You are on page 1of 6

2015 IEEE International Symposium on Robotics and Intelligent Sensors (IEEE IRIS20I5)

System Identification and Predictive Functional


Control for Electro-hydraulic Actuator System

N.H. Izzuddin, Ahmad 'Athif Mohd Faudzi, Mohd Khairuddin Osman


Ridzuan Johari Faculty of Electronic and Computer Engineering
Faculty of Electrical Engineering Universiti Teknikal Malaysia Melaka
Universiti Teknologi Malaysia Melaka, Malaysia
lohor Bahru, Malaysia
izzuddin5585@gmail.com

Abstract- Nowadays, electro-hydraulic actuator (ERA) is in [4-9] to obtain linear mathematical model of EHA. These
widely applied in industries. This study presents the modeling studies have shown that a high degree of precision can be
and development of a predictive functional control (PFC) approximate using ARX model.
algorithm for position control of EHA. System identification As the EHA suffer nonlinearities, controller can be
(SI) approach is used to obtain the linear transfer function of
implemented to solve the issue. Large varieties of
the system in discrete form. PFC is proposed based on its
controllers has been introduced to date which involved
ability to predict the future outputs of the actual plant over the
predictive horizon and computes the control effort over the
linear control, non-linear control and artificial intelligence
control horizon at every sampling instance. Numerical approaches such as PID [5,8,10], adaptive control [2, 11-
simulation and real-time experiment are conducted to study 14], sliding mode control (SMC) [2,15-16] and model
the PFC performance with respect to an optimized PID predictive control (MPC). Over the past two decades, there
controller tuned by using particle swarm optimization (PID­ has been significant interest on model predictive control
PSO) for several position tracking inputs. The result shows (MPC) [17].
that the PFC algorithm has better performance in term of Predictive functional control (PFC) is one of the most
overshoot and integral absolute error (IAE) as compared to the
widely implemented MPC. It has low online computation
optimized PID.
because the control law only consists of the basic function.
Besides, the control parameter is also easier to tune [18].
Keywords-predictive functional control; electro-hydraulic
actuator;position control; PlD controller; PSO optimization These advantages have made the controller becomes popular
and successfully applied to various processes with low
I. INTRODUCTION sampling time to very high sampling time such as guided
Compared to other popular actuators, EHA has many missile [17].
advantages such as high force delivered by the actuator over In this paper, identification of the EHA model is
the weight and size. Unlike electrical actuator, the EHA can obtained using two types of input signals. The model is used
maintain high loading capabilities for a longer period of for PFC algorithm design. The controller is then deployed
time. These advantages contribute to the vast successful into EHA system. The performance of PFC algorithm is
applications of the EHA to date. However, there are few compared with PID controller tuned by using PSO
issues that are associated with EHA system. algorithm. The PSO algorithm is widely used in optimizing
The issues that causing dynamic behavior of EHA are controller parameters such as in [19].
uncertainties, time-varying and nonlinearities caused by II. MODEL IOENTIFICAT10N
nonlinear flow and pressure characteristics, backlash in
control valve, actuator friction and variation in fluid volume This section will be discussing parametric identification
due to piston motion and fluid compressibility [1-2]. These using non-recursive estimation for a linear discrete-time of
issues cause the modeling and controller designs of EHA the EHA system. Third order model will also be presented
becoming more complex. based on literature study to prove that the model structure
Mathematical model of EHA system can be obtained selection is sufficient for the developed EHA system
using physical modeling or identification techniques. The workbench. The ARX model was selected because of the
forms of models using system identification techniques are good result obtained as compared with other model
linear models, non-linear models and intelligent models [3]. structures. The equation of ARX model is given as
Linear model is the most popular representation of the
relationship between input and output due to its simplicity A (z)y(k) = z-nB(z)u(k) + e (k) (1)
as compared with other forms. Auto-regressive Exogenous
(ARX) that is one of the widely used form has been utilized where

978-1-4673-7124-7/15/$31.00 ©2015 IEEE


138
2015 IEEE International Symposium on Robotics and Intelligent Sensors (IEEE IRIS20I5)

A (z) = 1 + al z-1 + a z z-z + ... + aka z-ka


(2)
B(z) = bo + b z-1 + b z z-z + ... + bka z-(kb-l)
1

A. Input Signal
The schematic diagram of the EHA system as shown in
Fig. 1 described the interface between the computer and the
hydraulic components. There are several types of input
signals that are commonly used in system identification
process such as Pseudo Random Binary Sequence (PRBS), "k-o ------;-------;------;�--�--+.,, ------!,-----
- f,-----
- +.---+.-----o!
Time (second)
sinusoidal, multi-sine, step and many more. Fig. 3. Continuous step input signal

B. Model Validation
Basically, model validation can be conducted using two
approaches; graphical approach and statistical approach.
The graphical approach generally involves common sense in
deducing the model's goodness based on observations from
the output of the systems [2]. Block diagram for the
graphical approach validation is shown in Fig. 4. The
closed-loop output of the models obtained using multi-sine
and continuous step model are compared with the real plant
output to find the most precise model representation.
Fig. 1. EHA system schematic

Multi-frequency sine signal have been used by


researchers [3,6] for the identification of EHA system. In
these studies, multi-sine signal with three different
frequencies is chosen as one of the inputs for system
identification process as shown in Fig. 2 and is given as

y = sine2n( O.OSt)t + sine2n(O.2)t


+ sine2n(1)t (3)

Fig. 4. Block diagram for graphical approach validation

The statistical approach is important in system


identification process for selecting a good model and
produces results that are statistically acceptable. Some
popular statistical approach validations are best fitting
criterion, histogram of residuals, correlation analysis and
30�--���--�----� root-mean-square-error (RMSE).
30 �--�" ----��--��--�
Time (second)

Fig. 2. Multi-sine input signal III. CONTROL STRATEGY

Meanwhile Osman [20] has used continuous step input This study proposed the implementation of PFC on EHA
signal to obtain position model of a pneumatic actuator system due to the ability of the controller to emulate a first
system. Thus, the second stimulus signal used to excite the order system without overshoot and ensure a smooth
EHA system for parameter estimation is based on the transient response. In MPC, predictions of system behavior
continuous step input signal as shown in Fig. 3. The can be made using state-space models and transfer function
sampling time for the experiments is set to 20ms. model. The state-space approach is preferable in modern
control applications because it can handle systems with non­
zero initial conditions and can represent multiple-input,
multiple-output systems compactly. The state-space
prediction model [21] is given as

139
2015 IEEE International Symposium on Robotics and Intelligent Sensors (IEEE IRIS20I5)

Xk+1 A These equivalences are called coincidence points. In this


Xk+Z AZ research, only one coincidence point is considered. Thus, at
Xk+3 A3 Xk + a single coincidence point and using Equation (7) and (9),
the control law can be determined by:
Xk+n An Yk+n = Wk+n = rk - (rk - Yk) 1jJi (10)
Xk Pxx (4)
B 0 0 0 Uk Hence, by substituting (8) into (10):
AB B 0 0 Uk+l
AZB AB Yk+n = PXk + HUk-l = rk - (rk - Yk) 1jJi (11)
B 0 Uk+Z
Assuming Uk+l = Uk, the control can be written as
An-1B An-ZB An-3 B B Uk+n-l
following
Hxx Ukl
(12)

Yk+l CA
IV. RESULTS AND DISCUSSION
Yk+Z CAZ
Yk+3 CA3 Xk + A. Modelldentification
The first step in model identification is to obtain input
Yk+n CAn and output signals of the EHA system. The multi-sine input
Yk P and the corresponding output as shown in Fig. 5 was
CB 0 0 0 Uk (5)
divided equally into two sets within 50s; the first set for
CAB CB 0 0 Uk+l
estimation and the second set for the validation process.
CAZB CAB B 0 Uk+Z
Three model orders were used for estimation; ARX221,
ARX331, and ARX441.
CAn-1B CAn-ZB CAn-3 B CB Uk+n-l
H Uk-l

Pxx, Hxx, P and H are matrices and vectors of the right


dimension. Thus, the state model, Xk and measured output
model,Yk are defined as i�t� o 5 10 15 l ?5 .
Valldaliondala

-EstimatiOOdata
r 35 40 45 50

iJ�
Xk = PxxXk + HxxUk-l (6)
Yk = PXk + HUk-l

The main concept in PFC is to realize closed-loop


Time (s)
behavior close to first order system with delay [22]. This is
achieved by placing the desired closed-loop dynamic into Fig. 5. Multi-sine Input and output signals

the reference trajectory. Given the actual set point is r, the


The estimated models were then simulated and compared
loop set point w is a first order lag which can be defined as
with the measured model output as shown in Fig. 6.
MaaSlI"ed and simulatEd modeloutp.Jt

(7)
'" :�\_"\

where 1jJ (0 < 1jJ < 1) is a tuning parameter setting the


desired closed-loop pole and Yk is the most recent E
.s o
N1�\, ;' . ' �
measured output. The 1jJ is related to the implied time
c
constant, Td and sample rate, T for the lag through o
Ow
·z

o
r: '--/\;.!\.�\
50

c.. \ r
-T \J - - -, ARX221: 73.25<>,{
1jJ= e Td (8) ...•... ARX331: 95.86°/9
...•... ARX441: 94.09o,{1
-Measured
PFC control law is obtained by using the degree of -150 '0-
25
--- ----!:c------+.�--____:;;__---C;---______:.
freedom (OOF) to enforce equality of the predictions and Time(s)
the reference trajectory at a number of points. By solving Fig. 6. Measured and simulated model output for multi-sine input signal
the control moves such that:
(9)

140
2015 IEEE International Symposium on Robotics and Intelligent Sensors (IEEE IRIS20I5)

ARX331 was selected to represent the model for multi-sine The output responses in Fig. 9 clearly shows that the model
input as it has the highest best-fit value as tabulated in Table obtained using continuous step input signal gives almost
1. similar output as the real-time experiment, compared with
the model obtained using multi-sine input signal. Thus, the
Table I: Model order selection for multi-sine input signal model obtained using continuous step input signal is chosen
and is given as
Model Best fit FPE MSE
B CZ)
ARX221 73.25% 0.003669 0.003648 GCz) = 2
A2CZ)
ARX331 95.86% 0.00316 0.003131 (13)
0.05326z2 + 0.1695z - 0.1818
ARX44 I 94.09% 0.003024 0.002987
Z3 - 2.447z2 + 1.954z - 0.5069
The second type of input signal used to obtain the EHA - - Reference
-
model was continuous step. Fig. 7 shows the data collected
for estimation and validation. The first half of data collected 100
I .j
�-....-"""--""'''''
-1 '
Real-time
---'5imulation . cant. step input 51
was used for estimation using three model orders; ARX221, I .
Simulation - multi-sine input SI
,0
.. · ...

ARX331, and ARX441. E


E-
o:
:;::;
i�� �.
0
0
.. ..
'- - -� �
!'!'!'!"! ••••!!'!'!'
••••'!'!!'!' � ...
'iii \:-- n.:-y,--,.,....�
0
.. .... .....

0.. ,0 "

0
,f
0
"
Time(s)

(rJ1i/illm
o 2 4 Tim:(s)
Fig. 7. Continuous step input and output signals
8 10 12
B. Control Strategy
Fig. 9. Output of a step reference signal

In this section, EHA posltlon control using PFC and


PID-PSO are compared for step and multi-step reference
The models estimated were validated with the other half of
positions. Fig. 10 illustrated the simulation output responses
the data as shown in Fig. 8. ARX331 model was selected
of the step input for PFC algorithm and PID-PSO controller.
due to the highest best-fit value as shown in Table 2.
The target position is set to 100 mm for lOs.
Measu,edandslrrUated modeloulpUl

E
\110024
.s
g
0:

.�
0..
-21
1908
•••• ARX221: 86.9% 96 0:.6 0.8
........ ARX441: 89.7%
- ARX331: 94.67°;'
-Measured
_.•.

.,.D,!-----
;-----;-------;;-
,

Time(s)
--
--j;;-----'=�;;;;�
;;; ��co!.
°O�-7_-�-__;_-�-�5�-7_-�-__;_-�-�
1.2
Time (s)
Fig. 8. Measured and simulated model output for continuous step input Fig. 10. Step response simulation of PFC VS. PID
signal
The time response of the graph is tabulated in Table 3.
Table 2: Model order selection for continuous step input signal
The rise time and settling time for both PFC algorithm and
Model order Best fit FPE MSE PID-PSO controller show almost the same result.
ARX221 86.9% 0.04462 0.04361
Meanwhile for steady-state error, PID-PSO controller has
higher value compared to the PFC algorithm. The most
ARX331 94.67% 0.01959 0.0189
obvious difference between the controllers performance is in
ARX44 I 89.7% 0.01638 0.01561
term of overshoot value. PID-PSO controller has a much
higher of overshoot value as compared with the PFC
In order to select the best model to represent the EHA controller's output response. Fig. 11 illustrated the
system, graphical validation approach is used with the block simulation output response of the controllers when
diagram is shown in Fig. 4. Simulation outputs using both subjected to multi-step positioning. It can be clearly seen
models obtained were compared with real-time experiment. that the PID-PSO controller has more overshoot when the

141
2015 IEEE International Symposium on Robotics and Intelligent Sensors (IEEE IRIS20I5)

actuator extends and retracts as compared with by using the The multistep signal was used to test the validity of the
PFC algorithm. controller over wide range input signal. Fig. 13 shows that
the PID-PSO controller has more overshoot when the
Table 3: Simulation time response of PFC and PID actuator extends and retracts as compared with the PFC
algorithm.
Specifications PFC PID
Reference
Rise time, T,(s) 0.4801 0.4800
!'
Settling Time, T,(s)
Steady State error, e,,(%)
0.6496
0.0023
0.6327
0.0129 \ . � /!
_

:�
• :

.:"
.1
: ...... PFC
...... ' PID-PSO

r ·I�\ ':r j \
Overshoot, OS (%) 8.2117x10'4 0.6991
Integral Absolute Error, IAE (Step) 35.37 35.55

-ReferEr1ce
� 60
0

;
!

i ':r-1 \ I 1\ o I \ I \\ ro ,
" ..·...... "'"

l !
7 72


74

.
. _
76

.. ur
r'--"'

::
\I I
---'PFC

..
_._" Plo-PSO

: : 83.5 :
._.....

i ......
°
',0 . 1 :
: ��.�� . .� ......... .
E
..s,oo 0
°
co

§ I Time (5)
� 80 ; Fig. 13. Multistep response experiment PFC vs. PID
� I
60
I
i
Fig. 14 shows the time response of PFC algorithm and
" Ii
PID-PSO controller for simulation and real-time
20 1
experiment. It can be clearly seen that there are some
°o�--7---�--�---7--�,�o--����- variations in time response between the simulation and
Time (5) experiment that have been obtained earlier.
Fig. 11. Multistep response simulation PFC vs. PID
1.4
The step response of real-time experiment for position 1.2
control of EHA using PFC algorithm and PID-PSO 1
controller is shown in Fig. 12. Table 4 tabulated the time 0.8
• PFe (simulation)
response result for the experiment. As expected from the 0.6
simulation conducted, the experiment result obtained shows 0.4 • PFe (experiment)
that the rise time and settling time for PFC algorithm and 0.2 • PID (simulation)
PID-PSO controller give almost the same result. The steady­ o
• PID (experiment)
state error for PFC is lower than the PID-PSO controller. In
term of overshoot, the PFC algorithm has much lower
overshoot percentage as compared with the PID-PSO
controller.

I
R,I",,,,,
---- PFC
_._ •• PID-PSO
Fig. 14. PFC and PID time response for simulation and experiment
'00 I I
I
I 104 These results could mainly cause by the linearization of
I
I the non-Iinearities during the modelling process. Even
E 102
.s I though minor differences were noted in the time response,
,. ,\ --
c: 60 i 100 --
}Il;
0
the simulation shows a close characteristic with the
:�
0..
iI 98 experimental works. Thus, the mathematical model obtained
"
0

,
I
I 96
I
is acceptable to represent the workbench in the performance
i j
'0 study.
94
iI 0.6 0.8 1.2 1.4 1.6 1.8
° ·I----�--�--�--�--�� 5 --�--�--�--�--�
0
Time (5) V. CONCLUSION
Fig. 12. Step response experiment of PFC vs. PID In this study, identification and control of EHA are
presented. Two types of inputs were used for the SI process.
Table 4: Time response experiment result
Using graphical approach, the model obtained using
Specifications PFC PID continuous step input signal is selected to represent the
Rise time, T,(s) 0.4680 0.4682 EHA system. Numerical simulation and real-time
Settling Time, T,(s) 0.6367 0.6269 experiment of EHA position control show the ability of PFC
Steady State error, e,,(%) 0.1096 0.1744
algorithm to give lower steady-state error and overshoot as
Overshoot, OS (%) 0.1457 1.1812
Integral Absolute Error, IAE (Step) 35.98 36.06
compared to PID-PSO.

142
2015 IEEE International Symposium on Robotics and Intelligent Sensors (IEEE IRIS20I5)

ACKNOWLEDGMENT [11] C. Guan and S. Pan, "Adaptive sliding mode control of electro­
hydraulic system with nonlinear unknown parameters, " COl1lrol
The authors would like to thank Universiti Teknologi Eng. Pract., vol. 16, no. 11, pp. 1275-12S4, 200S.
Malaysia, Universiti Sains Islam Malaysia, and Ministry of [12] J. M. Lee, H. M. Kim, S. H. Park, and J. S. Kim, "A position
Education (MOE) Malaysia under FRGS Grant No. control of electro-hydraulic actuator systems using the adaptive
control scheme, " Asian Control Conference, 2009, vol. 7, pp. 21-
R.J130000.7809.4F371, for their support. 26.
[13] K. K. Ahn, D. Ngoc, C. Nam, and M. Jin, "Adaptive
REFERENCES Backstepping Control of an Electrohydraulic Actuator, "
[1] C. Guan and S. Pan, "Nonlinear Adaptive Robust Control of IEEElASME Transactions on Mechatronics, 2014, vol. 19, no. 3,
Single-Rod Electro-Hydraulic Actuator With Unknown pp. 9S7-995.
Nonlinear Parameters, " IEEE Transactions on Control Systems [14] A. R. Plummer and N. D. Vaughan, "Robust adaptive control for
Technology, 200S, vol. 16, no. 3, pp. 43�45. hydraulic servosystems, " ASME 1. Dynam. Syst., Meas., Contr.,
[2] R. Ghazali, "Adaptive Discrete Sliding Mode Control of an vol. liS, no. 2, pp. 237-244, 1996.
Electro-hydraulic Actuator System, " 2013. [15] S. H. Cho and K. A. Edge, "Adaptive sliding mode tracking
[3] Zulfatman and M. F. Rahmat, "Application of Self-Tuning Fuzzy control of hydraulic servosystems with unknown non-linear
PID Controller On Industrial Hydraulic Actuator Using System friction and modelling error," Proc. Inst. Mech. Eng. Part I, .J.
Identification Approach, " flu. J. Smart Sens. Intell. Syst., vol. 2, Syst. Control Eng., vol. 214, no. February, pp. 247-257, 2000.
no. 2, pp. 246-261, 2009. [16] Y. Lin, Y. Shi, and R. Burton, "Modeling and robust discrete­
[4] H. Yanada and K. Furuta, "Adaptive control of an time sliding-mode control design for a fluid power
electrohydraulic servo system utilizing online estimate of its electrohydraulic actuator (EHA) system, " IEEElASME Trans.
natural frequency, " Mechatronics, vol. 17, no. 6, pp. 337-343, Mechatronics, vol. IS, pp. 1-10, 2013.
2007. [17] J. Richalet and D. O'Donovan, "Elementary Predictive
[5] S. M. Rozali, M. Rahmat, N. A. Wahab, and R. Ghazali, "PID Functional Control : A Tutorial, " International Symposium on
controller design for an industrial hydraulic actuator with servo Advanced Control of Industrial Processes, 20 l l , pp. 306-313.
system, " IEEE Student Co'!ference on Research and [IS] J. Y. Nagase, K. Hamada, T. Satoh, N. Saga, and K. Suzumori,
Development (SCOReD), 2010, pp. 21S-223. "Comparison between PFC and PID control system for tendon­
[6] M. F. Rahmat, S. M. Rozali, N. A. Wahab, Zulfatman, and K. driven balloon actuator, " Industrial Electronics Co'!ference
Jusoff, "Modeling and Controller Design of an Electro-Hydraulic (IECON), 2013, pp. 339S-3403.
Actuator System, " Am. J. Appl. Sci., vol. 7, no. S, pp. 1100-1 lOS, [19] L. Ramli, Y. Sam, Z. Mohamed, M. K. Aripin, and M. F. Ismail,
2010. "Composite Nonlinear Feedback Control with Multi-objective
[7] T. G. Ling, M. F. Rahmat, A. R. Husain, and R. Ghazali, "System Particle Swarm Optimization for Active Front Steering System, "
identification of electro-hydraulic actuator servo system, " IEEE .J. Teknol., vol. 2, pp. 13-20, 2015.
Int. CO'!f. Mechatronics (ICOM), pp. 1-7, May 2011. [20] K. Osman, "Control Algorithm for Intelligent Pneumatic
[S] N. Ishak, M. Tajjudin, H. Ismail, M. Hezri Fazalul Rahiman, Y. Actuator Applied in Rehabilitation Device, " Universiti Teknologi
Md Sam, and R. Adnan, "PID Studies on Position Tracking Malaysia, 2015.
Control of an Electro-Hydraulic Actuator, " Int. 1. Control Sci. [21] J. A. Rossiter, Model Based Predictive COl1lrol: A Practical
Eng., vol. 2, no. 5, pp. 120-126, 2012. Approach, vol. 42. Florida: CRC Press LLC, 1995.
[9] T. G. Ling, M. F. Rahmat, and A. R. Husain, "A Comparative [22] J. Rossiter and R. Haber, "The Effect of Coincidence Horizon on
Study of Linear ARX and Nonlinear ANFIS Modeling of an Predictive Functional Control, " Processes, vol. 3, pp. 25-45,
Electro-Hydraulic Actuator System, " J. Teknol., vol. 5, pp. I-S, 2015.
2014.
[10] K. M. Elbayomy, Z. Jiao, and H. Zhang, "PID controller
optimization by GA and its performances on the electro­
hydraulic servo control system, " Chinese J. Aeronaut., vol. 21,
no. 4, pp. 37S-3S4, 200S.

143

You might also like