control, rediction, attern recognition, classification, seech and vision ANNs have been trained to solve nonlinear and comlex roblems that are not exactly modeled mathematically [7] ANNs eliminate the limitations of the classical aroaches by extracting the desired information using the inut data Alying ANN to a system needs sufficient inut and outut data instead of a mathematical equation Furthermore, it can continuously re-train for new data during in oeration, thus it can adat to changing of the system Also, ANNs can be used to deal with the roblems with incomlete and imrecise inut data [8,9] In this study, an ANN has been used for redicting the erformance and soot formation in diesel engines The ANN redicted and results are extensively comared under different oerating conditions Exerimental aaratus and rocedure The exeriments in the resent study were conducted by oerating a direct injection diesel engine The general secifications of the engine are shown in Table A Leclasrege Electriou brand electrical dynamometer was used for the tests Soot formation was measured by means of VLT 600 S brand diesel emission device having 00% accuracy The schematic view of the test equiments is shown in Fig Table General secifications of the test engine Item Secification Engine tye Direct injection, Diesel Stroke (mm) 00 Bore (mm) 98 Dislacement (cc) 754 Cycle Four stroke Max Power 7 kw at 800 rm Comression ratio 7: Fig Schematic of the test facility The exeriments were erformed at full load oerating conditions The engine was loaded by the electrical dynamometer During the exeriments, engine seed was changed from 900 rm to 900 rm with 00-rm intervals Injection ressure was changed from 5 bar to 50 bar with 5 bar intervals Injection ressure is changed by means of adjusting the injector sring tension During the exeriments, the average ambient temerature and atmosheric ressure were o C and 75 mm-hg, resectively The tests were conducted after the engine reached the working temerature of 80 o C Fig shows the variation of engine torque with resect to the engine seed at different injection ressures As the injection ressure increases, the engine torque also increases Deending on the increase in injection ressure, drolet size becomes smaller and air-fuel mixture formation becomes better An increase in the engine torque can be seen as the injection ressure at a certain level (5 bar) is taken into consideration After this oint the engine torque decreases with the increasing value of the injection ressure Furthermore, deending uon the air-fuel mixture formation, the engine torque decreases drastically at low injection ressures

3 Fig Variation of engine torque as a function of engine seed Fig shows the variation of ower outut with resect to the engine seed at different injection ressures The increase of injection ressure causes the engine ower to increase at a certain level This trend is similar to of the engine torque Power outut decreases at low injection ressures Fig4 Variation of SFC as a function of engine seed Fig5 shows the variation of soot formation with resect to the engine seed at different injection ressures Deending on the increase in injection ressure, drolet size becomes smaller and air-fuel mixture formation becomes better A considerable reduction in soot formation is obtained when the injection ressure is increased Soot formation increases drastically at low injection ressures Fig Variation of ower outut as a function engine seed Fig4 shows the variation of SFC with resect to the engine seed at different injection ressures As the injection ressure increases, the SFC decreases SFC increases drastically at low injection ressures Fig5 Change of soot formation as a function of engine seed Alication of ANN There are many tyes of ANN architectures in the literature; however, multi-layer feed-forward neural-network is the most widely used for rediction A multi-layer feed-forward neuralnetwork tyically has an inut layer, an outut

4 layer, and one or more hidden layers [0] In multi-layer feed-forward networks, neurons are arranged in layers and there is a connection among the neurons of other layers The inut signals are alied to the inut layer, the outut layer contributes to the outut signal directly Other layers between inut and outut layers are called hidden layers Inut signals are roagated in gradually modified form in the forward direction, finally reaching the outut layer One neuron can receive signals from other neuron and transfer outut signal into other nodes using transfer function as an inut A sigmoid function is widely used for transfer function [] whose outut lies between zero and unity and is defined as f ( x) () x + e The function is differentiable throughout its domain During learning, the weights of the neurons are adjusted according to the generalized delta rule which is the learning algorithm for a back-roagation multi-layer feed-forward network The error is the sum of the squares of the overall errors of the network and is minimized by the generalized delta rule, defined as E ( y o ) () where E is the square errors, is the index of attern in the training set, o is the desired outut and y is the calculated outut of network The weight modification for a neuron is done in roortion to the gradient of E with resect to the neuron weights [] In this way, each udated weight in a layer deends on all the error terms of the outut layer Thus, the error of the outut layer is roagated back to each layer Faster learning can be done by changing the learning-rate constant, but imroer learning rate constant may cause the weights to bounce around the local minima, thus failing to learn roerly A four layers ANN is alied to the system to redict of torque, ower, SFC, and soot formation under different injection ressures The ANN structure used in this alication is shown in Fig6 seed injection ressure torque ower SFC soot formation Fig6 ANN architecture used for estimation of torque, ower, SFC, and soot formation The ANN has four layers namely, an inut, an outut, and two hidden layers The inut layer consists of two neurons, the outut layer consists of four neurons, and each of hidden layers consists of 5 neurons The inut variables in the network are the seed (n) and the ressure The outut variables are the torque, the ower, the SFC, and the soot formation The back-roagation algorithm has been imlemented to calculate errors and adjust weights of the hidden layer neurons In order to avoid long training time or network being traed in local error minima, various learning rate constants are tried The ANN structure and number of neurons in each of hidden layers have been selected by using an evolutionary algorithm All of the data have been normalized in the range of [0, +] Sigmoid function is chosen for transfer function, with 05-threshold value as defined, f ( x) () 4( x 05) + e Figs-5 show a arity lot between and comuted data by ANN for torque, ower, SFC, and soot formation The redictions have R -values equal to for torque, 0998 for ower, 0994 for SFC, and for soot formation It can be clearly seen from Figs7-0, the develoed ANN gives a very 4

5 accurate reresentation of R -values over the all range or working conditions Torque Soot Formation R R Fig7 Comarison of measured and redicted values for the engine torque Power R Fig8 Comarison of measured and redicted values for the ower Fig0 Comarison of measured and redicted values for the soot formation Since results are very close to the calculated values that can be obtained by using ANN, those cannot be grahically shown together For this reason, the following equations (Eqs4-7) are used to calculate the deviation values, and these values have been shown grahically dtorque dower dsfc torque ANN torque (4) torque owerann ower (5) ower SFCANN SFC (6) SFC SFC R 0994 soot_formation ANN soot_formation dsoot_formation soot_formation (7) The standard deviations for torque, ower, SFC, and soot formation are illustrated in Figs Fig9 Comarison of measured and redicted values for the SFC 5

6 Fig Variation of the dtorque as a function of seed at different injection ressures Fig4 Variation of the dsoot Formation as a function of seed at different injection ressures According to the results, maximum deviations in torque (dtorque) is 66%, in ower (dpower) is %, in SFC (dsfc) is 89%, and in soot formation (dsoot formation) is 9% Table shows the minimum and maximum deviations for each of the outut These results rove that the roosed ANN can be used successfully for the rediction of erformance and soot formation in diesel engines Fig Variation of the dpower as a function of engine seed at different injection ressures Table Maximum and minimum deviations of torque, ower, SFC, and soot formation Outut Min/ Max n (rm) ) Pressure (bar) Deviations (%) Torque Min Exerimental Value Torque Max Power Min Power Max SFC Min SFC Max Soot formation Min Soot formation Max Fig Variation of the dsfc as a function of seed at different injection ressures 4 Conclusions In this study, an artificial neural network is used for rediction of erformance and soot formation in diesel engines Engine erformance and soot formation are measured for stroke single cylinder, 754cc direct injection diesel engine 6

7 Measurements are conducted for each of the injection ressures 5, 50, 75, 00, 5, and 50 bar The deviations for torque, ower, SFC, and soot formation for different injection ressures are obtained by using ANN The maximum deviations for all ressures are 66% for torque, % for ower, 89% for SFC, and 47% for soot formation The statistical coefficients are above 099 This degree of accuracy shows that the roosed ANN can be used for obtained the engine erformance and soot formation To sum u, this study is considered to be helful in redicting the erformance of the diesel engine References: [] Lida, N and Sato, G T, Temerature and Mixing Effects on NOx and Particulate, SAE Paer, 88044, 988 [] Wyman, J A, An Investigation of The Mixing of Fuel and Air in The Pre chamber of An IDI Comression Ignition Engine Using a Water Analog Model, Msc Thesis, Deartment of Mechanical Engineering, University of Toronto, Canada, 994 [] Celikten, I, An Exerimental Investigation of the Effect of the Injection Pressure on Engine Performance and Exhaust Emission in Indirect Injection Diesel Engines, Alied Thermal Engineering, Vol, No6, 00, [4] Lida, N, Suzuki, Y, Sato, G T, Sawada, T, Effects of Intake Oxygen Concentration on the Characteristics of Particulate Emissions from a DI Diesel Engines, SAE Paer, 86, 986 [5] Rakooulos, C D, and Hountalas, D T, A simulation analysis of a DI diesel engine fuel injection system fitted with a constant ressure valve, Energy Conversion and Management, Vol7, No, 996, 5-50 [6] Rakooulos, C D, Rakooulos, D C, Giakoumis, E G, and Kyritsis, D C, Validation and sensitivity analysis of a two zone Diesel engine model for combustion and emissions rediction, Energy Conversion and Management, Vol45, No9-0, 004, [7] Mohandes, M, Rehman, S, Halawani, TO, Estimation of Global Solar-Radiation Using Artificial Neural Networks, Renew Energy, 4, 4, 79 84, 998 [8] Jang, JSR, Sun, CT, Mizutani, E, Neuro-Fuzzy and Soft Comuting: A Comutational Aroach to Learning and Machine Intelligence, Prentice-Hall International, 997 [9] Jang, JSR, Sun, CT, Neuro-Fuzzy Modeling and Control, The Proceedings of the IEEE, Vol8, 995, [0] Kalogirou, SA, Alications of Artificial Neural-Networks for Energy Systems, Al Energy, Vol67, 000, 7 5 [] Palau, A, Velo, E, Puigjaner, L, Use of Neural-Networks and Exert Systems to Control a Gas/Solid Sortion Chilling Machine, Int J Refrig, Vol, 999, [] Reddy, KS, Ranjan, M, Solar Resource Estimation Using Artificial Neural- Networks and Comarison with Other Correlation Models, Energy Convers Manage, Vol44, 00,