You are on page 1of 15

Article

Journal of Vibration and Control


1–15

System identification and robust position ! The Author(s) 2017


Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
control for electro-hydraulic servo system DOI: 10.1177/1077546317721417
journals.sagepub.com/home/jvc
using hybrid model predictive control

Hai-Bo Yuan1, Hong-Cheol Na1,2 and Young-Bae Kim1

Abstract
This paper examined system identification using grey-box model estimation and position-tracking control for an
electro-hydraulic servo system (EHSS) using hybrid controller composed of proportional-integral control (PIC) and
model predictive control (MPC). The nonlinear EHSS model is represented by differential equations. We identify
model parameters and verify their accuracy against experimental data in MATLAB to evaluate the validity of this
mathematical model. To guarantee improved performance of EHSS and precision of cylinder position, we propose a
hybrid controller composed of PIC and MPC. The controller is designed using the Control Design and Simulation module
in the Laboratory Virtual Instrumentation Engineering Workbench (LabVIEW). A LabVIEW-based experimental rig is
developed to apply the proposed controller in real time. Then, the validity and performance superiority of the hybrid
controller were confirmed by comparing them with the MPC and PIC results. Results of real-life experiments show
improved robustness and dynamic and static properties of EHSS.

Keywords
Electro-hydraulic servo system, grey-box system identification, model predictive control, real-time position control,
MATLAB, Laboratory Virtual Instrumentation Engineering Workbench

In previous studies, despite high nonlinearity of EHSS,


1. Introduction many researchers used linear control methods, such as
An electro-hydraulic servo system (EHSS) mainly con- pole placement (Lim, 1997; Plummer and Vaughan,
sists of power supply unit, servo valve, double-ended 1997) and classical feedback control (Elbayomy et al.,
hydraulic cylinder, position and force sensors, and 2008), around an operating point. However, traditional
other auxiliary system parts. EHSS is widely used in control theory does not track the desired reference
the industry because of its high-speed power/mass satisfactorily because of the parametric uncertainties
ratio, fast response, and high stiffness. Hydraulic fluid and nonlinearities in the system, thus limiting linear
can act as a heat exchanger to remove heat generated controller ability. To improve control precision of actu-
in the system (Merritt, 1967). EHSS exhibits some ator position, some nonlinear controllers were examined.
parametric uncertainties and uncertain nonlinearities An optimal tuning nonlinear proportional-integral-deri-
(Merritt, 1967; Yao et al., 2000). Parametric variations vative (PID) controller was introduced in Liu and Daley
include the hydraulic bulk modulus, pressure-flow (2009), and optimal parameters were provided using an
characteristics, and oil temperature in different working
conditions (Guo et al., 2016). Uncertain nonlinearities 1
Department of Mechanical Engineering, Chonnam National University,
include effect of the compressibility, friction, internal Gwangju, Republic of Korea
2
leakage, and slight dead zone in servo valve, leakage JKS Co., Gwangju, Republic of Korea
in actuator, and time delay caused by the interaction of Received: 31 October 2016; accepted: 15 June 2017
actuator and external load (Kalyoncu and Haydim,
Corresponding author:
2009). For position control, nonlinearity becomes Young-Bae Kim, Department of Mechanical Engineering, Chonnam
especially significant when the actuator operates with National University, Gwangju, Republic of Korea.
large displacement command. Email: ybkim@chonnam.ac.kr
2 Journal of Vibration and Control 0(0)

estimated plant model with inverse of dead zone. Other is discussed by comparing hybrid controller results
authors (Wos and Dindorf, 2013; Yao et al., 2000) pro- with those of MPC and PIC.
posed an adaptive robust controller for position control The rest of the paper is organized as follows. Section
of the EHSS, considering not only the nonlinearities but 2 describes a detailed nonlinear state-space model and
also effects of parameter variations. The Lyapunov-based grey-box model identification using experimental data
nonlinear controller using the back-stepping design strat- in MATLAB. In section 3, a hybrid controller design is
egy was suggested by Sirouspour and Salcudean (2000), derived, and controller parameters are adjusted
and the predictive control strategy was applied by Balau through simulation in LabVIEW. In section 4, a pro-
et al. (2011) to reduce effects of delay on control perform- posed controller for position control is implemented on
ance. Kim et al. (2011) used quantitative feedback theory a real-time platform, and experimental results are ana-
to optimize the control results with parameter uncertain- lyzed. Finally, section 5 summarizes the key points of
ties. Not only nonlinear and linear controllers but also this paper.
some hybrid controllers were used to optimize actuator
position control. Kalyoncu and Haydim (2009) developed 2. Nonlinear model and system
nonlinear hybrid controller consisting of the fuzzy con-
troller and PID controller and obtained good tracking
identification of EHSS
performance. Guo et al. (2015) proposed a nonlinear cas- The EHSS is difficult to control because of the nonlinear-
cade controller consisting of sliding model control and ity and complexity of its mathematical models. A simpli-
back-stepping technique. In consideration of external dis- fied diagram of this system is depicted in Figure 1.
turbance, the proposed controller performed excellently.
Li and Xiong (2008) designed a combination of foreword
2.1. Nonlinear state-space model of EHSS
model controller and feedback fuzzy controller. Faced
with system nonlinearities, many researchers considered Before establishing a mathematical model of the EHSS,
improving the control result by compensating for friction we formulate several assumptions, which are as follows
(Alleyne and Liu, 2000; Tafazoli et al., 2002), time delay (Qian et al., 2014):
(Balau et al. 2011; Qian et al., 2014), leakage (Kalyoncu
and Haydim, 2009), and dead zone (Liu and Daley, 2009) 1. Fluid properties (density, bulk modulus, etc.) are
in the servo valve. Some scientists (Guo et al., 2015; Wos constant.
and Dindorf, 2013) constructed disturbance observer to 2. Servo valve is not saturated.
optimize position control. 3. In pipelines and servo valve, oil volumes are not
Zad et al. (2016) proposed a theory of controlling the considered.
model predictive control (MPC) used in EHSS and con- 4. Time delay and friction force can be modeled as
structed a simulation model. Researchers did not present negative damping.
experimental data to verify whether the constructed simu-
lation model worked well. Marusak and Kuntanapreeda With the above assumptions, we derive mathematical
(2011) introduced constrained MPC for force control of models of servo valve, actuator cylinder, and test load.
EHSS and emphasized importance of constraints in con-
troller design. Despite the simplicity, practicality, and 2.1.1. Servo valve model. In this paper, the servo valve
robustness of the MPC controller, only few researchers chosen is a closed-centred, four-way sliding spool valve
focused on using MPC on EHSS, as examining MPC is with symmetric orifice containing a dead zone. However,
challenging. this design is simple and rugged to achieve dependable,
In the present paper, a hybrid controller consisting long-life operation, especially in preventing leakage losses,
of proportional-integral control (PIC) and MPC is which increase with wear and tear. According to manu-
investigated to improve tracking performance of actu- facturers’ datasheet information, we use a first-order
ators. First, a state-space EHSS model is developed, model to describe dynamics of this servo valve (valve
and model accuracy is verified against experimental type G761-3005 of MOOG), which can provide the high
data. Second, the proposed controller design is verified quality and high operational bandwidth of a used valve
in the Laboratory Virtual Instrumentation Engineering (Yao et al., 2000). Related equation is as follows:
Workbench (LabVIEW) environment. Later, a real-
time hybrid controller is used to optimize tracking pre- 1 kv
x_v ¼  xv þ u ð1Þ
cision of actuator position. This research contributes in  
developing a hybrid controller based on a reasonable
nonlinear state-space model of EHSS and in realizing where xv is spool displacement,  is time constant of
controller design in LabVIEW for actuator position servo valve, kv is direct current gain of servo valve, and
control. Finally, the control result of the experiment u is input voltage.
Yuan et al. 3

Figure 1. Schematic diagram of the electro-hydraulic servo system.

The following equation expresses orifice flow equa- equivalent damping of time delay), and k is the spring
tion in servo valve: constant.
Combining the above equations of each sub-system,
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Ps  sgnðxv ÞPL we obtain an open-loop nonlinear state-space equation
QL ¼ Cd !xv ð2Þ of the EHSS:

x_1 ¼ x2
where QL is load flow, Cd is discharge coefficient, ! is k b A
x_2 ¼  x1  x2 þ x3  g
spool valve area gradient, Ps is supply pressure, PL is m m m
load pressure, and  is fluid mass density. 4Ae 4Ctm e 4Cd e ! pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x_3 ¼  x2  x3 þ pffiffiffi Ps  x3 sgnðx4 Þ  x4
Vt Vt Vt 
2.1.2. Hydraulic cylinder model. Flow-pressure continuous 1 kv
x_4 ¼  x4 þ u
equation describes properties of the cylinder as follows:  
ð5Þ
Vt _
PL ¼ Ax_  Ctm PL þ QL ð3Þ
4e where x1 is actuator piston position, x2 is actuator
piston velocity, x3 is load pressure, and x4 is spool dis-
where Vt is total actuator volume, e is effective bulk placement in servo valve.
modulus, A is actuator ram area, Ctm is total coefficient
of leakage.
2.2. Grey-box model identification
2.1.3. Test load model. In our experiment, we do not con- In the previous section, we obtained a fourth-order
sider calculation of friction, which can influence the nonlinear state-space model of EHSS using known
dynamics of actuator. For simplicity, we consider fric- physical laws and hydraulic theories. That is, we have
tion as part of damping. Related equation is as follows: good information about characteristics of EHSS.
However, we cannot ascertain values of some param-
PL A ¼ m x€ þbx_ þ kx þ mg ð4Þ eters, without which we could not sufficiently describe
EHSS dynamics. Thus, this paper presents a system
where the m is mass of actuator rod and all moving identification method called nonlinear grey-box model
parts, b is damping coefficient of system (including identification to estimate parameters of an existing
4 Journal of Vibration and Control 0(0)

mathematical model using input and output data col- Some parameters can be clustered together to sim-
lected from the experiment. plify identification. The new state-space model is pre-
For the obtained nonlinear state-space model, struc- sented as follows:
ture and materials have some parameters that are
known or can be measured. However, most of them x_1 ¼ x2
are unknown. Thus, for grey-box model identification, x_2 ¼ a1 x1  a2 x2 þ a3 x3  g
pre-estimation should be performed to determine initial pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð6Þ
x_3 ¼ a4 x2  a5 x3 þ a6 Ps  x3 sgnðx4 Þ  x4
values of each unknown parameter according to other
papers (Guo et al., 2016; Kaddissi et al., 2007; Kim x_4 ¼  1 x4 þ a7 u
et al., 2011; Maneetham and Afzulpurkar, 2010;
Sirouspour and Salcudean, 2000), though some esti- where, a1 ¼ mk , a2 ¼ mb , a3 ¼ m
A
, a4 ¼ 4A 4Ctm e
Vt , a5 ¼ Vt ,
e

mated parameters have large errors. Tables 1–3 list all


a6 ¼ 4C effiffi! kv
Vt  , a7 ¼ 
dp
parameters.
Table 4 lists values of simplified parameters.
For time-domain testing input signal, we select one
Table 1. Physical parameters obtained from manufacturers’
chirp signal with frequency range of 0–20 Hz and 5 mm
catalogues.
amplitude as excitation. Figure 2 presents testing
Parameters Unit Description Value signal and response of open-loop system. As shown in
Figure 2, scales are different from practical unit, as
Ps Pa Supply pressure 1e7
input signal acts as voltage to the servo valve amplifier,
 s Time constant of the servo valve 0.067
and 7.5-scale is identified between real distance and
input voltage (u) to amplifier.
After identifying ordinary differential equations with
Table 2. Calculated physical parameters of the electro-hydraulic unknown parameters, we construct nonlinear grey-box
servo system. modeling of EHSS using System Identification ToolTM
commands in MATLAB. Output of system is simulated
Parameters Unit Description Value using Runge–Kutta 45 solver with adaptive step length
m kg Mass of the actuator 20 adjustment: absolute and relative error tolerances
rod and all moving parts are 1e-6 and 1e-5, respectively. To estimate model par-
A m2 Actuator ram area 1.25664e-3 ameters, the search method utilized is Trust-Region
Cd – Discharge coefficient 0.62 Reflective Newton method of nonlinear least-squares.
! – Spool valve area gradient 0.03399 In the method, cost is sum of squares of errors between
measured and simulated outputs. After several iter-
ations, we obtain optimal parameters fitting 96.34%
of real-time response data; the mean squared error is
Table 3. Pre-estimated parameters from other researches. 1.159e-06, as shown in Figure 3. Thus, we ensure that
obtained parameters can describe the system character-
Parameters Unit Description Value istics well.
b N  s=m Damping coefficient 2500 Table 5 presents parameters estimated from the
k N=m Spring constant 0 program.
e N=m2 Effective bulk modulus 5e8
To verify the veracity of the estimated mathematical
model, we use one square wave signal with 15 mm amp-
Ctm m4  s=kg Total coefficient of leakage 1.5e-11
litude and 3 Hz frequency to test real system and compare
 kg=m3 Fluid mass density 820
results with those of simulation model constructed in
kv m=v Direct current gain 0.0265e-4 MATLAB/Simulink. In Figure 4 (a) and (b) respectively
of the servo valve
we display the testing signal and comparison results.
Note: a in the operation, the cylinder is not always touching the spring; As indicated by the comparison result, large
hence, we set the value of k to 0 first. errors are not found between response of simulation

Table 4. The simplified pre-estimated parameters.

Parameters a1 a2 a3 a4 a5 a6 a7

Pre-estimated values 0 125 6.02832e-5 1.25664e10 150 7.359e9 1.65625e-4


Yuan et al. 5

Figure 2. (a) Test signal for system identification; (b) real-time response.

Figure 3. Grey-box system identification result.

Table 5. The estimated parameters.

Parameters a1 a2 a3 a4 a5 a6 a7

Estimated values 21.136 124.883 2.4107e-5 1.25664e10 149.898 7.359e9 0.0331967

model and actual system. Therefore, we can ascertain The tuning method is performed by setting the integral
that this model is accurate in predicting real system and derivative gains to zero. Proportional gain, KP is
behaviors. then increased (from zero) until it reaches ultimate gain
KPu , at which output of control loop has stable and
consistent oscillations. We then measure the ultimate
3. Controller design period Pu . Through experiment, tuning step response
is represented by Figure 5, from which ultimate gain
3.1. PI controller design (KPu ¼ 4:12) and ultimate period (Pu ¼ 0:08s) are
In this paper, the closed-loop Ziegler–Nichols PID obtained. Using tuning rules listed in Table 6, we
tuning method (Ziegler and Nichols, 1942) is utilized acquire controller parameters. PI controller behaves
to adjust proportional integral (PI) controller. And the better than another two controllers, and values of Kp
selected operating point is controlling actuator at 5 mm. (proportional gain) and Ti (integral time) are 1.854 and
6 Journal of Vibration and Control 0(0)

Figure 4. (a) Test signal; (b) comparison of the real-time and simulation response.

Figure 5. Tuning step response of real system.

Table 6. Tuning rules of closed loop Ziegler–Nichols method.

Parameters Kp Ti Td

P 0.5 KPu 1 0
PI 0.45 KPu Pu =1:2 0
PID 0.6 KPu Pu =2 Pu =8

0.0667s, respectively. We then conclude the acceptable


stability tuned by the closed loop Ziegler–Nichols PID
tuning method and step responses as illustrated by
Figure 6.
Figure 6. Step response of real system with tuned propor-
tional-integral control (PIC).
3.2. Introduction of MPC
Model predictive control is a robust controller com- practicality. Figure 7 illustrates the model structure
monly used in control engineering, and it is widely stu- used in the MPC controller.
died by the industrial and academic communities. The objective of the MPC is to create noise-free
Many successful applications were realized simply and plant output (yp ) to track reference (R) by manipulat-
efficiently because of its simplicity, richness, and ing signals (manipulated variable (uO )) in the plant.
Yuan et al. 7

Figure 7. Basic structure of model predictive controller.

ymin  y^ ðk þ ijkÞ  ymax ð8Þ

uO min  uO ðk þ ijkÞ  uO max ð9Þ

uO min  uO ðk þ ijkÞ  uO max ð10Þ

where
P is the prediction horizon
M is the control horizon
k is the discrete time
i is the index along the prediction horizon
q is the output error weighting
Figure 8. The state of model predictive control at instant r is the manipulated variable change weighting
k þ 1. y^ ðk þ ijkÞ is the predicted plant output at time k þ i
Rðk þ ijkÞ is the plant output reference at time k þ i
The controller also deals with measured and unmeas- uO ðk þ ijkÞ is the predicted optimal manipulated
ured disturbance. To better track (R) signal, measured variable at time k þ i
noise (electrical noise, sampling errors, etc.) should be uO ðk þ ijkÞ is the predicted rate of change in
considered to estimate the yp , and MPC provides feed- manipulated variable at time k þ i
forward compensation for measured and unmeasured The constraints specified in the manipulated variable
disturbances. According to the above basic structure, and plant output are set to prevent the overshoot of
the working process of MPC should be understood. the performance. The MPC controller considers
Normally, MPC relies on one linear-time-invariant, all these constraints and calculates a sequence of
discrete-time, state-space system model to predict future manipulated variables to minimize the cost
future response of the system. Horizons, including function.
prediction horizon (P) and control horizon (M), are
important parts of the definition of MPC. P is
number of samples to predict the plant output, and
3.3. MPC and hybrid controller design
M specifies number of samples to affect the medium Applying MPC, a discrete-time state-space mathemat-
voltage. Value of M must be less than that of P. As ical model is needed for MPC design. Other parameters
shown in Figure 8, MPC at time k predicts plant output including sampling time, prediction horizon (P), con-
for time k þ P to track R. In this process, uO is used to trol horizon (M), output error weighting (q), and
predict plant output for time k þ 1 and to minimize cost manipulated variable change weighting (r) also play
function JðkÞ. important roles in excellent control performance.
JðkÞ is nonlinear quadratic function with constraints Sampling time selected in this manuscript is 0.01 sec-
specified by Equations (8)–(10). For scalar system, JðkÞ onds, which is short enough to decrease response
is represented by Equation (7). time. Ample prediction horizon (P) is needed to guar-
antee the closed-loop stability of control system, but
X
P X
M1 the larger prediction may cause no further increment
JðkÞ ¼ q  ½y^ðk þ ijkÞ  Rðk þ ijkÞ2 þ r  ½uO ðk þ ijkÞ2 on control result. Control horizon (M) and manipu-
i¼1 i¼0
lated variable change weighting (r) can both affect
ð7Þ
how aggressive or conservative the control performance
8 Journal of Vibration and Control 0(0)

is. In addition, the larger control horizon (M) number


will increase the computational load. Output error
weighting (q) dictates the accuracy of control result
and has a significant effect on forcing output to track
set point.
Garriga JL and Soroush M (2010) reviewed tuning
methods for different kinds of MPC, in which they com-
pared with dynamic matrix control (DMC) and general-
ized predictive control (GPC). However, there are
limited papers studying the explicit tuning algorithm
on the general state-space representation of MPC. Lee
JH and Yu ZH (1994) studied tuning rules about general
state-space representation MPC, and provided tuning Figure 9. Simulation results of model predictive control (P ¼ 7,
principle to obtain nominal stability, then to detune M ¼ 1, q¼ 1:5, r¼ 1).
robust performance for infinite horizon MPC. Zhang
and Li (2007) focused on performance assessment of
MPC, and an MPC-achievable benchmark was pro- in which,
posed on subspace identification. They used trade-off
2 3
curves to find potential benefit by tuning weighting par- 0:1758 0:001673 324:1 7:803e  9
ameters and by a different combination of P and M in 6 7:916e  5 0:9998 5:862 6:479e  11 7
6 7
MPC-achievable benchmark, then optimal horizon A¼6 7
40 0 0:04394 0 5
numbers can be selected. Nagrath et al. (2000) intro- 4:068 7:138e5 5:005e9 0:184
duced finite horizon MPC based on state-space repre- 2 3
25:95
sentation and used default value of P (10) and fixed 6 0:1537 7
value of manipulated variable change weighting r (1) 6 7
B¼6 7
to tune a continuous-stirred tank reactor with excellent 4 0:01354 5
results. Edouard et al. (2005) adopted a state-space rep- 7:503e7
 
resentation MPC controller to catalytic reverse flow C¼ 0 1 0 0
reactor, in which the authors set M to 1 to induce the D¼0
computational burden of the constrained optimization
problem. In summary the key of MPC tuning is to Next, we develop a simulation model using the
develop one appropriate process model. Control Design and Simulation Module in LabVIEW
Generally speaking, a state-space representation to test the control method. Nevertheless, in the simula-
structure typically requires offline simulation and tion the controlled position cannot track the desired
some trial and error tuning method which is cumber- set-point and there exists a small gap. The optimal
some because of the overlapping effect of each param- simulation result is represented by Figure 9 with the
eter setting. Based on the previous experience on MPC, tuned parameters.
the present research considers the function of P and q
mainly to guarantee the stability and robustness of the 3.3.2. Hybrid controller. For highly nonlinear EHSS, con-
control action. With default value of M equaling to 1, sidering that linearizing the nonlinear model at initial
small computational load is guaranteed. Setting the condition is unsuitable for MPC controller design, we
fixed value of r (1) meets trade-off between robust suggest initially using typical PI controller (PIC) which
and aggressive control performance which is the same is commonly used for position control of hydraulic sys-
situation to M. If the robustness cannot be realized, we tems in industries, to make actuator piston approxi-
will detune M and r and all these works will be done in mately track desired trajectories. Then, we apply
simulation. MPC to improve the dynamic and static performances
of EHSS. That is, MPC works as outer controller and
3.3.1. MPC. For MPC application in LabVIEW, one the traditional controller (PIC) acts as inner controller.
linear discrete-time state-space mathematical model is PIC adjusts u in response to change between inner
obtained using Linear Analysis Tool in MATLAB, reference input and plant output, and outer controller
which is represented by Equation (11). MPC predicts and adjusts uO , which is reference
input to inner control loop. For the whole system, non-
linear plant with control of PIC can be treated as new
x_ ðk þ 1Þ ¼ AxðkÞ þ BuðkÞ plant. Figure 10 presents a diagram of the hybrid
ð11Þ
yðkÞ ¼ CxðkÞ þ DuðkÞ controller.
Yuan et al. 9

Figure 10. The diagram of the hybrid controller.

Figure 11. Simulation results of hybrid controller: (a) tuning P with M ¼ q ¼ r ¼ 1; (b) tuning q with P ¼ 12, M ¼ 1, r ¼ 1.

The nominal linearized discrete-time state-space response cannot be obtained after q tuning. Through the
model is indicated by Equation (12) comparison of simulation result, the optimum final value
of P and q is 12 and 0.02 respectively. We can see from
x_ ðk þ 1Þ ¼ A1 xðkÞ þ B1 uo ðkÞ that the simulation result which is acceptable, hence we
ð12Þ
yðkÞ ¼ C1 xðkÞ þ D1 uo ðkÞ do not need to detune M and r.

2 3 2 3
0:1404 47:5687 548:3871 1:2708e  8 651:2788 47:5736
6 7 6 7
6 0:00023 0:7106 5:3036 4:7657e  11 4:1426 7 6 0:2892 7
6 7 6 7
A1 ¼ 6
6 5:8674e  6 0:0235 0:0867 2:0444e  12 0:3134 7 7; B1 ¼ 6
6 0:0235 7
7
6 7 6 7
4 4:6525e6 8:6678e7 7:994e9 0:1536 5:8442e8 5 4 8:7457e7 5
6:2103e  6 0:0091 0:0337 6:8095e  13 0:987 0:0091
 
C1 ¼ 0 1 0 0 0 ; D1 ¼ 0

With the fixed values of M and r (both equaling to 1), Piston position range depends on manufactory, and
we initially set q to 1 and tune P value. After acquiring after measurement of the real hydraulic system, we
the stable performance in simulation, then we adjust the obtain piston position range of –75 mm to 60 mm.
optimal response by tuning q value with the obtained P. Considering that spring is present in the system,
The tuning result is represented by Figure 11 (a) and (b). under 10 mm, piston can move without elastic spring
During the simulation process, we find that even though force, and maximum amplitude of piston position is
a small P value can guarantee fast response, smooth step limited to 60 mm, at which point spring shows
10 Journal of Vibration and Control 0(0)

maximum compression. As for uO , which is a reference and 1200 rpm specification. Also, Intel E7300 with
input to inner control loop in hybrid controller, we use 2.67 GHz clock speed is used for the central processing
input voltage range of servo valve to set its constraints. unit. Figure 12 illustrates the real facility of present study.
Plant output is also constrained by real position range. For experiments using real-time MPC, sampling
The constraints of the controller are time should be equal to the model discrete sampling
time, and time loop is necessary to guarantee success
 10 V  uO ðk þ ijkÞ  8 V of the experiment. We design the MPC real-time con-
 0:075 m  y^ðk þ ijkÞ  0:060 m troller using obtained parameters from simulation in
the NI-PXI system and test the system using set-point
profile. In the experiment, PIC, MPC and hybrid con-
4. Simulations and experimental results troller are used to control the system. For stability and
LabVIEW is used to design the controller and project robustness verification of the proposed controllers, we
for the experiment. LabVIEW contains variety of tools test EHSS under three different dynamical conditions.
for simulation, data acquisition, instrument control,
measurement analysis, and data presentation. This pro-
gram can easily deal with inputs, outputs, and controller
4.1. Controlling actuator position at 5 mm
design in real time using graphic user interface modules. In the present research, we design the PI controller, MPC
In this paper, the NI PXI system implements data meas- and hybrid controller at operating point of 5 mm, hence
urement and controller designing parts; system type is we use sinusoidal signal with 1 Hz frequency and 5 mm
NI PXIe-1062Q. Position sensor of linear variable amplitude to detect system tracking ability with proposed
differential transformer (LVDT) is Series 210 from controllers. Figure 13 (a) and (b) present identical simu-
Trans-Tek Co., USA, and hydraulic power unit is lation and experimental results obtained using MPC and
from JKS Co., Korea with 100 bars pressure, 5 H.P., the hybrid controller respectively. Figure 14 illustrates

Figure 12. Real-time facility of the electro-hydraulic servo system.

Figure 13. Comparison of the experimental and simulation control results for sinusoidal input: (a) model predictive control; (b)
hybrid controller.
Yuan et al. 11

experimental control results of three proposed control- stability and robustness of proposed controllers (when
lers. PIC tracks sinusoidal signal with small overshoot using random signal to test positions, spring is not yet
and there is a tiny phase lags existing in control response installed into system). For different position changes,
of hybrid controller. The root-mean- square errors EHSS shows nonlinear behaviors because of friction,
(RMSEs) of proposed PIC, MPC and hybrid controller internal leakage, and oil compressibility. Hence, a big
are 0.2771, 0.5646 and 0.5201 respectively. This result challenge for controllers is realizing robust control in
indicates that PIC and hybrid controller can track wide position range. Figure 15 (a) and (b) present con-
desired signal well, while MPC controller does not trol results of MPC and hybrid controller in simulation
realize good performance, which is the same as simula- and real time, which indicate that simulation result can
tion result. provide correct information for experimental response.
Figure 16 illustrates comparison of control perform-
ance between PIC, MPC and hybrid controller. PIC
4.2. Controlling different actuator positions
can follow reference signal but with large overshoot,
For different actuator positions, we use random signal undershoot, and oscillations. And MPC cannot track
with position range from -60 mm to 60 mm to verify the desired input reference. In addition, PIC and hybrid

Figure 14. Experimental control results of proportional-integral control (PIC), model predictive control (MPC) and hybrid
controller.

Figure 15. Comparison of the experimental and simulation control results for random input: (a) model predictive control; (b) hybrid
controller.
12 Journal of Vibration and Control 0(0)

Figure 16. Experimental control results of proportional-integral control (PIC), model predictive control (MPC) and hybrid
controller.

Figure 17. The input voltage u controlled by proportional-integral control (PIC) and hybrid controller.

controller have RMSEs of 7.6251 and 4.5971, respect- Controlled voltage in inner control loop is much
ively. Controlling large position variations, hybrid con- smoother and smaller, resulting in better plant output.
troller shows more stable and robust behaviors than
another two controllers. 4.3. Controlling the fixed actuator position
Figure 17 shows u variations controlled by PIC and at 5 mm in different operating
hybrid controller, clearly illustrating that PIC performs
badly for controlled voltage with large overshoot,
temperature (OT)
undershoot, and oscillations. For hybrid controller, in For EHSS, temperature plays a significant role in con-
outer control loop, MPC adjusts new reference input trolling precise position. Oil properties, such as density
signal to inner control loop based on original reference and compressibility, change significantly with vari-
input and plant output, as illustrated in Figure 18. ations in OT, causing uncertainty in systems. In present
Yuan et al. 13

Figure 18. The manipulated variable uO controlled by model predictive control in hybrid controller.

Figure 19. Experimental results of proportional-integral control (PIC) and hybrid controller in different operating temperature
(OT): (a) PIC; (b) hybrid controller.

Table 7. RMSEs of step responses in different OT. control results of PIC and hybrid controller. Table 7
summarizes RMSE from two control methods. We con-
Proportional-integral Hybrid
clude that both controllers have robust control per-
Temperature control controller
formance in different OT, and hybrid controller
Operating temperature shows more stable behavior than PIC as shown by pre-
(OT) ¼ 25 C 0.8602 0.5809 sented figures and tables in this paper.
OT ¼ 30 C 0.8582 0.5780
OT ¼ 35 C 0.8644 0.5798
5. Conclusions
OT ¼ 40 C 0.8498 0.5782
Based on EHSS, this research examines EHSS
modeling, system identification, controller design, and
real-time control implementation. First, nonlinear
research, 30 C normal temperature is utilized for con- state-space model of EHSS is derived by same mathem-
troller design and actual experiments. To verify robust- atical equation for each sub-system with some assump-
ness and stability of designed controllers (PIC and tions, and simulation model is created in MATLAB/
hybrid controller), we test system step response with Simulink. Grey-box system identification method is
5 mm amplitude under different OTs. Figure 19 shows introduced to ascertain parameter values existing in
14 Journal of Vibration and Control 0(0)

mathematical model, and validity of model is verified Kalyoncu M and Haydim M (2009) Mathematical modeling
against experimental data. Second, we design hybrid and fuzzy logic based position control of an electrohydrau-
controller consisting of PIC and MPC. Nonlinear lic servosystem with internal leakage. Mechatronics 19(6):
system model controlled by PIC is treated as new 847–858.
Kim JW, Xuan DJ and Kim YB (2011) Robust control appli-
plant model and linearized using Linear Analysis
cation for a three-axis road simulator. Journal of
Tool in MATLAB. Then, based on new state-space
Mechanical Science and Technology 35(1): 221–231.
model, MPC is applied as outer controller to improve Lee JH and Yu ZH (1994) Tuning of model predictive con-
control result. Simulation and controller design are rea- trollers for robust performance. Computers & Chemical
lized in LabVIEW environment. In experiments, both engineering 18 (1): 15–37.
PIC and hybrid controller manifest robust control per- Li H and Xiong S (2008) Fuzzy internal model control for
formance, while hybrid controller shows much stable electro-hydraulic servo systems. In: 27th Chinese Control
behaviors than PIC in different operating condition. Conference, 16–18 July 2008, pp.372–376. IEEE.
Lim TJ (1997) Pole placement control of an electrohydraulic
servo motor. In: Power Electronics and Drive Systems,
Declaration of Conflicting Interests
1997. Proceedings of the Seond International Conference
The author(s) declared no potential conflicts of interest with on Power Electronics and Drive Systems, 26–29 May
respect to the research, authorship, and/or publication of this 1997. IEEE, pp.350–356.
article. Liu GP and Daley S (2009) Optimal-tuning nonlinear PID
control of hydraulic systems. Control Engineering Practice
Funding 8(9): 1045–1053.
Maneetham D and Afzulpurkar N (2010) Modeling, simula-
The author(s) disclosed receipt of the following financial sup-
tion and control of high speed nonlinear hydraulic servo
port for the research, authorship, and/or publication of this
system. Journal of Automation Mobile Robotics and
article: This work was supported in part by the National
Intelligent Systems 4(1): 94–103.
Research Foundation of Korea (15H1C1A1035825 and
Marusak PM and Kuntanapreeda S (2011) Constrained
15R1A4A1041746) and in part by Korea Electric Power
Cooperation (KEPCO: 16-07). model predictive force control of an electrohydraulic actu-
ator. Control Engineering Practice 19(1): 62–73.
Merritt HE (1967) Hydraulic control systems. New York, NY:
References Wiley.
Alleyne A and Liu R (2000) A simplified approach to force Nagrath D, Prasad V, Bequette BW, et al. (2000) Model pre-
control for electro-hydraulic systems. Control Engineering dictive control of open-loop unstable cascade systems. In:
Practice 8(12): 1347–1356. Proceedings of the 2000 American Control Conference,
Balau AE, Caruntu CF and Lazar C (2011) Simulation and 28–30 June 2000. IEEE, pp.3747–3752.
control of an electro-hydraulic actuated clutch. Mechanical Plummer AR and Vaughan ND (1997) Decoupling pole-pla-
Systems and Signal Processing 25(6): 1911–1922. cement control, with application to a multi-channel elec-
Edouard D, Dufour P and Hammouri H (2005) Observer tro-hydraulic servosystem. Control Engineering Practice
based multivariable control of a catalytic reverse flow 5(3): 313–323.
reactor: Comparison between LQR and MPC approaches. Qian Y, Ou G, Maghareh A, et al. (2014) Parametric identi-
Computers & chemical engineering 29(4): 851–865. fication of a servo hydraulic actuator for real-time hybrid
Elbayomy KM, Jiao Z and Zhang H (2008) PID controller simulation. Mechanical Systems and Signal Processing
optimization by GA and its performances on the electro- 48(1): 260–273.
hydraulic servo control system. Chinese Journal of Sirouspour MR and Salcudean SE (2000) On the nonlinear
Aeronautics 21(4): 378–384. control of hydraulic servo-systems. In: Proceedings 2000
Garriga JL and Soroush M (2010) Model predictive control ICRA. Millennium Conference. IEEE International
tuning methods: A review. Industrial & Engineering Conference on Robotics and Automation. Symposia
Chemistry Research 49(8): 3505–3515. Proceedings, 24–28 April 2000. IEEE, pp. 1276–1282.
Guo K, Wei J, Fang J, et al. (2015) Position tracking control Tafazoli S, de Silva CW and Lawrence PD (2002) Tracking
of electro-hydraulic single-rod actuator based on an control of an electrohydraulic manipulator in the presence
extended disturbance observer. Mechatronics 27(1): 47–56. of friction. IEEE Transactions on Control Systems
Guo Q, Sun P, Yin J, et al. (2016) Parametric adaptive esti- Technology 6(3): 401–411.
mation and backstepping control of electro-hydraulic Wos P and Dindorf R (2013) Adaptive control of the electro-
actuator with delayed memory filter. ISA Transactions hydraulic servo-system with external disturbances. Asian
62: 202–214. Journal of Control 15(4): 1065–1080.
Kaddissi C, Kenne JP and Saad M (2007) Identification and Yao B, Bu F, Reedy J, et al. (2000) Adaptive robust motion
real-time control of an electrohydraulic servo system based control of single-rod hydraulic actuators: theory and
on nonlinear backstepping. IEEE/ASME Transactions on experiments. IEEE/ASME Transactions on Mechatronics
Mechatronics 12(1): 12–22. 5(1): 79–91.
Yuan et al. 15

Zad HS, Ulasyar A and Zohaib A (2016) Robust model pre- Zhang Q and Li S (2007) Enhanced performance assessment
dictive position control of direct drive electro-hydraulic servo of subspace model-based predictive controller with param-
system. In: 2016 International Conference on Intelligent eters tuning. The Canadian Journal of Chemical Engineering
Systems Engineering (ICISE). IEEE, pp.100–104. 85(4): 537–548.
Ziegler JG and Nichols NB (1942) Optimum settings for
automatic controller. Transactions of the American Socity
of the Mechanical Engineers 64(11): 759–765.

You might also like