- Email: [email protected]
Modeling and simulation of the thermodynamic cycle of the Diesel Engine using Neural Networks
14-th IFAC Symposium on Control in Transportation Systems 14-th IFAC Symposium on Turkey Control in Transportation Systems May 18-20, 2016. Istanbul, 14-th IFAC Symposium Symposium on Turkey Control in in Transportation Transportation Systems Systems 14-th IFAC on May 18-20, 2016. Istanbul, 14-th IFAC Symposium on Control Control in Available Transportation onlineSystems at www.sciencedirect.com May 18-20, 18-20, 2016. 2016. Istanbul, Istanbul, Turkey Turkey May May 18-20, 2016. Istanbul, Turkey
ScienceDirect IFAC-PapersOnLine 49-3 (2016) 221–226 Modeling Modeling and and simulation simulation of of the the thermodynamic thermodynamic cycle cycle of of the the Diesel Diesel Engine Engine using using Modeling thermodynamic cycle of the Diesel Engine using Networks Modeling and and simulation simulation of of the theNeural thermodynamic cycle of the Diesel Engine using Neural Networks Neural Networks Neural Networks Ali Rida*, Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Ouladsine*****
Ali Rida*, Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Ouladsine***** Ali Rida*, Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Ouladsine***** Ali Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Ali Rida*, Rida*, Hassan Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Mustapha Ouladsine***** Ouladsine***** *Lebanease University, FOE, Beyrouth Hadath ([email protected]) *Lebanease University, FOE, Beyrouth Hadath ([email protected]) **Aix Marseille University, LSIS, Marseille ([email protected]) *Lebanease University, FOE, Beyrouth Hadath ([email protected]) *Lebanease University, FOE, Hadath *Lebanease University, FOE, Beyrouth Beyrouth Hadath ([email protected]) ([email protected]) **Aix Marseille University, LSIS, Marseille ([email protected]) *** Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) **Aix Marseille University, LSIS, Marseille ([email protected]) **Aix Marseille University, Marseille ([email protected]) *** Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) **Aix Marseille University, Marseille ([email protected]) ****Lebanease University, FOE, Beyrouth Hadath ([email protected]) *** Lebanease Lebanease University, University, FOE, FOE, Beyrouth Beyrouth Hadath Hadath ([email protected]) ([email protected]) *** ****Lebanease University, FOE, Beyrouth Hadath ([email protected]) *** Lebanease University, FOE, Beyrouth Hadath ([email protected]) ***** Aix Marseille University, Marseille ([email protected]) ****Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) ****Lebanease University, FOE, Beyrouth Hadath ([email protected]) ***** Aix Marseille University, Marseille ([email protected]) ****Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) ***** Aix Marseille University, LSIS, Marseille ([email protected]) Aix Marseille ([email protected]) Abstract: In ***** this paper, a unique University, single zoneLSIS, combustion model is proposed to predict Diesel engine’s Aix Marseille Marseille Marseille ([email protected]) Abstract: In ***** this paper, a unique University, single zoneLSIS, combustion model is proposed to predict Diesel engine’s performance, pressure, and temperature based on the conservation of massto and energy. In engine’s order to Abstract: In this paper, a unique single zone combustion model is proposed predict Diesel Abstract: paper, aa unique single model predict Diesel Abstract: In In this this paper, and unique single zone zone combustion model is is proposed proposed toand predict Diesel engine’s performance, pressure, temperature basedcombustion on the conservation of massto energy. In engine’s order to simulate all phases of combustion, the proposed model takes in consideration the dynamics of the intake performance, pressure, and temperature based on the conservation of mass and energy. In order to performance, pressure, and based on conservation of and energy. order to performance, pressure, and temperature temperature based model on the the takes conservation of mass massthe and energy. ofIn Inthe order to simulate all phases of combustion, the proposed in consideration dynamics intake and exhaust gas through the valves, the ignition delay, the instantaneous change in gas properties, the simulate all phases of combustion, the proposed model takes in consideration the dynamics of the intake simulate all of the proposed in the of simulate all phases phases of combustion, combustion, thethe proposed model takes in consideration consideration the dynamics dynamics of the the intake intake and exhaust gas through the valves, ignitionmodel delay,takes the instantaneous change in gas properties, the properties of gas the burned and the the heatignition losses by the walls. Validation ofchange this model hasproperties, been realized and exhaust exhaust gas throughfuel, the valves, valves, the ignition delay, the instantaneous instantaneous change in gas gas properties, the and through the delay, the in the properties of gas the burned fuel, and the the heatignition losses by the walls. Validation ofchange this model hasproperties, been realized and exhaust through the valves, delay, the instantaneous in gas by experimental data. Important issue has been recognized that the physicalof model takes too been muchrealized timethe in properties of the burned fuel, and the heat losses by the walls. Validation of this model has been realized properties of the burned fuel, and the heat losses by the walls. Validation this model has properties of the data. burned fuel, andissue the heat losses by the walls. this model by experimental Important has been recognized that Validation the physicalofmodel takes has too been muchrealized time in calculation. For this purpose, a Feed-Forward Neural Network (FFNN) model is developed and validated by experimental data. Important issue has been recognized that the physical model takes too much time by experimental data. Important issue has been recognized that the physical model takes too much time in by experimental data.purpose, Important issue has beenNeural recognized that (FFNN) the physical model takes too and much time in in calculation. For this a Feed-Forward Network model is developed validated experimentally to predict the aapressure and temperature in the cylinder in nominal and faulty operations. calculation. For this purpose, Feed-Forward Neural Network (FFNN) model is developed and validated calculation. For this purpose, Feed-Forward Neural Network (FFNN) model is developed and validated calculation. Forto this purpose, Feed-Forward Neural Network (FFNN) in model is developed and validated experimentally predict the apressure and temperature in the cylinder nominal and faulty operations. Finally, the influence of some possible and faults that may be produced on the diesel engine cycleoperations. during the experimentally to predict the pressure and temperature in the cylinder in nominal and faulty operations. experimentally to the temperature in the in and experimentally to predict predict the pressure pressure temperature in produced the cylinder cylinder in nominal nominal and faulty faulty Finally, the influence of some possible and faults that may be on the diesel engine cycleoperations. during the operation has been analyzed. Finally, the influence of some possible faults that may be produced on the diesel engine cycle during the Finally, of Finally, the influence of some some possible possible faults faults that that may may be be produced produced on on the the diesel diesel engine engine cycle cycle during during the the operationthe hasinfluence been analyzed. operation has been analyzed. operation has been analyzed. Keywords: Diesel Engine, Thermodynamic cycle, Artificial Neural Network Modeling, Faultyreserved. operation © 2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights operation has been analyzed. Keywords: Diesel Engine, Thermodynamic cycle, Artificial Neural Network Modeling, Faulty operation mode. Keywords: Diesel Diesel Engine, Thermodynamic cycle, Artificial Neural Network Modeling, Faulty operation Keywords: Keywords: Diesel Engine, Engine, Thermodynamic Thermodynamic cycle, cycle, Artificial Artificial Neural Neural Network Network Modeling, Modeling, Faulty Faulty operation operation mode. mode. mode. mode. Thus, the FFNN are trained using the experimental data. 1. INTRODUCTION Thus, the FFNN are trained using the experimental data. Network and learning rateexperimental parameters data. are 1. INTRODUCTION Thus, the the architecture FFNN are are trained trained using the the experimental data. Thus, FFNN using Thus, the FFNN are trained using the Network architecture and learning rateexperimental parameters data. are 1. INTRODUCTION INTRODUCTION 1. optimized using Back-propagation algorithm and LevenbergDiesel engine is a very complicated mechanical system. It is Network 1. INTRODUCTION Network architecture and learning rate parameters are architecture and learning rate parameters are Network architecture and learning rate parameters are using Back-propagation algorithm and LevenbergDiesel engine is a very complicated mechanical system. It is optimized Marquardt algorithm to optimize algorithm the performance function characterized by thevery ruggedness in construction, simplicity in optimized using Back-propagation algorithm and LevenbergDiesel engine is a very complicated mechanical system. It is optimized using Back-propagation and LevenbergDiesel engine is a complicated mechanical system. It is optimized using Back-propagation algorithm and Levenbergalgorithm to optimize the performance function Diesel engine by is athevery complicated mechanical simplicity system. It in is Marquardt characterized ruggedness in construction, (Hagan et al., 2014). The main contribution of this paper can operation andbyease of maintenance. It has become quite Marquardt algorithm to optimize optimize the performance performance function characterized the ruggedness ruggedness in construction, construction, simplicity in Marquardt algorithm to the function characterized the in simplicity in Marquardt algorithm to optimize the performance function (Hagan et al., 2014). The main contribution of this paper can characterized byease the ruggedness in construction, simplicity in operation andby of maintenance. It has become quite be resumed by the development of neural network model that popular in transportation and agriculture sector, because of (Hagan et et al., al., 2014). 2014). The The main main contribution contribution of of this this paper paper can can operation and and ease ease of of maintenance. maintenance. It It has has become become quite quite (Hagan operation (Hagan et al., The main contribution of this paper can resumed by2014). the development of neural network model that operation ease of maintenance. It has become quite popular in and transportation and agriculture sector, because of be estimates the variation of temperature and pressure in that the higher efficiency and longer operational time. In because this paper, be resumed by the development of neural network model that popular in transportation and agriculture sector, because of be resumed by the development of neural network model popular in transportation and agriculture sector, of resumedthe by variation the development of neuraland network model estimates of temperature pressure in that the popular in transportation andoperational agriculturetime. sector, of be higher efficiency and longer In because this paper, cylinder in normal and faulty operation mode. In our previous we present a physical model of thermodynamic cycle of estimates the variation of temperature and pressure in the higher efficiency and longer operational time. In this paper, variation of and in higher efficiency and operational time. paper, estimates the variation of temperature temperature and pressure pressure in the the cylinder inthe normal and faulty operation mode. In our previous higher efficiency and longer longer operational time. In In this this paper, we present a physical model of thermodynamic cycle of estimates works (Nahim et al., 2015a), we have developed a diesel diesel engine. During the engine cycle, gas temperature, cylinder in normal and faulty operation mode. In our previous we present a physical model of thermodynamic cycle of in operation In we present aa physical model of cycle cylinder in normal normal and faulty operation mode. In our our previous previous works (Nahim et and al., faulty 2015a), we havemode. developed a diesel we present physical model of thermodynamic thermodynamic cycle of of cylinder diesel engine. During the engine cycle, gas temperature, engine simulator that 2015a), simulates operation in pressure and mass can bethe found by applying the conservation works (Nahim (Nahim et al., al., 2015a), we the haveengine developed diesel diesel engine. engine. During the engine cycle, gas gas temperature, works et we have developed aaa diesel diesel During engine cycle, temperature, works et al., we have developed diesel engine (Nahim simulator that 2015a), simulates the engine operation in diesel During engine cycle, gas temperature, pressureengine. and mass can bethe found by applying the conservation presence of failure (Nahim et al., 2015b). The developed of energy and mass (Xin, 2013) and by the application of the engine simulator that simulates the engine operation in pressure and mass can be found by applying the conservation engine simulator that simulates the engine operation in pressure be found the simulates engineThe operation in presencesimulator of failurethat (Nahim et al.,the 2015b). developed pressure and mass can be 2013) found by by applying the conservation conservation of energyand andmass masscan (Xin, andapplying by the application of the engine model canof befailure used to(Nahim simulate the influence of thedeveloped fault on ideal gas and equations in 2013) the cylinder. Initial condition of presence presence of failure (Nahim et al., 2015b). The developed of energy and mass (Xin, 2013) and by the application of the et al., 2015b). The of energy mass (Xin, and by the application of the et the al., influence 2015b). The model canofbefailure used to(Nahim simulate of thedeveloped fault on of energy mass (Xin, and by the application of the ideal gas and equations in 2013) the cylinder. Initial condition of presence the global it can bethe used in real time simulation pressure, andthe can be Initial estimated from the model can system, be used usedand to simulate simulate the influence of the the fault on on ideal gas gas temperature equations in in themass cylinder. Initial condition of model can be to influence of fault ideal equations cylinder. condition of model can be used to simulate influence of the fault on the global system, and it can bethe used in real time simulation ideal gas equations in cylinder. condition of pressure, temperature andthe mass can be Initial estimated from the to detect any failureand in the engine. intake manifold parameters, and theestimated computation is the the global system, and it can be used in real time simulation pressure, temperature and mass can be estimated from the global system, it can be used in real time simulation pressure, temperature and mass can be from the globalany system, it can be used in real time simulation to detect failureand in the engine. pressure, temperature and mass and can betheestimated from the intake manifold parameters, computation is the conducted on a crank angle basis.and A zero-dimensional model to detect any any failure in in the engine. engine. intake manifold manifold parameters, and the computation computation is to intake parameters, the is to detect detect any failure failure in the the engine. intake manifold parameters, the computation is conducted on a crank angle basis.and A zero-dimensional model combined a single zone model is used in this work, in conducted with on aaa crank crank angle basis. A zero-dimensional zero-dimensional model conducted on angle basis. A model conducted on crank angle basis. A zero-dimensional model combined with a single zone model is used in this work, in 2. COMBUSTION MODEL order to have a fast and accurate analysis of the engine combined with a single zone model is used in this work, in 2. COMBUSTION MODEL combined with zone model used in combined with aaa single single zoneaccurate model is isanalysis used in inofthis this work, in order to have fast and thework, engine 2. COMBUSTION COMBUSTION MODEL 2. performance et al., 2012). The model then order to to have have(Basbous fast and and accurate analysis of the the isengine engine COMBUSTION MODEL MODEL order aaa fast accurate analysis of order to have fast and accurate analysis of the performance (Basbous et al., 2012). The model isengine then The purpose of 2. a zero-dimensional, single zone combustion implemented in Matlab/Simulink in a complete diesel engine performance (Basbous et al., 2012). The model is then The purpose of a zero-dimensional, single zone combustion performance (Basbous et The then performance (Basbous et al., al., 2012). 2012). The model model isengine then The implemented in Matlab/Simulink in a complete dieselis model developed in this work is tozone simulate the The purpose of a zero-dimensional, single zone combustion purpose of a zero-dimensional, single combustion simulator proposed by (Nahim etin al., 2015a), diesel and then the The implemented in Matlab/Simulink in a complete diesel engine purpose of a zero-dimensional, single combustion model developed in this work is tozone simulate the implemented in Matlab/Simulink a complete engine implemented in Matlab/Simulink a complete engine simulator proposed by (Nahim etinal., 2015a), diesel and then the model thermodynamic cyclein from the work beginning ofto thesimulate intake phase model developed in this work is to simulate the developed this is the global model is validated experimentally. But due to the fact simulator proposed by (Nahim et al., 2015a), and then the model developed in this work is to simulate the thermodynamic cycle from the beginning of the intake phase simulator proposed by (Nahim et al., 2015a), and then the simulator proposed by (Nahim et al., 2015a), andtothen the thermodynamic global model is validated experimentally. But due the fact to the end of thecycle exhaust phase. The model is derived based thermodynamic cycle from the beginning of the intake phase from the beginning of the intake phase that the simulation time of one engine cycle is too long, we global model model is is validated validated experimentally. experimentally. But But due due to to the the fact fact thermodynamic fromphase. the beginning of is thederived intake based phase to the end of thecycle exhaust The model global global is validated But the fact that themodel simulation time ofexperimentally. one engine cycle isdue tootolong, we to on theend conservation of phase. mass and energyis al., to the the end of the the exhaust exhaust phase. The model model is(Awad derived etbased based of The derived transformed the physical model of thermodynamic cycle into that the simulation time of one engine cycle is too long, we to the end of the exhaust phase. The model is derived based on the conservation of mass and energy (Awad et al., that the simulation time of one engine cycle is too long, we that the simulation time of one engine cycle is toocycle long,into we on transformed the physical model of thermodynamic 2013)(Verhelst and Sheppard, 2009). on the conservation of mass and energy (Awad et al., the conservation of mass and energy (Awad et al., FFNN modelthe andphysical we validated itof experimentally. transformed the physical model of thermodynamic cycle into on the conservation of mass and energy (Awad et al., 2013)(Verhelst and Sheppard, 2009). transformed model thermodynamic cycle into transformed modelitof thermodynamic cycle into 2013)(Verhelst FFNN modelthe andphysical we validated experimentally. 2013)(Verhelst and Sheppard, 2009). The neural are used it the field of diesel engine. 2013)(Verhelst and FFNN modelnetworks and we we validated validated itin experimentally. and Sheppard, Sheppard, 2009). 2009). FFNN model and FFNN model and we validated experimentally. The neural networks are used it inexperimentally. the field of diesel engine. Several studies on the control of turbocharged diesel engine The neural networks are used in the field of diesel engine. The neural networks used in the The neural networks are used of in turbocharged the field field of of diesel diesel engine. Several studies on theare control diesel engine. engine 2.1 Modelling assumptions using networks presented by (Omrandiesel et al.,engine 2008) 2.1 Modelling assumptions Severalneural studies on the the are control of turbocharged turbocharged diesel engine Several studies on control of Several studies on the control of turbocharged using neural networks are presented by (Omrandiesel et al.,engine 2008) 2.1 2.1 Modelling Modelling assumptions assumptions and (Dovifaaz, 2002). are (Michael, 2013) presentset a al., study on 2.1 using neural networks are presented by (Omran et al., 2008) assumptions using neural networks presented by (Omran 2008) using neural networks presented by (Omran 2008) and (Dovifaaz, 2002). are (Michael, 2013) presentseta al., study on ThisModelling section gives a description of the main hypotheses on the estimation of NOx emissions Using Artificial Neural and (Dovifaaz, (Dovifaaz, 2002). 2002). (Michael, (Michael, 2013) 2013) presents presents aaa study study on on This section gives a description of the main hypotheses on and and (Dovifaaz, 2002). presents on the estimation of NOx(Michael, emissions2013) Using Artificialstudy Neural which the thermodynamic model isof suchhypotheses as the energy This section gives a description ofbased, the main hypotheses on gives the on Networks. (Ghobadian etemissions al., 2009)presents a work Neural on the This the estimation of NOx emissions Using Artificial Neural This section gives aa description description the main main on whichsection the thermodynamic model isofbased, suchhypotheses as the energy the estimation of NOx Using Artificial the estimation of NOx etemissions Using Artificial Networks. (Ghobadian al., 2009)presents a work Neural on the which and mass balances and the elements required tothe calculate which the thermodynamic model is based, such as the energy the thermodynamic model is based, such as energy diesel engine performance and the exhaust emission analysis Networks. (Ghobadian (Ghobadian et et al., al., 2009)presents 2009)presents aaa work work on on the the which the thermodynamic model is based, such astothe energy and mass balances and the elements required calculate Networks. Networks. (Ghobadian et al., work on the diesel engine performance and 2009)presents the exhaust emission analysis these balances (Xin, 2013) such as: and mass mass balances and the the elements required to to calculate calculate balances and elements required using an artificial neural and network. In this work analysis we are and diesel engine performance and the exhaust emission analysis and mass balances and the elements required to calculate these balances (Xin, 2013) such as: diesel engine performance the exhaust emission diesel performance the exhaust emission using engine an artificial neural and network. In this work analysis we are these Single zone model. these balances (Xin, 2013) such such as: balances (Xin, interested create aneural neural network thatwork simulates the 1. using an an to artificial neural network.model In this this work we are are these balances (Xin, 2013) 2013) such as: as: 1. Single zone model. using artificial network. In we using an artificial network. In this we are interested to create aneural neural network model thatwork simulates the 1. 2. Zero-dimensional state parameters. 1. Single zone model. zone combustion in normal and faulty conditions. interested to to cycle createoperation neural network network model that simulates simulates the 1. Single zone model. model.state parameters. 2. Single Zero-dimensional interested create aaa neural model that the interested to create neural network model that simulates the combustion cycle operation in normal and faulty conditions. 2. Zero-dimensional state state parameters. combustion cycle cycle operation operation in in normal normal and and faulty faulty conditions. conditions. 2. 2. Zero-dimensional Zero-dimensional state parameters. parameters. combustion combustion cycle operation in normal and faulty conditions.
Copyright © 2016, 2016 IFAC 221Hosting by Elsevier Ltd. All rights reserved. 2405-8963 © IFAC (International Federation of Automatic Control) Copyright © 2016 IFAC 221 Peer review under responsibility of International Federation of Automatic Copyright © 2016 IFAC 221Control. Copyright © 221 Copyright © 2016 2016 IFAC IFAC 221 10.1016/j.ifacol.2016.07.037
IFAC CTS 2016 222 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
du u dT u d dT u d . Cv . . . d T d d d d
3. State equations of Ideal gases
PV . m.r.T
(1)
(7)
Substituting these relationships into equation (2), the energy conservation equation becomes: dQ fuel dQwall dm dV P. hin . in d d d dT 1 d (8) . d m.Cv dmex u d dm h u m . . . . ex d d d
4. The quasi-steady-state process for the gas flowing into and out of the cylinder. 5. Ignoring the kinetic energy of the intake and exhaust gas. 6. The fuel is injected in the combustion chamber at a constant temperature and it is immediately burned following Wiebe law (Stone, 2012). 7. The heat transfer occurs through the five boundary limits (cylinder head, piston, cylinder wall, exhaust valve and admission valve) that are at a constant and uniform temperature.
The mass conservation equation can be expressed as: dm dmin dmex dm fuelB (9) d d d d
m fuelB is the fuel mass injected into the cylinder.
2.2 Model equations
Where
The equation of the conservation of energy applied in the cylinder can be expressed as (Stone, 2012)(Xin, 2013): dm j dQi dU dW hj (2) d d d d i j
2.3 Sub-models 2.3.1 Heat transfer through cylinder walls The heat transfer terms in the conservation of energy can be derived as follows (Woschni, 1967):
Where is the crank angle (degrees), U is the internal dQwall ,i dQwall 1 . g . Awall ,i .(T Twall ,i ) (10) energy of the gas in the cylinder, W is the mechanical work 6. d d NE i i acting on the piston, Qi is the heat exchanged through the Where g is the instantaneous spatial-average heat transfer system boundary and fuel combustion, and h j .m j is the coefficient from the cylinder gas to the inner cylinder wall, energy brought into and out of the cylinder by the intake and N E is the engine speed (rpm), Awall is the heat transfer exhaust gas flow. Each term of the equation (2) is further
expressed as:
dU d (m.u) dm du u. m. d d d d dW dV P. d d dQi dQ fuel dQwall d d d i
(3)
area, Twall is the spatial-average temperature of the cylinder wall surface. g is critical for heat transfer calculation. In 1967 Woschni
(4)
has developed a formula that describe heat transfer coefficient. 2.3.2 Mass transfer model
(5)
The intake mass flow rate entering the cylinder is given by the Baré Saint-Venant model (Basbous et al., 2012):
Where V is the cylinder instantaneous volume, m is the mass 2 1 of the gas in the cylinder, Q fuel is the heat energy released C f ,in . Ain .Pin dmin 2. P P . from fuel combustion, and Qwall is the heat transfer through d 6. N E . Rin .Tin 1 Pin Pin the walls of the cylinder head, the piston and the liner. (11) The energy of the intake and exhaust gas exchange mass flow is given by: Where N E is the engine speed (rpm), C f ,in is the intake dm j dm dm h hin . in hex . ex (6) valve flow coefficient, Pin and Tin are the pressure and j. d d d j temperature in the intake port just before the intake valve, Where min is the intake gas mass flowing into the cylinder, is the ratio of specific heat capacities of the intake gas flow. mex is the exhaust gas flowing out of the cylinder, hin and The exhaust gas mass flow rate out of the cylinder is given below, when: hex are the specific enthalpies of the gases at the intake and Pex 2 1 exhaust valves. (12) P 1 Since the specific internal energy for ideal gases can be
expressed as: u u (T , ) where α is the excess air- fuel ratio. The following relationship is obtained:
The valve flow is subsonic and can be described as:
222
IFAC CTS 2016 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
2.3.5 Gas properties model
2 1 2. Pex Pex . . 6. N E . Rex .T 1 p p (13)
C f ,ex . Aex .P
dmex d
The gas properties is a function of temperature and composition (Basbous et al., 2012) (“JANAF THERMOCHEMICAL TABLES,” 1989). Where hi is the molar enthalpy for each constituent, aij is
2 1 P When ex P 1
found in JANAF thermodynamic Tables.
(14)
u i (T ) hi (T ) Rmol .T
1
dmex C f ,ex . Aex .P 2 1 2. . d 1 6. N E . Rex .T 1
Where
(15)
hi (T ) Rmol
)
(20)
global internal energy is calculated using the fraction method: m
u(T )
x .u (T ) i
(21)
i
i 0
The Volume delimited by the piston, the cylinder wall and the cylinder head can be calculated as a function of the crank angle position (Heywood, 1988): 2 .BE 2 SE SE r r L 1 cos( ) . 1 sin 2 ( ) . 4 1 2 L L r
L sin(2 ) .BE .S E sin( ) 2 8 2.r l 2 1 .sin ( ) r
2.4 Experimental validation Figures 1 and 2 show the comparison between simulated data and experimental data obtained from the test bench. The test bench used in this work is a marine diesel engine manufactured by the company SIMB under the reference 6M26SRP1 (Nahim et al., 2015a). The results show a very good coincidence but the simulation time takes 40 seconds per cycle.
(16)
2
is the engine geometric compression ratio,
the cylinder bore diameter,
1200
(17) Temperature (k)
j
ij
ui is the internal energy of each constituent. The
Where
Pex is the
2.3.3 Kinematic model
Where
(a .T j 0
pressure in the exhaust port just behind the exhaust valve.
dV d
(19)
n
C f ,ex is the exhaust valve flow coefficient, Aex is the
instantaneous exhaust valve flow area, and
V
223
BE is
S E is the engine stroke, L is the
experimental T simulated T
1000 800 600 400 200 0
connecting rod length, and r is the crank radius.
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 1. Experimental validation of cylinder Temperature at 1500 rpm.
2.3.4 Burn rate and combustion process model
6
The most complex process to be considered when performing the energy balance is combustion. Different proposals can be found in the literature for the modeling of this phenomena (Cosine law, Watson's law, and Wiebe law) (Basbous et al., 2012). In this study, the global combustion model can be simplified by the consideration of two Wiebe laws. The first one describes the pre-mixed combustion and the second one describes the diffusion combustion. Each Wiebe law allows the calculation of the burned fuel ratio as (Nahim et al., 2015a) (Payri et al., 2011) :
Pressure (Pa)
15 x 10
experimental Pressure simulated pressure
10
5
0 0
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 2. Experimental validation of cylinder pressure at 1500 rpm.
1
X fuelb
SOC Ccom . com 1 e
Figure 3 show the variation of mass of air in the cylinder at each phase of the engine in one cycle. The mass of air increase in the admission phase and then decrease in the combustion phase, the decreasing quantity is equal to 14.7 times quantity of fuel injected, the excess of air is rejected out of the cylinder in exhaust phase. Figure 4 and figure 5 shows respectively the variation of mass of burned air and fuel in the cylinder during the cycle. A comparison between
(18)
Where is a non-dimensional shape factor. A smaller produces a faster burning rate, com is the combustion duration, SOC is the crank angle at the start of combustion,
Ccom is a constant equal to 6.908.
223
IFAC CTS 2016 224 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
increasing of intake pressure for the same intake temperature increases the maximum pressure. This is due to the higher air quantity admitted.
-3
x 10
mass of air
0.8 0.6
6
15 x 10
0.4 0.2 0 0
100
200
300 400 500 Crank angle (degree)
600
Pressure (Pa)
mass of air (kg)
1
700
Fig. 3. Variation of mass of air in the cylinder at 1500 rpm.
Pressure
10
5
mass of burned gas (kg)
-4
6 x 10
0 0
mass of burned gas
4
3
3.5 -4 x 10
3. NEURAL NETWORK MODEL 100
200
300 400 500 Crank angle (degree)
600
700
Neural networks are non linear computer algorithms and can model the behavior of complicated non linear processes. They produce results very fast because of their property of working in parallel to solve a specific problem. The true power and advantage of ANN lies in their way to analyze data and to recognize patterns within that data (Rawlins, 2005). The basic processing element of a neural network is a neuron. Neural network is created by a means of interconnection of neurons and usually organized into layers. On the other hand, they have the ability to learn the relationship between the input and the output (T. Hari Prasad, 2010). In the present work, a multilayer feed-forward neural network with one hidden layer (Figure 9) is used and trained with the backpropagation algorithm in a supervised manner. Levenberg-Marquardt algorithm is used to optimize the performance function or the mean square error function.
-5
x 10
mass of fuel
2.5
mass of fuel (kg)
1.5 2 2.5 3 Volume (m )
2
Fig. 4. Variation of mass of burned gas in the cylinder at 1500 rpm.
2 1.5 1 0.5 0
340
360
380 400 420 Crank angle (degree)
440
460
Fig. 5. Variation of mass of fuel in the cylinder at 1500 rpm. 6
15 x 10
Pressure (Pa)
1
Fig. 8. P-V diagram at 1500 rpm.
0 0
Firing Pressure Motoring Pressure
10
5
3.1 Multilayer feed-forward networks
0 0
100
200
300 400 500 Crank angle (degree)
600
Multilayer networks are the simplest and most commonly used neural network architecture (Hagan et al., 2014). In our study the cylinder pressure is estimated by two layer feedforward network with sigmoid non linear transfer function as illustrated in Figure 10, the hidden layer is composed of ten neurons and one neuron in the output layer.
700
Fig. 6. Comparison between cylinder firing pressure and motoring pressure at 1500 rpm. Maximum pressure (Pa)
7
x 10 2
305
Intake Temperature (k)
0.5
1.8
300
1.6
295
1.4
290
1.2
285
1
280
0.8
275 1.8
2
2.2 2.4 Intake pressure (Pa)
2.6
2.8 5 x 10
3
Fig. 9. Architecture of neural network that estimate the cylinder pressure.
Fig. 7. Maximum gas pressure function of intake pressure and temperature for a fixed exhaust pressure of 2.5 bar at 1500 rpm.
Multilayer feed-forward networks have three distinct characteristics:
firing and motoring pressure is done in Figure 6. The results show the difference of pressure in the combustion phase. Regarding maximum cylinder pressure, we observe in Figure 7 that reducing the intake temperature for the same intake pressure increases the maximum cylinder pressure and the
1.
224
They have hidden units that include nonlinear activation functions that are smooth (or differentiable everywhere).
IFAC CTS 2016 May 18-20, 2016. Istanbul, Turkey
3.
They contain one or more layers of hidden units that are not part of the input or output of the network. They have a high degree of connectivity between the units.
1500
Temperature (k)
2.
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
The combination of these characteristics, together with the ability to learn from experience through training, add a significant computing power to multilayer feed-forward networks. It has been shown that a two-layer feed-forward network with sigmoid nonlinearity can approximate any function with arbitrary accuracy (Hagan et al., 2014).
225
T neural network T experimental
1000
500
0 0
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 12. Comparison of ANN result temperature with experimental result.
3.2 Back propagation algorithm 4. SIMULATION OF CYCLE WITH FAULT The training algorithm used in this work is back propagation algorithm, which is used to train the network in a supervised manner until it can approximate a function with high accuracy. The performance index of this algorithm is the mean square error. It can be described as: 1. Initialize of weights and biases arbitrary. 2. Propagate the input forward through the network. 3. Propagate the sensitivities backward through the network. 4. Update the weights and biases.
The main aim of this study is to simulate the thermodynamic cycle using neural network in normal and faulty condition. The first step was the simulation of internal pressure and temperature using neural network in normal conditions. The second step is to simulate the cylinder pressure and temperature using neural network in faulty conditions, that's why we re-trained the network to estimate the cylinder pressure and temperature in these conditions. 4.1 Cooling fault
3.3 Results and discussion
The cooling system is responsible of maintaining the temperature in the cylinder at an optimum operating temperature. If any fault occurs in this system, the temperature of cooling water entering the cylinder and the temperature of the metallic component of the engine increase (cylinder liner, intake and exhaust valves…), then the temperature of the gases flowing through the cylinder increases, and as a consequence the flow of air entering the cylinder decreases, then the pressure in the cylinder decreases and the efficiency of the engine decreases as shown in figure 13 and 14.
The aim of using the FFNN models in this work is to estimate the cylinder pressure and temperature for a diesel engine and to make the simulation time of one cycle equivalent to that of the real diesel engine. The model has two subsystems. The first estimates the cylinder temperature and the second estimates the cylinder pressure (Figure 10). Figure 11 and 12, shows a comparison between the cylinder pressure and temperature estimated by ANN with the experimental data show a good coincidence and the duration of one cycle is 0.5 second comparing to 40 second with the physical model.
Temperature (k)
2500
T without fault T with fault
2000 1500 1000 500 0 0
6
x 10
300 400 500 Crank angle (degree)
600
700
6
15
x 10
P neural network P experimental
15
Pressure (Pa)
Pressure (Pa)
200
Fig. 13. Comparison of the cylinder temperature with and without fault in cooling system at 1500 rpm.
Fig. 10. ANN model structure for pressure and temperature.
10
5
0 0
100
100
200
300 400 500 Crank angle (degree)
600
700
10
5
0 0
Fig. 11. Comparison of ANN result Pressure with experimental result.
P without fault P with fault
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 14. Comparison of the cylinder Pressure with and without fault in cooling system at 1500 rpm.
225
IFAC CTS 2016 226 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
4.2. Injection time error
Dovifaaz, X., 2002. Neural modeling for diesel engine control, in: Basañez, L. (Ed.), . pp. 1523–1523. doi:10.3182/20020721-6-ES-1901.01525 Ghobadian, B., Rahimi, H., Nikbakht, A.M., Najafi, G., Yusaf, T.F., 2009. Diesel engine performance and exhaust emission analysis using waste cooking biodiesel fuel with an artificial neural network. Renew. Energy 34, 976–982. doi:10.1016/j.renene.2008.08.008 Hagan, M., Demuth, H., Beale, M., Jesus, O.D., 2014. Neural Network Design (2nd Edition). Martin Hagan. Heywood, J., 1988. Internal Combustion Engine Fundamentals. McGraw-Hill Education. JANAF THERMOCHEMICAL TABLES, 1989. . Anal. Chem. 61, 1327A–1327A. doi:10.1021/ac00198a726 Michael, F., 2013. Transient NOx Estimation Using Artificial Neural Networks, in: Taketoshi, K. (Ed.), . pp. 101– 106. doi:10.3182/20130904-4-JP-2042.00006 Nahim, H.M., Younes, R., Nohra, C., Ouladsine, M., 2015a. Complete modeling for systems of a marine diesel engine. J. Mar. Sci. Appl. 14, 93–104. doi:10.1007/s11804-015-1285-y Nahim, H.M., Younes, R., Shraim, H., Ouladsine, M., 2015b. Oriented review to potential simulator for faults modeling in diesel engine. J. Mar. Sci. Technol. 1– 19. doi:10.1007/s00773-015-0358-6 Omran, R., Younes, R., Champoussin, J.-C., 2008. Neural networks for real-time nonlinear control of a variable geometry turbocharged diesel engine. Int. J. Robust Nonlinear Control 18, 1209–1229. Payri, F., Olmeda, P., Martín, J., García, A., 2011. A complete 0D thermodynamic predictive model for direct injection diesel engines. Appl. Energy 88, 4632–4641. doi:10.1016/j.apenergy.2011.06.005 Rawlins, M.S., 2005. Diesel engine performance modelling using neural networks (Thesis). Stone, R., 2012. Introduction to Internal Combustion Engines. Palgrave Macmillan. T. Hari Prasad, D.K.H.C.R., 2010. Performance and Exhaust Emissions Analysis of a Diesel Engine Using Methyl Esters of Fish Oil with Artificial Neural Network Aid. Int. J. Eng. Technol. 2, 23–27. doi:10.7763/IJET.2010.V2.94 Verhelst, S., Sheppard, C.G.W., 2009. Multi-zone thermodynamic modelling of spark-ignition engine combustion – An overview. Energy Convers. Manag. 50, 1326–1335. doi:10.1016/j.enconman.2009.01.002 Woschni, G., 1967. A Universally Applicable Equation for the Instantaneous Heat Transfer Coefficient in the Internal Combustion Engine. SAE 670931. SAE Tech. Pap. doi:10.4271/670931 Xin, Q. (Ed.), 2013. Copyright, in: Diesel Engine System Design. Woodhead Publishing, p. iv.
When AI decrease, the quantity of fuel entering the cylinder decreases, so the maximum pressure and temperature in the cylinder decrease (Figure 15 and 16), and thus, the power delivered by the engine decrease. Temperature (k)
2000
T AI= 1 degree T AI= 3 degree T AI= 5 degree
1500 1000 500 0 0
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 15. Comparison of the cylinder temperature for different injection timing at 1500 rpm. 7
x 10
P AI=1 degree P AI= 3 degree P AI= 5 degree
Pressure (Pa)
1.6 1.4 1.2 1 0.8 0.6 320
340
360 380 Crank angle (degree)
400
420
Fig. 16. Comparison of the cylinder Pressure for different injection timing at 1500 rpm.
5.
CONCLUSION
In the proposed thermodynamic model, the instantaneous gas states in the combustion chamber are obtained using the mass and energy conservation equations. A detailed description of the elements used for the calculation of each term in the energy balance has been provided. The main aim of this paper is to show the possibility of using artificial neural network for the prediction of a diesel engine performance of pressure and temperature in the cylinder in normal and faulty operation mode. A reduction of calculation time 80 times (from 40 seconds to 0.5 seconds) between the physical model and the NN model has been realized. This presented work is a part of a complete project for the realization of a diesel engine simulator, that simulates the variation of parameters of the engine in normal and faulty conditions. REFERENCES Awad, S., Varuvel, E.G., Loubar, K., Tazerout, M., 2013. Single zone combustion modeling of biodiesel from wastes in diesel engine. Fuel 106, 558–568. doi:10.1016/j.fuel.2012.11.051 Basbous, T., Younes, R., Ilinca, A., Perron, J., 2012. Pneumatic hybridization of a diesel engine using compressed air storage for wind-diesel energy generation. Energy 38, 264–275. doi:10.1016/j.energy.2011.12.003
226
ScienceDirect IFAC-PapersOnLine 49-3 (2016) 221–226 Modeling Modeling and and simulation simulation of of the the thermodynamic thermodynamic cycle cycle of of the the Diesel Diesel Engine Engine using using Modeling thermodynamic cycle of the Diesel Engine using Networks Modeling and and simulation simulation of of the theNeural thermodynamic cycle of the Diesel Engine using Neural Networks Neural Networks Neural Networks Ali Rida*, Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Ouladsine*****
Ali Rida*, Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Ouladsine***** Ali Rida*, Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Ouladsine***** Ali Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Ali Rida*, Rida*, Hassan Hassan Moussa Nahim**, Rafic Younes***, Hassan Shraim****, Mustapha Mustapha Ouladsine***** Ouladsine***** *Lebanease University, FOE, Beyrouth Hadath ([email protected]) *Lebanease University, FOE, Beyrouth Hadath ([email protected]) **Aix Marseille University, LSIS, Marseille ([email protected]) *Lebanease University, FOE, Beyrouth Hadath ([email protected]) *Lebanease University, FOE, Hadath *Lebanease University, FOE, Beyrouth Beyrouth Hadath ([email protected]) ([email protected]) **Aix Marseille University, LSIS, Marseille ([email protected]) *** Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) **Aix Marseille University, LSIS, Marseille ([email protected]) **Aix Marseille University, Marseille ([email protected]) *** Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) **Aix Marseille University, Marseille ([email protected]) ****Lebanease University, FOE, Beyrouth Hadath ([email protected]) *** Lebanease Lebanease University, University, FOE, FOE, Beyrouth Beyrouth Hadath Hadath ([email protected]) ([email protected]) *** ****Lebanease University, FOE, Beyrouth Hadath ([email protected]) *** Lebanease University, FOE, Beyrouth Hadath ([email protected]) ***** Aix Marseille University, Marseille ([email protected]) ****Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) ****Lebanease University, FOE, Beyrouth Hadath ([email protected]) ***** Aix Marseille University, Marseille ([email protected]) ****Lebanease University, FOE,LSIS, Beyrouth Hadath ([email protected]) ***** Aix Marseille University, LSIS, Marseille ([email protected]) Aix Marseille ([email protected]) Abstract: In ***** this paper, a unique University, single zoneLSIS, combustion model is proposed to predict Diesel engine’s Aix Marseille Marseille Marseille ([email protected]) Abstract: In ***** this paper, a unique University, single zoneLSIS, combustion model is proposed to predict Diesel engine’s performance, pressure, and temperature based on the conservation of massto and energy. In engine’s order to Abstract: In this paper, a unique single zone combustion model is proposed predict Diesel Abstract: paper, aa unique single model predict Diesel Abstract: In In this this paper, and unique single zone zone combustion model is is proposed proposed toand predict Diesel engine’s performance, pressure, temperature basedcombustion on the conservation of massto energy. In engine’s order to simulate all phases of combustion, the proposed model takes in consideration the dynamics of the intake performance, pressure, and temperature based on the conservation of mass and energy. In order to performance, pressure, and based on conservation of and energy. order to performance, pressure, and temperature temperature based model on the the takes conservation of mass massthe and energy. ofIn Inthe order to simulate all phases of combustion, the proposed in consideration dynamics intake and exhaust gas through the valves, the ignition delay, the instantaneous change in gas properties, the simulate all phases of combustion, the proposed model takes in consideration the dynamics of the intake simulate all of the proposed in the of simulate all phases phases of combustion, combustion, thethe proposed model takes in consideration consideration the dynamics dynamics of the the intake intake and exhaust gas through the valves, ignitionmodel delay,takes the instantaneous change in gas properties, the properties of gas the burned and the the heatignition losses by the walls. Validation ofchange this model hasproperties, been realized and exhaust exhaust gas throughfuel, the valves, valves, the ignition delay, the instantaneous instantaneous change in gas gas properties, the and through the delay, the in the properties of gas the burned fuel, and the the heatignition losses by the walls. Validation ofchange this model hasproperties, been realized and exhaust through the valves, delay, the instantaneous in gas by experimental data. Important issue has been recognized that the physicalof model takes too been muchrealized timethe in properties of the burned fuel, and the heat losses by the walls. Validation of this model has been realized properties of the burned fuel, and the heat losses by the walls. Validation this model has properties of the data. burned fuel, andissue the heat losses by the walls. this model by experimental Important has been recognized that Validation the physicalofmodel takes has too been muchrealized time in calculation. For this purpose, a Feed-Forward Neural Network (FFNN) model is developed and validated by experimental data. Important issue has been recognized that the physical model takes too much time by experimental data. Important issue has been recognized that the physical model takes too much time in by experimental data.purpose, Important issue has beenNeural recognized that (FFNN) the physical model takes too and much time in in calculation. For this a Feed-Forward Network model is developed validated experimentally to predict the aapressure and temperature in the cylinder in nominal and faulty operations. calculation. For this purpose, Feed-Forward Neural Network (FFNN) model is developed and validated calculation. For this purpose, Feed-Forward Neural Network (FFNN) model is developed and validated calculation. Forto this purpose, Feed-Forward Neural Network (FFNN) in model is developed and validated experimentally predict the apressure and temperature in the cylinder nominal and faulty operations. Finally, the influence of some possible and faults that may be produced on the diesel engine cycleoperations. during the experimentally to predict the pressure and temperature in the cylinder in nominal and faulty operations. experimentally to the temperature in the in and experimentally to predict predict the pressure pressure temperature in produced the cylinder cylinder in nominal nominal and faulty faulty Finally, the influence of some possible and faults that may be on the diesel engine cycleoperations. during the operation has been analyzed. Finally, the influence of some possible faults that may be produced on the diesel engine cycle during the Finally, of Finally, the influence of some some possible possible faults faults that that may may be be produced produced on on the the diesel diesel engine engine cycle cycle during during the the operationthe hasinfluence been analyzed. operation has been analyzed. operation has been analyzed. Keywords: Diesel Engine, Thermodynamic cycle, Artificial Neural Network Modeling, Faultyreserved. operation © 2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights operation has been analyzed. Keywords: Diesel Engine, Thermodynamic cycle, Artificial Neural Network Modeling, Faulty operation mode. Keywords: Diesel Diesel Engine, Thermodynamic cycle, Artificial Neural Network Modeling, Faulty operation Keywords: Keywords: Diesel Engine, Engine, Thermodynamic Thermodynamic cycle, cycle, Artificial Artificial Neural Neural Network Network Modeling, Modeling, Faulty Faulty operation operation mode. mode. mode. mode. Thus, the FFNN are trained using the experimental data. 1. INTRODUCTION Thus, the FFNN are trained using the experimental data. Network and learning rateexperimental parameters data. are 1. INTRODUCTION Thus, the the architecture FFNN are are trained trained using the the experimental data. Thus, FFNN using Thus, the FFNN are trained using the Network architecture and learning rateexperimental parameters data. are 1. INTRODUCTION INTRODUCTION 1. optimized using Back-propagation algorithm and LevenbergDiesel engine is a very complicated mechanical system. It is Network 1. INTRODUCTION Network architecture and learning rate parameters are architecture and learning rate parameters are Network architecture and learning rate parameters are using Back-propagation algorithm and LevenbergDiesel engine is a very complicated mechanical system. It is optimized Marquardt algorithm to optimize algorithm the performance function characterized by thevery ruggedness in construction, simplicity in optimized using Back-propagation algorithm and LevenbergDiesel engine is a very complicated mechanical system. It is optimized using Back-propagation and LevenbergDiesel engine is a complicated mechanical system. It is optimized using Back-propagation algorithm and Levenbergalgorithm to optimize the performance function Diesel engine by is athevery complicated mechanical simplicity system. It in is Marquardt characterized ruggedness in construction, (Hagan et al., 2014). The main contribution of this paper can operation andbyease of maintenance. It has become quite Marquardt algorithm to optimize optimize the performance performance function characterized the ruggedness ruggedness in construction, construction, simplicity in Marquardt algorithm to the function characterized the in simplicity in Marquardt algorithm to optimize the performance function (Hagan et al., 2014). The main contribution of this paper can characterized byease the ruggedness in construction, simplicity in operation andby of maintenance. It has become quite be resumed by the development of neural network model that popular in transportation and agriculture sector, because of (Hagan et et al., al., 2014). 2014). The The main main contribution contribution of of this this paper paper can can operation and and ease ease of of maintenance. maintenance. It It has has become become quite quite (Hagan operation (Hagan et al., The main contribution of this paper can resumed by2014). the development of neural network model that operation ease of maintenance. It has become quite popular in and transportation and agriculture sector, because of be estimates the variation of temperature and pressure in that the higher efficiency and longer operational time. In because this paper, be resumed by the development of neural network model that popular in transportation and agriculture sector, because of be resumed by the development of neural network model popular in transportation and agriculture sector, of resumedthe by variation the development of neuraland network model estimates of temperature pressure in that the popular in transportation andoperational agriculturetime. sector, of be higher efficiency and longer In because this paper, cylinder in normal and faulty operation mode. In our previous we present a physical model of thermodynamic cycle of estimates the variation of temperature and pressure in the higher efficiency and longer operational time. In this paper, variation of and in higher efficiency and operational time. paper, estimates the variation of temperature temperature and pressure pressure in the the cylinder inthe normal and faulty operation mode. In our previous higher efficiency and longer longer operational time. In In this this paper, we present a physical model of thermodynamic cycle of estimates works (Nahim et al., 2015a), we have developed a diesel diesel engine. During the engine cycle, gas temperature, cylinder in normal and faulty operation mode. In our previous we present a physical model of thermodynamic cycle of in operation In we present aa physical model of cycle cylinder in normal normal and faulty operation mode. In our our previous previous works (Nahim et and al., faulty 2015a), we havemode. developed a diesel we present physical model of thermodynamic thermodynamic cycle of of cylinder diesel engine. During the engine cycle, gas temperature, engine simulator that 2015a), simulates operation in pressure and mass can bethe found by applying the conservation works (Nahim (Nahim et al., al., 2015a), we the haveengine developed diesel diesel engine. engine. During the engine cycle, gas gas temperature, works et we have developed aaa diesel diesel During engine cycle, temperature, works et al., we have developed diesel engine (Nahim simulator that 2015a), simulates the engine operation in diesel During engine cycle, gas temperature, pressureengine. and mass can bethe found by applying the conservation presence of failure (Nahim et al., 2015b). The developed of energy and mass (Xin, 2013) and by the application of the engine simulator that simulates the engine operation in pressure and mass can be found by applying the conservation engine simulator that simulates the engine operation in pressure be found the simulates engineThe operation in presencesimulator of failurethat (Nahim et al.,the 2015b). developed pressure and mass can be 2013) found by by applying the conservation conservation of energyand andmass masscan (Xin, andapplying by the application of the engine model canof befailure used to(Nahim simulate the influence of thedeveloped fault on ideal gas and equations in 2013) the cylinder. Initial condition of presence presence of failure (Nahim et al., 2015b). The developed of energy and mass (Xin, 2013) and by the application of the et al., 2015b). The of energy mass (Xin, and by the application of the et the al., influence 2015b). The model canofbefailure used to(Nahim simulate of thedeveloped fault on of energy mass (Xin, and by the application of the ideal gas and equations in 2013) the cylinder. Initial condition of presence the global it can bethe used in real time simulation pressure, andthe can be Initial estimated from the model can system, be used usedand to simulate simulate the influence of the the fault on on ideal gas gas temperature equations in in themass cylinder. Initial condition of model can be to influence of fault ideal equations cylinder. condition of model can be used to simulate influence of the fault on the global system, and it can bethe used in real time simulation ideal gas equations in cylinder. condition of pressure, temperature andthe mass can be Initial estimated from the to detect any failureand in the engine. intake manifold parameters, and theestimated computation is the the global system, and it can be used in real time simulation pressure, temperature and mass can be estimated from the global system, it can be used in real time simulation pressure, temperature and mass can be from the globalany system, it can be used in real time simulation to detect failureand in the engine. pressure, temperature and mass and can betheestimated from the intake manifold parameters, computation is the conducted on a crank angle basis.and A zero-dimensional model to detect any any failure in in the engine. engine. intake manifold manifold parameters, and the computation computation is to intake parameters, the is to detect detect any failure failure in the the engine. intake manifold parameters, the computation is conducted on a crank angle basis.and A zero-dimensional model combined a single zone model is used in this work, in conducted with on aaa crank crank angle basis. A zero-dimensional zero-dimensional model conducted on angle basis. A model conducted on crank angle basis. A zero-dimensional model combined with a single zone model is used in this work, in 2. COMBUSTION MODEL order to have a fast and accurate analysis of the engine combined with a single zone model is used in this work, in 2. COMBUSTION MODEL combined with zone model used in combined with aaa single single zoneaccurate model is isanalysis used in inofthis this work, in order to have fast and thework, engine 2. COMBUSTION COMBUSTION MODEL 2. performance et al., 2012). The model then order to to have have(Basbous fast and and accurate analysis of the the isengine engine COMBUSTION MODEL MODEL order aaa fast accurate analysis of order to have fast and accurate analysis of the performance (Basbous et al., 2012). The model isengine then The purpose of 2. a zero-dimensional, single zone combustion implemented in Matlab/Simulink in a complete diesel engine performance (Basbous et al., 2012). The model is then The purpose of a zero-dimensional, single zone combustion performance (Basbous et The then performance (Basbous et al., al., 2012). 2012). The model model isengine then The implemented in Matlab/Simulink in a complete dieselis model developed in this work is tozone simulate the The purpose of a zero-dimensional, single zone combustion purpose of a zero-dimensional, single combustion simulator proposed by (Nahim etin al., 2015a), diesel and then the The implemented in Matlab/Simulink in a complete diesel engine purpose of a zero-dimensional, single combustion model developed in this work is tozone simulate the implemented in Matlab/Simulink a complete engine implemented in Matlab/Simulink a complete engine simulator proposed by (Nahim etinal., 2015a), diesel and then the model thermodynamic cyclein from the work beginning ofto thesimulate intake phase model developed in this work is to simulate the developed this is the global model is validated experimentally. But due to the fact simulator proposed by (Nahim et al., 2015a), and then the model developed in this work is to simulate the thermodynamic cycle from the beginning of the intake phase simulator proposed by (Nahim et al., 2015a), and then the simulator proposed by (Nahim et al., 2015a), andtothen the thermodynamic global model is validated experimentally. But due the fact to the end of thecycle exhaust phase. The model is derived based thermodynamic cycle from the beginning of the intake phase from the beginning of the intake phase that the simulation time of one engine cycle is too long, we global model model is is validated validated experimentally. experimentally. But But due due to to the the fact fact thermodynamic fromphase. the beginning of is thederived intake based phase to the end of thecycle exhaust The model global global is validated But the fact that themodel simulation time ofexperimentally. one engine cycle isdue tootolong, we to on theend conservation of phase. mass and energyis al., to the the end of the the exhaust exhaust phase. The model model is(Awad derived etbased based of The derived transformed the physical model of thermodynamic cycle into that the simulation time of one engine cycle is too long, we to the end of the exhaust phase. The model is derived based on the conservation of mass and energy (Awad et al., that the simulation time of one engine cycle is too long, we that the simulation time of one engine cycle is toocycle long,into we on transformed the physical model of thermodynamic 2013)(Verhelst and Sheppard, 2009). on the conservation of mass and energy (Awad et al., the conservation of mass and energy (Awad et al., FFNN modelthe andphysical we validated itof experimentally. transformed the physical model of thermodynamic cycle into on the conservation of mass and energy (Awad et al., 2013)(Verhelst and Sheppard, 2009). transformed model thermodynamic cycle into transformed modelitof thermodynamic cycle into 2013)(Verhelst FFNN modelthe andphysical we validated experimentally. 2013)(Verhelst and Sheppard, 2009). The neural are used it the field of diesel engine. 2013)(Verhelst and FFNN modelnetworks and we we validated validated itin experimentally. and Sheppard, Sheppard, 2009). 2009). FFNN model and FFNN model and we validated experimentally. The neural networks are used it inexperimentally. the field of diesel engine. Several studies on the control of turbocharged diesel engine The neural networks are used in the field of diesel engine. The neural networks used in the The neural networks are used of in turbocharged the field field of of diesel diesel engine. Several studies on theare control diesel engine. engine 2.1 Modelling assumptions using networks presented by (Omrandiesel et al.,engine 2008) 2.1 Modelling assumptions Severalneural studies on the the are control of turbocharged turbocharged diesel engine Several studies on control of Several studies on the control of turbocharged using neural networks are presented by (Omrandiesel et al.,engine 2008) 2.1 2.1 Modelling Modelling assumptions assumptions and (Dovifaaz, 2002). are (Michael, 2013) presentset a al., study on 2.1 using neural networks are presented by (Omran et al., 2008) assumptions using neural networks presented by (Omran 2008) using neural networks presented by (Omran 2008) and (Dovifaaz, 2002). are (Michael, 2013) presentseta al., study on ThisModelling section gives a description of the main hypotheses on the estimation of NOx emissions Using Artificial Neural and (Dovifaaz, (Dovifaaz, 2002). 2002). (Michael, (Michael, 2013) 2013) presents presents aaa study study on on This section gives a description of the main hypotheses on and and (Dovifaaz, 2002). presents on the estimation of NOx(Michael, emissions2013) Using Artificialstudy Neural which the thermodynamic model isof suchhypotheses as the energy This section gives a description ofbased, the main hypotheses on gives the on Networks. (Ghobadian etemissions al., 2009)presents a work Neural on the This the estimation of NOx emissions Using Artificial Neural This section gives aa description description the main main on whichsection the thermodynamic model isofbased, suchhypotheses as the energy the estimation of NOx Using Artificial the estimation of NOx etemissions Using Artificial Networks. (Ghobadian al., 2009)presents a work Neural on the which and mass balances and the elements required tothe calculate which the thermodynamic model is based, such as the energy the thermodynamic model is based, such as energy diesel engine performance and the exhaust emission analysis Networks. (Ghobadian (Ghobadian et et al., al., 2009)presents 2009)presents aaa work work on on the the which the thermodynamic model is based, such astothe energy and mass balances and the elements required calculate Networks. Networks. (Ghobadian et al., work on the diesel engine performance and 2009)presents the exhaust emission analysis these balances (Xin, 2013) such as: and mass mass balances and the the elements required to to calculate calculate balances and elements required using an artificial neural and network. In this work analysis we are and diesel engine performance and the exhaust emission analysis and mass balances and the elements required to calculate these balances (Xin, 2013) such as: diesel engine performance the exhaust emission diesel performance the exhaust emission using engine an artificial neural and network. In this work analysis we are these Single zone model. these balances (Xin, 2013) such such as: balances (Xin, interested create aneural neural network thatwork simulates the 1. using an an to artificial neural network.model In this this work we are are these balances (Xin, 2013) 2013) such as: as: 1. Single zone model. using artificial network. In we using an artificial network. In this we are interested to create aneural neural network model thatwork simulates the 1. 2. Zero-dimensional state parameters. 1. Single zone model. zone combustion in normal and faulty conditions. interested to to cycle createoperation neural network network model that simulates simulates the 1. Single zone model. model.state parameters. 2. Single Zero-dimensional interested create aaa neural model that the interested to create neural network model that simulates the combustion cycle operation in normal and faulty conditions. 2. Zero-dimensional state state parameters. combustion cycle cycle operation operation in in normal normal and and faulty faulty conditions. conditions. 2. 2. Zero-dimensional Zero-dimensional state parameters. parameters. combustion combustion cycle operation in normal and faulty conditions.
Copyright © 2016, 2016 IFAC 221Hosting by Elsevier Ltd. All rights reserved. 2405-8963 © IFAC (International Federation of Automatic Control) Copyright © 2016 IFAC 221 Peer review under responsibility of International Federation of Automatic Copyright © 2016 IFAC 221Control. Copyright © 221 Copyright © 2016 2016 IFAC IFAC 221 10.1016/j.ifacol.2016.07.037
IFAC CTS 2016 222 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
du u dT u d dT u d . Cv . . . d T d d d d
3. State equations of Ideal gases
PV . m.r.T
(1)
(7)
Substituting these relationships into equation (2), the energy conservation equation becomes: dQ fuel dQwall dm dV P. hin . in d d d dT 1 d (8) . d m.Cv dmex u d dm h u m . . . . ex d d d
4. The quasi-steady-state process for the gas flowing into and out of the cylinder. 5. Ignoring the kinetic energy of the intake and exhaust gas. 6. The fuel is injected in the combustion chamber at a constant temperature and it is immediately burned following Wiebe law (Stone, 2012). 7. The heat transfer occurs through the five boundary limits (cylinder head, piston, cylinder wall, exhaust valve and admission valve) that are at a constant and uniform temperature.
The mass conservation equation can be expressed as: dm dmin dmex dm fuelB (9) d d d d
m fuelB is the fuel mass injected into the cylinder.
2.2 Model equations
Where
The equation of the conservation of energy applied in the cylinder can be expressed as (Stone, 2012)(Xin, 2013): dm j dQi dU dW hj (2) d d d d i j
2.3 Sub-models 2.3.1 Heat transfer through cylinder walls The heat transfer terms in the conservation of energy can be derived as follows (Woschni, 1967):
Where is the crank angle (degrees), U is the internal dQwall ,i dQwall 1 . g . Awall ,i .(T Twall ,i ) (10) energy of the gas in the cylinder, W is the mechanical work 6. d d NE i i acting on the piston, Qi is the heat exchanged through the Where g is the instantaneous spatial-average heat transfer system boundary and fuel combustion, and h j .m j is the coefficient from the cylinder gas to the inner cylinder wall, energy brought into and out of the cylinder by the intake and N E is the engine speed (rpm), Awall is the heat transfer exhaust gas flow. Each term of the equation (2) is further
expressed as:
dU d (m.u) dm du u. m. d d d d dW dV P. d d dQi dQ fuel dQwall d d d i
(3)
area, Twall is the spatial-average temperature of the cylinder wall surface. g is critical for heat transfer calculation. In 1967 Woschni
(4)
has developed a formula that describe heat transfer coefficient. 2.3.2 Mass transfer model
(5)
The intake mass flow rate entering the cylinder is given by the Baré Saint-Venant model (Basbous et al., 2012):
Where V is the cylinder instantaneous volume, m is the mass 2 1 of the gas in the cylinder, Q fuel is the heat energy released C f ,in . Ain .Pin dmin 2. P P . from fuel combustion, and Qwall is the heat transfer through d 6. N E . Rin .Tin 1 Pin Pin the walls of the cylinder head, the piston and the liner. (11) The energy of the intake and exhaust gas exchange mass flow is given by: Where N E is the engine speed (rpm), C f ,in is the intake dm j dm dm h hin . in hex . ex (6) valve flow coefficient, Pin and Tin are the pressure and j. d d d j temperature in the intake port just before the intake valve, Where min is the intake gas mass flowing into the cylinder, is the ratio of specific heat capacities of the intake gas flow. mex is the exhaust gas flowing out of the cylinder, hin and The exhaust gas mass flow rate out of the cylinder is given below, when: hex are the specific enthalpies of the gases at the intake and Pex 2 1 exhaust valves. (12) P 1 Since the specific internal energy for ideal gases can be
expressed as: u u (T , ) where α is the excess air- fuel ratio. The following relationship is obtained:
The valve flow is subsonic and can be described as:
222
IFAC CTS 2016 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
2.3.5 Gas properties model
2 1 2. Pex Pex . . 6. N E . Rex .T 1 p p (13)
C f ,ex . Aex .P
dmex d
The gas properties is a function of temperature and composition (Basbous et al., 2012) (“JANAF THERMOCHEMICAL TABLES,” 1989). Where hi is the molar enthalpy for each constituent, aij is
2 1 P When ex P 1
found in JANAF thermodynamic Tables.
(14)
u i (T ) hi (T ) Rmol .T
1
dmex C f ,ex . Aex .P 2 1 2. . d 1 6. N E . Rex .T 1
Where
(15)
hi (T ) Rmol
)
(20)
global internal energy is calculated using the fraction method: m
u(T )
x .u (T ) i
(21)
i
i 0
The Volume delimited by the piston, the cylinder wall and the cylinder head can be calculated as a function of the crank angle position (Heywood, 1988): 2 .BE 2 SE SE r r L 1 cos( ) . 1 sin 2 ( ) . 4 1 2 L L r
L sin(2 ) .BE .S E sin( ) 2 8 2.r l 2 1 .sin ( ) r
2.4 Experimental validation Figures 1 and 2 show the comparison between simulated data and experimental data obtained from the test bench. The test bench used in this work is a marine diesel engine manufactured by the company SIMB under the reference 6M26SRP1 (Nahim et al., 2015a). The results show a very good coincidence but the simulation time takes 40 seconds per cycle.
(16)
2
is the engine geometric compression ratio,
the cylinder bore diameter,
1200
(17) Temperature (k)
j
ij
ui is the internal energy of each constituent. The
Where
Pex is the
2.3.3 Kinematic model
Where
(a .T j 0
pressure in the exhaust port just behind the exhaust valve.
dV d
(19)
n
C f ,ex is the exhaust valve flow coefficient, Aex is the
instantaneous exhaust valve flow area, and
V
223
BE is
S E is the engine stroke, L is the
experimental T simulated T
1000 800 600 400 200 0
connecting rod length, and r is the crank radius.
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 1. Experimental validation of cylinder Temperature at 1500 rpm.
2.3.4 Burn rate and combustion process model
6
The most complex process to be considered when performing the energy balance is combustion. Different proposals can be found in the literature for the modeling of this phenomena (Cosine law, Watson's law, and Wiebe law) (Basbous et al., 2012). In this study, the global combustion model can be simplified by the consideration of two Wiebe laws. The first one describes the pre-mixed combustion and the second one describes the diffusion combustion. Each Wiebe law allows the calculation of the burned fuel ratio as (Nahim et al., 2015a) (Payri et al., 2011) :
Pressure (Pa)
15 x 10
experimental Pressure simulated pressure
10
5
0 0
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 2. Experimental validation of cylinder pressure at 1500 rpm.
1
X fuelb
SOC Ccom . com 1 e
Figure 3 show the variation of mass of air in the cylinder at each phase of the engine in one cycle. The mass of air increase in the admission phase and then decrease in the combustion phase, the decreasing quantity is equal to 14.7 times quantity of fuel injected, the excess of air is rejected out of the cylinder in exhaust phase. Figure 4 and figure 5 shows respectively the variation of mass of burned air and fuel in the cylinder during the cycle. A comparison between
(18)
Where is a non-dimensional shape factor. A smaller produces a faster burning rate, com is the combustion duration, SOC is the crank angle at the start of combustion,
Ccom is a constant equal to 6.908.
223
IFAC CTS 2016 224 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
increasing of intake pressure for the same intake temperature increases the maximum pressure. This is due to the higher air quantity admitted.
-3
x 10
mass of air
0.8 0.6
6
15 x 10
0.4 0.2 0 0
100
200
300 400 500 Crank angle (degree)
600
Pressure (Pa)
mass of air (kg)
1
700
Fig. 3. Variation of mass of air in the cylinder at 1500 rpm.
Pressure
10
5
mass of burned gas (kg)
-4
6 x 10
0 0
mass of burned gas
4
3
3.5 -4 x 10
3. NEURAL NETWORK MODEL 100
200
300 400 500 Crank angle (degree)
600
700
Neural networks are non linear computer algorithms and can model the behavior of complicated non linear processes. They produce results very fast because of their property of working in parallel to solve a specific problem. The true power and advantage of ANN lies in their way to analyze data and to recognize patterns within that data (Rawlins, 2005). The basic processing element of a neural network is a neuron. Neural network is created by a means of interconnection of neurons and usually organized into layers. On the other hand, they have the ability to learn the relationship between the input and the output (T. Hari Prasad, 2010). In the present work, a multilayer feed-forward neural network with one hidden layer (Figure 9) is used and trained with the backpropagation algorithm in a supervised manner. Levenberg-Marquardt algorithm is used to optimize the performance function or the mean square error function.
-5
x 10
mass of fuel
2.5
mass of fuel (kg)
1.5 2 2.5 3 Volume (m )
2
Fig. 4. Variation of mass of burned gas in the cylinder at 1500 rpm.
2 1.5 1 0.5 0
340
360
380 400 420 Crank angle (degree)
440
460
Fig. 5. Variation of mass of fuel in the cylinder at 1500 rpm. 6
15 x 10
Pressure (Pa)
1
Fig. 8. P-V diagram at 1500 rpm.
0 0
Firing Pressure Motoring Pressure
10
5
3.1 Multilayer feed-forward networks
0 0
100
200
300 400 500 Crank angle (degree)
600
Multilayer networks are the simplest and most commonly used neural network architecture (Hagan et al., 2014). In our study the cylinder pressure is estimated by two layer feedforward network with sigmoid non linear transfer function as illustrated in Figure 10, the hidden layer is composed of ten neurons and one neuron in the output layer.
700
Fig. 6. Comparison between cylinder firing pressure and motoring pressure at 1500 rpm. Maximum pressure (Pa)
7
x 10 2
305
Intake Temperature (k)
0.5
1.8
300
1.6
295
1.4
290
1.2
285
1
280
0.8
275 1.8
2
2.2 2.4 Intake pressure (Pa)
2.6
2.8 5 x 10
3
Fig. 9. Architecture of neural network that estimate the cylinder pressure.
Fig. 7. Maximum gas pressure function of intake pressure and temperature for a fixed exhaust pressure of 2.5 bar at 1500 rpm.
Multilayer feed-forward networks have three distinct characteristics:
firing and motoring pressure is done in Figure 6. The results show the difference of pressure in the combustion phase. Regarding maximum cylinder pressure, we observe in Figure 7 that reducing the intake temperature for the same intake pressure increases the maximum cylinder pressure and the
1.
224
They have hidden units that include nonlinear activation functions that are smooth (or differentiable everywhere).
IFAC CTS 2016 May 18-20, 2016. Istanbul, Turkey
3.
They contain one or more layers of hidden units that are not part of the input or output of the network. They have a high degree of connectivity between the units.
1500
Temperature (k)
2.
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
The combination of these characteristics, together with the ability to learn from experience through training, add a significant computing power to multilayer feed-forward networks. It has been shown that a two-layer feed-forward network with sigmoid nonlinearity can approximate any function with arbitrary accuracy (Hagan et al., 2014).
225
T neural network T experimental
1000
500
0 0
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 12. Comparison of ANN result temperature with experimental result.
3.2 Back propagation algorithm 4. SIMULATION OF CYCLE WITH FAULT The training algorithm used in this work is back propagation algorithm, which is used to train the network in a supervised manner until it can approximate a function with high accuracy. The performance index of this algorithm is the mean square error. It can be described as: 1. Initialize of weights and biases arbitrary. 2. Propagate the input forward through the network. 3. Propagate the sensitivities backward through the network. 4. Update the weights and biases.
The main aim of this study is to simulate the thermodynamic cycle using neural network in normal and faulty condition. The first step was the simulation of internal pressure and temperature using neural network in normal conditions. The second step is to simulate the cylinder pressure and temperature using neural network in faulty conditions, that's why we re-trained the network to estimate the cylinder pressure and temperature in these conditions. 4.1 Cooling fault
3.3 Results and discussion
The cooling system is responsible of maintaining the temperature in the cylinder at an optimum operating temperature. If any fault occurs in this system, the temperature of cooling water entering the cylinder and the temperature of the metallic component of the engine increase (cylinder liner, intake and exhaust valves…), then the temperature of the gases flowing through the cylinder increases, and as a consequence the flow of air entering the cylinder decreases, then the pressure in the cylinder decreases and the efficiency of the engine decreases as shown in figure 13 and 14.
The aim of using the FFNN models in this work is to estimate the cylinder pressure and temperature for a diesel engine and to make the simulation time of one cycle equivalent to that of the real diesel engine. The model has two subsystems. The first estimates the cylinder temperature and the second estimates the cylinder pressure (Figure 10). Figure 11 and 12, shows a comparison between the cylinder pressure and temperature estimated by ANN with the experimental data show a good coincidence and the duration of one cycle is 0.5 second comparing to 40 second with the physical model.
Temperature (k)
2500
T without fault T with fault
2000 1500 1000 500 0 0
6
x 10
300 400 500 Crank angle (degree)
600
700
6
15
x 10
P neural network P experimental
15
Pressure (Pa)
Pressure (Pa)
200
Fig. 13. Comparison of the cylinder temperature with and without fault in cooling system at 1500 rpm.
Fig. 10. ANN model structure for pressure and temperature.
10
5
0 0
100
100
200
300 400 500 Crank angle (degree)
600
700
10
5
0 0
Fig. 11. Comparison of ANN result Pressure with experimental result.
P without fault P with fault
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 14. Comparison of the cylinder Pressure with and without fault in cooling system at 1500 rpm.
225
IFAC CTS 2016 226 May 18-20, 2016. Istanbul, Turkey
Ali Rida et al. / IFAC-PapersOnLine 49-3 (2016) 221–226
4.2. Injection time error
Dovifaaz, X., 2002. Neural modeling for diesel engine control, in: Basañez, L. (Ed.), . pp. 1523–1523. doi:10.3182/20020721-6-ES-1901.01525 Ghobadian, B., Rahimi, H., Nikbakht, A.M., Najafi, G., Yusaf, T.F., 2009. Diesel engine performance and exhaust emission analysis using waste cooking biodiesel fuel with an artificial neural network. Renew. Energy 34, 976–982. doi:10.1016/j.renene.2008.08.008 Hagan, M., Demuth, H., Beale, M., Jesus, O.D., 2014. Neural Network Design (2nd Edition). Martin Hagan. Heywood, J., 1988. Internal Combustion Engine Fundamentals. McGraw-Hill Education. JANAF THERMOCHEMICAL TABLES, 1989. . Anal. Chem. 61, 1327A–1327A. doi:10.1021/ac00198a726 Michael, F., 2013. Transient NOx Estimation Using Artificial Neural Networks, in: Taketoshi, K. (Ed.), . pp. 101– 106. doi:10.3182/20130904-4-JP-2042.00006 Nahim, H.M., Younes, R., Nohra, C., Ouladsine, M., 2015a. Complete modeling for systems of a marine diesel engine. J. Mar. Sci. Appl. 14, 93–104. doi:10.1007/s11804-015-1285-y Nahim, H.M., Younes, R., Shraim, H., Ouladsine, M., 2015b. Oriented review to potential simulator for faults modeling in diesel engine. J. Mar. Sci. Technol. 1– 19. doi:10.1007/s00773-015-0358-6 Omran, R., Younes, R., Champoussin, J.-C., 2008. Neural networks for real-time nonlinear control of a variable geometry turbocharged diesel engine. Int. J. Robust Nonlinear Control 18, 1209–1229. Payri, F., Olmeda, P., Martín, J., García, A., 2011. A complete 0D thermodynamic predictive model for direct injection diesel engines. Appl. Energy 88, 4632–4641. doi:10.1016/j.apenergy.2011.06.005 Rawlins, M.S., 2005. Diesel engine performance modelling using neural networks (Thesis). Stone, R., 2012. Introduction to Internal Combustion Engines. Palgrave Macmillan. T. Hari Prasad, D.K.H.C.R., 2010. Performance and Exhaust Emissions Analysis of a Diesel Engine Using Methyl Esters of Fish Oil with Artificial Neural Network Aid. Int. J. Eng. Technol. 2, 23–27. doi:10.7763/IJET.2010.V2.94 Verhelst, S., Sheppard, C.G.W., 2009. Multi-zone thermodynamic modelling of spark-ignition engine combustion – An overview. Energy Convers. Manag. 50, 1326–1335. doi:10.1016/j.enconman.2009.01.002 Woschni, G., 1967. A Universally Applicable Equation for the Instantaneous Heat Transfer Coefficient in the Internal Combustion Engine. SAE 670931. SAE Tech. Pap. doi:10.4271/670931 Xin, Q. (Ed.), 2013. Copyright, in: Diesel Engine System Design. Woodhead Publishing, p. iv.
When AI decrease, the quantity of fuel entering the cylinder decreases, so the maximum pressure and temperature in the cylinder decrease (Figure 15 and 16), and thus, the power delivered by the engine decrease. Temperature (k)
2000
T AI= 1 degree T AI= 3 degree T AI= 5 degree
1500 1000 500 0 0
100
200
300 400 500 Crank angle (degree)
600
700
Fig. 15. Comparison of the cylinder temperature for different injection timing at 1500 rpm. 7
x 10
P AI=1 degree P AI= 3 degree P AI= 5 degree
Pressure (Pa)
1.6 1.4 1.2 1 0.8 0.6 320
340
360 380 Crank angle (degree)
400
420
Fig. 16. Comparison of the cylinder Pressure for different injection timing at 1500 rpm.
5.
CONCLUSION
In the proposed thermodynamic model, the instantaneous gas states in the combustion chamber are obtained using the mass and energy conservation equations. A detailed description of the elements used for the calculation of each term in the energy balance has been provided. The main aim of this paper is to show the possibility of using artificial neural network for the prediction of a diesel engine performance of pressure and temperature in the cylinder in normal and faulty operation mode. A reduction of calculation time 80 times (from 40 seconds to 0.5 seconds) between the physical model and the NN model has been realized. This presented work is a part of a complete project for the realization of a diesel engine simulator, that simulates the variation of parameters of the engine in normal and faulty conditions. REFERENCES Awad, S., Varuvel, E.G., Loubar, K., Tazerout, M., 2013. Single zone combustion modeling of biodiesel from wastes in diesel engine. Fuel 106, 558–568. doi:10.1016/j.fuel.2012.11.051 Basbous, T., Younes, R., Ilinca, A., Perron, J., 2012. Pneumatic hybridization of a diesel engine using compressed air storage for wind-diesel energy generation. Energy 38, 264–275. doi:10.1016/j.energy.2011.12.003
226