Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Proceedingsof
ofthe
theASME 2010 Internal Combustion Engine Division Fall Technical Conference
ASME2010
ICEF2010
September12-15,
September 12-15,2010,
2010,San
SanAntonio,
Antonio,Texas
Texas,- USA
ICEF2010-35134
ICEF2010-
ABSTRACT
SA): this effect must be taken into account in order to avoid
Future emission regulations could force manufacturers to over-advanced calibrations.
install in-cylinder pressure sensors on production engines. The SA is then controlled by means of a PID (Proportional
availability of such a signal opens a new scenario in terms of Integer Derivative) controller, fed by an error that is defined
combustion control: many settings that previously were based on the previous considerations: a contribution is related
optimized off-line, can now be monitored and calibrated in real- to the MFB50-IMEP distribution, and a second contribution is
time. One of the most effective factors influencing performance related to the net Cumulative Heat Release (CHRNET)-IMEP
and efficiency is the combustion phasing: in gasoline engines distribution. The latter is able to take into account for heat
Electronic Control Units (ECU) manage the Spark Advance losses.
(SA) in order to set the optimal combustion phase. Firstly, the methodology has been tested on in-cylinder
SA optimal values are usually determined by means of pressure data, collected from different SI engines; then, it has
calibration procedures carried out on the test bench by changing been implemented in real-time, by means of a programmable
the ignition angle while monitoring Brake and Indicated Mean combustion analyzer: the system performs a cycle-to-cycle
Effective Pressure (BMEP, IMEP) and Brake Specific Fuel combustion analysis, evaluating the combustion parameters
Consumption (BSFC). The optimization process relates BMEP, necessary to calculate the target SA, which is then actuated by
IMEP and BSFC mean values with the control setting (SA). the ECU. The approach proved to be efficient, reducing the
However, the effect of SA on combustion is not deterministic, number of engine cycles necessary for the calibration to less
due to the cycle-to-cycle variation: the analysis of mean values than 1000 per operating condition.
requires many engine cycles to be significant of the
performance obtained with the given control setting. INTRODUCTION
This paper presents a novel approach to SA optimization,
with the objective of improving the performance analysis Spark Ignition engines performance are influenced by
robustness, while reducing the test time. The approach can be combustion phasing and duration: these parameters are related
either used in the calibration phase or in on-board applications, to the Spark Advance setting. The SA is usually controlled in
if the in-cylinder pressure signal is available: this would allow open loop: a calibration phase is therefore required, to
maintaining the optimization active throughout the entire determine optimal SA values, that will be used by the ECU in
engine life. open loop control during engine operation. The process is
The methodology is based on the observation that, for a usually carried out on the test bench, keeping the engine in
given running condition, IMEP can be considered a function of steady conditions for many engine cycles (sometimes many
a single combustion parameter, represented by the 50% Mass thousands), to filter out the effect of cycle-to-cycle variation.
Fraction Burned (50%MFB). Due to cycle-to-cycle variation, Data are usually collected following a speed-load matrix: for
many different MFB50 and IMEP values are obtained during a each breakpoint defining the engine operating condition many
steady state test carried out with constant SA, but these values tests are carried out, with different SA values. SA sweeps must
are related by means of a unique relationship. The distribution be performed for different engine speed, load, and, sometimes,
on the plane IMEP-MFB50 forms a parabola, therefore the oil/water temperature, gasoline temperature, etc. The mapping
optimization could be carried out by choosing SA values operation is time-consuming, and, especially in racing
maintaining the scatter around the vertex. Unfortunately the applications, very expensive, due to the short engine life and
distribution shape is slightly influenced by heat losses (i.e., by
MFB5 [°]
WOT@4000rpm, cyl 2
13 -10
12.8 ADV=29°
ADV=32°
12.6 ADV=35° -15
ADV=38°
12.4 ADV=41°
ADV=44° -20
12.2 -5 0 5 10 15 20 25
IMEP [bar]
38
12
36
11.6 34
11.4
32
11.2
30
11
-5 0 5 10 15 20 25 30
MFB50 [°]
28 ADV=29°
ADV=32°
Figure 1: relationship between IMEP and MFB50 26 ADV=35°
-5 0 5 10 15 20 25
distributions ADV=38°
70
ADV=41°
Moreover, the indicated work is remarkably influenced by ADV=44°
65
heat losses: this phenomenon is not entirely represented by
Rohr Peak [J/°]
IMEP [bar]
ADV=5° 7
9.4
ADV=6.5°
ADV=8° 6.5
9.2
ADV=9.5°
9 ADV=11°
6
ADV=12.5°
8.8 ADV=14°
ADV=15.5° 5.5
10 15 20 25 30 35 40
8.6 120
10 15 20 25 30 35
MFB50 class [°]
100
10.2
Engine Cycles
10 80
9.8 60
mean IMEP [°]
9.6 40
9.4 20
9.2 0
10 15 20 25 30 35 40
9 MFB50 class [°]
IMEP [bar]
add terms appearing in the Wiebe function, such as duration ADV=5°
(∆θ) or start of combustion (θSOC) but other solutions should be ADV=6.5°
taken into account, in order to maximize the amount of 9 ADV=8°
information added to the model. ADV=9.5°
ADV=11°
Generally speaking, the combustion effect on IMEP can be 8.5 ADV=12.5°
synthesized by means of the heat release equation ([11]): ADV=14°
15.5°
dQch dQht dm γ dV 1 dp (2)
− − (h'−u + cvT ) ⋅ cr = p + V 8
900 950 1000 1050 1100 1150
dθ dθ dθ γ − 1 dθ γ − 1 dθ
CHRnet [J]
The first member of equation (2) is often referred to as ‘net
1150
heat release’: it can be evaluated based on the indicated
pressure information, and it is influenced by the combustion
energy release rate (i.e., by the Wiebe function parameters), by 1100
the heat transferred to the cylinder wall, by the crevices filling-
emptying effects and by the combustion inefficiency. Equation
(2) can then be re-written as: 1050
CHRnet [J]
dQn 1 dp γ dV (3)
− V = p
dθ γ − 1 dθ γ − 1 dθ 1000
CHRnet = ∫ dQn (5) The reason of the non-linear relationship between MFB50
and CHRnet can be ascribed to the different heat losses effects
This means that CHRnet, together with MFB50, could be a on the two parameters: on the CHRnet side the effect is linear,
suitable parameter for the application of the trend analysis on the MFB50 side not only the effect is non-linear (heat losses
methodology. The net Cumulative Heat Release adds important depend on temperature, that is non-linearly related to
information to the combustion analysis carried out by means of combustion phase), but there is also a lack of sensitivity, due to
MFB50. CHRnet is intrinsically sensitive to combustion the fact that the majority of the heat losses take place in the
inefficiencies, causing the term dQch to decrease. In the same second part of the combustion.
dθ
10
SA CONTROLLER
9.9
A fast and robust calibration requires the SA to be changed
9.8 rapidly and coherently with IMEP variations: the operation is
carried out by means of a PID [14, 15] controller, whose input
9.7 must be sensitive to the distance from the optimal SA, and to
9.6
the direction the SA shall be changed. As a result of the
preceding discussion, the error input of the PID controller
9.5 should depend both on MFB50 and CHRnet, and it should tend
950 1000 1050 1100
CHR [J]
to zero as the IMEP is maximized (optimal SA). Since the
net
calibration operation should be fast, the number of engine
10.4 cycles available to build the distributions will be low;
10.3 moreover, the data will be affected by experimental errors.
Finally, the calculations must be carried out in real-time: the
10.2 definition of an error based on a complex IMEP model is not
worthy. The IMEP can then be expanded in Taylor series near
10.1
the values (MFB50*, CHRnet*) and stopped at the first term:
IMEP [bar]
10 ∂IMEP ∂IMEP
IMEP = IMEP * +
∂MFB50
(
⋅ MFB50 − MFB50* + )
∂CHRnet
(
⋅ CHRnet − CHRnet
*
)
9.9
(6)
9.8
The expression should be applied for small MFB50 and
9.7 CHRnet variations, so the data used to evaluate the error should
be obtained with the same SA. Equation (6) can be
9.6 reformulated as
*
9.5 IMEP − IMEP * ∂IMEP ∂IMEP CHR net − CHR net
10 12 14 16 18 20 22 24 = + ⋅
MFB50 [°] MFB 50 − MFB 50 * ∂MFB 50 ∂CHR net MFB 50 − MFB 50 * (7)
Figure 6: IMEP-CHRnet and IMEP-MFB50 trends The first member of equation (7) quantifies the IMEP
variation with respect to MFB50 within the linearization
Figure 6 shows that the information gained from the two interval: it takes into account for the effect of CHRnet, that is
distribution is complementary: the IMEP-MFB50 trace is still neglected in the partial derivative (first term of the second
not perfectly centered, suggesting a SA increase (corresponding member), and it is useful to evaluate the direction and amount
to a MFB50 decreasing), the IMEP-CHRnet trace is already of the SA shift needed to improve IMEP. Positive values of this
descending, pointing out that for further SA increasing (causing ratio mean that IMEP can be increased by retarding the
CHRnet to decrease) the IMEP would decrease. A possible combustion (i.e., increasing MFB50), negative values require to
interpretation is that the IMEP component taking into account advance the combustion (lowering MFB50). As long as MFB50
the conversion from actual energy to indicated work would is related to SA, the input error of the SA PID controller can be
benefit of a further advance, while the component taking into defined as the first term of equation (7), changed in sign.
account for the energy actually released would be
disadvantaged due to the increase in heat losses.
[bar/°]
The terms mIMEP,MFB50 and mIMEP,CHRnet are the angular
coefficients of the interpolating lines: they must be evaluated 0
25
on the basis of several engine cycles. The error will always
e
refer to the last N cycles, being N the number of cycles used for -0.05
the interpolations. The mean error over N cycles is represented
by the following expression: -0.1
*
1 CHRnet − CHRnet
e N = − mIMEP,MFB 50 N
− m IMEP,CHRnet
N
⋅
N
∑ MFB50 − MFB50 * 0 200 400 600 800 1000
(10) WOT@4000rpm, N=75 Cycles
0.15
The last term in equation (10) is the average angular
coefficient of the line connecting the point representing a given 0.1
cycle to that representing the mean value of the last N cycles, in
the plane MFB50-CHRnet. This term could reach very high 0.05
values when the MFB50 is very close to the reference MFB50*.
e75 [bar/°]
eN = −mIMEP,MFB 50 − mIMEP,CHRnet ⋅ mCHRnet ,MFB 50 (11) 0 200 400 600 800 1000
N N N
WOT@4000rpm, N=125 Cycles
Finally, the error should be normalized with respect to the 0.15
mean IMEP, evaluated over the N cycles used for the
interpolations: in this way the error will be representative of the 0.1
IMEP relative variation obtained advancing or retarding the
combustion. 0.05
e125 [bar/°]
Cyl5
Cyl6 RESULTS
0 The SA PID controller has been designed based on sample
data, then it has been implemented in a real-time controller.
The controller receives combustion data from an indicating
-0.05 system cycle by cycle, evaluates the interpolation based on a
configurable number of engine cycles, and finally requires SA
variations to the ECU. The automatic calibration system has
-0.1 been applied to correct open loop SA values of a 1.2 liters
30 32 34 36 38 40 42 44 gasoline engine running on the test bench. The results shown in
SA [°]
the followings have been obtained setting to 100 the number of
cycles used for the interpolations.
Figure 8: IMEP and error trend for different SA values
Figure 9 shows the controller action in three different
The average error value for a given test is representative of engine running conditions: even without a careful tuning of the
the benefits obtained reducing (or increasing) SA: figure 8 controller settings, the calibration system proved to be effective
shows a comparison between the typical IMEP(SA) trace and in the determination of the optimal SA. From the moment the
the mean error(SA) trace. They give the same information, controller is activated, the optimal SA is reached in a few
2
10
0
0 200 400 600 800 1000
Engine Cycles
5
Figure 10: controller behavior in different AFR conditions
0
0 5 10 15 Figure 11 shows how the controller acts on SA, modifying
35°@3500rpm, AFR=14.6
20 the combustion phase, and therefore the IMEP.
15°@1500rpm AFR=14.6
20
15
deltaSA [°]
15
10
SA [°]
10
5
5
0
0 5 10 15
38°@4500rpm, AFR=14.6
10 0
0 200 400 600 800 1000 1200
40
8
deltaSA [°]
6 30
MFB50 [°]
4 20
2
10
0
0 5 10 15
Time [s] 0
0 200 400 600 800 1000 1200
carried out with the same speed and load condition, but with
different AFR. The effect of AFR on the optimal SA is 4
evident: the leaner the composition, the slower the combustion, 3.8
the higher the optimal SA. The system can thus be used for
complex calibration operations, with multi-dimensional control 3.6
matrixes; SA optimization over a speed, load, AFR matrix is a 3.4
possible application, but other examples can be added: injection 0 200 400 600 800 1000 1200
Engine Cycles
pattern in direct injection engines VVT and EGR settings,
influence the combustion process, and their calibration tasks Figure 11: effects of controller action on IMEP and MFB50
could benefit of the proposed approach.