Sunteți pe pagina 1din 6

ICROS-SICE International Joint Conference 2009

August 18-21, 2009, Fukuoka International Congress Center, Japan

A Design of Electromechanical Valve Control System


using Model Predictive Controller
Addy Wahyudie1 , Taizo Nakao1 , Masakazu Mukai1 and Taketoshi Kawabe1
1

Graduate School of Information Science and Electrical Engineering, Kyushu University, Fukuoka, Japan
(Tel: +81-92-802-3691; E-mail: see http://terra.ees.kyushu-u.ac.jp)

Abstract: This paper considers an implementation of model predictive controller for an electromechanical valve. At
each sampling time instants, the controller sends optimum control input to the system. In order to simplify the calculation process within controller, we solve the control input as algebraic calculation rather than two-point boundary value
problem. We also use a disturbance observer to provide information of armature valve velocity, and to improve EMV
robustness. The proposed control scheme is demonstrated on experimental testbed. The performance model predictive
controller with a disturbance observer is compared experimentally to model predictive controller with a regular full order
observer.
Keywords: Electromechanical valve train control, Optimal control, Robust control, Model Predictive Control, Automotive engine control.

1. INTRODUCTION

abling it to move at high speed. In open valve operation, the current of the at the right-side electromagnet is
stopped, which make armature moves to the left electromagnet by spring power. When armature moves close to
the left electromagnet, current ows this electromagnet,
so that pull control force is available. The closing valve
operation, the electromagnets system works conversely.
The springs are attached to make the armature balanced
at middle position. This motion of engine valve can be
use a conventional way by the application of voice coil.
However, the conventional way needs a large amount of
power, that will enlarge the size of engine valve.
The armature moves a short distance (9 mm) over a
short amount of time (3 msec or 4 msec). This short
amount of time is desirable in order to meet maximum
engine speeds of 5000 6000 rpm. Impact/collision velocity between armature to magnet surface also one of
big issue. Impact velocity should be controlled less than
0.1 m/s for maintaining acceptable engine acoustic noise
level, and ensuring valve seating and wear requirements
are met. However, there are many difculties of controlling EMV, some of them are follows:
1. Nonlinearity of electromagnet characteristic.
2. The push-pull control is not available from the electromagnets.
3. Magnetic force is only effective when the armature is
very close to the pulling / catching magnet.
4. The delay of response of electromagnet.
5. The valve should be open/closed within the desired
time in synchronization with the engine revolution.
6. Control input should be calculated over a short amount
of time.
In this paper we will implement Model Predictive
Controller (EMV) [6] to control laboratory-scale EMV
system. This approach is pioneer on the eld, in term of
controlling EMV by using the advantages of MPC control
strategy. This solution approach is quite challenging, because we have very short time control time to control the
EMV. The simplicity of control strategy is the important

Electromechanical valve (EMV) actuator is studied in


recent years [1-4]. It has many performance benets over
conventional camshaft engines [5]. This actuator can potentially improve engine performance via exibility in
valve timings at all engine conditions. Unlike conventional camshaft driven systems, EMV system afford valve
timings that are fully independent of crank-shanks position. The additional exibility in valve timing gives excellent cycle-to-cycle control of cylinder air charge and
residual gas fraction. Fuel economy can be improved
through un-throttled load control and cylinder deactivation. Internal residuals together with appropriate valve
actuation strategy can be used to lower exhaust emission
below engines with a camshaft. At low-to-moderate engine speeds, valve timings can be optimized to improved
full load torque.
Construction of EMV is shown in Fig. (1). The EMV
valve actuator consists of two opposing electromagnets
and an armature suspend between them. The armature
travel between the two electromagnets. The armature
is suspended by the springs with large spring rate, en-

Fig. 1 EMV control system conguration.

- 3559 -

PR0002/09/0000-3559 400 2009 SICE

Authorized licensed use limited to: Universidad Nacional de Colombia. Downloaded on April 19,2010 at 02:22:51 UTC from IEEE Xplore. Restrictions apply.

the following performance function

1 tf 2
u (t)dt,
J=
2 t0

MPC

v
OBSERVER

subject to state equation (2). Here, t0 and tf is the initial


time and the terminal time. The initial state x(t0 ) and
terminal state x(tf ) are dened as follows:

z0
zf
, x(tf ) =
,
x(t0 ) =
z0
0

EMV

Fig. 2 EMV control system conguration.


key so that its can be implemented in real time (avoiding overrun process). Based on optimal control policy,
MPC generates sequence optimal control law, and update
it in each sample instants. The previous simulation results of MPC to control EMV is presented in [7]. Paper
[8] presents output feedback MPC for control EMV system. We also use a disturbance observer to provide information of armature valve velocity and to improve EMV
robustness due to un-precise model dynamics and noise.
This paper is organized as follows. Section 2 presents
a mathematical model of EMV system. Section 3, provides detail of control strategy and the structure of observers. System identication of EMV model and computer simulation results, are given in Section 4. Section 5
present experimental setup and results. Finally, Section 6
provides conclusions.

where z0 is the initial position of the armature, z0 is initial velocity, and zf is the desired position of armature at
terminal time.
The necessary condition of our optimal control problem are given in the following equation.

1 = k 2 ,
m
(3)
2 = 1 + c 2 ,
m
2
= 0,
(4)
u+
m
x = f.
1 and 2 are costate variables that can be found by
solving co-state equation (3), as the following.

1 = A1 e1 t + A2 e2 t ,
m
m
2 = 1 = (A1 1 e1 t + A2 2 e2 t ).
k
k

2. ELECTROMECHANICAL VALVE
MODEL
The mathematical model of the EMV system can be
represented as simple single mass-spring-damper system,
written in following equation.
m
z + cz + kz = u + d.

A1 and A2 are constants, decided base on initial condition


of 1 and 2 . Moreover, 1 and 2 are the following
constants.

c + c2 4mk
c c2 4mk
1 =
, 2 =
.
2m
2m

(1)

Here m, k, and c represent total mass, spring constant


and viscosity coefcient of friction, respectively. z denotes the position of armature. The control input u is
electromagnetic force produce by pulling magnet, and d
is disturbance of the EMV system. Equation (1) can be
transform in state space equation, as equation (2).
x = Ax + B(u + d)

0
z
x=
, A=
k
z
m

1
c
m

, B=

0
1
m

The optimal input u can be calculated by substitute 2


into equation (4), written as
u =

1
2
= (A1 1 e1 t + A2 2 e2 t ).
m
k

(5)

Constants A1 and A2 are obtained by the following


procedure. The armature trajectory z is derived by using
system equation (1) and optimal control input in equation (5), as follows.

(2)

z = B1 e1 t + B2 e2 t + A1 e1 t + A2 e2 t .

3. CONTROLLER DESIGN

(6)

Here,

c2 4mk
.
2m

The control system conguration for EMV is depicted


in Fig. 2. Each elements of controller are describe in the
following sections.

1 =

3.1 Model Predictive Controller

The expressions B1 and B2 are calculated later. The relation among A1 , A2 , A1 , and A2 are satised the follow.

At each time sampling instant, MPC sends optimum


control input to the EMV system for the next sampling
instant. In order to obtain an optimal control input for
system, we use optimal control design with xed terminal
state and xed terminal time. The objective is minimize

c +

c2 4mk
,
2m

A1 =

2 =

1 A1
1
c
2
km 1 + m
1 +

A2 =

2 A2
1
c
km 22 + m
2 +

1
A1 ,
2kc

k
m

k
m

1
A2 .
2kc

- 3560 Authorized licensed use limited to: Universidad Nacional de Colombia. Downloaded on April 19,2010 at 02:22:51 UTC from IEEE Xplore. Restrictions apply.

Table 1 Experiment parameters of the EMV

MPC output Force (N)

Parameter

Value

Armature mass (m)

0.285

Kg

Spring coefcient (k)

1330

N/m

Viscous friction coefcient (c)

1.6239

N sec/m

MPC sampling time

1.1 104

Armature position when control start

Larger

mm

Armature position when control end

3.900

mm

Armature target position

3.980

mm

Magnet force constant (k1 )

3.59

Magnet force constant (k2 )

1.3779

Final catching current

Final time

0.058075

sec

2

Smaller

0.005

0.01

0.015

0.02

0.025
0.03
Time (sec)

0.035

0.04

0.045

sec

2.980

tf

4
0

Unit

0.05

Fig. 3 The relationship between tf and u.


3.3 Full order observer
The control system conguration using full order observer is the same with general conguration of EMV
system, shown in Fig. (2). The observer structure is in
the form of following equation.

The derivative of equation (6) will give us,


z = B1 1 e1 t + B2 2 e2 t + A1 1 e1 t + A2 2 e2 t .
Bring all results, we can have the following matrix
equation

= (A LC)
x
x + L(y y) + Bu,
C = [ 1 0 ].

T w = [ z0 z0 zf 0 ] .

The gain L is designed by using pole placement procedure. The real parts of closed loop poles should be designed large enough to make velocity estimation faster
than the response of the MPC.

Where,
T

w = [ B1 B2 A1 A2 ] ,

e1 t0
e2 t0
e1 t0

t
1 e 1 0 2 e 2 0 1 e1 t0
T =
e1 tf
e2 tf
e1 tf
1 tf
2 tf
1 e
2 e
1 e1 tf

with

e2 t0
2 e2 t0
.
e2 tf
2 e2 tf

3.4 Disturbance Observer


In order to provide system robustness (i.e., due to uncertain model parameter or disturbance from outside),
we design a disturbance observer. EMV control system
conguration with a disturbance observer, is depicted in
Fig. (4). The disturbance observer structure is given by
the following equation.

Initial time t0 , terminal time tf , and terminal state zf are


xed, matrix T is not zero, and T 1 exist. Now, unknown
parameter in w can be obtained by solving

= (A LC )
x
x + L (y y) + B u.

w = T 1 [ z0 z0 zf 0 ] .

Here,

As the result, optimum control u can be decided


uniquely.
From numerical simulation result in Fig. (3), we
found important fact that: small predictive horizon makes
u negative between t0 and tf . A large predictive horizon
makes make u always positive between t0 and tf . According to the relationship, the following presumption is
obtained. When u is negative it is possible to make u
positive by enlarging tf . In the application, the calculation of tf that ensure u always positive, can be vary in
each time step by using online calculation; or kept xed
in each time instants, while the value of tf decided before
by using computer simulation information.

A =
x
=

A
0

B
0

z z d

B =

B
0

C = [ C 0 ] ,

y = C x
.

4. COMPUTER SIMULATION RESULTS


We conducted system identication in order to obtained EMV parameters. Using the obtained parameu

MPC

3.2 Observer
We use an disturbance observer to obtain information
the velocity (z or v) of the EMV armature. We will use
two type observers, i.e., full order observer and disturbance observer.

DISTURBANCE
OBSERVER

u'

EMV

Fig. 4 EMV control system with disturbance observer.

- 3561 Authorized licensed use limited to: Universidad Nacional de Colombia. Downloaded on April 19,2010 at 02:22:51 UTC from IEEE Xplore. Restrictions apply.

ters, we conduct simulation and experiment for the EMV


setup.

4.1 EMV parameter identication


The EMV model parameters are obtained by system
identication. Armature position for free oscillation are
experimentally measured. We only use the rst swing
of armature position (as shown in Fig.(5)), because we
only control on this region. Then, we use a least square
method to estimate the value of m, k, and c. The results
of EMV parameter identication, are given in Table 1.

4.2 Magnet static force characteristic experiment


We model the pulling magnet static force as

k1 i2
.
F =
2(x + k2 )2

Fig. 6 Static pulling magnet force characteristic.

Here, i is current that apply to the the magnet, and x is


the distant of armature from the pulling magnet. The parameters k1 and k2 is obtained by conduct least squares
procedure to the experiment data, as depicted in Fig.(6).
The obtained parameters are given in Table 1.

to ensure low impact velocity between the armature and


pulling magnet. We can see that, our approach ideally can
reduce impact velocity very near to zero m/sec. If we select tf too small, it not just make control input from MPC
become negative, but also the system require a larger control input compares with ideal condition. In other hand, if
we select tf too large, it will make velocity become even
larger when MPC starts to work, compare with ideal condition or in condition where the armature swing without
any control action. Therefore, selection process of value
of tf , is involving a consideration positiveness and how
large control input that we can afford.

4.3 Various condition of simulation results


In this section we explains the various conditions that
we can obtain in simulation results. We only control
the EMV within last one millimeter of valve-train. We
xed terminal time tf so that control input of EMV is
always positive between t0 and tf . Searching optimal
tf automatically in each sample time in online process,
can make system overrun due the heavy calculation of
it. When MPC stop working, nal current sends to the
pulling magnet to hold the armature. The various simulation results are depicted in Fig. (7). Here, the information
MPC output (u),
of position (z), observed velocity (z),
shown from up to bottom. In an ideal condition (i.e., a
good choice of tf and good modeling parameter), when
MPC start control action, velocity of armature become
decrease. As the EMV armature travel more closely to
desired target position, control input decrease its value

5. EXPERIMENTAL SETUP
The experiment setup is shown in Fig. (8). A realtime implementation of control law is coded in C by using dSPACE DS1104 control board. The experiment parameters are given in Table 1. Within sampling instants,
the control board reads armature position as analog in calculate the
put, give prediction of armature velocity (z),
MPC algorithm, and send the control input to EMV system. We xed tf as stated in Table 1. tf decided by using
simulation results to ensure optimal control command al-

3

x 10

5
Position(m)

Ideal tf
Too small tf
Too large tf

5
0
Velocity(m/sec)

0.01

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03

0.04

0.05

0.06

5
0

0.01

0.02

0.03
Time(sec)

0.04

0.05

0.06

0.4

0.2

Force(N)

10

5
0

Fig. 7 Various simulation condition results for EMV via


MPC based on selection of tf .

Fig. 5 Spring motion for EMV parameter identication.

- 3562 Authorized licensed use limited to: Universidad Nacional de Colombia. Downloaded on April 19,2010 at 02:22:51 UTC from IEEE Xplore. Restrictions apply.

ten times of real part of eigenvalues matrix A. We can


see control input u and current input (information from
current sensor) to the magnet, are kept positive by MPC.
From magnet input information, we can see that there is
a lag between response of magnet and a commanding input from MPC. This phenomenon has been remark on the
Section 1. Impact velocity of EMV armature is successfully kept low by MPC.
More superior result is shown in Fig. (11). This gure is experimental result of MPC with disturbance observer. The information of position (z), observed veloc MPC output (u), MPC total force (u ) i.e., comity (z),
bination of u and disturbance (d ), and input current, are
given from top to bottom on Fig. (10). The disturbance
observer compensate the unidentied dynamic and noise
that occur in EMV system. Disturbance observer gains
L are designed so that desired closed loop poles of observer are 2p
0.09, 2p 0.091, and 2 0.092,
where p = k/m. More lower impact velocity is obtained by using this control conguration. Comparison
of velocities for EMV system controlled by MPC with
full order observer and MPC with disturbance observer,
as well as armature velocity when there is no control action (just apply 4 A current to catch the armature), is depicted in Fig. (12). The quantities of impact velocities of
these three condition is given on Table 2.
Forty-nine experimental tests are conducted using
MPC with disturbance observer in order to verify its ability to repeatedly maintain low impact velocity. An average impact velocity of 0.0233 m/s with a standard deviation of 0.0075 m/s is obtained, as depicted in Fig. (13).
A small discrepancy that appears in standard deviation,
come from the following sources. First, there is discrepancy of desired target position zf and the real target position, due to un-uniformly of pulling magnet force in each
simulation cycle. Second, our pulling magnet consists
two magnet which connected paralelly. This type of construction give the advantage in term of small total static
impedance value, but produce un-uniformly magnet force
due to un-identical these magnets. Because of this unidentical pulling magnets, whole armatures side is not

Position Sensor
EMV System

Current Sensor

Audio Amplifier
Power Supply
DSpace System

Fig. 8 EMV experimental device setup.


ways positive between t0 and tf .
The control input send by control board is amplify by
using audio amplier. We use high-performance LM3886
audio amplier. We design the amplier so that it has approximately 0.6 Mhz bandwidth gain and sufcient DC
gain. The DC gain is used to amplify the nal pulling
force. The pulling force is provided by two parallel
hand made electromagnets, with total static resistance is
4 Ohm.
We use a hand made DC power supply, which can provide us 50V / 5A supply for our audio amplier. Armature displacement are measured by an eddy current position sensor (EX-422V). A HCS-20-AP device is used
for current sensor. Detail of EMV system is shown in
Fig. (9).
5.1 Experimental result
Experimental result for EMV control via MPC with
full order observer, is shown in Fig. (10). The informa MPC output
tion of position (z), observed velocity (z),
(u), and input current, are given from top to bottom on
Fig. (10). Here, the observer gains L are designed so
that desired closed loop observer poles are approximately
140 mm

Position sensor

3

9 mm
Pos. (m)

Armature
shaft

x 10

2.5
0
2.5
5
0

EMV armature

65 mm

0.01

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03
Time (sec)

0.04

0.05

0.06

0.4
0.3
0.2
0.1
0
8

Pushing
magnet
Pulling
magnet

u (N)

Lower
spring

Vel. (m/sec)

Upper
spring

6
4
2
0

Input (A)

4
3
2
1
0

Fig. 10 Experiment results for EMV control via MPC


with full order observer.

Fig. 9 Electromechanical valve system.

- 3563 Authorized licensed use limited to: Universidad Nacional de Colombia. Downloaded on April 19,2010 at 02:22:51 UTC from IEEE Xplore. Restrictions apply.

Table 2 Impact velocities of various control


conguration
0.01

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03

0.04

0.05

0.06

u (N)

Vel. (m/sec)

Pos. (m)

3

x 10
5
2.5
0
2.5
5
0
0.4
0.3
0.2
0.1
0
0
6
4
2
0
0
6
4
2
0
0
0
0.2
0.4
0.6
0
4
3
2
1
0
0

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03

0.04

0.05

0.06

0.01

0.02

0.03
Time (sec)

0.04

0.05

0.06

Impact velocity (m/sec)

MPC with disturbance observer

0.0233

MPC with full order observer

0.0584

No control

0.5098

6. CONCLUSION

Input (A)

d (N)

u (N)

0.01

Control method

Fig. 11 Experiment results for EMV control via MPC


with disturbance observer.
touch pulling magnet at same time. A better production
of pulling magnet will be used in future experiment, including a single circular shape pulling magnet, will ensure better experiment result. However, using present
pulling magnet, Fig. (13) shows that how our control
strategy consistently provide soft landing for the EMV
armature.

REFERENCES

0.7
With disturbance observer
With full order observer
Without control

0.6

Velocity (m/sec)

0.5

0.4

0.3

0.2

0.1

0.01

0.02

0.03
Time (sec)

0.04

0.05

0.06

Fig. 12 Experiment results comparison for impact velocities under various control conguration.
18
Mean Velocity = 0.0233 (m/sec)

16

Standar Deviation = 0.0075 (m/sec)


14

Frequency (m/sec)

12

10

0.005

0.01
0.015
0.02
0.025
The EMV Armatures Impact Velocity (m/sec)

0.03

The implementation of MPC for an EMV has been


present. Sequence of optimum control input sent by MPC
in every sample instants. In order to get estimation velocity and improve robustness performance, we proposed
a disturbance observer to be attach with MPC. Experiment results shows that the MPC with a disturbance observer, give superior performance in term of impact velocity, compared with MPC with regular full order observer. Consistent low impact performance is demonstrated for the proposed control. The MPC with can keep
impact low impact velocity with a very narrow standard
deviation.

[1] T. Kawabe, Initial condition-adaptive Robust Control for a High-speed Magnetic Actuator, Control
Engineering Practice, vol. 11, no.6, pp. 675685,
2003.
[2] S. K. Chung, et. al, Flatness-Based Feedback Control of an Automotive Solenoid Valve, IEEE Journal
of Contr. Sys. Technology, Vol. 15, No. 2, pp. 394
401, 2007.
[3] K. Uchida, et. al, Sliding Mode Servo Control
for Electromagnetic Engine Valve, Proc. of SICEICASE Int. Joint Conf. 2006, pp. 36583663, 2006.
[4] M. Montanari, et. al, Control of Camless Engine
Electromechanical Actuator: Position Reconstruction and Dynamic Performance Analysis, IEEE
Trans. on Industrial Electronics, Vol. 51, No. 2, pp.
299311, 2004.
[5] K. Yan, et. al, Modeling and Control of Electromechanical Valve Actuator, SAE 2002-01-1106, 2002.
[6] D. Q. Mayne, et. al,Constrained Model Predictive Control: Stability and Optimality, Automatica,
vol.36, no.6, pp. 790814, 2000.
[7] M. Mukai, et. al, A Model Predictive Control
Method for a High-speed Magnetic Actuator, Proc.
in Chinese Control Conference 2007, pp. 623626,
2007.
[8] M. Mukai, et. al, Output Feedback Model Predictive Control for a Electromechanical Valve Actuator, Proc. in SICE Annual Conf. 2007, pp. 1734
1738, 2007.

0.035

Fig. 13 Distribution of armature impact velocity for


49 cycle experiments via MPC with disturbance observer.

- 3564 Authorized licensed use limited to: Universidad Nacional de Colombia. Downloaded on April 19,2010 at 02:22:51 UTC from IEEE Xplore. Restrictions apply.

S-ar putea să vă placă și