Professional Documents
Culture Documents
Modeling and Identification of Hydraulic-Electric Actuator
Modeling and Identification of Hydraulic-Electric Actuator
Modeling and Identification of Hydraulic-Electric Actuator
ABSTRACT In this work, a dynamic modeling and system identification process have been conducted
on an hydraulic-electric actuator driven system. The work aims to design the system model in term
of parameters that can be easily identified in time and frequency domains, using some known system
identification methods, such as the Least Square Estimation, and the Matlab System Identification Toolbox.
The estimate model is validate through open-loop numerical simulation of the real and estimate models.
Finally, an application built in Matlab App Designer has been proposed in order to illustrate the modeling
analysis, as well as the identification process.
INDEX TERMS Dynamic modeling, Hydraulic-electric actuator, System identification, Matlab App
Design.
I. INTRODUCTION of the real system is known. While a grey box has only a part
A model is a set of mathematical equations describing ad- of known system parameters.
equately a physical system [1]. A physical system can be The current work aims to identify the parameters of a
a robotic, biological, chemical, social, and economical. A system in time and frequency domains. A propose a designed
system can also be an existing real problem, caused by hu- application based on Matlab App Designer that can be used
man, continuous or discrete system, deterministic or random a didactic material. The system considered in this work is na
system. G.Box stated that all models are wrong, but some hydraulic-electric actuator driven system. The choice of such
are useful [2]. Einstein stated that everything should be make a system is motivated by its wide application such as lifting,
simple as possible, but no too simple. Thus an adequate conveyor system, hexapod system, car partaking system, and
system is represented by equation that gives sufficiently a manipulator arm.
precise prediction of system properties without being hard The layout of the work is as follows: the section (II)
to be evaluated. A real problem can be classified i nto three describe the dynamic model and the modeling analysis of
categories [3]: forward, inverse and design. In a forward the hydraulic-electric actuator, the section (III) focuses on
problem, one assumes the system parameter are known, and the system identification techniques, and the section (IV)
the problem is reduced to the prediction only. In inverse proposes a Matlab App designer applications to demonstrate
problem, the problem focus on the determination system the modeling analysis and the system identification process.
parameters from known input and output data, while a design
problem aims to regulate a certain state of the system for II. DYNAMIC MODEL OF AN HYDRAULIC-ELECTRIC
given its parameters. ACTUATOR
The system dynamics is characterized by the following equa-
In real situation, the system parameters, even in a design
tions (1) [5] [6].
problem, are unknown. Therefore, the system can be identi-
fied b y a n e quivalent o r e stimate s ystem w orking c loser to M ẍ + B ẋ + Af S(ẋ) + d(t) = A1 p1 − A2 p2 (1a)
the real system. Such an identified system can be white box,
βe V̇1 βe Q1
black box, or a grey box [4]. A white box represents a system ṗ1 = − + (1b)
derived from physics law and experimental observations and V1 V1
expressed in mathematics form, including variables, parame- βe V̇2 βe Q2
ṗ2 = − − (1c)
ters and uncertain terms. A black box assume none parameter V2 V2
1
CIANYI Y.KAZADI, READING PAPER
δy = C δξ + D δu
0 -0.1
T 0 1 2 3 4 5 0 1 2 3 4 5
0 λ5 14
8
0 λ4 0 0 −λ6 12
−1 T 6 10
and C = I4×4 , D = 04×1 , and d(t) = [0, −M , 0, 0] 0 1 2 3 4 5 0 1 2 3 4 5
(12)
with λi the system parameters given in Tab.(2) with Figure 2: Simulation results of the hydraulic-electric actuator.
Case 1: u = 2 sin(4πt) - d = 5.0e3 sin(2πt), Case 2:
Table 2: System parameters u = 2 sin(4πt) - d = 0, Case 3: u = 6 sin(4πt) -
Parameters formula d = 5.0e3 sin(2πt), Case 4: u = 6 sin(4πt) - d = 0
λ1 (B + 1800π −1 Af )/M
λ2 A1 /M
λ3 A2 /M
λ4 βe /L0 4 simulation results because in each couple of cases the
βe kq1 p disturbance d(t) is divided by the big value of mass M ,hence
λ5 ∆ξ3,e
A1 (L0 + ξ1,e ) it does not affect much.
βe kq2 p
λ6 ∆ξ4,e The open-loop numerical simulations of the linearized
A2 (L0 − ξ1,e )
model have resulted in the following response (3), under
the cases as during nonlinear system open-loop numerical
∆ξ3e = ps − ξ3e ; u > 0 simulations.
∆ξ3e = ξ3e ; u < 0
(13) 0.04 0.1
∆ξ 4e = ξ4e ; u > 0 case_1 case_1
case_2 case_2
x = ∆ξ1 + ξ1,e 0.01 -0.05
v = ∆ξ + ξ
2 2,e 0 -0.1
(14) 0 1 2 3 4 5 0 1 2 3 4 5
p 1 = ∆ξ 3 + ξ 3,e
p2 = ∆ξ4 + ξ4,e
25 35
case_1 case_1
case_2 30 case_2
20
Consider the initial state variables of system at x(t = 0) = case_3
case_4 25
case_3
case_4
ẋ(t = 0) = 0, p1 (t = 0) = 12 M P a, and p2 (t = 0) = 15 20
A1 10 15
3
CIANYI Y.KAZADI, READING PAPER
The results from figure (3) shown how proportionally lin- 0.04 0.1
ear are the outputs from low level to high as input oscillation 0.03 0.05
10-3 25 35
15 0.04
0.03 30
20
0.02 25
10
15 20
0.01
0 10 15
5 10
-0.01
5
5
-0.02
0 0
0 -0.03 0 1 2 3 4 5 0 1 2 3 4 5
0 1 2 3 4 5 0 1 2 3 4 5
16 24
Figure 6: Nonlinear model versus linearized model for case
22
3: u = 6 sin(4πt) - d = 5.0e3 sin(2πt)
14
20
12 18 0.04 0.1
16
0.03 0.05
10
14
0.02 0
8 12
0 1 2 3 4 5 0 1 2 3 4 5
0.01 -0.05
25 35
0.014 0.04 30
20
0.012 0.03 25
15 20
0.01 0.02
0.008 0.01 10 15
0.006 0 10
5
0.004 -0.01 5
0.002 -0.02 0 0
0 1 2 3 4 5 0 1 2 3 4 5
0 -0.03
0 1 2 3 4 5 0 1 2 3 4 5
12 18
requires the nonlinear system (1) to be expressed in term of
10
16
linear regressor form (15).
14
8
0 1 2 3 4 5
12
0 1 2 3 4 5
y1 (t) = ϕ1 (t)T θ1 + ∆1 (15a)
T
y2 (t) = ϕ2 (t) θ2 (15b)
Figure 5: Nonlinear model versus linearized model for case
2: u = 2 sin(4πt) - d = 0 where ϕ(t) is the regressor matrix given by ϕ1 (t) =
T
2 ẋ
ẍ, ẋ, tan−1 (900 ẋ), 1 and ϕ2 (t) = − +
From the above results Fig.(4) - Fig.(7), the comparison √π L 0 +x
reveals that there is close behavior of nonlinear and linearized Kq1 ∆p1
u. θ is the unknown vector parameters given by
system. This reveals also how good is the linearization being A1 L0 + x T
done. θ1 = M, B, Af , do and θ2 = βe . ∆ is the noise given by
d(t) = d0 +∆t ( with d0 is the slow time-varying component
III. SYSTEM IDENTIFICATION OF of the disturbance d(t)), and y(t) is the output function given
HYDRAULIC-ELECTRIC ACTUATOR by y1 (t) = A1 p1 − A2 p2 and y2 (t) = ṗ1 .
A. TIME DOMAIN SYSTEM IDENTIFICATION As noticed, a stable filter H(s) with relative order more
Consider a Least Square Estimation method for time do- than one (17) should be applied to both sides of (15) in order
main system identification. The application of such method to cancel out some noise attached to the measurable data x, ẋ,
4
CIANYI Y.KAZADI, READING PAPER
and ẍ, which are obtained through sensors. The acceleration B. FREQUENCY DOMAIN SYSTEM IDENTIFICATION
ẍ is obtained from differentiation, and through filtering, all Consider the linearized model (11), with given initial states
the signals are expected to be in the same phase. Therefore x0 = 0, ẋ0 = 0, p1o = 12e6 M P a, p2o = 12e6 M P a,
(15) becomes (16) u0 = 0, and d = 0. And let define new state variable such
that δξ = Z and δ ξ˙ = Ż, (11) can be written as (19).
yf (t) = ϕ(t)Tf θ + ∆f (16)
ż = Az + Bu (19a)
1 y = Cz + Du (19b)
H(s) = (17)
τf s + 1 where A, B et C are given in (12), with λi the system
where τf = 0.01 for quick stability. The estimate vector parameters.
parameter θ̂ if then obtained using the LSE formula [4]: The transfer function from input u to output y of the model
−1 T (19) is given by
θ̂ = ϕTf ϕf
ϕf yf (18)
P (s) = C(sI − A)−1 B + D (20)
The physical parameters for the system obtained with no In term of system parameters, (20) can be written as (21).
regard to the disturbance term ∆f are given in Tab.(3).
X(s) λ5 λ2 + λ6 λ3
P (s) = = 3 (21)
Table 3: LSE parameter estimate under no disturbance U (s) s + λ1 s2 + (λ2 λ4 + λ3 λ4 )s
Parameters Units Nominal Value Estimated Value It is obvious that the open-loop transfer function P (s) has
M kg 4000 4000 a one of the poles at zero, which make P (s) to be unstable.
B N m/s 1500 1500
Af N 3000 3000 Since the purpose of the linearization is to ensure that the
βe Pa 1.1e9 9.15e8 system variables stay no far away from their initial values,
Dc m 0.12 0.12 therefore one can use a proportional controller C(s) = Kp
dc m 0.07 0.07
d0 - 0 so that the closed-loop transfer function H(s) (22) is stable.
236.4
P̂ (s) = (26)
s3 + 9.073s2 + 1.711e4s + 8.598e − 7
which by identification with (21), leads to the following
estimated system parameters:
B̂ + 1800π −1 Âf
= λ̂1 (27a)
M̂
Â1 /M = λ̂2 (27b)
Â2 /M = λ̂3 (27c)
β̂e /L̂0 = λ̂4 (27d)
Figure 8: data input generated using idint Matlab command
β̂e kq1 p
∆P1o = λ̂5 (27e)
Â1 L̂0
β̂e kq2 p
∆P2o = λ̂6 (27f)
Â2 L̂0
From (27a) B̂ + 1800π −1 Âf = 9.073 M̂ . Using the stability
condition of (21), B̂ + 1800π −1 Âf > 0.
From (27b), (27c), (27e), and (27f), one can estimated βe
as
L̂0 M̂
β̂e = √ √ ≈ 1.1793e9
kq1 ∆P1o + kq2 ∆P2o
where Â1 , Â2 and M̂ = 4000 being measurable
0.015
0.01
0.005
-0.005
-0.01
-0.015
-0.02
-0.03
0 1 2 3 4 5 6 7 8 9 10
References
[1] R. Aris, Mathematical modelling techniques. Courier Corporation, 1994.
[2] G. E. Box and N. R. Draper, Empirical model-building and response
surfaces. John Wiley & Sons, 1987.
[3] T. W. Thomas Witelski and M. B. Mark Bowen, “Methods of mathematical
modelling: Continuous systems and differential equations,” 2015.
[4] Y. B. Zhu Xiaocong, Mechatronics control I. Zhejiang University, Lecture
notes, 2020.
[5] M. A. Avila, A. G. Loukianov, and E. N. Sanchez, “Electro-hydraulic
Figure 12: Frequency response in Bode diagram actuator trajectory tracking,” in Proceedings of the 2004 American Control
Conference, vol. 3, pp. 2603–2608, IEEE, 2004.
[6] B. Šulc and J. Jan, “Non linear modelling and control of hydraulic actua-
tors,” Acta Polytechnica, vol. 42, no. 3, 2002.
nonlinearity of the real system. Two system identification
[7] L. Ljung and R. Singh, “Version 8 of the matlab system identification
methods have been used: the least square estimation method toolbox,” IFAC Proceedings Volumes, vol. 45, no. 16, pp. 1826–1831, 2012.
for the frequency domain identification, a nd t he M atlab SI [8] N. H. Harun, H. Hambali, M. G. Hassan, and K. N. Karim, MATLAB APP
toolbox for frequency domain identification. The results have Designer: Learn By Example (UUM Press). UUM Press, 2017.
shown a good estimation in time-domain. However, despite
input disturbance the estimation in frequency domain has
also shown remarkable results. For didactic material, GUI APPENDIX. STATE-SPACE LINEARIZED MATRICES
application, has been proposed, in order to illustrate the mod- The first order Taylor expansion of (6) around the working
eling analysis, as well as the system identification process. point (10) gives the following matrices A, B, and d.
7
CIANYI Y.KAZADI, READING PAPER
where
p
kq1 ps − ξ3,e ; u > 0
∂f1 ∂f1 ∂f1 ∂f1 ∂Q1
= (34a)
∂ξ1 ∂ξ2 ∂ξ3 ∂ξ4 ∂u ξ=ξe k pξ ; u < 0
ξ=ξe ξ=ξe ξ=ξe ξ=ξe q1 3,e
∂f2 ∂f2 ∂f2 ∂f2
−k ue
q1
∂ξ
1 ∂ξ2 ∂ξ3 ∂ξ4
p = 0 ;u > 0
ξ=ξe ξ=ξe ξ=ξe ξ=ξe 2 p − ξ3,e
A=
s
∂f3 ∂f3 ∂f3 ∂f3 ∂Q1
= (34b)
∂ξ1
∂ξ2 ∂ξ3 ∂ξ4
∂ξ3 ξ=ξe
kq1 ue
ξ=ξe ξ=ξe ξ=ξe ξ=ξe
= 0 ;u < 0
∂f4 ∂f4 ∂f4 ∂f4 p
2 ξ3,e
∂ξ1 ξ=ξe ∂ξ2 ξ=ξe ∂ξ3 ξ=ξe ∂ξ4 ξ=ξe in addition,
(28)
T
∂f1 ∂f2 ∂f3 ∂f4 ∂f4 βe Q2e
B= =− = 0; (35a)
∂u ξ=ξe ∂u ξ=ξe ∂u ξ=ξe ∂u ξ=ξe ∂ξ1 ξ=ξe A2 L20
(29) ∂f4 βe
T = (35b)
∂f1
∂f2 ∂f3 ∂f4 ∂ξ2 ξ=ξe L0
d=
∂d ξ=ξe ∂d ξ=ξe ∂d ξ=ξe ∂d ξ=ξe
∂f4 ∂f4
= =0 (35c)
(30) ∂ξ3 ξ=ξe ∂d ξ=ξe
where
with Q2e the operating point flow
∂f1 ∂f1 ∂f1 ∂f1 ∂f1
= = = = =0 (31a)
∂ξ1 ∂ξ3 ∂ξ4 ∂u ∂d ξ=ξe
βe ∂Q2 ∂Q2
∂f1 ∂f4 (Q4 (ξ4 , u)) = − δu + δξ4
=1 (31b) A2 L0 ∂u ∂ξ4
∂ξ2 ξ=ξe ξ=ξe
ξ=ξe
(36)
where
p
kq2 ξ4,e
;u > 0
∂Q2
∂f2 = (37a)
=0; (32a) ∂u ξ=ξe k pp − ξ
∂ξ1 q2 s 4,e ;u < 0
ξ=ξe
k u
∂f2 B 2Af q2 e
= 0 ;u > 0
=− − 900
; (32b) p
2 ξ4,e
∂ξ2 M πM
ξ=ξe ∂Q2
= (37b)
∂f2 A1 ∂ξ4 ξ=ξe
kq2 ue
= (32c)
− p =0 ;u < 0
∂ξ3 ξ=ξe M 2 ps − ξ4,e
∂f2 A2
=− (32d) By substituting the equations (32) to (37) into the state
∂ξ4 ξ=ξe M
space equations (28) to (30), the matrices A and B in the
∂f2 equation (11) are given in (12).
=0; (32e)
∂u ξ=ξe
∂f2
= − M −1 ; (32f)
∂d ξ=ξe
∂f3 βe Q1e
=− = 0; (32g)
∂ξ1 ξ=ξe A1 L20
∂f3 βe
=− (32h)
∂ξ2 ξ=ξe L0
√
with Q1e = kq1 ∆ξ3 ue = 0 the operating point flow.
βe ∂Q1 ∂Q1
∂f3 (Q1 (ξ3 , u)) = δu + δξ3
A1 L0 ∂u ξ=ξe ∂ξ3 ξ=ξe
(33a)
∂f3 ∂f3
= =0 (33b)
∂ξ4 ξ=ξe ∂d ξ=ξe
8
CIANYI Y.KAZADI, READING PAPER
Clic to input
parameters
and initial
Display input responses conditions Time domain SI
9
CIANYI Y.KAZADI, READING PAPER
10