7th IFAC Symposium on Advances in Automotive Control

The International Federation of Automatic Control

September 4-7, 2013. Tokyo, Japan

Air/Fuel Ratio Estimation of SI Engine Using

Higher Order Sliding Mode

Muhammad Amin Akram* Aamer Iqbal Bhatti** Qadeer Ahmed***

*Control & Signal Processing Research Group, M. A. Jinnah

University, Islamabad, Pakistan. ([email protected]).
**([email protected]) *** ([email protected])

Abstract: Optimized fuel economy and lower pollutant emission require accurate control of air-fuel ratio
in both transient and steady state region. Most of the existing AFR control schemes use oxygen sensor
output as feedback, having inherent transport delays and poor performance in cold start. To overcome
these problems a model based air-fuel ratio estimation scheme using HOSM is developed, to provide a
feedback signal to AFR control scheme that is independent of the problems mentioned above. The
methodology adopted in this paper utilizes engine speed to estimate the air-fuel ratio of the SI engine.
Data is taken from a production vehicle equipped with ECU complaint to OBD-II standards. The
experimental results show that the estimated signal is in good agreement with the value found in
literature. The proposed scheme can be used as virtual sensor for monitoring AFR and is simple enough
to be easily implemented online.
Keywords: Parameter estimation, higher order sliding mode, air/fuel ratio

transients. To overcome this delay issue some authors avoid

1. INTRODUCTION
oxygen sensor for finding AFR and use alternative
Modern vehicles have the facility of onboard diagnosis using approaches (Tunestal and Hedrick 2003, Lee et al. 2001). The
ECU, to enhance fuel economy and lower emission of methods solve the transport delay problem but still have cost
harmful gases as regulated by EPA. The air-fuel ratio is an and reliability issues due to extra components installation.
important parameter in this regard. The AFR of in-cylinder The approach adopted in this paper uses existing elements to
mixture should be as close to its stoichiometric value as meet the task. For this purpose the existing setup is analyzed
possible. Several techniques are found in literature to control to produce an implementable scheme. The proposed scheme
the value of AFR around its stoichiometric value (Jones et al. relies on the system model and uses data from already
1995, Powell et al. 1998, Yoon and Sunwoo 2001). Most of installed speed sensor to estimate the AFR. This estimated
these techniques use binary oxygen sensor output as feedback signal can be used for monitoring air-fuel ratio in the system,
or install an extra sensor for the same purpose. Installing an reducing the AFR excursions caused by gas transport delay
extra sensor is a costly solution and also reduces system while using oxygen sensor. Furthermore during cold start
reliability by adding an additional element to monitor. oxygen sensor is not at the optimal temperature to produce
the required measurement, so during this phase open loop
TABLE I scheme is adopted for fuel supply, resulting in poor AFR
ACRONYMS USED IN ARTICLE control. The AFR estimation scheme developed in this paper
can be used in cold start phase when the exhaust chamber is
Acronym Description at low temperature for oxygen sensor to produce correct
AFR Air fuel ratio output. Also the proposed methodology is computationally
ECU Electronic Control Unit cheap and can be easily implementable online.
EPA Environmental Protection Agency
HOSM Higher Order Sliding Mode Sliding mode techniques are in use for many years for
OBD-II On-board Diagnostic-II parameter estimation due to its robustness and finite time
SI Spark Ignition convergence properties. Sliding mode design consists of two
SMO Sliding Mode Observer stages. First a sliding surface is designed. Then a
SOSMO Second Order Sliding Mode discontinuous control is found such that the system states
Observer reach the sliding surface (reaching phase) and then slides to
the origin (sliding phase). Rich literature is available on use
of sliding mode in parameter estimation. Using sliding mode
Furthermore it also involves transport delays of exhaust gases observer online estimation of indicated torque of Internal
and sensor response delay in using oxygen sensor. The Combustion (IC) engine is done in (Rizzoni et al. 1996) and
inherent transport delays causes AFR excursions during fast this estimation is further utilized in fault diagnosis. In

978-3-902823-48-9/2013 © IFAC 501 10.3182/20130904-4-JP-2042.00049

IFAC AAC 2013
September 4-7, 2013. Tokyo, Japan

(McCann et al. 2001) SMO is used for speed and position The rest of the paper is organized as: Section 2 describes the
estimation of switched reluctance motor to control the engine model used in the present work. In Section 3 Second
conduction angle of motor. Shraim et al. 2006 used SMO for Order Sliding Mode Observer (SOSMO) design for speed
the estimation of vehicle parameters In (Floquent et al.2006) dynamics and parameter estimation is discussed. Section 4
a Higher Order Sliding Mode Observer (HOSMO) is used to gives the experimental results and at the end Section 5
estimate more than one parameter in one equation of a Linear summarizes the whole work.
Time Invariant system. The same approach is used for the
parameter estimation of a nonlinear system model in (Butt 2. MODEL DESCRIPTION
and Bhatti 2008). In (Draknuov and Law 2007) SMO is
developed and demonstrated for parameter estimation of a The mean value engine model used in (Ahmed and Bhatti,
biokinetics model. Parameter estimation of uncertain 2010) is utilized here for estimation of AFR (Fig. 1). The
nonlinear system using sliding mode differentiator observer manifold pressure dynamics are modelled by filling and
is discussed in (Muhammad et al. 2010). (Ahmed and Bhatti, emptying of inlet manifold air behaving as perfect gas (Butt
2010) used SOSMO for estimating SI engine efficiencies and and Bhatti 2009). It is assumed that air flow is only in one
parameters. Traditional sliding mode technique, however, has direction and is frictionless. The process is considered to be
intrinsic problem of chattering (Utkin 1999) which is solved adiabatic and temperature and pressure of out-flowing gas is
by using HOSM (Levant 2001, 2003). So instead of using taken same as that of the gas in the manifold using lumped
first order sliding mode, the present work uses the higher parameter approach (Guzzella and Onder, 2004). The speed
order sliding mode for the parameter estimation to overcome dynamics is modelled in terms of brake torque using
the unwanted chattering effect. The advantage of using combustion process modelled by Otto cycle. The net
HOSMO is twofold; it overcomes the unwanted chattering available torque is the difference of other torques present in
problem in estimated process and also retains the robustness the system. These unavoidable torques are present due to the
to uncertainties and disturbances. physical phenomena occurring in the system e.g. combustion,
pumping, friction and internal and external loads.
TABLE II
SYMBOLS USED IN THIS ARTICLE
Symbol Description
Pm Manifold Pressure
ωe Engine Speed
ηvol Volumetric Efficiency
ηc Combustion Efficiency
ηth Thermal Efficiency
R Universal Gas Constant
Fig. 1. Block diagram of mean value engine model
Vm Manifold Volume
AE Throttle Effective Area The equations representing the engine manifold pressure and
speed dynamics are given below
Pa Ambient Pressure
Cd Throttle Discharge Coefficient Pm = A1 f ( Pm ) − A2 Pmωeηvol ⎫
γ Heat capacities ratio ⎪
1 ⎬ (1)
Ta Ambient Temperature ω e = (Ti − Tp − T f − Tl ) ⎪
Je ⎭
D Inlet Diameter
α cl Throttle Angle at Closed Position Where the terms used in (1) are explained in the Table II
α Throttle Angle and A1 , A2 , f ( Pm ) and other expressions are defined in
Vd Displaced Volume Appendix A. Model given in (1) is validated by (Ahmed and
Vm Manifold Volume Bhatti, 2010) against data obtained from a 1.3L commercial
vehicle equipped with ECU complaint to OBD-II.
Q Fuel Heat Value
Cγ Compression Ratio 3. AFR ESTIMATION SCHEME
Cv Specific Heat at Constant Volume
Sliding mode techniques are in use for many years by control
Je Engine Inertia
community with applications in parameter estimation due to
Ti Indicated Torque its simple design and robustness properties. In this paper,
Tf Friction Torque SOSMO proposed in (Saif et al. 2008) is applied on engine
model given in (1). The nonlinear system under study is
Tp Pumping Torque
described in state space form as
Tl Load Torque

502
IFAC AAC 2013
September 4-7, 2013. Tokyo, Japan

x1 = x2 ⎫ ⎡ −5.69 × 10−5 − 4.66 × 10−14 x1 0 ⎤

⎪ ⎢ ⎥
x2 = f (t , x1 , x2 , u ) + ξ (t , x, u ) ⎬ J PO =⎢ ∂f 2
(2) −5
− 5.69 × 10 − 4.66 × 10 −14
x2 ⎥
⎢ ∂f1 ⎥
y = x1 ⎪ ⎣ ⎦
⎭

where ⎡ V × imep ⎤
⎢ − d Pm ⎥
4π J e ( AFR) 2
 
[ x1 , x2 ]T is state vector, f(t,x1,x2,u) is nominal system J PI = ⎢ ⎥
⎢ 5.69 × 10−5 Vd × imep 2 x1 Vd × imep ⎥
dynamics, ξ(t,x,u) represents internal uncertainties, y is ⎢− Pm − Pm ⎥
⎣⎢ Je 4π J e ( AFR ) 2
J e 4π J e ( AFR) 2
⎥⎦
system output.
Based on system dynamics (2) the second order sliding mode
observer is As the above matrices J PO and J PI consists of physical
   system parameters that are bounded in nature and non-zero in
x1 = x2 + z1 x1 (0) = x1 ⎫ steady state condition, so theses matrices are full ranked in
    ⎪⎪
x2 = f (t , x1 , x2 , u ) + z2 x2 (0) = 0 ⎬ (3) steady state as
 ⎪
y = x1 ⎡ −5.8507 ⎤
⎪⎭
J PO = 3.2376 × 10−9 ≠ 0 and ⎡⎣ J PI ⎤⎦ = ⎢ ⎥.
  ⎣ −936.117 ⎦
where [ x1 , x2 ]T is state estimation vector, z1 and z2 are
correction variables defined as Therefore the speed dynamics defined above is observable
and identifiable with respect to the parameter to be estimated,
sign( x1 ) + ν 1 ⎫⎪ so we can now design SOSMO for engine rotational
1
z1 = λ1 x1 2
⎬ (4) dynamics using (3) and (6) as
ν1 = α1 sign( x1 ) ⎪⎭
 
x1 = x2 + z1 ⎫
and ⎪
 Vd ⎬ (7)
x2 = ( A1 f ( Pm ) − A2 Pm x1ηvol ) + z2 ⎪
z2 = 0 if x1 ≠ 0, x1 ≠ 0 ⎫ 4π J e ⎭
⎪
1 ⎪
= λ2 z1 2
sign( z1 ) + ν 2 if x1 = 0, x1 = 0 ⎬ (5) where terms used are already defined. The gains α1 and λ1 are
⎪
ν2 = α 2 sign( z1 ) ⎪⎭
selected on the basis of uncertainty bound and the initial state
estimation error ( x10 ) in worst case as defined in (Saif et al.

x1 = x1 − x1 2008)
with 
x2 = x2 − x2
α1 > k1 x1 + ξ1+ 0

α i , λi with i = 1, 2 are observer gains. The above second 4α1

λ1 >
order sliding mode observer uses anti-peaking structure α1 − ξ1+
where x1 and x2 reach sliding manifold one by one (Saif et
al. 2008). where the uncertainty function defined in (2) satisfies

3.1 SOSMO for Engine Speed Dynamics ξ1 ( x, t , u ) < ξ1+

To estimate the AFR signal using the speed dynamics given ξ1 ( x, t , u )
in (1), we redefine speed equation described in Section 2 in < δξ1+
dt
the form given in (2).
Similar procedure as adopted for the selection of gains
x1 = x2 ⎫ α1 and λ1 can be used for the gains α 2 and λ2 . The choice of
⎪
Vd imep ⎬ (6) constant k1 is done as described in observer convergence
x2 = ( + 1)( A1 f ( Pm ) − A2 Pm x1ηvol ) ⎪
4π J e AFR ⎭ analysis in (Saif et al. 2008).

where x1 = ωe and x2 = ω e 3.2 Parameter Estimation

Using (6) and (7) we define error dynamics as
ωe is engine rotational speed measured from a 1.3L
commercial vehicle equipped with ECU. To check whether x1 = x2 − z1
the system (6) is identifiable and observable with respect to
the estimated parameter we analyse the observability and V imep
identifibility of the above dynamics as defined in x2 = d ( A1 f ( Pm ) − A2 Pm x1ηvol ) − z2
4π J e AFR
(Muhammad et al. 2010), so we get

503
IFAC AAC 2013
September 4-7, 2013. Tokyo, Japan

Here error variable given above is taken as sliding surface, so

when system is in sliding mode x and therefore x
approaches zero.
 
where x = x − x and x = x − x

so we get

Vd imep
( A1 f ( Pm ) − A2 Pm x1ηvol ) = z2 , which gives
4π J e AFR

Vd imep
AFR = ( A1 f ( Pm ) − A2 Pm x1ηvol ) (8)
4π J e z2

The convergence analysis of the observer can be found in Fig. 2. Parameter estimation block diagram
(Saif et al. 2008).

4. EXPERIMENTAL RESULTS Throttle Pattern

12
This Section describes the results and experimental setup
Degree
10
used to implement the estimation scheme discussed in the 8
previous Section. Data is acquired from 1.3L production 0 50 100 150 200 250
vehicle equipped with electronic control unit complaint to Engine Angular Speed
OBD-II. Different built-in sensor readings are available in the 100
rad/sec

ECU. Out of these, we get data of our interest i.e. engine 90

speed, manifold pressure and throttle pattern using OBD-II 80
0 50 100 150 200 250
data scanner and relevant software. (Fig. 3) shows the sensors
reading obtained from the 1.3L production vehicle having 4
x 10 Manifold Pressure
ECU facility in it. These are used as inputs to the proposed 3
KPa

estimation scheme. The vehicle is in idle position and no

2.5
extra load is applied on it. Throttle is kept constant 0 50 100 150 200 250
throughout the experiment. SOSMO is given the required Time (sec)
parametrical data and experimental data obtained from
vehicle (Fig. 2). Observer loopback is based on the difference Fig. 3. Engine angular speed and engine block temperature
between engine rotational speed acquired from the production
vehicle and the speed estimated by the observer. The outputs
of observer are system states and the estimated parameter Engine Speed
which in present case is AFR. The estimated parameter can 100
further be used for system monitoring or other control 90
purposes. For parameter estimation the observer should
80
converge to minimize the error between actual and estimated
70
values, so firstly observer convergence is attained. (Fig. 4)
shows the result of observer convergence along-with 60

estimation error. We can see in (Fig. 4) the observer is

rad/sec

50
tracking the actual value quite efficiently even in the presence 40
Engine speed
Observer response
of variations in the actual signal, thus indicating the Error
30
robustness of the scheme employed for estimation. Once the
observer convergence is achieved the computed injection 20

signals of the observer are used in estimating the required 10

parameter. The estimated AFR is given in (Fig. 5) that is in 0
good agreement with the value of AFR available in the 0 50 100 150 200 250
Time (sec)
literature (Tunestal and Hedrick 2003). This estimated signal
can act as virtual sensor for AFR monitoring and control, Fig. 4. Second Order Sliding Mode Observers convergence
providing several advantages over existing techniques using
oxygen sensor as discussed earlier.

504
IFAC AAC 2013
September 4-7, 2013. Tokyo, Japan

1
20
Air Fuel Ratio
ηth (Cγ 2 −γ )(Cγ γ −1 − 1)ηc Q(1 − )
Cγ γ −1
19 imep =
(γ − 1)(Cγ − 1)CvTm
18

17
Vd
16 Tp = ( Pa − Pm )
AFR
4π
15

14 T f = 11.72 + 5.69 × 10−5 ωe + 2.33 × 10−14 ωe2

13

12 REFERENCES
11 Ahmed, Q. and Bhatti, A.I. (2010). Estimating SI engine
10
50 100 150 200 250
efficiencies and parameters in second order sliding
Time (sec) modes. Accepted in IEEE Trans. Industrial Electronics
Butt, Q.R. and Bhatti, A.I. (2008). Estimation of gasoline
Fig. 5. Estimated values of AFR engine parameters using higher order sliding mode. IEEE
Trans. Industrial Electronics, volume. (55), 3891–3898.
5. CONCLUSIONS Butt, Q.R. and Bhatti, A.I. (2009). HOSM based online
In this paper AFR estimation of automotive engine through estimation of gasoline engine parameters for MVM
HOSM is discussed. The estimation is performed to improvment. PhD Report, Centre for Advanced Studies
overcome the problems involved in using oxygen sensor to in Engineering, Pakistan
control AFR, viz. exhaust gas transport delay and bad Drakunov, S. and Law, V.J. (2007). Parameter estimation
response of oxygen sensor in cold start. This work uses using sliding mode observers: application to monod
existing setup without installing any extra component for kinetic model. Chemical product and processing
implementation of proposed estimation scheme. Using this modelling. volume. (2), article 21
estimation scheme as virtual sensor bypasses oxygen sensor Floquet, T., Twiddle, J.A. and Spurgeon, S.K. (2006).
to control AFR that overcomes the inherent transport delays. Parameter estimation via second order sliding modes
This estimated signal can also be used as feedback in cold with application to thermal modelling in a high speed
start to control AFR since oxygen sensor is not producing rotating machine. In Proc. IEEE-ICIT
correct readings in this phase. The proposed methodology is Guzzella, L, and Onder, C. (2004). Introduction to modelling
simple enough to be easily implemented online. and control of internal combustion engine systems. ETH
Zurich: Springer
APPENDIX A Jones, V.K., Ault, B.A., Franklin, G.F and Powell, J.D.
(1995). Identification and air-fuel ratio control of a spark
ignition engine . IEEE Trans. on Control. Systems
RTm Technology, volume. (3), 14–21
A1 = AE Pa Cd γ c
Vm Lee, B., Guezennec, Y.J and Rizzoni, G (2001). Estimation
of cycle-resolved in-cylinder pressure and air-fuel ratio
1 2 γ −1
γ +1 using spark plug ionization current sensing. Int J Engine
γc = γ( ) Research, volume. (2), 263–276
RTa γ +1 Levant, A. (2001). Universal SISO sliding mode controllers
with finite time convergence. IEEE Trans. Automatic
D2 α + α cl Control, volume. (46), 1447–1451.
AE = π (1 − cos( )) Levant, A. (2003). Higher-order sliding modes differentiation
4 α cl
and output feedback control. Int. J. Control, volume.
(76), 924–941
Vd McCann, R.A., Islam, M.S. and Hussain, I. (2001).
A2 =
4π Vm Application of sliding mode observer for position and
speed estimation in switched reluctance motor drives.
(9(
Pm
−1))
IEEE Trans. Industry Applications, volume. (37), 51–58.
f ( Pm ) = (1 − e Pa
) Muhammad, I. Bhatti, A. Ayubi, S. and Khan, Q. (2010).
Robust parameter estimation of nonlinear systems using
Vd × imep sliding mode differentiator observer. Accepted in IEEE
Ti = Pm Trans. Industrial Electronics
4π × AFR Powell, J.D, Fekete, N.P., and Chang, C.F. (1998). Observer
based air fuel ratio control. IEEE Control. Systems,
volume. (18), 72–83

505
IFAC AAC 2013
September 4-7, 2013. Tokyo, Japan

Rizzoni G., Draknuov S. and Wang Y.Y., “Online estimation

of indicated torque in IC engines via sliding mode
observers,” in Proc. ACC, 1996, pp 2123–2127.
Saif, M. Chen, W. and Wu, Q. (2008). High Order Sliding
Mode Observers and Differertiators- Applications to
Fault Diagnosis Problem. In M. Thoma, M. Morari (ed.),
Modern Sliding Mode Control Theory, 321–344.
Springer, Berlin, Germany
Shraim, H., Ananou, B., Fridman, L., Noura, H. and
Ouladsine, M. (2006). Sliding mode observers for
estimation of vehicle parameters, forces and states of the
centre of gravity. In Proc.45th IEEE–CDC.
Tunestal, P., and Hedrick, J.K. (2003). Cylinder air/fuel ratio
estimation using net heat release data. Control
Engineering Practice, volume. (11), 311–318
Utkin,.V. (1999). Sliding Mode Control in Electromechanical
Systems, Taylor and Francis, London, UK
Yoon, P., and Sunwoo, M. (2001). An adaptive sliding mode
controller for air-fuel ratio control of spark ignition
engines. Journal of Automobile Engineering 2001,
volume. (215), 305–315

506