2012-01-0162

Published 04/16/2012
Copyright © 2012 SAE International
doi:10.4271/2012-01-0162
saepcelec.saejournals.org

Iterative Learning Control for a Fully Flexible Valve Actuation

in a Test Cell
Hai Wu, Jyh-Shin Chen, Meng-Feng Li and Russell P. Durrett
General Motors Company

Wen Chen
Wayne State University

Kevin L. Moore
Colorado School of Mines

ABSTRACT
An iterative learning control (ILC) algorithm has been developed for a test cell electro-hydraulic, fully flexible valve
actuation system to track valve lift profile under steady-state and transient operation. A dynamic model of the plant was
obtained from experimental data to design and verify the ILC algorithm. The ILC is implemented in a prototype controller.
The learned control input for two different lift profiles can be used for engine transient tests. Simulation and bench test are
conducted to verify the effectiveness and robustness of this approach. The simple structure of the ILC in implementation
and low cost in computation are other crucial factors to recommend the ILC. It does not totally depend on the system
model during the design procedure. Therefore, it has relatively higher robustness to perturbation and modeling errors than
other control methods for repetitive tasks.

CITATION: Wu, H., Chen, J., Li, M., Durrett, R. et al., "Iterative Learning Control for a Fully Flexible Valve Actuation in a
Test Cell," SAE Int. J. Passeng. Cars - Electron. Electr. Syst. 5(1):2012, doi:10.4271/2012-01-0162.
____________________________________

INTRODUCTION structure introduces the difficulties in seating speed control

because of the nonlinearity. The electro-hydraulic systems,
Fully flexible valve actuation (FFVA) is a key technology such as the Ford ‘camless’ system [6], the Sturman system
of next generation internal combustion engines. It is [7] and the Lotus system [8] also provide fully flexible valve
beneficial to fuel economy, emission control and drivability event control. The electro-hydraulic systems can achieve
over the fixed geometry of a camshaft. It can also simplify higher lift control accuracy; but their drawback is the energy
engine structure by eliminating the throttle and enable consumption and structure complexity. A new VVA system
advanced combustion auto-ignited (CAI) engines. using a stiff hydro-mechanical internal-feedback mechanism
Mechanical, electro-mechanical and electro-hydraulic is proposed in [9]. A built-in physical feedback is designed,
systems have been introduced to achieve fully flexible valve which is activated and deactivated by two on/off valves. The
actuation. BMW VALVETRONIC, for example, is a concise structure can reduce the cost and it was claimed to be
mechanical fully variable valve control system which was feasible for a mass production engine. In this paper, an FFVA
introduced to BMW's range of spark-ignition (SI) engines [1]. system for a test cell is the focus. The main purpose of this
It controls the load from idle to full load instead of using a system is to achieve accurate and repeatable valve lift for
throttle. Besides improvement of fuel consumption and both steady-state and transients. An electro-hydraulic
emission, it improves power output, engine speed capability, structure is adopted to guarantee the accuracy and
response, and economy. The cost, complexity, and durability repeatability.
of mechanical VVA, however, are not favorable. Electro- VVA control design has become a very active research
mechanical variable valve systems, such as the GM R&D area in the past few years for the development of both
electromagnetic valve actuators [2], FEV [3], Siemens [4], laboratory systems [10, 11, 12, 13] and production engine
and Aura [5] are implemented by electromagnetic mechanism systems [5, 8, 14, 15, 16]. From the frequency point of view,
with relatively low cost. Their simple ‘catch and hold’
 Wu et al / SAE Int. J. Passeng. Cars - Electron. Electr. Syst. / Volume 5, Issue 1(May β01β)

perfect tracking requires the system to have a wide bandwidth

corresponding to the spectrum of desired lift, and the
bandwidth must be traded off with the noise and disturbance
rejection capability. Because of the nonlinearity, non-
minimum phase and time delay characteristic of practical
electro-mechanical and fluid systems, system inverse-based
tracking control is not applicable and generic PID control
cannot meet the design criteria [17]. Usually, a feedback
controller is designed to stabilize the system, and then a feed-
forward compensator is designed to improve the tracking
performance. Model reference control [18], repetitive control
[17, 19] and iterative learning control [20] are employed to
utilize input information and the engine cyclic characteristics
to achieve better tracking performance. Among them,
repetitive control has been extensively investigated for
laboratory VVA system [17]. With a periodic signal Figure 1. Typical lift profiles of intake and exhaust valve
generator, it can asymptotically track a periodic reference
signal with high accuracy based on the internal model
principle. Perfect tracking is reported from experimental data. Test Bench Setup
But repetitive control is a model based method. The
controller structure and computation cost limits its robustness A test bench is set up for the FVVA system analysis,
to modeling error and need powerful controller. control design and verification purpose. The structure of the
In this research, a new iterative learning control algorithm FFVA is summarized in Figure 2. The system consists of
is proposed for FFVA control. The objective is to loosen the hydraulic components, actuator components and an electronic
constraints to reference signal so that we can apply it to a controller, which is implemented in a dSPACE DS1103
wide range of engine speed and transient situation. The control board. A servo current amplifier drives the torque
robustness and performance are investigated with simulation motor of a two-stage servo valve to regulate the fluid flow,
and hardware-in-loop test. which is proportional to spool displacement. The pressure
difference between the two sides of the actuator cylinder
SYSTEM DESIGN output force movement of the piston of the hydraulic actuator
against the valve spring to accomplish valve lift according to
System Design Requirements the command profile. The piston rod of the actuator is
The FFVA controller is required to accurately track lift separate from valve stem. It has a linear variable
profiles. The operation covers engine speed from idle up to displacement transducer (LVDT) mounted on the other end.
3000 rpm. Considering the valve wear and noise, the seating The low voltage conditioner (LVC) provides low distortion
speed needs to be limited to a given level. Valve opening and sine wave excitation for the LVDT and employs a
closing timing is controlled because of its importance in synchronous demodulator to convert the LVDT's AC output
combustion investigation. Safety functions, such as collision signal to a DC signal proportional to core position. This
avoidance between valves and piston, and emergency stop position signal is fed back to the amplifier and system
when tracking error exceeds tolerance limits also need to be controller to be compared with the command profile.
implemented. Meanwhile, the engine encoder generates crank angle signal
The valve actuation system consists of electro-hydraulic to synchronize the valve lift with movement of the engine
servo valves, cylinders and two intake and two exhaust cylinder piston. During the operation, the hydraulic actuator
valves. The actuator system moves the intake and exhaust pushes the valve downward against valve the spring to open
valves from initial closed position to final closed position. the valve. The energy is stored in spring and later used to
The valve periodically follows the reference signal in push the valve back just like the situation in a mechanical
duration and height based on crank angle. The lifting profile cam. The actuator needs to control the seating speed to avoid
is the same for every intake and exhaust strokes at steady damage and reduce the wear and noise. Once the valve is
engine operating points. A typical valve lift profiles is shown seated, the piston rod and valve stem can be separated to
in Figure 1. During load transient, the valve lifting profile ensure the valve closure. The impact between actuators and
changes according to combustion requirement. valves stem should be also considered. The system has an
internal position feedback loop closed at the servo amplifier
to stabilize the system, minimize the initial condition
difference and sensitivity to different mechanical offsets,
which are the necessary conditions to utilize the ILC.
 Wu et al / SAE Int. J. Passeng. Cars - Electron. Electr. Syst. / Volume 5, Issue 1(May β01β)

(γ)

(4)
From the zeros and poles location indicated in Figure 4,
Figure 2. Schematics of FFVA system of test bench the FFVA system is a non-minimum phase system with a
delay of 6 samples times.

Iterative Learning Control for the FFVA

System Modeling
An iterative learning control (ILC) is chosen for FFVA
To obtain the system model of the FFVA, the frequency tracking because the system is required to repeat the same
response is measured using a swept sine signal. The system operation over a finite time interval (crank angle duration).
Bode diagram is plotted in Figure 3. The frequency response For every individual cycle, the valve starts from the same
shows jittering behavior of the system beyond 500 Hz (3140 initial position, which is the closed condition and ends with
rad/s). The beginning of the fluctuation in both the magnitude the closed condition, again. The same exact sequence of lift
and phase diagram is from the spring resonance, as the spring height and opening/closing time must be followed on each
has its lowest resonance frequency around 280 - 460Hz (1750 cycle during engine steady-state operation. In both laboratory
- 2900 rad/s). The following jump part may be from the and production engines, steady-state operation dominates the
separation between valve stem and actuator rod. operating condition, accompanying certain transients between
two different conditions

Figure 3. FFVA system frequency response

Figure 4. Pole-zero plot of FFVA
System identification is carried out with an autoregressive
moving average (ARX) model The learning and memory capability of the ILC can be
utilized to improve the tracking performance while running.
After a single execution of the lift control is completed (off-
(1)
line or online during calibration), the input control action and
where q is the standard shift operator. A system order of five, output response produced are available to update the control
with a pure time delay of six samples was chosen at a input for the next cycle and thereby sequentially improve the
sampling rate of the 3000 Hz, which is chosen to have 30 tracking performance. Once the satisfied control sequences
samples during system rising time of 0.01s. The system are learned, they can be stored for later recall under the same
parameters are listed in Equations (2) and (3). The condition. Two different lift profiles can be used later for fast
corresponding discrete transfer function model is given in transient control.
Equation (4) and has a good fit with the original frequency The simple structure of ILC in implementation and low
response measurement data, as shown in Figure 3. cost in computation are other crucial factors to recommend
the ILC. It does not totally depend on the system model
during design procedure. Therefore, it has relatively higher
(β) robustness to perturbation in signal and modeling errors than
 Wu et al / SAE Int. J. Passeng. Cars - Electron. Electr. Syst. / Volume 5, Issue 1(May β01β)

that of other control methods. The ILC is proved to be an

effective method in control theory [21] and practical
application [22].
Consider a given discrete-time, linear time-invariant
(LTI), single input single output (SISO) system

Figure 5. Over-damping system of FFVA

(5)
where k is the time index, i is the iteration index, q is the The over-damped new system model becomes
forward time-shift operator qx(k) ≡ x(k + 1), yi is the output,
ui is the control input, and d is an exogenous signal that
repeats during each iteration. Repetitive disturbances, like the
friction which is hard to model, can be captured in d. The
plant G(q) is a proper rational function of q and has delay, or
(8)
equivalently relative degree of m. The plant needs to be
asymptotically stable. Consider the N sample sequences of The comparison between the two system's step responses
inputs and outputs is illustrated in Figure 6. The zeros and poles locations do not
change much with a few more stable poles being added, as
shown in Figure 7, by this approach.

Two-step ILC with Forgetting Factors

Adding a forgetting factor into the ILC algorithm, as
shown in equation (9) below, was proposed by
and the desired lift output Heinzinger[23], Arimoto[24], and Freeman[25, 26], among
others. The use of a forgetting factor is found to be successful
to overcome the problem of instability at the expense of
increased steady-state error. Based on the error analysis
with the performance or output-tracking error is defined by observed above when using a PD ILC, a correcting term is
introduced to improve the performance in a two-step design.

(6)
For such a systems a PD-type ILC algorithm has the form

(7)
where 1,2 is the learning gain. Considering the system delay
or relative degree, u(k) has effects on system output after m
step. Therefore, ui+1 (k) is corrected from previous learned
ui(k) with error ei(k + m) and its derivative. When a PD-type
ILC is applied to the system model (4), the valve can track
the lift profile in the short term, but becomes unstable over a
long period. The system character of non-minimum phase Figure 6. FFVA system after damping introduced
and time delay might cause the instability of general ILC
algorithm. Safety is critical in valve control, and no lift
overshoot is permissible in engine operation. A robust and
monotonically convergent ILC algorithm is preferred for the
FFVA system.
It is noticed that the overshoot of the dynamic behavior
makes the system more sensitive and usually deteriorates the
tracking performance. A positive feed-back is first applied to
the original system to increase damping and reduce the
overshoot, as shown in Figure 5. The gain is k = 0.45.
 Wu et al / SAE Int. J. Passeng. Cars - Electron. Electr. Syst. / Volume 5, Issue 1(May β01β)

(14)

Step 2:
Therefore, the ideal ILC control should be in form of

(15)

where both and (k) can be obtained online or offline from

the previous step.
One beneficial fact from this ILC algorithm is that for
proportional tracking the profile yd1 = ϕyd, the corresponding
Figure 7. Zeros/poles diagram of FFVA after damping
ideal control can be obtained from re-setting
The proposed ILC can be designed in the following steps. in controller (15) directly. Then, the
Step 1: first step can be saved; the optimal control will be
An ILC with forgetting factor α is designed as follow

(16)
(9)
This inheriting feature is very useful for FFVA lift
Let u*(k) represent the control sequence as yi(k) → yd. control. The learning time can be saved for other similar
ėi(k) = 0, when ui(k) converges. But the ILC in (8) can only profiles.
obtain suboptimal control (k) instead of u*(k).
SIMULATION RESULTS
The new ILC is applied to the damped valve's model for
(10) verification. The most critical intake lift profile of 8mm at
3000 rpm is used as the tracking command. The simulation
Substituting (6) into (10) gives results are shown in Figure 8. The desired tracking profile is
indicated (yellow solid line). The system tracking profile
increases monotonically from 30 iterations (black solid line)
to 50 iterations (rid solid line) and close to the target at the
(11) 80th iteration (blue dot line). After 100 cycles tracking can
It is worth noting that once the ILC (9) stabilizes the reach the target lift and stay there, in Figure 8 and Figure 9.
tracking, the suboptimal control is proportional to yd in The 5000th iteration gives the same tracking profile. The 100
iterations in engine operation corresponds to 4s to 12s for
Equation (11) and the steady-state error has the following
3000rpm to1000 rpm. The simulation results illustrate that
form
the new ILC algorithm stabilizes the tracking and has
reasonable accuracy in steady state. At present, the learning
speed is the issue that needs to be improved. Once the
optimal control is obtained, it can be stored for later
(1β) application and transient control purposes.
A half (4mm) height lift tracking simulation result by
The expected system output can be expressed as using 0.5 in the controller (16) directly is illustrated in
Figure 9. The satisfactory tracking was obtained after 80 -
100 iterations. The forgetting factor, learning gains, and
constant correction term are given as α = 0.96, 1 = 0.009, 2
(1γ) = 0.07, and = 0.121 respectively. Note that in the results
presented, the advance by m steps shown in the ILC
From principle of superposition of LTI systems, the algorithms above was not used. Future results will
corresponding control that produces the second term of (13) incorporate this time advance.
should be
 Wu et al / SAE Int. J. Passeng. Cars - Electron. Electr. Syst. / Volume 5, Issue 1(May β01β)

error is exaggerated after valve closure because the actuator

is separated from the valve. The actuation system dynamic
changes after the separation. In the experiment, the ILC has
moderate learning speed of 3-7s to reach steady state. It is
noticed that the tracking trajectory has a delay at the top half
part. This might be caused by the introduced over-damping
effect. Also note that in the experiments results, the advance
by m steps shown in the ILC algorithms above was not used.
This leads to the limited performance under high engine
speed. They will be dealt with in further testing.

Figure 8. Simulation verification with new ILC

Figure 10. Experiment result of ILC with forgetting

factor under 2600 rpm engine speed

Figure 9. Simulation of half lift tracking using 0.5

from previous case

EXPERIMENT RESULTS
The ILC with forgetting factor is implemented in a
dSPACE DS1103 board to control a hydraulic valve actuation
test bench system. The experiment sampling rate is 10000 Hz
to have more control for high engine speed because of the
low computation cost of ILC algorithm. The hydraulic system
pressure is 3000 psi. A series of tests are carried out to verify
the controller stability and performance from 1000 to 3000
rpm. The test result with lift height of 3.5 mm, duration of
16ms (2600 rpm) is shown in Figure 10. The dotted line Figure 11. Experiment result of tracking speed of intake
indicates the desired valve lift profile. The negative 0.15 mm valve (opening, topping, closing points inside dots
traveling of actuator is used to guarantee the valve closure circles)
simulating the valve lash in a mechanical lobe mechanism.
The valve accelerates following the speed requirement, as
shown in Figure 11, and then decelerates to the top of profile. CONCLUSIONS
After acceleration, the actuator decelerates before closing at The ILC algorithm is an effective candidate for FFVA
16 ms. The seating speed is one key parameter to evaluate the tracking control even though the system has the
valve actuation system. The lifting speed follows the desired characteristics of non-minimum phase and several sampling
profile and has an error about 5 cm/s at dash circled opening, delays. The damping feature introduced into the system can
peak and closing points, shown in Figure 11. The tracking help improve the tracking performance and stability. The new
 Wu et al / SAE Int. J. Passeng. Cars - Electron. Electr. Syst. / Volume 5, Issue 1(May β01β)

ILC algorithm with a forgetting factor provides satisfactory 21. Ahn, H. S., Moore, K. L., and Chen, Y. Q., “Iterative Learning Control:
Robustness and Monotonic Convergence for Interval Systems,”
tracking performance with guaranteed stability. The Springer, New York, ISBN 3-540-40173-3, 2003.
controller can monotonically track the desired profile without 22. Bien, Z., and Xu, J. X., “Iterative Learning Control- Analysis, Design,
Integration and Application,” Kluwer Academic Publisher, Boston,
exceeding tracking limits and convergence is monotonic. The ISBN 978-0-7923-8213-3, 1998.
beneficial feature of the controller's learning capability is 23. Heinzinger, G., Fenwick, D., Paden, B., and Miyazaki, F., “Robust
Learning Control,” presented at Decision and Control, December 13-15,
very useful for lift control situation. The learned tracking- Tampa, FL, USA, 1989
control commands are available for later fast transient 24. Arimoto, S., Naniwa, T., and Suzuki, H., “Robustness of P-type
Learning Control with a Forgetting Factor for Robotic Motions,”
operation; the transient performance is much improved. The presented at Decision and Control 1990, Honolulu, HI, USA, Dec 5-7,
learning while running character of the ILC makes it easier in 1990.
25. Freeman, C. T., Lewin, P. L., and Rogers, E., “Experimental Evaluation
calibration and implementation. of Iterative Learning Control Algorithms for Non-minimum Phase
Plants.” International Journal of Control 78(11): 826-846, 2005.
REFERENCES 26. Freeman, C. T., Lewin, P. L., and Rogers, E., “Further Results on the
Experimental Evaluation of Iterative Learning Control Algorithms for
1. Flierl, R., and Kluting, M., “The Third Generation of Valvetrains - New Non-minimum Phase Plants.” International Journal of Control 80(4):
Fully Variable Valvetrains for Throttle-Free Load Control,” SAE 569-582, 2007.
Technical Paper 2000-01-1227, 2000, doi:10.4271/2000-01-1227.
2. Theobald, M. A., Lequesne, B., and Henryet, R., “Control of Engine
Load via Electromagnetic Valve Actuators,” SAE Technical Paper ACKNOWLEGEMENTS
940816, 1994, doi: 10.4271/940816.
3. Boie, C., Kemper, H., Kather, and Corde, G., “Method for Controlling a The authors are appreciated to Steven Dsouza, Gerald P.
Electromagnetic Actuator for Activating a Gas Exchange Valve On a
Reciprocation Internal Combustion Engine,” U.S. Patent, 6 340 008 B1, Malta, Mark Lavine and Barry Brown at Warren GM R&D
2000. Center for the suggestive discussion and their helps in the
4. Hartke, A., and Koch, A., “Method for Controlling a Electromechanical
Actuating Drive for a Gas Exchange Valve of an Internal Combustion experiment preparation.
Engine,” U.S. Patent 6 371 064 B2, 2002.
5. Schneider, L., “Electromagnetic Valve Actuator with Mechanical End

6.
Position Clamp or Latch,” U.S. Patent, 6 267 351 B1, 2001.
Wright, G., Schecter, N. M., and Levin, M. B., “Integrated Hydraulic
CONTACT INFORMATION
System for Electro-hydraulic Valvetrain and Hydraulically Assisted Hai Wu; Jyh-Shin Chen, Research and Development Center
Turbocharger,” U.S. Patent 5 375 419A, 1994.
7. Sturman, O. E., “Hydraulic Actuator for an Internal Combustion General Motors Corporation
Engine,” U.S. Patent 5 638 781, 1997. hai.wu@gm.com
8. Allen, J., and Law, D., “Production Electro-Hydraulic Variable Valve-
Train for a New Generation of I.C. Engines,” SAE Technical Paper
2002-01-1109, 2002, doi:10.4271/2002-01-1109. jason.chen@gm.com
9. Gillella, P., Sun, Z., “Design, Modeling, and Control of a Camless
Valve Actuation System with Internal Feedback,” IEEE/ASME
Transactions on Mechatronics, 16(3): 527-539. 2011.
10. Chen, J. S., “Electro-Hydraulic Fully Flexible Valve Actuation System
for Engine Test Cell,” SAE Technical Paper 2010-01-1200, 2010, doi:
10.4271/2010-01-1200.
11. Butzmann, S., Melbert, J., and Koch, A., “Sensorless Control of
Electromagnetic Actuators for Variable Valve Train,” SAE Technical
Paper 2000-01-1225, 2000, doi:10.4271/2000-01-1225.
12. Henry, R. R., “Single-Cylinder Engine Tests of a Motor-Driven
Variable Valve Actuator,” SAE Technical Paper 2001-01-0241, 2001,
doi:10.4271/2001-01-0241.
13. Parlikar, T. A., Chang, W. S., Qiu, Y. H., Seeman, M. D., Perreault, D.
J., Kassakian, J. G., and Keim, T. A., “Design and Experimental
Implementation of an Electromagnetic Engine Valve Drive,” IEEE/
ASME Transaction on Mechatronics, 10(5): 482-494, 2005.
14. Parlikar, T. A., Chang, W. S., Qiu, Y. H., Seeman, M. D., Perreault, D.
J., Kassakian, J. G., and Keim, T. A., “Design and Experimental
Implementation of an Electromagnetic Engine Valve Drive,” IEEE/
ASME Trans. Mechatronics, 10(5): 482-494, 2005.
15. Denger, D., and Mischker, K., “The Electro-hydraulic Valvetrain
System EHVS-System and Potential,” SAE Technical Paper
2005-01-0774, 2005, doi:10.4271/2005-01-0774.
16. Battistoni, M., Foschini, L., Postrioti, L. and Cristiani, M.,
“Development of an Eelectro-Hydraulic Camless VVA System,” SAE
Technical Paper 2007-24-0088, 2007, doi:10.4271/2007-24-0088.
17. Liao, H. H., Roelle, M. J., Chen, J. S., Park, S., and Gerdes, J. C.,
“Implementation and Analysis of a Repetitive Controller for an Electro-
Hydraulic Engine Valve System,” IEEE Transactions on Control
Systems Technology 19(5): 1102-1113, 2011
18. Yamada, Y., Machida, K., and Yamazaki, T., “Development of
Continuous Variable Valve Event and Lift Control System for SI Engine
(VEL),” SAE Int. J. Engines 1(1):949-959, 2009, doi:
10.4271/2008-01-1348.
19. Sun, Z., and Kuo, T. W., “Transient Control of Electro-Hydraulic Fully
Flexible Engine Valve Actuation System” IEEE Transactions on Control
Systems Technology 18(3): 613-621, 2010.
20. Hoffmann, W., Peterson, K., and Stefanopoulou, A. G., “Iterative
Learning Control for Soft Landing of Electromechanical Valve Actuator
in Camless Engines,” Control Systems Technology, IEEE Transactions
on 11(2): 174-184, 2003.