Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY 1

Hardware Implementation of an ADRC Controller


on a Gimbal Mechanism
Behzad Ahi and Amin Nobakhti, Senior Member, IEEE

Abstract— Active disturbance rejection control (ADRC) is


applied to a one-axis gimbal mechanism. The dynamic model
of the system is derived and validated from mathematical
modeling and practical experiments. Disturbances acting on the
complete model of the gimbal mechanism are introduced via
base lateral acceleration and angular motion. The ADRC is
designed by utilizing an extended state observer for observing
and suppressing the effects of external disturbances and internal
parameter uncertainties. A PID controller is used to form a basis
of comparison for set-point tracking and disturbance rejection
performance. The effects of identification errors and observer
bandwidth are experimentally investigated. Based on simulation
and experimental results, the effectiveness of ADRC in the
presence of external disturbances and parameter uncertainties
is illustrated. Fig. 1. Single-axis gimbal configuration.
Index Terms— Active disturbance rejection control (ADRC),
gimbal mechanism, inertially stabilized platform (ISP), system
uncertainty.

I. I NTRODUCTION

I N MANY applications, the motion of sensing instruments,


such as camera, radar, laser, and navigation equipment
mounted on a platform, should be isolated from the base Fig. 2. ISP control loop.
angular motion. For this purpose, inertially stabilized plat-
form (ISP), typically involving one or two axes gimbal assem-
blies, is used, as shown in Fig. 1. ISPs usually comprise two
in common with the center of rotation). Abdo et al. [5] inves-
loops, the inner one, referred to as the stabilization loop and
tigate dynamic mass imbalance effects in a two-axis gimbal
an outer tracking loop [1], as shown in Fig. 2. The tracking
system, and a self-tuning PID-type fuzzy controller is proposed
loop must have sufficient bandwidth to track the sensor line
to eliminate torque disturbances. In [6], an internal model
of sight kinematics [2], and generates the reference signal
controller is designed to counter static mass imbalance torque
for the stabilization loop. The overall performance is mainly
in a three-axis gyro stabilized system. In [7], a multilayer
dependent on the ability of the inner loop to reject external
neural network is utilized to control a one-axis gimbal system
disturbances, which is the problem specifically addressed in
in the presence of various sources of disturbances, such as
this brief.
static mass imbalance, gimbal friction, and restoring torque.
The most critical performance metric for an ISP is torque
In [8], a robust H∞ controller is used for the control of a
disturbance rejection [3]. In [4], equations of motion for
two-axis gimbal sensor system with a diagonal inertia matrix.
the two-axis gimbal configuration are presented. The model
An observer-based sliding mode control is implemented in [9].
includes the effects of dynamic mass imbalance arising from
Friction torque, another sources of ISP disturbances, is studied
an asymmetrical mass distribution. However, it is assumed the
in [10] and [11], and a new friction identification approach
gimbal has no static mass imbalance (mass center of gimbal is
based on dual-relay feedback is introduced with application
Manuscript received June 13, 2016; revised May 17, 2017; accepted to ISPs in [12]. Cable restraint torques, geometric errors,
August 14, 2017. Manuscript received in final form August 20, 2017. and measurement sensor imperfections have been studied
Recommended by Associate Editor H. Fathy. (Corresponding author: in [13] and [14].
Amin Nobakhti.)
The authors are with the Department of Electrical Engineering, Sharif Uni- This brief outlines the design and implementation of an
versity of Technology, Tehran 11365-11155, Iran (e-mail: ahi@ee.sharif.edu; active disturbance rejection control (ADRC) system proposed
nobakhti@sharif.edu). by Han [15]. ADRC has been successfully applied to several
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. applications (see [16]–[19]) and has been further extended
Digital Object Identifier 10.1109/TCST.2017.2746059 into various forms, such as compound ADRC [20], adaptive
1063-6536 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

ADRC [21], dynamic decoupling ADRC [22], back-stepping


ADRC [23], predictive ADRC [24], and neural ADRC [25].
Traditionally, multi-axis gimbal mechanisms are considered
as coupled SISO systems with the interaction terms taken as
the external disturbances. Since ADRC uses an extended state
observer (ESO) to estimate and subsequently reject the total
disturbance, it is reasonable to expect it leads to improvements
in comparison with traditional PID controls.
We derive a complete model of the gimbal mechanism
considering various sources of disturbances in addition to
the back EMF effect. A linear ADRC (LADRC) [26] is
subsequently designed and implemented on a real gimbal
system and carefully compared with a PID controller.
II. BASIC P ROBLEM F ORMULATION Fig. 3. Geometric configuration of a gimbal.

A. Dynamic Mass Imbalance


The method of [4] is adopted to model dynamic mass The y-component of the total torque is obtained by substi-
imbalance arising from asymmetrical mass distribution of the tuting (2) and (6) into (4)
gimbal. In Fig. 1, two reference frames are specified. Frame 
P is fixed to the platform with axes (x, y, z) and frame B is T py = Jy ω̇ py + (Jx − Jz ) ω px ω pz + Jx z ω2pz − ω2px
fixed to the base with axes (i, j, k). The frame axes conform + Jyz (ω̇ pz − ω py ω px ) + Jx y (ω̇ px + ω py ω pz ) (7)
at ψ = 0. The axis of rotation for both frames is similar in this
where T py is equal to the external torque applied to the y-
section. The case in which the axes are different is considered
axis and ω̇ py is the angular acceleration of platform about the
subsequently. The frame transformation matrix is
⎡ ⎤ y-axis. According to Newton’s second law for rotation, one
cosψ 0 −sinψ can assume the remaining terms of (7) as torque disturbances
L PB = ⎣ 0 1 0 ⎦ (1) caused by the base motion
sinψ 0 cosψ 
TD1 = (Jx − Jz ) ω px ω pz + Jx z ω2pz − ω2px
for a vector ωb expressed by its coordinates in frame B, and
L PB ωb gives the coordinates of the same vector in frame P. + Jyz (ω̇ pz − ω py ω px ) + Jx y (ω̇ px + ω py ω pz ). (8)
The angular velocity vector of the platform and base related Consistent with the mechanical constraints illustrated in Fig. 1,
to the inertial space in frames P and B are, respectively the relation between angular velocities ω̄ P/I and ω̄ B/I is
 T  T ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ω̄ P/I = ω px , ω py , ω pz , ω̄ B/I = ωbi , ωbj , ωbk . ω px ωbi 0
(2) ⎣ω py ⎦ − L P B ⎣ωbj ⎦ = ⎣ψ̇ ⎦ . (9)
The output of the stabilization loop in Fig. 2 is ω py , which ω pz ωbk 0
is measured by a gyro mounted on the platform. The inertia Manipulating (8) and (9), TD1 can be stated in terms of base
matrix of the platform, which includes the inertia moment of angular velocities related to inertial space, angle ψ, and their
all the rotational parts, is time derivatives. For instance, assuming POIs to be zero yields
⎡ ⎤
Jx Jx y Jx z  2
TD1 = (Jx − Jz ) ωbi − ωbk2
sinψ cosψ
J P = ⎣ Jx y J y Jyz ⎦ (3)
Jx z Jyz Jz + ωbi ωbk (cos2 ψ − sin2 ψ) . (10)
in which Jx , Jy , and Jz are platform moments of inertia
about the x, y, and z axes and Jx y , Jx z , and Jyz are called B. Static Mass Imbalance
gimbal moments products of inertia (POI) originating from the In Section II-A, the center of gravity (CG) of the rotating
dynamic mass imbalance. According to the moment equation parts is assumed to be precisely on the pivot point. In reality,
for a rotating frame, the total torque applied to the platform an offset between gimbal pivot point and platform CG is
is inevitable (Fig. 3). This slight offset can cause significant
d torque disturbances about the pivot point. By modifying the
T̄ P = H̄ P + ω̄ P/I × H̄ P (4)
dt approach presented in [7], the torque disturbance due to the
in which the angular momentum H̄ P is defined as static mass imbalance, assuming α = θbj + ψ + θ0 , can be
expressed as
H̄ P = J P ω̄ P/I . (5)
TD2 = mr1r0 θ̈bj cos(α − θbj ) + mr1r0 θ̇bj
2
sin(α − θbj )
Substituting (2) and (3) into (5) yields
⎡ ⎤ ⎡ ⎤ + mr0 v b θ̇v cos(α − θv ) (11)
H px Jx ω px + Jx y ω py + Jx z ω pz
H̄ P = ⎣ H py ⎦ = ⎣ Jx y ω px + Jy ω py + Jyz ω pz ⎦ . (6) where v b is the base linear velocity, m is the total mass of
H pz Jx z ω px + Jyz ω py + Jz ω pz platform, r0 is the offset distance, θv is the angle of linear
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

AHI AND NOBAKHTI: HARDWARE IMPLEMENTATION OF AN ADRC CONTROLLER ON A GIMBAL MECHANISM 3

TABLE I
DC M OTOR S PECIFICATIONS

Fig. 5. Simplified block diagram of gimbal mechanism.

Fig. 4. Block diagram of motor considering base motion.

velocity vector of base related to inertial reference axis, θbj is


the base angle related to the reference axis, and θ0 is angle
between the z axis and platform CG. The term v b θ̇v in (11)
can be interpreted as the lateral acceleration of the base.
Furthermore, manipulating (9) permits θbj + ψ to be found
by integrating the output of stabilization loop ω py .

C. Friction and Back EMF Effect


The DC motor transfer function that takes into account Fig. 6. Input and output signals used for system identification.
viscous friction torque can be written as [5]
ω kt required for modeling the system have been defined with
= (12)
V (Ls + R)(J s + b) + kt ke the exception of the viscous friction coefficient b. While,
where b is the viscous coefficient and other parameters are according to (12), the DC motor is a second-order system,
introduced in Table I. In (12), ω is the angular velocity of in many instances, the rotor inductance can be neglected.
platform relative to the base. This is equal to time derivative Subsequently, the motor transfer function with respect to the
of ψ, as shown in Fig. 1. Accordingly, from (9) and (12), platform angle becomes
the block diagram of the DC motor with base angular motion ψ kt
is shown in Fig. 4. Shifting the disturbances to the output and = . (13)
V s(R J s + b R + kt ke )
considering TD as the combination of TD1 and TD2 and any
other unmodeled torque disturbances leads to the arrangement The MATLAB system identification toolbox is used to
shown in Fig. 5, illustrating the effects of friction and back identify the parameters of (13) based on real measurements.
EMF on the gimbal mechanism. This model is subsequently Experiments revealed that using the 17-b absolute encoder
used in Section V. data instead of the MEMS gyro as the output of the system
resulted in less modeling error. The first half of the data shown
in Fig. 6 is used for identification, and the second half is used
III. S TABILIZATION L OOP C OMPONENTS
for validation. For the sampling rate of 5 ms, the following
A. DC Motor Specifications transfer function has been identified:
A DC torque motor made by Racing Electric with a full- ψ 0.1739
bridge pulse width modulation driver (L6225D) made by ST = . (14)
V s(0.0835s + 1)
is used. The specifications of the motor are listed in Table I.
The inertia matrix of gimbal computed by SolidWorks is The model has best fits of 83.88% and FPE = 4.44 × 10−5 ,
⎡ ⎤ ⎡ ⎤ which confirms that the desired accuracy has been achieved.
Jx Jx y Jx z 106 2 −3 By substituting numerical values of Table I in (13) and
J P = ⎣ Jx y Jy Jyz ⎦ = 10−4 ⎣ 2 100 5 ⎦ Kg.m2 . comparing it with (14), two values for b are derived
Jx z Jyz Jz −3 5 92
b1 = 0.129, b2 = 0.116 (15)
Inertia moment J in (12) is equals to Jy = Jl + Jm
(see Fig. 1), where Jl and Jm are the y-components of inertia the values differ by 10%, which is acceptable. For a controller
moment for load and motor, respectively. All the parameters design, the mean values is used.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

TABLE II where u 0 (t) is the control law, which achieves the satisfactory
G YROSCOPE C HARACTERISTICS performance. In LADRC, this is usually chosen to be propor-
tional to estimated tracking error
u 0 (t) = k p (v(t) − x̂ 1 (t)) (21)
where v(t) is the reference signal. Regarding impeccable
observations, the closed-loop system can be derived as follows:
1
ẏ(t) + y(t) = v(t). (22)
kp
Therefore, k p can be tuned by choosing the desired 2% settling
B. Measurement Sensors Specifications time k p = −s CL ≈ 4/Tsettle, where s CL is the desired closed-
loop pole. Applying the procedure to the system (17) yields
The digital MEMS gyroscope (Table II) used in this appli-
an nth-order closed-loop system with n tuning parameters.
cation is made by Analog Devices (ADIS16137). Prior to the
Observer gains are specified via bandwidth parameteriza-
analog-to-digital converter, this rate gyro goes through a two-
tion [26], which suggests choosing observer poles at the same
pole, low-pass filter. The first pole is at 429 Hz, and the second
place
pole is at 1595 Hz. Accordingly, the gyro transfer function is
27 × 106 l1 = −2s ESO , l2 = (s ESO )2 (23)
G gyro (s) = 2 (16)
s + 12710 s + 27 × 106 in which the observer bandwidth is determined by
averaging and decimating filters are set to default. Out- s ESO = kob s CL . kob is referred to as the observer coeffi-
put data rate is 2048 SPS. Moreover, a 17-b absolute cient. Increasing kob leads to a more accurate observation,
encoder (5853FS3) from Kubler is used to measure ψ. and improved control. Practical issues that limit how much
the observer coefficient can be increased are discussed in
IV. C ONTROLLER D ESIGN Section VI.
A. Design LADRC in Continuous Time Domain
Assume the following linear system with known order n: B. Stability Analysis
α0 The expressions for closed-loop error dynamics ec (t) =
G(s) = (17)
βn s + βn−1 s
n n−1 + · · · + β1 s + 1 v(t) − x 1 (t) and eo (t) = x(t) − x̂(t) can be obtained by
an LADRC can be designed for (17) by knowing only the substituting (20) and (21) into (18) as follows:
parameter b = α0 /βn . We show that only a rough estimate of b
ėc (t) = −k p (ec (t) + eo1 (t)) − eo2 (t) + v̇(t) (24)
denoted by b0 is sufficient for LADRC to deliver a superior
performance compared with PID control. Taking the angular with reference to (19), one can show that the overall closed-
velocity as the output and comparing (13) with (17), it is loop dynamic takes the form
evident that b = (kt /R J ), implying that viscous coefficient ⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
has almost no effect on b, and by consequence, the control ėc (t) −k p −k p −1 ec (t) v̇(t)
⎣ėo1(t)⎦ = ⎣ 0 −l1 1 ⎦ ⎣eo1(t)⎦ + ⎣ 0 ⎦ . (25)
law. Furthermore, if the parameters kt , R, and J are known,
the LADRC can be implemented without performing any ėo2 (t) 0 −l2 0 eo2 (t) f˙(t)
system identification. Consider Since in the general case f˙(t) is not independent of the
ẋ 1 (t) = f (t) + b0 u(t) system states, it must be assessed for stability. Assuming b =
(18) b0 + b in which b is the unknown part of b, f (t) becomes
y(t) = x 1 (t)
1 b
where y(t) is the measured angular velocity, u(t) is the f (t) = (α0 d(t)−x 1(t))− (x̂ 2 (t) − k p (v(t)− x̂ 1 (t)))
β1 b0
input voltage, and f (t), referred to as the total disturbance,
(26)
is a time-varying function that depends on the states, inter-
nal uncertainties, unmodeled dynamics, and external distur- in which d(t) is the external input disturbance. Accordingly,
bances. To estimate f (t), the system is extended by defining from (25) and (26)
x 2 (t) = f (t). Taking L = [l1 l2 ]T and x̂(t) as the estimate of ⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
x(t) = [x 1 (t) x 2 (t)]T , the observer for an extended system is ėc (t) −k p −k p −1 ec (t) D
⎣ėo1 (t)⎦ = ⎣ 0 −l1 1 ⎦ ⎣eo1 (t)⎦ + ⎣ 0 ⎦ (27)
˙ = A x̂(t) + Bu(t) + L(y(t) − ŷ(t))
x̂(t) ėo2 (t) A B C eo2 (t) E
(19)
ŷ(t) = C x̂(t) where A, B, C, D, and E are defined as follows:
   
where A = 00 10 , B = b00 and C = [ 1 0 ]. Based on the kp b 2 kp b 
A=− − k , B=− − l2 + k 2p + l1 k p − l2
online estimation of ESO, the control law becomes β1 b0 p β1 b0
−1 1 b α0
u(t) = (x̂ 2 (t) − u 0 (t)) (20) C = − , D = v̇(t), E = k p v̇(t) + ḋ(t).
b0 β1 b0 β1
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

AHI AND NOBAKHTI: HARDWARE IMPLEMENTATION OF AN ADRC CONTROLLER ON A GIMBAL MECHANISM 5

From the system matrix of (27), with the assumption of


positive Tsettle and kob , it is straightforward to verify the
asymptotic stability of the error dynamics for any b0 , which
satisfies b0 b > 0. The former indicates that knowing the sign
of b is sufficient to ensure the asymptotic stability of the
closed-loop system with the LADRC. Evidently, the bound-
edness of v̇(t) and ḋ(t) assures the bounded-input bounded-
output (BIBO) stability of (27). Furthermore, one can show
that by generalization of u 0 (t) at the absence of b
u 0 (t) = k p (v(t) − x̂ 1 (t)) + v̇(t) (28)
the BIBO stability can be inferred from the boundedness
of ḋ(t). Nevertheless, the first approach is used in this brief Fig. 7. Comparison of disturbance rejection behavior of LADRC with a
different observer coefficient versus PI.
so that the subsequent comparison of the LADRC with a
PI controller shall be fair.
V. S IMULATION A NALYSIS
C. Design LADRC in Discrete Time Domain A. Input Disturbance Rejection
The LADRC has been implemented on a digital signal The simulated input disturbance rejection ability of the
processor (DSP) board. Since the computation time can be controllers is compared in Fig. 7. An input step disturbance
neglected compared with the sampling time, a discrete current with an amplitude of 10 is applied from t = 0.5 s to t = 1 s
observer [27] is designed as follows: and an input ramp disturbance with a slope of 20 is applied
x̂(k + 1|k) = Ad x̂(k) + Bd u(k) from t = 1.5 s thereafter. As mentioned in Section IV,
 increasing the observer gain from 5 to 20 has led to better
x̂(k + 1) = x̂(k + 1|k) + L d y(k + 1) − Cd x̂(k + 1|k)
estimation and, consequently, improved disturbance rejection.
where Ad , Bd , and Cd refer to discrete time versions of The effect of the observer coefficients is further investigated
matrices in (19) and Ad − L d Cd Ad determines the observer in Section VI. As expected, the transient responses of LADRC
dynamic. In [28], it has been shown that for current form and PI are similar in conformity to the design objectives.
observers, ZOH discretization leads to the most accurate ESO. According to (27), asymptotic stability of the estimation error
Consequently dynamics for both constant input disturbances and BIBO
1 stability for ramp input disturbances is confirmed in Fig. 7.
l1 = 1 − 2 (z ESO )2 , l2 = (1 − z ESO )2
Ts
B. Output Disturbance Rejection
in which Ts is the sampling time and z ESO is observer pole
location that can be found by mapping s ESO to the Z -plane In this section, a sinusoidal output disturbance is simulated.
via z ESO = es .Ts . Finally, the discrete time domain control
ESO ESO performance will decrease with the increasing frequency
law becomes of the sinusoidal disturbance. The performance of the LADRC
controller is superior to that of PI for disturbances with
1
u(k) = (x̂ 2 (k) − k p (v(k) − ŷ(k))). (29) frequencies up to half the observer bandwidth. For instance,
b0 if kob = 10 and Tsettle = 0.1, the observer bandwidth is
around 64 Hz, and LADRC has a superior performance up
D. Comparative PID and Antiwindup Strategy to 32 Hz. For illustration, two disturbances at 5 and 20 Hz are
For comparison purposes, an IMC-based PID controller with represented in Fig. 8. The PI performance is consistent for a
the same bandwidth as LADRC is designed. Based on the wide range of frequencies, and for the frequencies above half
methodology of [29], the gains are computed as follows: of the observer bandwidth, there are no noticeable differences
4β1 4 in performance.
KP = , KI = , K D = 0. (30)
α0 Tsettle α0 Tsettle
The backward Euler discretization method is utilized to imple- C. Complete Model of Gimbal Mechanism
ment the PI controller. Assuming perfect ESO, the PID gives The performance of the gimbal mechanism has been simu-
a similar nominal closed-loop response as the LADRC. Note lated under various disturbances. The reference angular veloc-
that unlike the LADRC, both the steady-state gain value and ity is set to 40°/s. Base angular velocities are effective during
the time constant of the system are required to tune the PI. the period from t = 0 s to t = 1.5 s, assuming ωbk = 2°/s
Due to actuator saturation, it is necessary to implement an and ωbi = ωbj = S1 sin(4π) + S2 sin(10π), where S1 and S2
antiwindup strategy. In the LADRC, a simple and effective are varied between 0 and 100°/s. The lateral acceleration is
means is feeding the state observer with the limited actua- applied from t = 2 s to t = 3 s. It is assumed to reach its
tion [30]. For the PI, conditional integration is used, which maximum value in each simulation run with variable amplitude
stops integration when actuator is saturated, and ec (t) and u(t) between 0 and 30 g. The product of the gimbal mass and its
have the same sign. offset distance is assumed to be 1.5 × 10−3 kg·m. The output
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

Fig. 8. Comparison of disturbance rejection behavior for sinusoidal distur-


bance.

Fig. 10. Tracking comparison of LADRC with a different observer coefficient


Fig. 9. Step response of gimbal mechanism under different base angular versus PI. (a) Velocity. (b) Voltage.
velocities and lateral acceleration disturbances.
1) Observer Coefficient Effects on Set-Point Tracking:
Increasing the observer coefficient (when it is chosen greater
angular velocity in the LADRC with observer coefficient equal than 5) does not significantly vary the set-point tracking per-
to 20 and settling time equal to 0.1 s is shown in Fig. 9. formance of the LADRC. However, it does lead to increased
When the lateral acceleration exceeds 29 g, a steady-state oscillations in the controller output, as shown in Fig. 10.
error occurs as a result of actuator saturation. In other cases, As expected from Section IV-D, the closed-loop responses of
the proposed LADRC successfully rejects the disturbances. the LADRC and PI controllers are similar.
2) Observer Coefficient Effects on Disturbance Rejection:
An external torque disturbance is applied to the gimbal mech-
VI. E XPERIMENTAL R ESULTS
anism by setting ωbi and ωbk to be zero and ωbj to have
Actual experimental results for both set-point tracking and an amplitude of 100°/s and a frequency varying from 2 to
disturbance rejection are illustrated in the presence of identi- 2.4 Hz. ωbj is measured by a second gyroscope fixed to the
fication error and variation of observer coefficient. The con- base. The output angular velocity ω py is shown in Fig. 11,
trollers are implemented on a TI DSP board (TMS320F2810) which confirms the superior ability of the LADRC approach
with a sample time of 1 ms. The controller outputs are in countering the external disturbance.
limited to 13 V to evaluate the effectiveness of the antiwindup In this test, the actuator is saturated by setting the observer
strategies. The closed-loop bandwidth in both approaches is set coefficient (kob ) to 25. Increasing kob to a higher value will
to 40 rad/s. Note that while increasing the bandwidth directly lead to a deterioration of the LADRC performance. According
translates into improved disturbance rejection, typically the to [30], increasing measurement noise (with fixed observer
structural resonances of ISPs limit how much this can be coefficient) leads to more oscillations of controller output.
increased [3]. On the other hand, according to Fig. 10, increasing kob (with
fixed measurement noise) also leads to increased controller
output oscillations. Therefore, in order to be able to increase
A. Observer Coefficient Effects
the observer coefficient, it is necessary to use a sensor with
As described in Section V, the observer coefficient has less measurement noise.
a significant effect on a controller performance. A simple
rule of thumb is to set the observer coefficient between 3 B. Identification Error Effects
and 10 [30]. However, varying factors, such as disturbance In the remainder, a fixed observer coefficient of 13 is used to
dynamic, measurement noise, and actuator saturation, should investigate the effects of identification error on both set-point
be considered in determining the appropriate coefficient value. tracking and disturbance rejection.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

AHI AND NOBAKHTI: HARDWARE IMPLEMENTATION OF AN ADRC CONTROLLER ON A GIMBAL MECHANISM 7

Fig. 11. Disturbance rejection comparison of LADRC controllers with a


different observer coefficient versus PI. Fig. 13. Disturbance rejection performance of LADRC in the presence
of 100% identification error.

Fig. 12. Investigating tracking ability of controllers in the presence of 100%


identification error. (a) LADRC. (b) PI. Fig. 14. Performance comparison of LADRC and PI in the presence of
disturbances. (a) Output velocity. (b) Input voltage.

1) Identification Error Effects on Set-Point Tracking: 2) Identification Error Effects on Disturbance Rejec-
Asymptotic stability of the error dynamic at the presence of tion: An external disturbance with the same characteristics
excessive identification error in coefficient b is demonstrated in Section VI-A2 is applied to the base. According to Fig. 13,
in Section IV-B. According to [15], a rough approxima- the identification error will decrease the disturbance rejection
tion of b within a ±50% range is adequate for ADRC to ability of LADRC. However, even with ±100% identification
deliver a superior performance. Here, the LADRC controller is error, the disturbance rejection ability of the LADRC is still
tuned assuming ±100% error in identification of b. Similarly, superior to the PI controller, as verified in Fig. 11.
the PI controller is tuned for an assumed ±100% error in
the identification of process time constant (β1 ). Fig. 12(a)
indicates that the tracking behavioral of LADRC controller is C. Comparison of Controllers
nearly unchanged under these excessive variations, in contrast To conclude the experimental results, simultaneous com-
to the PI controllers [Fig. 12(b)] in which β0 and β1 are, parison of tracking and disturbance rejection performance of
respectively, the identified and actual values. controllers with the assumption of square wave (with 40°/s
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

TABLE III [6] S. Li and M. Zhong, “High-precision disturbance compensation for a


Q UANTITATIVE C OMPARISON OF P ERFORMANCES three-axis gyro-stabilized camera mount,” IEEE/ASME Trans. Mecha-
tronics, vol. 20, no. 6, pp. 3135–3147, Dec. 2015.
[7] C.-L. Lin and Y.-H. Hsiao, “Adaptive feedforward control for distur-
bance torque rejection in seeker stabilizing loop,” IEEE Trans. Control
Syst. Technol., vol. 9, no. 1, pp. 108–121, Jan. 2001.
[8] S. B. Kim, S. H. Kim, and Y. K. Kwak, “Robust control for a two-
axis gimbaled sensor system with multivariable feedback systems,” IET
Control Theory Appl., vol. 4, no. 4, pp. 539–551, 2010.
[9] B. Tamhane, D. Singh, S. Kurode, and P. Parkhi, “Observer based control
of a missile seeker system using sliding modes: Seeker system,” Asian
J. Control, vol. 18, no. 2, pp. 784–787, 2016.
[10] J. Fang, R. Yin, and X. Lei, “An adaptive decoupling control for three-
amplitude and 0.34-Hz frequency) as the reference signal is axis gyro stabilized platform based on neural networks,” Mechatronics,
presented here. ωbj has an amplitude of 80°/s and a frequency vol. 27, pp. 38–46, Apr. 2015.
[11] Y. Zou and X. Lei, “A compound control method based on the adaptive
ranging from 1.6 to 1.8 Hz. The observer coefficient is set neural network and sliding mode control for inertial stable platform,”
to 13. The results are depicted in Fig. 14, which once again Neurocomputing, vol. 155, pp. 286–294, May 2015.
illustrate the improved performance of the LADRC controller. [12] X. Chen, F. Fang, and X. Luo, “A friction identification approach based
on dual-relay feedback configuration with application to an inertially
In order to quantify the analysis of the controller performances stabilized platform,” Mechatronics, vol. 24, no. 8, pp. 1120–1131, 2014.
in the presence of disturbances, three performance measures, [13] J. Liu, T. Zhao, R. Ning, and J. Liu, “Physics-based modeling and
T T
such as IAE = 0 |e(t)|dt, ITAE = 0 t|e(t)|dt, and simulation for motional cable harness design,” Chin. J. Mech. Eng.,
T vol. 27, no. 5, pp. 1075–1082, 2014.
ISCE = 0 |u(t)| dt, are computed for both closed loops,
2
[14] Q. Tang, X. Wang, Q. Yang, and F. Liu, “Calibration error analysis of
and the results are listed in Table III. inertially stabilized platforms using quaternions and octonions in rotation
decomposition,” Proc. Inst. Mech. Eng., B, Eng. Manuf., vol. 230, no. 9,
pp. 1771–1777, 2016.
VII. C ONCLUSION [15] J. Han, “From PID to active disturbance rejection control,” IEEE Trans.
An inclusive dynamic model of a one-axis gimbal mech- Ind. Electron., vol. 56, no. 3, pp. 900–906, Mar. 2009.
[16] C.-E. Huang, D. Li, and Y. Xue, “Active disturbance rejection control
anism was derived and simulated under the influence of for the ALSTOM gasifier benchmark problem,” Control Eng. Pract.,
various sources of disturbances including base angular motion, vol. 21, no. 4, pp. 556–564, Apr. 2013.
dynamic mass imbalance, static mass imbalance, and friction [17] S. M. Manojlović, Ž. P. Barbarić, and S. T. Mitrović, “A novel active dis-
turbance rejection based tracking design for laser system with quadrant
torque. The effects of the LADRC observer pole locations on photodetector,” Int. J. Control, vol. 88, no. 6, pp. 1246–1256, 2015.
set-point tracking and disturbance rejection have been fully [18] Z. Gao, “Active disturbance rejection control for nonlinear fractional-
investigated by the simulation and experimental results. It has order systems,” Int. J. Robust Nonlinear Control, vol. 26, no. 4,
pp. 876–892, 2016.
been shown that the LADRC is more effective in suppressing [19] N. Jinbiao and Y. Fengjun, “Compound control strategy based on active
output sinusoidal disturbances than the PI controller (for dis- disturbance rejection for selected catalytic reduction systems,” J. Dyn.
turbances with frequency lower than half of the observer band- Syst., Meas., Control, vol. 137, no. 5, p. 051008, 2015.
[20] B. Gao, J. Shao, and X. Yang, “A compound control strategy combining
width). Moreover, the effects of 100% identification error on velocity compensation with ADRC of electro-hydraulic position servo
the transient response and disturbance rejection performance control system,” ISA Trans., vol. 53, no. 6, pp. 1910–1918, 2014.
of the controllers were also studied. Comparative simulations [21] L. A. Castañeda, A. Luviano-Juárez, and I. Chairez, “Robust tra-
jectory tracking of a delta robot through adaptive active disturbance
and experimental results show that the control of gimbal rejection control,” IEEE Trans. Control Syst. Technol., vol. 23, no. 4,
system under the LADRC approach can effectively restrain pp. 1387–1398, Jul. 2015.
the influence of external disturbances and achieve desirable [22] Q. Zheng, Z. Chen, and Z. Gao, “A practical approach to disturbance
decoupling control,” Control Eng. Pract., vol. 17, no. 9, pp. 1016–1025,
transient and steady-state performance. This is more attractive 2009.
in reality, since that the ADRC can also resolve the need for [23] S. Xingling and W. Honglun, “Back-stepping active disturbance rejection
system identification in systems, which use DC motor as their control design for integrated missile guidance and control system via
reduced-order ESO,” ISA Trans., vol. 57, pp. 10–22, Jul. 2015.
actuator. [24] Q. Zheng and Z. Gao, “Predictive active disturbance rejection control for
processes with delay,” ISA Trans., vol. 53, no. 4, pp. 873–881, Apr. 2014.
R EFERENCES [25] G. Sun, X. Ren, and D. Li, “Neural active disturbance rejection output
control of multimotor servomechanism,” IEEE Trans. Control Syst.
[1] S. Sadhu and T. Ghoshal, “Sight line rate estimation in missile seeker Technol., vol. 23, no. 2, pp. 746–753, Mar. 2015.
using disturbance observer-based technique,” IEEE Trans. Control Syst. [26] Z. Gao, “Scaling and bandwidth-parameterization based controller tun-
Technol., vol. 19, no. 2, pp. 449–454, Mar. 2011. ing,” in Proc. Amer. Control Conf., Jun. 2003, pp. 4989–4996.
[2] P. J. Kennedy and R. L. Kennedy, “Direct versus indirect line of [27] G. F. Franklin, M. L. Workman, and D. Powell, Digital Control of
sight (LOS) stabilization,” IEEE Trans. Control Syst. Technol., vol. 11, Dynamic Systems. Reading, MA, USA: Addison-Wesley, 1997.
no. 1, pp. 3–15, Jan. 2003. [28] R. Miklosovic, A. Radke, and Z. Gao, “Discrete implementation and
[3] J. M. Hilkert, “Inertially stabilized platform technology concepts and generalization of the extended state observer,” in Proc. Amer. Control
principles,” IEEE Control Syst., vol. 28, no. 1, pp. 26–46, Feb. 2008. Conf., Jun. 2006, pp. 2209–2214.
[4] B. Ekstrand, “Equations of motion for a two-axes gimbal system,” IEEE [29] D. E. Seborg, D. A. Mellichamp, T. F. Edgar, and F. J. Doyle, III, Process
Trans. Aerosp. Electron. Syst., vol. 37, no. 3, pp. 1083–1091, Jul. 2001. Dynamics, Operations, and Control. Hoboken, NJ, USA: Wiley, 2010.
[5] M. M. Abdo, A. R. Vali, A. R. Toloei, and M. R. Arvan, “Stabilization [30] G. Herbst, “A simulative study on active disturbance rejection con-
loop of a two axes gimbal system using self-tuning PID type fuzzy trol (ADRC) as a control tool for practitioners,” Electronics, vol. 2,
controller,” ISA Trans., vol. 53, no. 2, pp. 591–602, Mar. 2014. no. 3, pp. 246–279, 2013.

You might also like