Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

16th international conference on Sciences and Techniques of Automatic control STA'2015-PID3810-RAC

& computer engineering - STA'2015, Monastir, Tunisia, December 21-23, 2015

Analysis of Middleware and ADRC based


Techniques for Networked Control Systems
Ammar Ahmed, Hafiz Asad Ullah, Ijlal Haider, Uzair Tahir, Hammad Attique
MEDICS Lab., Department of Electrical Engineering,
The University of Lahore
Lahore, Pakistan
ijlal.haider@ee.uol.edu.pk

Abstract— This paper presents two control techniques to In this paper two controllers have been model to maintain
cater issues of stabilization due to network delays in the network the stability of a network control system in the presence of
control systems (NCS). First technique is focused on the network-induced delay (sensor-controller delay and controller-
modifying a simple PI controller by adding a middle ware to actuator delay). The first technique comprises of a PI
adjust PI parameters in accordance with the delays. Next, a novel
technique Automatic Disturbance Rejection Control (ADRC) is
controller and an additional block namely Gain Scheduling
discussed to cater the issues due to delays. The ADRC stabilizes Middleware (GSM) is added for the smooth operation of the
the response of the system however the overall response is system. Middleware is the key component of the design. As
delayed. replacing existing controllers with the ones specially designed
for networked environment can be inconvenient, a middleware
Keywords— NCS, Middleware, Gain Scheduling, PI, ADRC that will enables existing control strategy to work in
networked environments can be more efficient. The proposed
I. INTRODUCTION middleware has two essential elements: network traffic
estimator and Gain. Random delays in the network are fixed to
The NCS or Network Control Systems are the systems a certain value by using buffer and then gain is tuned to get
comprising sensors, actuator, controllers and system or desired response [10].
systems to be controlled, where information is shared via
communication medium. They have several benefits over The second technique is ADRC, which was proposed by
traditional point to point wired networks and finds application Jingqing Han [7], is used to control the plant over-the-
in many areas. They find application in a number of areas network. Active Disturbance Rejection Control is a robust
including factory automation, vehicles and airplanes. Due to control methodology . It is another technique that is used to
insertion of medium system experiences various natures of minimize the effect of delays in network control systems [8].
side effects. ADRC is based on the extension of the system model with an
additional imaginary state variables; the virtual state (sum of
Control over-the-Network has emerged as a topic of external and internal disturbances or delays usually
significant interest in the control system community . It is represented as a "total disturbance"). This disturbance
fully understood that in many practical implementation of the rejection feature pave way for researcher to study the system
systems, the physical plant, sensor, controller, and actuator under consideration with a simpler model as their negative
are difficult to be located at the same place, so it is therefore implication of modeling uncertainty overcame in real time. As
required to transmit the signals from place to place. In the a result, the need of a precise analytical description of the
modern world industrial systems, the components are system is not required, as the unknown parts of dynamics can
frequently connected over network media (particularly digital be assumed as the internal disturbance in the plant. The
band-limited channels of serial communication), this adaptive ability and robustness of this technique make it an
eventually give rise to the so-called networked control systems efficient solution in different scenarios where the complete
(NCSs). information of the system is not available.
Difficulties in designing an NCS include, the delays Implementation and simulation results of both techniques
which is induced by the network [1] which exchanging the follows in subsequent sections. Section II describes the system
data from one medium to another, and packet losses during the under consideration and the delay modeling. Section III and
transferring of data, because of the unpredictable network IV presents analysis of GSM and ADRC based techniques on
transmission path, where packets or information not only the system described in Section II respectively. Conclusion are
suffer the problem of transmission delays but also lost during summarized in Section V.
transmission. Several techniques are now in existing to handle
the network delay effect the techniques like Gain scheduling
[2], introducing a buffer in between [3], and sampling time II. NETWORK DELAYS AND SYSTEM DESCRIPTION
scheduling [4], robust gain scheduling [5], using a robust
controller [6]. A. Delays
For the analysis of how to schedule the gains of PI
controller on the controller parameters (control agent) with

978-1-4673-9234-1/15/$31.00 ©2015 IEEE 628


respect to IP network traffic, let us formulate the problem x = [ ia ω]
T

scenario mathematically keeping the IP network delays Where


constant. A general single line networked control system is All motor parameters are mentioned in the table below.
formulated as shown in Fig. 1, where R (s), U (s), Y (s), and
E(s) = R(s)-Y(s) are the reference, control, output, and error
signals, respectively. TABLE 1 MOTOR PARAMETERS
Parameter Description Value
Ra Armature Resistance 4.67
La Armature Inductance 170e-3 H
J Moment of Inertia 42.6e-6 Kg-m2
F Viscous Friction 47.3e-6 N-
Coefficient m/rad/sec

Fig. 1. Simplified block diagram of NCS K Torque Constant 14.7e-6 N-m/A


Kb Back EMF Constant 14.7e-6 V-sec/rad
The action agent plant dynamics is expressed as a transfer
ia Armature Current --
function Gp(s), where the PI controller Gc(s) is described by:
Rotational Speed --
⎛ K ⎞
Kp ⎜s + i ⎟⎟ Plant dynamics are represented by the following transfer

Gc ( s ) = ⎝ Kp ⎠ = K p ( s + zc ) (1)
function
s s 2029.826
GP ( s) = (6)
Where KP and KI are the proportional gain and integral ( s + 26.29 )( s + 2.296 )
gain, respectively, and ZC = KI / KP is a constant. The IP
network delays for sending the control U(s) to the plant GP(s),
and for sending the system output Y(s) to the PI controller C. Desired Performance
GC(s), are represented by GDCP(s) and GDPC(s), respectively, of The plant’s step response can be modified using control
which the analytical forms are: techniques to get desired performance. The design criteria for
the plant under consideration, as set in the reference, are:
GDCP ( S ) = e −τDCP S (2)
• Percentage Overshot ≤5%
GDPC ( S ) = e −τ DPC S
(3) • Settling Time ≤0.309s
• Rise Time ≤0.117s
Where τ DCP and τ DPC are the delay from the controller to
the plant, and the delay from the plant to the controller in time
domain, respectively. The numerator degree zero to D. Controller Design
approximate the delays as follows [9]: PI controllers are easy to realize and implement. PI
controllers have a general form of
−n
−τ s ⎛ ⎛τ s ⎞⎞
e ≅ ⎜1 + ⎜ ⎟ ⎟ (4) KI
⎝ ⎝ n ⎠⎠ GC ( s ) = K P + (7)
s
B. System Description Where KP denotes the proportional gain and KI integral
gain.
The system which is used to analyze the technique is a DC
motor, whose parameters are known and according to those A simple PI controller is tuned to get desired performance.
parameters the controllers are designed or tuned. Following gain values fulfil design criteria:
A simple linear DC motor plant is used for simulations. • KP = 0.1701
The dc motor transfer function from [10] is used. The linear
• KI = 0.308
equations in state space and transfer function are as follows:

⎡ Ra Kb ⎤ III. GAIN SCHEDULING MIDDLEWARE


⎢− L − L ⎥ ⎡1⎤
x = ⎢ a a
⎥ x + ⎢ La ⎥ u A. Description
⎢ K f ⎥ ⎢ ⎥ (5) In various cases under some networked condition, it is
⎢⎣ J − ⎢⎣ 0 ⎥⎦ necessary for us to replace the existing controllers with the
J ⎥⎦ ones specially designed for networked environment. The
y = [ 0 1] x middleware will enable existing control strategy to work in

629
networked environments. The proposed middleware has two
essential elements: Network Traffic Estimator, and Gain
Scheduler. The network traffic estimator is used to send a
packet signal to calculate the round trip time and by using this
time the network traffic estimator calculate or estimates the
delay in the system. Delays appearing in the network are sent
to the gain scheduling block. The Gain Scheduler calculates
the optimized gain to counter the delay calculated in system
by network traffic estimator. By applying the appropriate gain
the signal is transmitted to the plant without deteriorating the
performance caused by delays in the system.

Fig. 4. Root Locus to find gain limits (Max. Gain = 0.676)

Fig. 2. Block Diagram of middleware based NCS


J c = w1 J1 + w2 J 2 + w3 J 3 + w4 J 4 (9)

Where J1denotes rise time, J2 settling time, J3overshoot


B. Gain Scheduling and J4 is steady state error.
The β gain needs to be greater than zero in order to avoid
⎧⎪tr − tr0 , tr < tr0 (10)
the repercussions caused by positive feedback. The β gain is J1 = ⎨
placed after the first PI controller is shown in Fig. 3. ⎪⎩ 0, tr0 ≥ tr0

⎧⎪ts − ts0 , ts < ts0


J2 = ⎨ (11)
⎪⎩ 0, t s ≥ tr0

⎧OS − OS0 , OS < OS0


Fig. 3. Simple gain scheduling J3 = ⎨ (12)
⎩ 0, OS ≥ OS0
The β adjust both KP and KI of the PI controller the
transfer function is given as ⎧ SSE − SSE0 , SSE < SSE0
J4 = ⎨ (13)
K ( s + zc ) ⎩ 0, SSE ≥ SSE0
β Gc ( s ) = β P (8)
s Are the cost functions, tr0,ts0,OS0,SSE0 are the nominal
Where KI and KP are the integral gain and proportional values and w1, w2, w3 and w4 are the weighting functions.
gain respectively and zC is a constant. zC is the ratio of KI and
KP (zC = KI / KP). C. Simulation Results
The performance criterion relies mainly on rise time, The system was tested with delay of 50 ms. The simulation
settling time and percentage overshoot of the system when an results obtained are presented in fig. 5, fig. 6 and fig. 7. In fig.
input is applied. An efficient technique would be one which 5 the plant and controller are connected without any network
ensures minimum deviation from the desired performance in the system, hence, there is no delay and the response could
criterion. The root locus is used to obtain the range of gains be termed as the nominal response. In the next stage, the
keeping the response in stable region as seen in fig. 4 and from system is connected over the network and a delay of 50ms is
this a suitable gain is selected to fit the system response. introduced, the result is shown in fig. 6. To reduce disturbance
and oscillation from the signal, middleware is introduced and
A cost function can be defined here which involves all
the effectiveness of middleware could be seen through fig. 7.
these system specification. Each one of this design
requirement can be taken as a cost function and overall cost Only settling time is increased and all the other parameters are
function is given as eq. 9 describes [10]. within the range.

630
IV. ACTIVE DISTURBANCE REJECTION CONTROL

A. Introduction
The PID(proportional-integral-derivative) controller is the
most extensively used controller occupying a dominant status
in industrial process control system. With the development in
the fields of science and technology, the importance of control
system mainly focuses on three major components including
rapid respond speed, control and accuracy, and also the high
ability to adapt the various changes in the environment
increasingly which the classical PID controller cannot obtain
due to some shortcomings with control performance. The
Active Disturbance Rejection Controller (ADRC) is a new
robust approach having the advantages of classical PID and
improved with respect to most of its defects and problems. In
the contemporary era, the ADRC controller has wide spread
applications in many domains such as chemical industry,
electricity generation plants, medical especially tele-surgery.
aerospace, aviation and other fields. This technique in many
Fig. 5. Step response of the system without network & delay
parameters such as transient system response, time and very
negligible overshoot are far better than classical PID.
ADRC is divided into three different parts:
• Tracking-Differentiator (TD)
• Extended State Observer (ESO)
• Nonlinear State Error Feedback (NLSEF)

B. Tracking Differentiator (TD)


The input is most likely to be given as a unit step function
which is not appropriate for most of the dynamic systems
since it is compelling the output and the control signal to make
a sudden jump [11]. To avoid set point jump, the transient
profile construction is very crucial that the output of the plant
can follow reasonably.

v1 = v2 (14)
Fig. 6. Step response of system with network delay (50 ms) in loop
⎛ v v ⎞
v2 = − r ⎜ v1 − v + 2 2 ⎟ (15)
⎝ 2r ⎠
v is the input and v1 is the desired trajectory of the step
input, v2 is input’s derivative. The parameter ‘r’ is speed
controlling parameter for the desired transient profile. The
parameter ‘r’ can be chosen so as to speed up or slow down
the transient profile. Fastest tracking is provided for input v
and its derivative, which is limited with respect of speed
variable r.

C. Non Linear State Error Feedback (NLSEF)


The control law of ADRC is defined by the Nonlinear
State Error Feedback. NLSEF [12] is the control law for the
actuator or the plant, which is the combination of tracking
signal and the differential signal which are the output of the
tracking signal and the real time signal generated by the ESO
Fig. 7. Step response of the system with delay (50ms) and middleware in loop
through a non-linear function. The control law described by
Jingqing Han [7] uses a nonlinear function. The nonlinear
function is defined as:

631
u0 = − fhan ( e1 , ce2 , r , h ) (16) But x3 = f ⇒ x2 = f + bu (24)
Where,
x3 = f (25)

⎪ 2 y = x1 (26)
⎪d = rh , a0 = hx2 , y = x1 + a0
⎪ This depicts the extended state of the state observer hence
⎪a1 = d ( d + 8 y ) named as Extended State Observer. Jingqing Han introduce
⎪ the implementation of ESO through nonlinear function which
⎪a2 = a0 + sign ( y )( a1 − d ) / 2 (17)
⎪ can be expressed through nonlinear function as

fhan ( e1 , ce2 , r , h ) = ⎨ s y =
sign ( y + d ) − sign ( y − d )
⎪ 2 e = z1 − y (27)
⎪a = ( a0 + y − a2 ) s y + a2
⎪ fe = fal ( e, 0.5, δ ) , fe1 = fal ( e, 0.5, δ ) (28)
⎪ sign ( a + d ) − sign ( a − d )
⎪ sa = 2 z1 = z2 − β 01e (29)

⎪ fhan = −r ⎛ a − sign a ⎞ s − rsign a
⎪⎩ ⎜ ( )⎟ a ( ) z2 = z3 + bu − β 02 fe (30)
⎝d ⎠
Here u0 is the output of NLSEF, r0and h0 are the controller z3 = − β 03 fe1 (31)
parameters which are set equal to the parameter h which is
sampling time. b0 is selected as 1.Finally the control law is In which the nonlinear function fal is described as
stated as
⎧ e
u −z ⎪ , x <δ
u= 0 (18) fal ( e, α , δ ) = ⎨ δ ( −α )
1
(32)
b0 ⎪ e α sign ( e ) , x ≥ δ

The gains β01, β02, β03 are selected through the equation.

1 2
β 01 = 1 β 02 = β 03 = 2 2 (33)
3h 8 h
Where h is the sampling period.

Fig. 8. Nonlinear state error feedback Simulink implementation

D. Extended State Observer (ESO)


The ESO states are here stated as ‘x1’ the first state
variable, ‘x2’ as the second state variable which is derivative
of x1, and ‘x3’ as the extended state variable that will be used
as a necessary part of the control system for the estimation of
error(f). This setup of the extended state observer makes a
third order observer for a second order system. Which in state Fig. 9. ESO Simulink implementation
space representation is given as[13][14]:

x1 = x (19)

x2 = x (20)

x3 = f (21)
For the desired output y can be obtained as given below.

x1 = x2 (22)

x2 = x3 + bu (23)

Fig. 10. ADRC Simulink model

632
E. Simulation Results [2] M.-Y. Chow and Y. Tipsuwan, “Gain adaptation of networked DC
motorcontrollers based on QoS variations,” IEEE Transaction on
The ADRC was implemented in Simulink for model Industrial Electronics, vol. 50,pp. 936–943, Oct. 2003.
system and tested with delays of 50 ms and 150 ms. The [3] R. Luck and A. Ray, “An observer-based compensator for distributed
simulation results obtained are presented in fig. 11 and fig. 12. delays,” Automatica, vol. 26, no. 5, pp. 903–908, 1990.
It can be clearly seen without the use of any middleware the [4] S. H. Hong, “Scheduling algorithm of data sampling times in the
controller cater the delays, and provides smooth functioning of integrated communication and control systems,” IEEE Transaction on
plant. Control System Technology, vol. 3, pp. 225–230, June 1995.
[5] A. Sano, H. Fujimoto, and M. Tanaka, “Gain-scheduled compensation
for time delay of bilateral teleoperation systems,” IEEE ICRA8,
V. CONCLUSION Belgium, 1998, pp. 1916–1923.
In the implemented Gain Scheduler Middleware (GSM) [6] Manh Dung Ngo, “robust controller for dc servo motor”, Southeast-
Asian Journal of Sciences, Vol. 2, No 1, pp. 71-78, 2013.
methodology, the system successfully overcomes the problems
due to the introduction of network in the close loop system. [7] J. Han, “Auto disturbances rejection controller and its applications,”
Control Decisions, vol. 13, no. 1, pp. 19–23, 1998.
This technique has enabled the practitioners to use the existing
[8] Z. Gao, “Active disturbance rejection control: A paradigm shift in
controller in the presence of network delays. The results show feedback control system design,”IEEE American Control Conference,
that system effectively reduced the rising overshoot of the 2006,
system although the settling time was a bit greater than the pp. 2399–2405.
nominal values. [9] B. C. Kuo, Automatic Control Systems, 5 ed. Englewood Cliffs, NJ:
Prentice-Hall, 1987.
In contrast to PI controller the ADRC does not need any
middleware to modify the control signal. ADRC has [10] Y. Tipsuwan,M.Y. Chow, “Gain Scheduler Middleware: A
Methodology toEnable Existing Controllers for Networked Control and
extraordinary dynamical configuration to overcome the issues Teleoperation—Part I: Networked Control”, IEEE Transaction on
of delay when applied to the network. The ESO serves the Industrial Electronics, vol. 51, Dec. 2004.
purpose of state estimation and rejection of the disturbances [11] J. Han and W.Wang, “Nonlinear tracking-differentiator,” System Science
and delays in ADRC. The results shows that ADRC in Math., vol. 14, no. 2, pp. 177–183, 1994.
presence of network delay works efficiently. There are some [12] J. Han, “Nonlinear PID controller,” Automatica., vol. 20, no. 4, pp.
parameter that crosses the nominal values like settling time but 487–490, 1994.
the rise time and overshoot are even better than the GSM. [13] J. Han, “Extended state observer for a class of uncertain plants,” Control
Decisions., vol. 10, no. 1, pp. 85–88, 1995.
[14] Gernot Herbst,” A Simulative Study on Active Disturbance Rejection
REFERENCES Control (ADRC) as a Control Tool for Practitioners”, Electronics 2013,
[1] Y. Tipsuwan and M.-Y. Chow, “Control methodologies in vol. 2, pp. 246-279.
networkedcontrol systems,” Control Engineering Practice, vol. 11, no.
10, pp. 1099–1111,2003.

Fig. 11. Output with ADRC at 50ms delay

Fig. 12. Output with ADRC at 150ms delay

633

View publication stats

You might also like