You are on page 1of 6

Copyright © IF AC Advances in Automotive Control ELSEVIER

Salerno, Italy, 2004

IFAC
PUBLICATIONS
www.elsevier.comllocatelifac

IN-CYLINDER PEAK PRESSURE POSITION


MODELLING

S.D.Carroll, A.T.Shenton, G.Triantos

Powertrain Control Group, Department of Engineering,


The University of Liverpool,Liverpool L69 3GH, UK

Abstract: Higher efficiency demands and decreasing cylinder pressure sensor costs
increase the opportunity for cylinder pressure based engine control. Control of
spark advance using cylinder pressure has been widely examined in the literature
and has been shown to be a suitable method of achieving minimum spark timing
for best torque in variable environmental conditions. This paper summarises the
peak pressure position control problem and the identification of forward and
inverse NARMAX models for a range of engine speeds for use in such control.
The models presented relate the spark advance, air bleed valve duty and the
crank angle at which the peak cylinder pressure occurs and are developed for a
single cylinder of a 1.6L Ford Zetec engine. The models established are valid over
a range of engine speeds at a single initial load condition. The forward model may
be used in controller evaluation and design. The inverse model may be used as
an open loop controller or in plant linearisation to allow the application of linear
feedback control techniques. Copyright © 2004 IFAC

Keywords: In-Cylinder Pressure, NARMAX, Inverse models, Identification,


Non-linear Spark Control.

1. INTRODUCTION way directly accessing the cylinder (Lancaster et


al. , 1975), or, infer the cylinder pressure by some
Ensuring the angle at which the peak cylinder other less direct method . Less direct methods in-
pressure occurs is at its optimum position ensures clude spark plug ionisation (Eriksson et al. , 1996)
the most efficient use of the available work in (Upadhyay and Rizzoni , 1998), Crank Shaft Ac-
the cylinder. Although the optimum peak pressure celeration (Kao and Moskwa, 1995) (Citron et
position (PPPopd varies with speed and changes al., 1989), analysis of vibration (Du et al. , 2001 ),
and, strain gauges mounted in a spark plug boss
for different engines (Powell, 1993), The constant
(Sellnau et al., 2000) or as a transducer mounted
value of 16 0 after TDC is a typical PPP opt for
under the spark plug (Sawamoto et al. , 1987).
many applications(Heywood, 1988).

Once a cylinder pressure reading is obtained, the


The in-cylinder pressure time history with respect
problem of controlling the SA using feedback of
to crank angle is affected by the spark advance and
the PPP remains . This has been examined before
the relative efficiency of a given combustion cycle.
(Powell, 1993) (Sawamoto et al. , 1987) (Cho and
There are two broad methods to gain a cylin-
Oh, 1993), but in this paper the issue of control is
der pressure reading. Either directly measure the
not explored directly. Non-linear modelling and
cylinder pressure with a pressure sensor in some

293
- gas motion throughout the cylinder during
combustion. Because the flow into the cylin-
der is turbulent the gas motion is different
every cycle .
• The mixture within the cylinder is non ho-
mogeneous and will vary between cycles, crit-
ically, in the vicinity of the spark plug. This
non-homogeneity, in and near the spark gap ,
'r significantly influences the initial flame de-
'r velopment.
,c
The CCV with specific regard to the PPP has
been found to increase with increasing engine
speed and decrease with increasing engine load.
Fig. 1. Cylinder Pressure Vs. Crank angle
Further, the skewness and kurtosis of this vari-
inverse modelling is considered as a precursor ation are speed and load dependent (Carroll,
to later design of feedforward and feedback con- September, 2003).
trollers.

3. PROBLEM DESCRIPTION
2. CYCLIC VARIATION
The relationship between engine speed, N, spark
advance, SA, and PPP is inherently non-linear,
Cycle by cycle variations (CCV) are an in-
therefore consideration of the interactions of these
escapable facet of internal combustion engines.
variables as inputs and output is valuable before
The PPP is susceptible to CCV and this may be
any identification begins. So given a fixed SA with
considered noise on the PPP output. The PPP
the engine running at constant speed and load
can be seen to vary between cycles as can be
the PPP for any combustion cycle is subject to
seen in figure [1 ] The nature and the degree of
CCV as discussed earlier. The PPP for a given
the normal cyclic variation effectively becomes a
SA will vary depending upon engine speed and
governing characteristic of the engine. In practice
load. However, there is an optimal PPP, PPPopt
there are only probabilistic methods to determine
which will be produced by optimal spark timing
whether a given combustion cycle will be fast or
Le. MBT. Having the PPP optimally positioned,
slow burning, a strong or a weak cycle. The fastest
for a given AF ratio and temperature, produces
burning combustion cycles limit the amount of
the greatest torque from the available work. The
spark advance which may be applied to the engine,
nature of the combustion process means that the
a strong combustion cycle will produce higher
rate of combustion will vary cycle to cycle and
temperatures and pressures and, if engine knock
so the PPP will also vary. This PPP variation
occurs, it will occur first on the strong combustion
alters the amount of torque produced by a given
cycles as pressures become so high that in regions
cylinder in a given combustion cycle. Therefore,
autoignition occurs (This is not the only cause of
as the PPP varies for a given SA so the engine
engine knock) . Weak, slow burning, cycles define
speed varies. This variation in the engine speed
the other end of the scale. The weak cycle becomes
will mean that the SA setting is no longer optimal
the misfire or, in less extreme cases, the weak
which will lead to an incorrectly positioned PPP
cycle is merely inefficient and produces excess
unless the next combustion cycle is a faster burn-
emissions. So in attempting to control PPP, an
ing one which will lead to an increase in engine
understanding of ccv and how this may alters the
speed. The effect of this behaviour in a single
PPP is important.
cylinder will be ameliorated by the other three
cylinders in the engine so a decrease in engine
Generally, cyclic variations within internal com- speed caused by poor combustion cycle in cylinder
bustion (IC) engines have been discussed by one may be compensated by a normal or strong
(Heywood, 1988) and (Roberts et al. , 1997) and combustion cycle in the next three combustion
(Ozdor et al., 1994). There are three critical fac- events in the other firing cylinders.
tors influencing the efficiency and strength of any
given combustion cycle:
It is apparent that a non-linear relationship exists
• The air fuel ratio (AFR) and exhaust gas between engine speed, SA, and PPP. The changes
recirculation (EGR) entering the cylinder in PPP affect the torque produced by a given
varies from cycle-to-cycle. combustion cycle and therefore the crank speed .
• The turbulent flow of the fuel air mixture The crank speed is an output of the PPP process
into the cylinder produces variations in the and so the air bleed valve (ABV), in the absence

294
of a throttle position sensor, as this input directly
alters speed independent ofPPP. An identification
of this non-linear relationship between SA. ABV
and PPP is presented. .

4. IDENTIFICATION

The pressure curves for a single cylinder, as in


figure [1], were processed on-line. The PPP is ob-
tained using an algorithm which logs the cylinder
pressure every 10 of crank angle and records the
crank angle at which the maximum cylinder pres-
sure occurs. The PPP is then output at the end Fig. 2. SA, ABV and PPP Input/ Output Data
of each combustion cycle. The ABV was randomly
varied to ensure the desired range of 1000 to 3000
rpm was covered and large transitions occurred to
emulate large changes in engine speed and excite The model is accordingly crank based and con-
as much of the system dynamics as possible. The tains 28 parameters, Pn . No attempt was made
SA was varied between 20 0 and 30 0 using a ran- at refining the model structure aside from limit-
dom number generator BTDC which led to a vari- ing the delay order to one based upon a priori
ation in the PPP of between 10 and 40 0 ATDC . knowledge of the combustion process.
The fuel strategy used by th engine management
is based on a heated exhaust gas oxygen sensors
and was not interrupted and so the correct AFR y(B) = PI + P2YI(B) + P3Yl(B -1) + P4UI(B) +
was maintained for all the ABV duties. The input
and output data is shown in figure [2] . The engine psul(B - 1) + P6U2(B) + P7U2(B -1) +
was placed under an initial fixed load using a low + P9U2(B)YI (B - 1) +
PSUl (B)YI (B - 1)
inertia dynarnometer (Dorey R.E. and Z. , 1995). PlOUl(B -1)YI(B -1) + PllU2(B)Yl(B -1) +
The initial EEC I load was 0.32, a mean torque
load of 85Nm, at a speed of 1000rpm. The engine P12Ui(B) + P13U~(B) +
data was initially sampled every 10 using dSpace PI4Ui(B - 1) + PI5U~(B - 1) +
DIO and A/ D conversion in an expansion box P16yr(B - 1) + P17yi(B - 2) +
programmed using Matlab/ Simulink and Real-
Time Workshop. Data logging was performed by PlSYl(B - I)Yl(B - 2) + P19UI(B)U2(B) +
streaming to hard disk through dSpace Control P20Ul(B)Ul(B - 1) + P2IU2(B)U2(B -1) +
Desk software. Engine data was processed on-line Pnul(B)U2(B - 1) + P23U2(B)UI(B - 1)) +
to produce and record the PPP, ABV, SA, engine
speed and combustion cycle number. Additional P24UI(B -1)U2(B - 1) + P2SUI(B)YI(B - 2) +
data such as torque load were noted. Approxi- P26UI(B -1)YI(B - 2) + P27U2(B)YI(B - 2) +
mately 6000 combustion cycles for cylinder one P2SU2(B -1)YI(B - 2)
were recorded for use in identification. The data
set was divided for separate use in model estima-
tion and validation. The experiment set-up of is
shown in figure [3] .
The parameters were estimated using an ordinary
least squares algorithm. The parameters produced
The MISO NARMAX model structure used in the
were then incorporated in a simulink model and
identification was similar to that used by Petridis
subjected to the validation data inputs to pro-
(Petridis and Shenton, 1999) and is shown in equa-
duce a simulated output. The simulated output
tion 1. The model used has a similar structure to
was compared with the validation data output.
the NAR..t\1AX form presented by (Leontaritis and
Additional cross correlation checks of the model
Billings, 1985) (Billings et al. , 1988) (Billings et
residuals with both inputs were found to be below
al. , 1989). This was used with two inputs, ABV,
0.05. The NAR..\1AX model fit and residuals are
UI , and SA, U2 , with PPP as the output Yb B
shown in figure 5 and the cross correlation checks
refers to the current combustion cycle of cylinder
are shown in figure 4. The remaining residuals
one, thus B-1 is the previous combustion cycle.
are due to the PPP reflecting the noise on the
output signal due to ccv. The model parameters
1 EEC load is a non-dimensional ratio of air charge over a are shown in table 1.
nominal standard air charge, representing universal load

295
1.6litre ZETEC Engine Experimental Set-Up
Identification and Data Collection

Low Inertia
Dynomometer

Vibrometer
Torque ' . Pressure
Speed
Transducers (x4)
:To1=~ E~~-- -" ,--- -:-- - - --\
I Fuel Control. " To Engtfle _ I
I Temp, Pressure,Speed : I Spst1< and Air SIHd I
\~~a<!!~~ ___ __ I \!aJ~~~t~~ __ )
Comparison Circuit
roe indication
Every Degree Signal
dSpace
RealtimeCcntrol
AID, 010

PC
Programming clSpaoe
PrtJSSure. Temp.
Monitoring
'nn.., CcndniOns OalllLogglng

Fig. 3. Experimental Setup

Parameter NARMAX INARMAX


PI -1495.465 269.312
P2 7.085 -8,758
P3 0.967 8 ,700
P4 -41.493 3633.636
Ps 830,489 -3504.512
P6 -6.727 -4.260
P-r 8,775 2.710

~P1::J
Ps -0.956 -2.127
P9 0,015 0.025
PlO 0.214 2,658
Pll -0.021 -0,002

- PI2 -265,219 -195.642


." • 10 12 ,. " " 20 PI3 -0.012 0.004
P l4 48.461 -445 .246
PIS -0.036 -0.005
Fig. 4. NARMAX Cross Correlation with Inputs
P I6 -0.011 0.022
5. I::--J'VERSE NARMAX P I7 -0.003 -0.002
PIB 0.006 -0.019
P I9 -6.277 -4.545
Identification of the inverse NARJ...1AX model P20 128.394 643,965
is performed by considering the SA, U2 , input P21 0.043 0.003
as an output in the manner of (Shenton and P22 -0.198 -5.458
Petridis, July 2001 ). The PPP output, Yl, is con- P23 3.978 4.246
P24 1.302 5.402
sidered as an input and then the identification is 2.079 0.215
P2S
performed in an identical manner to the earlier P26 ·3.027 -0.700
NARMAX case. This reversal of input and output PH 0.005 -0.015
is shown in figure [6]. The parameters calculated P2B -0.003 -0,005
for the IN AIL\1AX model are sho'wn in table 1. Table 1. Model parameters

The model fit and residuals are shown in figures


[7] and [8]. Once the model is given the current control techniques can then be applied across the
ABV setting and PPP demand for a fixed load inverse compensated plant in the manner demon-
condition, the output from the INARJ...IAX model strated for idle speed control by (Petridis and
is the SA required by the engine to produce the re- Shenton, 1999) (Shenton and Petridis, September,
quired PPP. The model when used in-line with the 1999) (Petridis, October, 2000) .
plant acts to linearise the plant. If desired, linear

296
Original Vs Model DATA
35
30
_25
~ 20 .
<
i 15 ·
10
5
0
0 3000 5000 6000

Resids
15
10
5
0
l -5
-10
-15
-20
-25
0 1000 2000 3000 4000 5000 6000
Cycle

Fig. 5. NARMAX fit of Model


_ID'" M C)OIt.er......,.,....,... .............
-u

J~~~~
.:.J ~
....I -
:

U1 ...I '

U2
:1 NARMA 1 .. Y1 ... . ,0..

)
"

U1 ..I-
I
_,L
.. U2
Y1 :I,NARMAI -tr
I

Fig. 6. Inputs/ Outputs of NARMAX/ INARMAX


-
_.Ie..-:--:----::---:--::;"--:,:;-,--:';---:'"_-:'-

Fig. 8. INARMAX Cross Correlation with Inputs


,, ---;;;-'

The validation of the INARMAX filter of sec-

I~!NW_~~~~~~~
tion 5 may be recognised as essentially a feed-
forward control process. The inverse model may

. ----- I accordingly be used as an open loop controller,


this would remove the need for spark look-up
tables. This would set optimum spark advance
~
- for a given ABV duty and desired PPP. Addi-
tionally the desired PPP may change depending
upon engine speed, coolant temperature and en-
vironmental conditions. An algorithm to calculate
E '• • •" ••••
the optimum PPP, requiring tuning for different

-
engines, was presented by Powell (Powell, 1993) .
-1'0:----::-==---=_=--_=----::':::-----:::::-----:::
In order for the option to be effective as an open
loop controller the inverse model would require
Fig. 7. L~ARMAX fit of Model an additional input to reflect engine load. Inlet
manifold pressure (MAP ) would be a suitable
6. OPEN LOOP CONTROL variable, or EEC load if available.

Current industrial practice is to define SA look-


up tables and a range of engine speed and load 7. CONCLUSIONS
conditions to give minimum spark advance for
best torque (MBT) spark timing without regard • A non-linear model of the ABV and SA
to the in-cylinder pressure (Gallas, 1999). to PPP relationship is presented which is
suitable for non-linear control studies.

297
• Control of SA using feedback of PPP v.rill Gallas, J. (1999). Spark and Fuel Settings De
allow the removal of SA look-up tables and termination and Identification. Ford M oto:
the associated calibration effort to produce Company, C.E. T.P. 03.0L-L-601 .
them. The identified model is acquired rela- Heywood, J . (1988) . Internal Combustion Engin.
tively quickly but as yet does not account for Fundamentals. McGraw-Hill.
variations in engine load. Kao, M. and J .J . Moskwa (1995). Nonlinear Diese
• An inverse NARMAX model which may act Engine Control and Cylinder Pressure Obser·
as a non-linear compensator or open loop vation. ASME Journal of Dynamic Systems
controller is presented. Measurement and Control 117, Jun, 183-
• Closed loop control of SA using feedback 192.
of PPP will enable the optimal PPP to be Lancaster, D.R., R.B . Krieger and J.H. Lienescl
maintained in changing load , engine wear, (1975 ). Measurement and Analysis of EnginE
and environmental conditions. Pressure Data. SAE, paper 750026.
• An inverse model including MAP could be Leontaritis, I.J. and S.A. Billings (1985). Input·
identified to provide SA for desired PPP at output parametric models for non-linear sys·
a more complete range of speed and load tems. LDeterministic non-linear systems. Int
conditions. J. Control 41(2) , 303-328.
• Future possible work might include design of Ozdor, N., M. Dulger and E. Sher (1994) . Cycli<
a controller across an inverse compensated Variability in Spark Ignition Engines A Lit·
system controlling SA to give optimal PPP erature Survey. SAE, paper 940987.
across a wider range of speeds and load Petridis, A.P. (October, 2000) . Non-linear robus:
conditions. control of S.I. engines. PhD thesis, Liverpoo
University.
Petridis, A.P. and A.T. Shenton (1999). Non·
REFERENCES Linear Robust Performance Control of an SI
Billings, S.A. , M.J. Korenberg and S. Chen Engine by Identified Inverse System Compen-
(1988). Identification of non-linear output- sation. Identification in Engineering Systems,
affine systems using an orthogonal least- Proceedings of the Second International Con·
squares algorithm. Int. J. Systems Sci. ference held in Swansea, March 1999 pp. 190-
19(8), 1559-1568. 199.
Billings, S.A., S. Chen and R.J . Backhouse (1989 ). Powell, J .D. (1993). Engine control Using Cylin-
The identification of linear and non-linear der Pressure: Past, Present, and Future.
models of a turbocharged automotive diesel ASME Journal of Dynamic Systems, Mea-
engine. Mechanical Systems and Signal Pro- surement and Control 115, 343-350.
cessing 3(2) , 123-142. Roberts, J .B. , J.C. Peyton Jones and KJ. Lands-
Carroll, S.D. (September, 2003) . Control of S.I. borough (1997). Cylinder Pressure Variatiom
engines using In-Cylinder Pressure. PhD the- as a Stochastic Process. SAE, paper 970059.
sis, Liverpool University. Sawamoto, K , Y. Kawamura, T . Kita and
Cho, D-LD. and H-K Oh (1993). Variable struc- K. Matsushita (1987). Individual Cylinder
ture control method for fuel-injected systems. Knock Control by Detecting Cylinder Pres-
Journal of Dynamic Systems, Measurement sure. SAE, paper 871911 .
and Control. Trans. of ASME 115, 475-48l. Sellnau, M.C. , F.A. Matekunas, P.A. Battis-
Citron, S.J ., J .E . O'Higgins and L.Y. Chen ton , C. Chang and D.R. Lancaster (2000).
(1989). Cylinder by Cylinder Engine Pressure Cylinder-Pressure-Based Engine Control Us-
and Pressure Torque Waveform Determina- ing Pressure-Ratio Management and Low-
tion Utilizing Speed Fluctuations. SAE, paper Cost Non-Intrusive Cylinder Pressure Sen-
890486. sors. SAE, paper 2000-01-0932.
Dorey R.E. , Maclay D., Shenton A.T. and Shafiei Shenton, A.T . and A.P. Petridis (July 2001 ).
Z. (1995). Advanced Powertrain Control Robust performance speed control of an SI
Strategies. IFAC Workshop on Advanced Au- engine by a non-linear 111S0 direct-inverse
tomotive Control, Asona. method. NCN Workshop, Sheffield.
Du, H., L. Zhang and X. Shi (2001). Reconstruct- Shenton, A.T . and A.P. Petridis (September,
ing cylinder pressure from vibration signals 1999). Total SI engine control by inverse dy-
based on radial basis function networks. Pro- namic compensation. UnICEG: Powertrain
ceedings of the Institution of Mechanical En- Integration and Control, University of Bath.
gineers, Part D: Journal of Automotive En- Upadhyay, D. and G. Rizzoni (1998). AFR Con-
gineering 215 No.6 , 761-767. trol on a single Cylinder Engine using the Ion-
Eriksson , 1., 1. Nielsen and J. Nytomt (1996). ization Current. SAE, paper 980203.
Ignition Control by Ionization Current Inter-
pretation. SAE, paper 960045.

298