Вы находитесь на странице: 1из 10

Applied Thermal Engineering 108 (2016) 660–669

Contents lists available at ScienceDirect

Applied Thermal Engineering


journal homepage: www.elsevier.com/locate/apthermeng

Research Paper

Volumetric efficiency optimization of a single-cylinder D.I. diesel


engine using differential evolution algorithm
Stephan Hennings Och a,⇑, Luís Mauro Moura a, Viviana Cocco Mariani a, Leandro dos Santos Coelho b,
José Antonio Velásquez c, Eric Domingues d
a
Mechanical Engineering Graduate Program, Pontifical Catholic University of Parana – PUCPR, Curitiba, PR, Brazil
b
Industrial and Systems Engineering Graduate Program, Pontifical Catholic University of Parana – PUCPR, Curitiba, PR, Brazil
c
Federal University of Technology of Parana – UTFPR, Curitiba, PR, Brazil
d
CORIA-UMR6614, Normandie Université, CNRS, Université et INSA de Rouen, 76800 Saint Etienne du Rouvray, France

h i g h l i g h t s g r a p h i c a l a b s t r a c t

 Three sets of optimization


calculations were conducted.
 It was found that optimal valve
timing produced a gain in volumetric
efficiency.
 The optimal length of the intake duct
diminishes as the engine speed rises.
 High volumetric efficiencies can be
attained by opening the exhaust valve
late.
 DE method was capable of finding the
global extremum in the defined
domain.

a r t i c l e i n f o a b s t r a c t

Article history: In this work, a mathematical optimization procedure was used to improve the gas exchange process of a
Received 31 December 2015 single-cylinder compression ignition naturally aspirated engine. Duct lengths and valve timing were cho-
Revised 4 July 2016 sen as optimization variables while volumetric efficiency was defined as the objective function.
Accepted 6 July 2016
Calculations were carried out using a parallelized computational code consisting of (i) a one-
Available online 6 July 2016
dimensional model for the unsteady compressible gas flow taking place in intake and exhaust ducts;
(ii) a single-zone combustion model for the in-cylinder processes; and (iii) an optimization routine based
Keywords:
on the Differential Evolution technique. Three sets of optimization calculations were conducted. In the
I.C. engine
Gas exchange process
first one, the intake duct length was the only optimization variable and it was found that optimal inlet
Differential evolution optimization method duct lengths vary becoming shorter as engine speed is increased. In the second set of calculations, both
Single zone combustion model intake and exhaust duct lengths have been taken as the optimization variables, and the resulting optimal
Volumetric efficiency intake duct lengths were quite similar to those of the first set. In addition, optimal exhaust duct lengths
resulted very close in value to optimal intake duct lengths, except at the highest speeds, when the
decreasing tendency as engine speed is raised was supplanted by the opposite tendency. In the third
set of calculations, the crank angles defining valve synchronism were the optimization variables. It
was found that optimal valve timing produced a gain in volumetric efficiency, which is similar to that
obtained with optimal duct lengths.
Ó 2016 Elsevier Ltd. All rights reserved.

⇑ Corresponding author.
E-mail addresses: stephan.och@pucpr.br (S.H. Och), luis.moura@pucpr.br (L.M. Moura), viviana.mariani@pucpr.br (V.C. Mariani), leandro.coelho@pucpr.br (L.S. Coelho),
velasquez@utfpr.edu.br (J.A. Velásquez), eric.domingues@coria.fr (E. Domingues).

http://dx.doi.org/10.1016/j.applthermaleng.2016.07.042
1359-4311/Ó 2016 Elsevier Ltd. All rights reserved.
S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669 661

1. Introduction exhaust emission. Costa et al. [12] reported experimental data


showing that long intake ducts with small diameters perform well
IC engines are widely used in transportation as well as in sta- at low engine speeds, while short ducts with large diameter do so
tionary applications, being responsible for an important part of at high speeds.
global fuel consumption and air pollution. This explains the inter- Articles dealing with the mathematical optimization of gas
est in research activities aimed to improve their operation [1–4]. exchange processes are still scarce. D’Errico et al. [13] applied
The purpose of the gas exchange processes taking place in IC engi- the MADS method and Genetic Algorithm for multi-objective opti-
nes is to remove burned gases from the cylinder and refill it with a mization techniques in order to find optimal duct lengths and
fresh charge for the next cycle. Efficient gas exchange processes valve timing to minimize the NOx emission and the specific fuel
would allow for the maximum amount of fresh charge to be consumption. The authors also optimize the volumetric efficiency
trapped within the cylinder with a minimum of pumping work as a function of the ducts length through a rectangular valve lift
[5]. Therefore, the effectiveness of such processes is influenced model, and friction and heat transfer are neglected. However, the
by the inertial and pulsating nature of the flow in the intake and lack of input data about the engine geometry and engine operating
exhaust ducts [2], as well as by the friction and heat transfer occur- conditions prevent the reproduction of the results they achieved.
ring between the flowing gas and the duct walls. Sher and Bar-Kohany [14] used MICE (Modeling Internal Combus-
At the end of the intake process inertial effects may cause addi- tion Engine) and experimental data of an unthrottled SI engine
tional gas inflow into the cylinder, even though the cylinder vol- with variable valve timing, thus obtaining an increase of 6% in
ume is being reduced by the piston movement. In order to take power and a reduction of 13% in brake specific fuel consumption.
advantage of this effect, inlet valves are usually closed down This study presents a graphical strategy to minimize the fuel con-
around 40–60 c.a. deg after the piston passes bottom dead center sumption and maximize the torque of engine in terms of intake
(BDC) [6]. In order to measure the effectiveness of the intake pro- valve duration, and opening and closing timings of the exhaust
cess, the volumetric efficiency is frequently used, which is defined valve. Zhao and Xu [15] used the GT-Power commercial software
as the ratio of the trapped mass of fresh charge to the mass of air coupled with a genetic algorithm code, in order to optimize the
that would occupy the volume displaced by the piston, under the gas exchange processes for the Atkinson cycle engine. The project
same pressure and temperature conditions found at the entrance variables are throttle valve, intake valve closure, exhaust valve clo-
of the intake duct. sure, spark angle and air-fuel-ratio, and the objective function was
Studies aiming to improve the gas exchange processes in IC the brake specific fuel consumption. Thus achieving a reduction of
engines involve the analysis of several variables characterizing 8% in engine fuel consumption.
the geometry of the ducts; the valves timing; and the displacement The goal of the present work is to find the maximum volumetric
laws of the valves. In these studies, the intake and exhaust systems efficiency of a single-cylinder naturally aspirated CI engine at dif-
design is modified to achieve specific goals, such as fuel consump- ferent engine speeds, which could be reached by varying intake
tion reduction or volumetric efficiency and torque increase [7]. and exhaust duct lengths and valve timing. Additional goals are
Traditionally, experimental tests on a physical engine have been to investigate the physics behaviors while the volumetric effi-
used for these studies, but such approach is usually expensive and ciency is maximum at these points, and extract a general guideline
demands an advanced stage in the development of the engine pro- for gas exchange optimization for comparison with other optimiza-
totype [5]. Because of this, computer simulation is recognized as an tion methods. The analysis was done considering a parabolic valve
alternative. Nevertheless, despite the power and speed of compu- lift curve, heat transfer and viscous friction in duct walls. For this
tational hardware having continuously risen during last years, purpose, a zero-dimensional model of the processes taking place
detailed multidimensional simulation of engine flows still inside the cylinder and a one-dimensional model of the induction
demands high computational time, thus preventing its use in rou- and exhaust flows is used, together with a stochastic optimization
tinely optimization studies. Being simpler than multidimensional technique called Differential Evolution.
models, one-dimensional gas dynamic models drastically reduce The remainder of this paper is organized as follows: Section 2
the simulation time, and usually do not require high- describes the mathematical model used to represent the gas flow
performance computers [8,9]. These models allow to simulate occurring in the intake and exhaust ducts (Section 2.1), as well
accurately the propagation of pressure waves along the ducts as the in-cylinder processes (Section 2.2). In Section 3, the funda-
and are appropriate for studies involving mathematical optimiza- mentals of the differential evolution optimization technique are
tion, in which hundreds of alternative designs are usually assessed. explained. Results of the optimization calculations are presented
Technical literature reports a few experimental and theoretical and discussed in Section 4, while conclusions are given in
studies dealing with the improvement of the engine gas exchange Section 5.
process. In some studies, this improvement is achieved without the
utilization of a formal optimization. Thus, they do not access the 2. Mathematical model
potential that the variation on optimization of each parameter
can offer. 2.1. Gas dynamics of duct flow
Kesgin [10] used a computational model to perform a
theoretical study on the design of the exhaust system of a station- The gas flow in induction and exhaust ducts of the engine is
ary natural gas engine. He analyzed the effects on engine perfor- considered compressible, one-dimensional, non-stationary, and
mance caused by parameters such as length and diameter of the non-isentropic [8,14]. Besides that, variations of the cross
duct system, valve timing and valve lift profile; and made a proper sectional area along the ducts are allowed; and friction and heat
sizing of the inlet and exhaust pipe systems as well as the inlet and transfer between the flowing gas and duct walls are taken into
exhaust valves. Ceviz and Akin [11] carried out an experimental account [9]. Thus, mass, momentum and energy conservation
study on the effects of intake duct length in a multi-cylinder spark equations can be written for the control volume shown in Fig. 1
ignition engine. These authors proposed a new intake system with as follows
variable length and concluded that variations in ducts length do
not affect directly engine emissions, but influence fuel consump- @U @F
þ ¼f ð1Þ
tion, thus causing variations in the total amount of produced @t @z
662 S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669

other. These eigenvalues are w, w + c and w  c, where c is the


speed of sound. Hyperbolic problems derived from conservation
laws are convective dominant, and their solution is quite sensitive
to stability and accuracy [16].
The Lax-Wendroff two-step scheme, also known as Richtmyer
scheme, is a second-order accurate procedure widely used in solv-
ing such problems. In this scheme, an intermediate step is accom-
plished following the Lax-Friedrich method (Eq. (3)), which is
explicit and first-order accurate. Afterwards, the second-order accu-
racy is retrieved using the leap-frog method, given by Eq. (4) [17,18].
U niþ1 þ U ni Dt  n  Dt  n n
U nþ1=2
iþ1=2 ¼  F  F ni þ f þ fi ð3Þ
2 2Dz iþ1 4 iþ1
Fig. 1. Control volume for duct flow. Dt  nþ12 nþ1
 Dt 
nþ1=2 nþ1=2

U nþ1 ¼ U ni  F iþ1  F i12 þ f iþ1=2 þ f i1=2 ð4Þ
i
Dz 2 2 2
where
The time step should be chosen to ensure stability of the solu-
2 3 tion by satisfying the Courant-Friedrich-Lewis condition (Eq. (5)).
q
6 qw 7 Dz
U¼4 5;
Dt 6 ð5Þ
P
þ qw2 ðjwj þ cÞmax
2 k1
3 2
qw Additionally, boundary conditions are applied using the
6 qw2 þ P 7
F¼6
4 
7
 5;
methods of characteristics [19] by the procedure presented by
2 ð2Þ El-Rahman et al. [20] and Broatch et al. [21]. In the present work,
kP
w k1 þ qw2
2 3 the boundary conditions comprise equations for open end, closed
qw dA
A dz
end as well as for flow through valves, as we can see in Table 1.
6 7 Where ce is the characteristic variable passing by the end node
6 qw2 dA
 qF f 7
f ¼6 7 of the duct, cA is the entropy level, w is the areas valve and duct
4  A dz
2
 5
w k1
kP
þ qw2 1 dA
A dz
þ qq_ ratio. The indices ‘‘ext” is external of duct, ‘‘n” is no-corrected value,
and ‘‘ref” is referential condition. The parabolic lift model calcu-
lates the lift curves from valves:
Here q is the local gas density; w is the local gas velocity; A is
8  2
the duct cross-sectional area; P is the local gas pressure; k is the >
> h
0 6 h 6 hnv
> 2nymax hv
>
specific heats ratio (considered constant); F f and q_ are, >
>  
< 2
respectively, friction force and heat transfer, both, per unit mass y ¼ r2n ymax hhv  12 þ r2n
ac hv
6 h 6 n1 hv ð6Þ
of flowing gas. >
> ac n n
>
>  
Eq. (1) can be written in the form @U=@t þ B  @U=@z ¼ f in order >
> 2
: 2ny
max 1  hv hv 6 h 6 hv
h n1
to show that it defines a hyperbolic equation system, as the eigen- n

values of the coefficient matrix B are real and different from each

Table 1
Boundary conditions.

Boundary condition Subsonic condition


  4k    2 
Outlet from duct to the valve cA;ext c k1
ðce  cÞ2 cA; cext  w2  k1 2
2 w c  cA;ext c ext
2 cA
¼0
  4k    2 
Outlet from duct to outside cA;ext c cA;
ðce  cÞ2
k1
cA cext  1  k1 2 c2  cA;ext cext ¼0
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   
Inlet from outside to duct  2  2
cA;n cA;n
c2e þk1
2 k1þ
2 cA;ext cA;ext c2ext c2ext ce;n

jwj ¼   2 
cA;n
k1
2
þ cA;ext

Inlet from valve to duct  k1  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi


12
Pext 2k 2 2
2K w w þ 4K  w  ce;n  k1
2 jwj ¼ 0
1
cA;n Pref
 2
k1 jwj
2 cext
K¼  2 2
jwj
1k1
2 cext

Valve close w¼0


Sonic condition
  2  4
Outlet from duct to valve c cA;ext c cA;ext k1
w2  kþ1k1
 k1
2
cext cA cext cA ¼0
  2  4
Outlet from duct to outside c cA;ext c cA;ext k1
1  kþ1
k1
 k1
2
cext cA cext cA ¼0
qffiffiffiffiffiffiffi
Inlet from outside to duct jwj ¼ cext kþ1 2

Inlet from valve to duct 2  2 3k1


2k
 k1 1k1 jwj   kþ1
Pext 2k 6 2 cext 2ðk1Þ 7
cA;n Pref 4w jwj
2
kþ1 5  ce;n  k1
2 jwj ¼ 0
cext

Valve close w¼0


S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669 663

Table 2 combustion reaction is represented by two Wiebe functions: one


Geometrics valve parameters. for the premixed combustion and another for the diffusive com-
Intake valve Exhaust valve bustion phase. The Wiebe function is given by [32]:
"
nþ1 #
ymax (mm) 9.0 10.0
h  hig
rac 5.0 5.0 v ¼ 1  exp 6:908 ð11Þ
Number of valves 2 1 Dh d
Valve bore (mm) 27.5 36.0
Seat bore (mm) 28.9 38.6 The volumetric efficiency (gv) is calculated as follows
Outside seat bore (mm) 31.4 38.6  
Stem bore (mm) 7.0 7.0 mc 1  /c =/cycle
gv ¼ ð12Þ
Seat angle (degrees) 30.0 45.0 qo V disp 1 þ /c ðF=AÞst
where mc is the mass of the in-cylinder gases at the end of the
where hv is the valve opening time, ymax is the high lift valve, rac is
intake process; qo is the density of the atmospheric air; Vdisp is
the acceleration ratio and n is calculated by 2–2rac. To this analysis,
the volume displaced by the piston; /c is the equivalence ratio of
it has chosen the following valves geometrics parameters (Table 2).
the in-cylinder gases at the end of the intake process; /cycle is the
equivalence ratio of the in-cylinder gases at the beginning of the
2.2. Model for in-cylinder processes
exhaust process and (F/A)st is the stoichiometric fuel-air mass ratio.

In this work, the processes taking place in the engine cylinder


were represented by a zero-dimensional [22] and a single-zone 3. Differential evolution method
combustion model [23]. According to this model, the fuel burns
instantaneously when entering the combustion chamber and the Optimization methods can be classified into deterministic and
working fluid is considered a homogeneous mixture of combustion stochastic. In deterministic methods, every obtained solution is
gases in chemical equilibrium. Therefore, phenomena such as further improved by searching for a new solution in the direction
atomization and vaporization of the fuel, ignition delay, as well of the gradient, and the procedure ends when the gradient is near
as pressure and temperature gradients are neglected. Besides that, zero. So, these methods demand the derivatives to be calculated
the combustion reaction is admitted to be given by [24,25] and, unfortunately, they may converge to a local optimum.
" # Stochastic methods are suitable for cases where the objective
n þ m4  2l function is discontinuous, non-differentiable, multimodal and diffi-
x13 C n Hm Ol Nk þ ðO2 þ aN2 þ bCO2 þ cH2 O þ dArÞ cult to represent. The evolutionary algorithms (EA) are stochastic
/
optimization algorithms, inspired by Darwinian natural evolution
! x1 H þ x2 O þ x3 N þ x4 H2 þ x5 OH þ x6 CO þ x7 NO þ x8 O2 [33]. According to the Darwinian Theory, only the best adapted
þ x9 H2 O þ x10 CO2 þ x11 N2 þ x12 Ar ð6Þ individuals survive natural selection and reproduce from one gen-
eration to another. One optimization term, the evolution, is
Using the mass and energy balance equations and the ideal gas reflected by an interactive process of searching for an optimal
state equation [26], the following expressions can be obtained: point in the design variables domain. The evolutionary optimiza-
dm X dmi tion process begins with the initialization step: an individual’s
¼ ð7Þ finite amount, xi, generally randomly chosen within the field of
dh dh
design variables, forms the initial population, P0. Then, the varia-


1 tion operator (mutation and crossover) is applied in order to create
dP 1 1 @R dT 1 dm 1 dV 1 @R d/ P @R
¼P þ þ  þ 1 ð8Þ a new set of individuals, called the children population (this step is
dh T R @T dh m dh V dh R @/ dh R @P
entirely stochastic). These children will be evaluated and combined


with their parents in order to decide which ones will replace cer-
dT dV A P d/ @u A @R
¼   þ tain parents and will be part of the next generation (selection
dh dh V m dh @/ R @/
 
1 stage). An important fact to consider in most cases of real-world
1 dQ X dmi A @u A @R applications is that the computational cost of evolutionary algo-
þ þ ðhi  u  AÞ þ þ
m dh dh T @T R @T rithms depends essentially on the evaluation of objective function.
ð9Þ For example, in order to evolve ten individuals for ten generations,
thousands of calculations of the objective function must be made
where at a usually high computational time.
P @u Storn and Price [34], proposed the method of Differential Evolu-
@P
A¼ ð10Þ tion, a simple yet efficient algorithm for real parameter optimiza-
1  RP @R
@P
tion over continuous spaces. It creates new candidate solutions
In Eqs. (7)–(10), P, V, T, m, u, h, h, R, /, and Q are pressure, vol- by combining the parent individual and several other individuals
ume, temperature, mass, internal energy, enthalpy, crank angle, of the same population. However, the differential evolution algo-
gas constant, fuel/air equivalence ratio and heat transfer, respec- rithm significantly differs from other evolutionary algorithms in
tively. The subscript i refers to mass flows through inlet and the information about the distance and direction of the current
exhaust valves or through the fuel injector. The rate of heat trans- population, used to guide the search process.
fer is obtained by the correlation proposed by Woschni [27,28], The method is divided into three steps: mutation, crossover and
which is one of the most used [29]. The equation system above is selection. For each target vector xi,G with i = 1, 2 . . .NP (NP is the
solved by the fourth-order Runge-Kutta method. The mass flow number of individuals), the vector mutation is created by the fol-
through valves is obtained from the model for flow in intake and lowing equation:
exhaust ducts and the effects of turbulence are taken into account
v i;Gþ1 ¼ xr ;G þ Fðxr ;G  xr ;G Þ
1 2 3
ð13Þ
through discharge coefficients, which were experimentally
obtained [30,31]. The step in crankshaft angle is the one In this equation, the random indices r1, r2 and r3 are chosen
corresponding to the same time step calculated in the model for using uniform distribution from the range 1, 2 . . .NP in such a
flow in intake and exhaust ducts. The heat released during the way that they must be distinct from each other, as well as different
664 S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669

Fig. 2. Schema of differential evolution method.

from the current index i. Therefore, NP should be greater than or


equal to four. The factor F is a positive real value between 0 and
2 that controls the amplification of (xr2,G  xr3,G). A schema of this
process can be visualized in Fig. 2. Fig. 3. Pressure measurement points for validation of the computational model.
The purpose of the crossing step is to increase the diversity of
the population, disrupting the vector mutation. Thus, the experi-
ment vector uji,G+1 is calculated by: 20
Experimental
v ji;Gþ1 ifðrandb ðjÞ 6 CRÞ j ¼ rnbrðiÞ; Engine running Measurement uncertainty
uji;Gþ1 ¼ motored at 2000 rpm
xji ifðrandb ðjÞ > CRÞ j–rnbrðiÞ ð14Þ Numerical model

j ¼ 1; 2; . . . ; D 15
Cylinder Pressure (bar)

In the above equation, D is the number of optimization 1.1


variables, randb(j) is the jth term in a uniform random number
1.0 1.1
generator between 0 and 1. CR is the crossover probability chosen 10
by the user between 0 and 1. randb(i) is used to randomly select 0.9
the index from the range 1, 2, . . .D, this ensures that uji,G+1 will 0.8 0.9
receive at least one parameter vi,G+1. The selection consists in
deciding which individual will become a member of the generation 5
359 239 119 120 240 360
G + 1. The experiment vector ui,G+1 is compared with the target ivo ivc
vector xi,G using the criterion of ambitiousness: if the vector uji, evc evo
G+1 produces a function value less costly than xi,G, then xi,G is equal
0
to uji,G+1, otherwise, the value is kept equal to xi,G. There are several 360 300 240 180 120 60 0 60 120 180 240 300 360
strategies for crossover; in the present work, we chose to use the so Crankshaft angle (deg. aTDC)
called DE/rand/1/bin [31], which is given by Eq. (13).
Choosing suitable control parameter values in DE approaches is, Fig. 4. Validation of the computational model: in-cylinder pressure.

frequently, a problem-dependent task. Suitable control parameters


are different for different function problems. The difficulty in the 4. Results and discussions
use of DE arises in that the choice of these is mainly based on
empirical evidence and practical experience. The F is a scaling fac- Prior to the optimization calculations, the computational
tor, which controls the length of the exploration vector. On the program based on the model described in Sections 2.1 and 2.2
other hand, the CR practically controls the diversity of the was validated by comparing numerical results with experimental
population. data measured in a AVL 5482 single-cylinder engine, which was
Additionally, the multiplier factor (mutation factor) F was running motored. Straight intake and exhaust ducts were built
assumed equal to 0.8 and the crossover probability CR equal to and mounted on this engine, then, pressure measurement were
0.5. Every optimization calculation was repeated for 20 trials (20 made using two sensors—an AVL GU22C piezoelectric sensor and
optimization experiments), each one made up of a succession of a Kistler 4007 piezoresistive sensor, the first ones is installed in
100 generations. the combustion chamber and the latter in the intake duct, as
Many of the most recent developments in differential evolution shown in Fig. 3. It can be seen in Figs. 4 and 5 that the computa-
algorithm design and applications can be found in Price et al. [35] tional model allowed to reproduce the experimental data with
and Das and Suganthan [36]. high accuracy, as the maximum difference between calculated
It is worth mentioning that the optimization procedure based and measured data was 172 mbar for in-cylinder pressure and
on the differential evolution technique demands the objective 5 mbar for pressure in intake duct. The experimental results shown
function to be evaluated several hundreds of times. Therefore, it in Figs. 4 and 5, is the average of 100 consecutive cycles, the blue
requires an intensive use of computational resources during this dotted lines represents the measurement uncertainty.
stage, as well as an extended computational time. By this reason, Once the computational code was validated, we proceeded to
the algorithm used in this work was parallelized using the MPI analyze the DI naturally aspirated diesel engine, whose main data
library [37] in order to take advantage of running the calculation are shown in Table 3. The values for the lengths of intake and
on multiple processors. So, it was possible to simultaneously exhaust ducts that appear in this table are taken as a reference
evaluate the objective function of different individuals of the pop- for comparison with the optimal values obtained later.
ulation and, therefore, the computational time was reduced Fig. 6 shows the calculated volumetric efficiency as a function of
considerably. the intake duct length for several engine speeds. It is possible to
S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669 665

1.10 variation range of the volumetric efficiency, resulting from changes


Experimental in intake duct length, becomes wider as the engine speed increases,
Measurement uncertainty
Numerical model and that the length of the intake duct that maximizes volumetric
Gas pressure inside the duct (bar)

1.05
efficiency becomes shorter when engine speed increases. The vol-
umetric efficiency values for the based case are 0.7665, 0.8130,
1.00 0.8142, 0.8375 and 0.8777 for the follow engine speeds: 1000,
2000, 3000, 4000 rpm, respectively.
Table 4 shows final results of the optimization calculation using
0.95 only one design variable — length of the intake duct. Reported val-
ues in Table 4 are given with a confidence interval of 95%, based on
0.90 a population of 20 optimization experiments for each engine
speed. Taking into account the values in Table 4, as well as the
intermediate values of volumetric efficiency calculated during
0.85 Engine running the optimization procedure (shown in Fig. 6) it is possible to con-
ivo ivc motored at 2000 rpm
clude that the Differential Evolution method was able to converge
360 300 240 180 120 60 0 60 120 180 240 300 360 towards the global optimum of the analyzed domain. It is worth
Crankshaft angle (degrees) noting that the optimal volumetric efficiency increases as the
engine speed rises up to 3500 rpm, and then diminishes as speed
Fig. 5. Validation of the computational model: pressure in the intake duct.
keeps rising. So, the optimal volumetric efficiency at 4000 rpm
was lower than the corresponding value at 3500 rpm.
Table 3 In-cylinder pressure and pressure at the inlet valve port are
Engine main data. shown in Fig. 7 as functions of crank angle. The baseline case
Engine type Compression ignition (Ladm = 0.38 m) and the optimized case (values from Table 4) are
Compression ratio 16:1
reported for 1000 rpm (Fig. 7a) and for 3000 rpm (Fig. 7b). The
Bore 93 mm mass flow rate of fresh charge into the cylinder varies continuously
Displacement 170 mm during the intake process, being influenced by the area, the density
Stroke 103 mm and the velocity of the fluid in the throat section of the inlet valve.
Injection Direct
When the optimization is performed by varying the lengths of the
Air metering Naturally aspirated
ducts, the instantaneous cross sectional area is the same for all
Intake system
considered cases (baseline case, optimized case and others) Thus,
Number of intake valves 2
Intake duct length 380 mm the optimal conditions should be searched by changing (indirectly)
Diameter of intake duct 48 mm the speed and the density of the fluid in the throat section. At the
Intake valve open (IVO) 10° bTDC beginning and the end of the intake process, when the throat area
Intake valve closed (IVC) 37° aTDC
is small, there may be significant differences between pressure
Exhaust system upstream of the valve and in-cylinder pressure. This difference
Number of exhaust valves 1
affects the fluid velocity in the throat section in such a way that
Exhaust duct length 1500 mm
Diameter of exhaust duct 50 mm
higher pressures upstream of the valve allow increasing the fluid
Exhaust valve open (EVO) 42° bBDC velocity at the throat section (as long as the pressure ratio is less
Exhaust valve closed (EVC) 10° aTDC than the sonic critical limit). Moreover, higher upstream valve
pressures also allow increasing the fluid density in this cross sec-
tion. Thus, an increase in the pressure upstream of the valve will
influence simultaneously the velocity and the density of the fluid
0.90
1000 rpm in the valve throat. Besides that, in the middle of the intake pro-
2000 rpm cess, when the valve throat sectional area is large, the in-cylinder
3000 rpm pressure closely follows the pressure upstream of the valve, as
0.85
4000 rpm can be observed in Fig. 7a and b. This is due to the proximity of
Volumetriceffciency

these two sections, as well as to the absence of a restriction


0.80 between them. Therefore, in the middle of the intake process it is
not feasible to increase the mass flow rate by acting on the pres-
sure difference across the valve. However, it is still possible to
0.75 increase the fluid density, which, as well known, is directly propor-
tional to pressure and inversely proportional to temperature.
During the optimization calculation, the intake duct length was
0.70 used as an optimization variable and by changing this parameter,
the pressure waves that propagate along the duct are modified in
both, frequency and amplitude. The longer the duct, the lower is
0.5 1.0 1.5 2.0 2.5 3.0 the frequency and the greater is the amplitude of the wave. This
Intake duct length (m) effect can be clearly noticed in Fig. 7a, where the pressure
upstream of the intake valve is shown for the baseline as well as
Fig. 6. Volumetric efficiency as a function of admission duct length. for the optimized case, which correspond to intake ducts of quite
different lengths (0.38 and 2.84 m, respectively). On the contrary,
this effect does not appear in Fig. 7b, as the ducts lengths in the
notice that volumetric efficiency exhibits local maxima and min- compared cases are close each to other.
ima, and this behavior could be a problem if optimization methods The duct length was optimized for each engine speed, and it
based on derivatives were used, as the iterative process could stop was found that the optimum values are those that make the wave
at these points. Additionally, it can be observed in Fig. 6 that the period to coincide with the duration of the intake process (as can
666 S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669

Table 4
Optimal values of intake duct length (valve timing and exhaust duct length were held unchanged as in baseline data).

Speed (rpm) Lint (m) Lexh (m) avo (deg bTDC) avc (deg aBDC) evo (deg bBDC) evc (deg aTDC) gv
1000 2.845 ± 0.009 1.5 10 37 42 10 0.82722 ± 0.00010
1500 1.8878 ± 0.0015 0.85171 ± 0.00005
2000 1.4023 ± 0.0011 0.8709 ± 0.0005
2500 1.1108 ± 0.0002 0.8722 ± 0.0004
3000 0.919 ± 0.004 0.89040 ± 0.00001
3500 0.7710 ± 0.0004 0.89249 ± 0.00003
4000 0.6600 ± 0.0001 0.87849 ± 0.00001

Fig. 7. Inlet port pressure as function of the crank angle (the angle domain on the x- Fig. 8. Mass flow rate as a function of the crank angle (the angle domain on the x-
axis is between IVO and IVC). axis is between IVO and IVC).

be seen in Fig. 7a and b). Thus, in the optimal cases a full period of is worth taking into account that along the wave the thermody-
this wave reaches the intake valve while it is open. Furthermore, namic properties of the fluid vary following closely an isentropic
this wave is synchronized with the intake process in such a way process: q / P1/k and q / T1/(k1). The fact that the temperature
that two crests coincide with the beginning and with the end of exponent is higher than that of the pressure shows that the density
the process, while a valley is positioned in the middle of the is more sensitive to temperature. Thus, the coldest regions of the
process. wave correspond to these of greater density.
This result meets the desired behavior as described in the pre- Air mass flow rate through intake valve is shown in Fig. 8 for
ceding paragraphs, since the presence of the crests at the beginning 1000 rpm (left side) as well as for 3000 rpm (right side) in the case
and at the end of the process increases the velocity and the density of optimization of the lengths of inlet and exhaust ducts. Notice
of the fluid, thus increasing the mass flow rate into the cylinder. that in the optimized case, the mass flow enters the cylinder from
Furthermore, the wave valley in the middle of the intake process the beginning of the intake process, while in the baseline case,
also increases the fluid density, as in this region occur the lowest some outflow precedes the inflow of air into the cylinder. The opti-
temperatures in the wave. In order to support this observation it mized case exhibits higher mean mass flow rate than the baseline
S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669 667

Table 5
Optimal values of intake and exhaust duct lengths (valve timing was held unchanged as in baseline data).

Speed (rpm) Lint (m) Lexh (m) avo (deg bTDC) avc (deg aBDC) evo (deg bBDC) evc (deg aTDC) gv
1000 2.8608 ± 0.0029 2.840 ± 0.019 10 37 42 10 0.83220 ± 0.00015
1500 1.8889 ± 0.0014 1.792 ± 0.015 0.85473 ± 0.00012
2000 1.4024 ± 0.0015 1.253 ± 0.007 0.87262 ± 0.00002
2500 1.1111 ± 0.0007 0.966 ± 0.004 0.88511 ± 0.00001
3000 0.9196 ± 0.0026 0.759 ± 0.012 0.89170 ± 0.00002
3500 0.7722 ± 0.0003 0.598 ± 0.004 0.89263 ± 0.00001
4000 0.6611 ± 0.0006 1.1189 ± 0.0024 0.88946 ± 0.00001

case in most of the intake process, and reduction of backflow in the


intake valve, leading thus, to a higher volumetric efficiency.
The difference between the optimized volumetric efficiency and
the baseline values was 8% at 1000 rpm and 9% at 3000 rpm. It is
worth mentioning that at the end of the intake process, mass
outflow through the inlet valve occurs in both, the optimized and
the baseline cases, however, it is smaller in the former case. At
low speed, the ram effect is less significant because the gas velocity
in the duct is lower. Then the goal is reducing the backflow in the
intake valve. For high speed, we have high gas velocity that charac-
terizes the ram effect. The optimization improves all these effects,
which are combined in the model that is hard to discern. In
Fig. 8a and b, we see a reduction in backflow in both rotations.
Observing Fig. 7a and b, the pressure amplitude is higher for the
optimized case due to the ram effect (inertia is transformed in
pressure). It can be observed that the ram effect is more significant
at high than low speed.
Table 5 shows results of the optimization calculation when two
design variables are considered—length of the intake duct (Ladm)
and length of the exhaust duct (Lexh). It can be observed in Table 3 Fig. 9. Final EVO value resulting from each optimization experiment (for 1000 rpm
that the largest 95% confidence interval of the optimized engine speed).
volumetric efficiency values was as small as 0.00015, thus evidenc-
ing that at each engine speed all the 20 optimization experiments
converged towards a single point, which is a global optimum. stand for intake valve opening and closing, respectively, while
Comparing the values reported in Table 4 with those in Table 3, EVO and EVC stand for exhaust valve opening and closing. It is
it can be seen that a further increase in volumetric efficiency was worth mentioning that this part of the optimization study began
obtained, with the highest gain being of 1.5% at 2500 rpm. Note analyzing the case of 1000 rpm engine speed, and that the series
also that the optimal intake duct lengths in Table 5 are quite close of twenty optimization experiments resulted in very scattered val-
to those in Table 4, which allows us to conclude that the volumet- ues for EVO and EVC; with EVO occurring either before or after the
ric efficiency is more sensitive to the intake duct length than to the piston has passed BDC. However, after plotting the obtained values
exhaust duct length. of EVO as shown in Fig. 9, it was observed that at this engine speed,
This is because the exhaust duct length only affects the exhaust the optimization experiments did not converge towards a single
gas process. Thus, its influence on the cylinder filling is indirect, as value of EVO, but two candidate ranges for optimal EVO can be
the greater part of this process occurs when the exhaust valve is identified. In one of these ranges, exhaust valve opening occurs
closed. very late, after piston has passed BDC, so that a high volumetric
Additionally, at engine speeds lesser than 3500 rpm, optimal efficiency is obtained by spending a large amount of pumping
exhaust duct lengths resulted close to optimal intake duct lengths, work, thus reducing the engine thermal efficiency, as shown in
following the same tendency to diminish as engine speed Fig. 10. In contrast, the second candidate range for optimal EVO
increases. However, this behavior changes as engine speed rises corresponds to valve opening occurring before piston passes BDC,
above 3500 rpm, thus an increase of the optimal exhaust duct thus without a penalty on engine thermal efficiency. Based on such
length was observed as engine speed passed from 3500 rpm to a result, subsequent optimization calculations were conducted
4000 rpm. imposing the additional restriction that EVO must occur before
Table 6 shows the results obtained when the intake and exhaust BDC.
ducts lengths are kept with their baseline values and the optimiza- Results reported in Table 6 show that in order to maximize vol-
tion is conducted varying the crank angles at which intake and umetric efficiency IVO and EVC must happen, respectively, earlier
exhaust valves are opened or closed. In this table, IVO and IVC and later than they did in the baseline case; while optimal IVC

Table 6
Optimal valve timing (duct lengths are held unchanged as in baseline data).

Speed (rpm) Lint (m) Lexh (m) ivo (deg bTDC) ivc (deg aBDC) evo (deg bBDC) evc (deg aTDC) gv gt
1000 0.38 1.50 49 ± 3 32 ± 6 47.9 ± 1.8 49.4 ± 1.6 0.8495 ± 0.0008 0.346
2000 46.9 ± 1.1 30.6 ± 1.2 77.0 ± 1.6 54 ± 1 0.8518 ± 0.0001 0.376
3000 55.7 ± 1.7 43.2 ± 1.6 8.2 ± 2.7 78.6 ± 2.7 0.8962 ± 0.0001 0.363
4000 31.1 ± 0.5 30.1 ± 0.4 54.4 ± 1.3 75.9 ± 1.9 0.8808 ± 0.0001 0.372
668 S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669

clusion already mentioned, that the influence of parameters, which


characterize the exhaust process, is secondary on the parameters of
the intake process.

5. Conclusion

In this work, we presented a numerical approach to increase the


volumetric efficiency of a single cylinder DI diesel engine. The
obtained results allow us to draw the following conclusions:

 The computational model used to simulate the intake and


exhaust processes occurring in the engine is able to represent
adequately these processes;
 The results of the optimization proved that the Differential Evo-
lution method was capable of finding the global extremum in
the defined domain;
 The obtained 95% confidence intervals are sufficiently narrow as
Fig. 10. Volumetric efficiency and corresponding engine thermal efficiency result- to consider that the number of optimization experiments, as
ing from each optimization experiment (for 1000 rpm engine speed). well as the number of generations used in each of them, were
adequate for this study;
 The optimal length of the intake duct diminishes as the engine
speed rises. This observation is completely in accordance with
other results found in literature;
 The optimal intake duct length makes the period of the pressure
wave in the intake duct to be a little bit shorter than the time
interval during which the inlet valve remains open, thus making
possible that two crests of pressure wave reach the intake valve
while it is open;
 The volumetric efficiency is more sensitive to the intake duct
length than to the exhaust duct length;
 When using parameters defining valve timing as design vari-
ables, high volumetric efficiencies can be attained at some
engine speeds by opening the exhaust valve very late, after pis-
ton has passed BDC. However, by doing so a large amount of
pumping work is spent and a reduction the engine thermal effi-
ciency can be observed;
 Volumetric efficiency values attained with optimal valve timing
are fairly close to those attained with optimal duct lengths.

In future work, new methods of optimization can be compared


Fig. 11. Optimized volumetric efficiency as a function of engine speed (for different by evaluating their ability to find optimal point of volumetric effi-
sets of design variables). ciency. Another subject of future research could be the optimiza-
tion of the shape of the valve opening, especially of the inlet
occurred close to the value of the baseline case. Additionally, opti- valve. The exhaust process could be excluded, in order to reduce
mal EVO values varied within a wide crank angle range (between 8 the search domain, since its influence in the induction process is
and 77 deg. bBDC) without exhibiting a monotonic behavior. It is less important.
worth emphasizing that optimal valve overlapping ranged
between 98 and 135 c.a. deg, thus becoming significantly longer Acknowledgements
than the baseline value (20 c.a. deg).
Fig. 11 shows volumetric efficiency as a function of engine This work has been supported by CAPES (process number
speed. Values obtained after each set of optimization calculations 9115/12-9) and CNPq (process numbers 303906/2015-4 and
are plotted there, as well as the baseline values. It can be noticed 303908/2015-7).
that the gain in volumetric efficiency obtained after optimization
with the intake duct length as the only design variable accounts References
for almost all the gain resulting when lengths of both ducts are
taken as design variables for optimization. Furthermore, volumet- [1] M.A. Jemni, G. Kantchev, M.S. Abid, Influence of intake manifold design on in-
ric efficiency values obtained with optimal valve timing are quite cylinder flow and engine performances in a bus diesel engine converted to LPG
gas fuelled, using CFD analyses and experimental investigations, Energy 36
close to those resulting when optimal duct length are used. (2011) 2701–2715.
It is worth emphasizing that before making the optimization by [2] H. Mezher, D. Chalet, J. Migaud, P. Chesse, Frequency based approach for
changing the intake and exhaust valves timing, only the optimiza- simulation pressure waves at the inlet of internal combustion engine using
parameterized model, Appl. Energy 106 (2013) 275–286.
tion of intake valves timing was made, keeping the exhaust valve
[3] D. Chalet, A. Mahe, J. Migaud, J.F. Hetet, A frequency modelling of the pressure
timing equal to the base case. The results show that the optimized waves in the inlet manifold of internal combustion engine, Appl. Energy 88
angles of opening and closing of the intake valves are quite close to (2011) 2988–2994.
those found by optimizing the intake valves and exhaust together [4] M. Deb, R. Banerjee, A. Majumder, G.R.K. Sastry, Multi objective optimization of
performance parameters of a single cylinder diesel engine with hydrogen as a
and the volumetric efficiency in this optimized case was 3% below dual fuel using pareto-based genetic algorithm, Int. J. Hydrogen Energy 38
the value obtained in the last study. This result supports the con- (2014) 8063–8077.
S.H. Och et al. / Applied Thermal Engineering 108 (2016) 660–669 669

[5] A.J. Torregrosa, J. Galindo, C. Guardiola, O. Varnier, Combined experimental and [22] M.S. Lounici, K. Loubar, M. Balistrou, M. Tazerout, Investigation on heat
modelling methodology for intake line evaluation in turbocharged diesel transfer evaluation for a more efficient two-zone combustion model in the
engines, Int. J. Automot. Technol. 12 (2011) 359–367. case of natural gas SI engines, Appl. Therm. Eng. 31 (2011) 319–328.
[6] J. Heywood, Internal Combustion Engine Fundamentals, McGraw-Hill, New [23] D. Descieux, M. Feidt, One zone thermodynamic model simulation of an
York, 1988. ignition compression engine, Appl. Therm. Eng. 27 (2007) 1457–1466.
[7] G. Fontana, E. Galloni, Variable valve timing for fuel economy improvement in [24] J.A. Velásquez, L.F. Milanez, Computational Model for Simulation of Processes
a small spark-ignition engine, Appl. Energy 86 (2009) 96–105. in Diesel Engines, SAE Technical Paper No. 952304, 1995.
[8] A. Albrecht, G. Corde, V. Knop, H. Boie, M. Castagne, 1D Simulation of [25] M. Lapuerta, R. Ballesteros, J.R. Agudelo, Effect of the gas state equation on the
Turbocharged Gasoline Direct Injection Engine for Transient Strategy thermodynamic diagnostic of diesel combustion, Appl. Therm. Eng. 26 (2006)
Optimization, SAE Technical Paper No. 2005-01-0693, 2005. 1492–1499.
[9] J. Galindo, F.J. Arnau, A. Tiseira, P. Piqueras, Solution of the turbocompressor [26] C.D. Rakopoulos, D.C. Rakopoulos, G.C. Mavropoulos, E.G. Giakoumis,
boundary condition for one-dimensional gas-dynamic code, Math. Comput. Experimental and theoretical study of the short term response temperature
Model. 52 (2010) 1288–1297. transientes in the cylinders walls of a diesel engine at various operating
[10] U. Kesgin, Study on the design of inlet and exhaust system of a stationary condition, Appl. Therm. Eng. 24 (2004) 672–702.
internal combustion engine, Energy Convers. Manage. 46 (2005) 2258–2287. [27] G. Woschni, A Universally Applicable Equation for the Instantaneous Heat
[11] M. Ceviz, M. Akin, Design of a new si engine intake manifold with variable Transfer Coefficient in the Internal Combustion Engine, SAE Technical Paper
length plenum, Energy Convers. Manage. 51 (2010) 2239–2244. No. 670931, 1967.
[12] R.C. Costa, S.M. Hanriot, J.R. Sodré, Influence of intake pipe and diameter on the [28] G. Woschni, Engine Cycle Simulation, An Effective Tool for the Development of
performance of a spark ignition engine, J. Braz. Soc. Mech. Sci. Eng. 36 (2014) Medium Speed Diesel Engine, SAE Technical Paper No. 870570, 1987.
29–35. [29] F. Ma, J.W. YuWang, S. Ding, Y. Wang, S. Zhao, Effects of combustion phasing,
[13] G. D’Errico, T. Cerri, G. Pertusi, Multi-objective optimization of internal combustion duration and their cyclic variations on spark-ignition (SI) engine
combustion engine by means of 1d fluid-dynamic models, Appl. Energy 88 efficiency, Energy Fuels 22 (2008) 3022–3028.
(2011) 767–777. [30] A.R. Ismail, R.A. Bakar, Semin, An investigation of valve lift effect on air flow
[14] E. Sher, T. Bar-Kohany, Optimization of variable timing for maximizing and coefficient of discharge of four stroke engines based on experiment, Am. J.
performance of an unthrottled SI engine – a theoretical study, Energy 27 Appl. Sci. 8 (2008) 963–971.
(2002) 757–775. [31] A.H. Shapiro, The Dynamics and Thermodynamics of Compressible Fluid Flow,
[15] J. Zhao, M. Xu, Fuel economy optimization of an Atkinson cycle engine using The Ronald Press, 1954.
genetic algorithm, Appl. Energy 105 (2013) 335–348. [32] J. Kim, C. Bae, G. Kim, Simulation on the effect of the combustion parameters
[16] M.K. Kadalbajoo, R. Kumar, A high resolution total variation diminishing on the piston dynamics and engine performance using the Wiebe function in a
scheme for hyperbolic conservation law and related problems, Appl. Math. free piston engine, Appl. Energy 107 (2013) 446–455.
Comput. 175 (2006) 1556–1573. [33] K. Atashkari, N. Nariman-Zadeh, M. Gölcü, A. Khalkhali, A. Jamali, Modelling
[17] R.J. Pearson, D.E. Winterbone, The simulation of gas dynamics in engine and multi-objective optimization of a variable valve-timing spark-ignition
manifolds using non-linear symmetric difference schemes, Proc. Inst. Mech. engine using polynomial neural networks and evolutionary algorithms, Energy
Eng. 211 (1997). Part C. Convers. Manage. 48 (2007) 1029–1041.
[18] F. Payri, J. Galindo, J. Serrano, F. Arnau, Analysis of numerical methods to solve [34] R. Storn, K. Price, Differential evolution – a simple and efficient heuristic for
one-dimensional fluid-dynamic governing equations under impulsive flow in global optimization over continuous spaces, J. Global Optim. 11 (1997) 341–
tapered ducts, Int. J. Mech. Sci. 46 (2004) 981–1004. 359.
[19] J.A. Velásquez, L.F. Milanez, Simulation of Admission and Exhaust Processes in [35] K. Price, R. Storn, J.A. Lampinen, Differential Evolution: A Practical Approach to
Diesel Engines, SAE Technical Paper No. 961124, 1996. Global Optimization, Springer-Verlag, Heidelberg, 2005.
[20] A.A. El-Rahman, A. Sabry, A. Mobarak, Non-linear simulation of single pass [36] D. Das, P.N. Suganthan, Differential evolution: a survey of the state-of-the-art,
perforated tube silencers based on the method of characteristics, J. Sound Vib. IEEE Trans. Evol. Comput. 15 (2011) 4–31.
278 (2004) 63–81. [37] ‘‘Open MPI: Open Source High Performance Computing”. <www.open-mpi.
[21] A. Broatch, J.R. Serrano, F.J. Arnau, D. Moya, Time-domain computation of org> (Retrieved April 20, 2015).
muffler frequency response: comparison of different numerical schemes, J.
Sound Vib. 305 (2007) 33–47.

Вам также может понравиться