You are on page 1of 6

Predictive PI Controller for Wireless Control System

with Variable Network Delay and Disturbance


Sabo Miya Hassan, Rosdiazli Ibrahim, Nordin Saad,Vijanth Sagayan Asirvadam, Tran Duc Chung
Department of Electrical & Electronic Engineering,
Universiti Teknologi PETRONAS
32610, Bandar Seri Iskandar, Perak, Malaysia
hsmiya2010@gmail.com, rosdiazli@petronas.com.my, nordiss@petronas.com.my,
vijanth sagayan@petronas.com.my, chungtran@acm.org

AbstractAs control using wireless technology is receiving easily be deployed in remote and hostile environments. They
increased attention, its often associated with network delays can as well be deployed on mobile and rotating equipment
and uncertainties. The use of PIDs proved inadequate to handle such as robots [11].
these challenges while model based controllers are complex. As
a compromise, this paper examines the use of Predictive PI Despite the enormous opportunities and advantages offered
(PPI) controller in a wireless networked control environment by the wireless technology in the industries for automation,
characterized by both network delay and process deadtime. their use in the industries is not without challenges [3]. The
As against deadtime compensators, the PPI allows for model use of wireless network and transmitters usually introduce
mismatch while still maintaining the simplicity of PIDs in terms into the control loop delay and uncertainties such as packet
of tunable parameters. Comparison of the simulation results
for the thermal chamber while considering input disturbance dropout and communication paths are sometimes affected by
and measurement noise showed that the PPI controller produce noise [2], [4]. The delay, which is the most critical among
improved performance in terms of systems overshoot, rise time the two problems listed, is capable of degrading the control
and integral absolute error in all cases. The PPI is still better performance of the system.
than the other two controllers in terms of settling time for most Key features of any good low level controller is to be simple,
cases.
easy to be manually tuned and with few tuning parameters. The
KeywordsWireless control, PPI controller, variable network PID control strategy is the most commonly used in the process
delay, disturbance, measurement noise. and control industry. Of the three variants of the PID structure
(i.e. PID, PD and PI), the PI structure is the most applied,
I. I NTRODUCTION since usually the derivative action (D) is switched off [12],
Recently, several attempts are being made to extend the [13]. This is because the derivative action is highly sensitive
use of wireless network for control [1][6]. This is as a to noise and requires additional filtering and tuning for its
result of rekindled hope by the advent of two industrial implementation [14]. However, the use of the PID controller in
wireless standards: WirelessHART and ISA100.11a [6]. The a delayed and uncertain environment is accompanied with poor
WirelessHART being an extension of the wired highway control performance and system instability. The PIDs have a
addressable remote transducer (HART) communication pro- very limited gain for them to perform well in a control loop
tocol was the first among the two standards to be approved characterized with system deadtime and/or network delay [15].
by the International Electrotechnical Commission (IEC) [7]. To compensate for the effects of delay, the use of model based
Thus by virtue of being an offshoot of the HART protocol, predictive controllers (MPCs) such as smith predictor and
the standard has the capability of being the most accepted internal model control (IMC) has been proposed by researchers
over the ISA100.11a standard [8]. This is because, there for application in a wireless networked environment [16].
are close to 30 million HART devices already installed in The drawbacks of the model based controllers is that they
the industry that require only little adaptation to work with require exact model of the system and the number of tunable
the new WirelessHART standard as against the ISA.100.11a parameters is increased compared to the PID only structure
standard [9]. Hitherto, the previous wireless standards such [17]. This adds to the design complexity.
as Bluetooth, Wi-Fi, and ZigBee did not meet the industry In this work, a special deadtime compensator; the predictive
requirement of security, device interoperability, scalability and PI (PPI) controller [18] is adopted for wireless networked
reliability [10]. control systems characterized by network delay and long
The advantages offered by the new wireless industrial stan- process deadtime. The controller can well serve as a settle-
dards over their wired counterparts includes the elimination of ment between the complex MPC and the simple but poor
cabling, reduced installation and maintenance time and reliable performance PID as far as long deadtime and varying network
communication paths. Additionally, the problem of flexibility delays are concerned [19]. The advantage of the PPI controller
in the wired systems due to fixed connection and infrastructure is that it allows for model mismatch, it can handle integrating
is solved by the wireless systems. The wireless systems can processes and has fewer tunable parameters than the other
R Y E U
-+ Gc(s) e-cas Gp(s)e-Lps KC +

e-scs e-sLp/(1+Tis)
Wireless
Network Fig. 2: Implementation of PPI controller
Fig. 1: Network delay representation in a single loop wireless
networked control system Assuming that the desired closed loop time constant is equal
to the open loop time constant, the desired closed loop transfer
function is specified as
model based compensators [14]. Furthermore, the controller is
guaranteed to improve systems control performance and can
() = (6)
work well in systems with varying delay. The implementation 1 +
of the controller is also cheaper compared to the model based It follows from (5) that the controller can be expressed as
controllers.
1 +
The reminder of the paper is organized as follows: Section () = (7)
II discusses the basic PPI controller and its adaptation (1 + )
to wireless control. The procedure for the random delay Expressing the controller () in (7) in terms of the relation-
estimation and model selection is presented in section III. In ship between its input () and its output () we have
Section IV, the simulation results are presented and discussed. 1
Lastly, conclusion is drawn in Section V. (1 + ) () = (1 + )() (8)

Thus, (8) can be expressed as follows:
II. PPI C ONTROLLER D ESIGN FOR W IRELESS N ETWORK 1
A. Typical Wireless networked control structure () = (1 + ){() (1 ) ()} (9)
1
It has been shown by researchers that the single loop Hence, 9 can be further simplified as follows:
wireless networked control system characterized by network 1 1
delay can be represented as shown in Fig. 1. Thus the total () = (1 + )() (1 ) () (10)

network delay, can be represented as follows:
From (10), it can be deduced that a PI controller acts upon the
= + (1) error signal () and the prediction is achieved via low-pass
filtering of the control signal (). Again, when the delay
where and are the controller-to-actuator and sensor- term = 0, the controller is reduced to a PI controller with
to-controller delays respectively. Assuming there is commuta- gain = 1
and = .
tivity, the process deadtime can be added to the network delay For the case when > 0 , (10) can be conveniently written
which now gives the total closed loop delay as as
1
= + (2) () = () + () (11)
1 +
The implementation of (11) is shown in Fig. 2.
Furthermore, (11) can be written in the form of the PPI
B. Typical PPI Structure controller transfer function and be factored in to () =
() () as follows:

() = (3) 1 1
1 + () = (1 + ){ } (12)
1 + 1
Consider a general first order deadtime (FODT) representation (1 )
1
of process with dead-time of (3), the closed loop transfer where, () = (1 + ), and () =
function of the process () controlled with () is 1
.
1+ 1 (1 )

() ()
() = (4) The behavior of the predictor is determined by the ratio of
1 + () ()
the system delay to the time constant / . Thus, it can
solving (4) for () we obtain be seen from Fig. 2 that the PPI controller gain approaches
1 () unity at high frequencies. This implies that the PPI controller
() = . (5) approaches at these frequencies.
() 1 0 ()
R E Y 3.5
-+ KC + e-cas Gp(s)e-Lps

Upstream delay(s)
3

2.5
e-sL/(1+Ts)
2

1.5
0 100 200 300 400 500 600 700 800 900 1000
e-scs
3.5

Downstream delay(s)
3
Wireless Network

Fig. 3: PPI controller in wireless network set-up 2.5

2
TABLE I: N ETWORK D ELAY S TATISTICS 0 100 200 300 400 500 600 700 800 900 1000

Statistics Fig. 4: Generated variable wirelessHART network induced


Delay type
Max Min Average Std. Dev. delay
Upstream (s) 5.476 1.981 2.6401 0.113
Downstream (s) 5.124 2.318 2.600 0.280 TABLE II: C ONTROL P ERFORMANCE WITHOUT D ISTUR -
BANCE OR N OISE

Control Strategy
C. Wireless Networked PPI structure Performance Parameter
PPI PI Smith pred.
For a wireless network characterised with both network IAE 60,618 89,061 125,2900
delay and process deadtime , the total delay in the network Rise time (s) 17.537 28.902 66.879
is given in (2). Thus, the PPI control structure for such network Settling time (s) 52.198 101.431 141.697
1
is given in Fig. 3. For = and = , and for the case Overshoot (%) 0.000 4.523 0.000
when > 0. PPI controller for such systems can be expressed Undershoot (%) 0.000 0.000 0.000
as
1
() = () + () (13)
1 +
Equation (13) can be implemented in the controller set-up of
Fig. 1 which will now become the structure of Fig. 3.
IV. R ESULTS & A NALYSIS
In this section, the simulation results of the thermal chamber
III. N ETWORK D ELAY E STIMATION AND P LANT M ODEL under three operating conditions i.e. with and without input
S ELECTION disturbance, then with measurement noise will be presented
A. Network Delay Estimation and discussed. The results with PPI controller will be com-
pared to those of Smith predictor and PI designed according
To estimate for the network induced delays and , an experi-
to Smith and Corripio as reported in [20].
ment was conducted using the Linear Technology Smart Mesh
WirelessHART network development kit. In the experiment, A. Simulation without Disturbance or Noise
the WirelessHART network induced delays (upstream and Considering the step response of the system devoid of both
downstream) for two way communication between a wireless noise and disturbance as shown in Fig. 5, it is clear that
node (mote) and a wireless gateway/controller were measured. the PPI controller is faster and without overshoot. The PPI
Each time a message is received at a destination (i.e. node or control action as shown in Fig. 6, is also faster and more
controller), a timestamp is created. The difference between aggressive compared to the other controllers. All these can
two executive timestamps is recorded as either of the network clearly be seen in crisp terms while observing Table II. From
induced delay. The average value of the upstream delay ( ) the table, PPI controller provides least integral absolute error
is 2.6401s and that of the downstream delay ( ) is 2.600. (IAE), shorter rise time, and faster settling time as compared
The statistical information on these delays shown in Table I to the PI and Smith predictor controllers. While only the PI
is used to generate variable delays as shown in Fig. 4. controller experiences overshoot of 4.5% it is still faster than
B. Plant Model Selection the Smith predictor. The settling time of PPI controller is the
parameter that has been improved the most, up to more than
In this work, a practical model of a thermal chamber in 50% reduction as compared to both PI and Smith predictor
[15] with a transfer function presented in 14 is used for controllers.
simulation. This model falls well within the class of processes
characterized by long dead-time. B. Simulation with Input Disturbance
8 When disturbance is injected at the input and the plant is
() = 4 (14) simulated to a step signal, the response and the control action
1 + 9.13
35 35

30 30

25 25
Temperature(C)

Temperature(C)
20 20

15 15

10 10

5 5
Setpoint Setpoint
PPI control PPI control
0 PID control 0 PID control
Smith predictor Smith predictor

-5 -5
0 50 100 150 200 250 300 350 400 450 500 0 50 100 150 200 250 300 350 400 450 500
Time(s) Time(s)

Fig. 5: Plant response without disturbance Fig. 7: Plant response with disturbance

6 6

5 5

4 4

Control signal
Control signal

3 3

2 2

1 1

PPI control 0 PPI control


0
PID control PID control
Smith predictor Smith predictor

-1 -1
0 50 100 150 200 250 300 350 400 450 500 0 50 100 150 200 250 300 350 400 450 500
Time(s) Time(s)

Fig. 6: Control signal of the plant without disturbance Fig. 8: Control signal of the plant with disturbance

in Figs. 7 and 8 are obtained. Similarly, the PPI controller 9. Despite all controllers produce overshoot, the PPI controller
proved yet again its better tracking ability as compared to is still better in terms of IAE and rise time. Whereas the PPI
the two other controllers. Just as in the previous case, the maintained lead in settling time before disturbance, the settling
numerical results in Table III showed that the PPI controller time after disturbance is significantly affected by the noise.
outperforms both the PI and Smith predictor controllers in This settling time stands at 80% and 62% higher than those of
terms of IAE, rise time. As seen from the table, although for PI controller and Smith predictor respectively. The percentage
settling time before the disturbance the PPI controller shows overshoot of PPI controller is lower (at around 3%) than that
an improvement of about 50% and 70% compared to the PI of PI controller (at around 6.5%), and is slightly higher than
and Smith predictor controllers respectively, the improvement that of Smith predictor of around 1.2%. As a result of the
margin is narrowed to about 5% compared to PI controller noise, all the three controllers considered produce insignificant
and to about 15% compared to smith predictor controller after undershoot of less than 1%. Thus, the measurement noise will
the disturbance is injected. In terms of control action, the PPI potentially affect the performance of all the controllers (i.e. PI,
controller is more responsive to changes at both the step time Smith predictor and PPI) compared. By implication, a filter
and at the disturbance time as seen in Fig. 8.
TABLE III: C ONTROL P ERFORMANCE WITH I NPUT D ISTUR -
C. Simulation with both Input Disturbance and Measurement
BANCE
Noise
Control Strategy
Taking into consideration the presence of both measurement Performance Parameter
PPI PI Smith pred.
noise and input disturbance in the system, the plant is simu-
IAE 70,053 102,060 143,650
lated to a step signal and the response of Fig. 9 and control
Rise time (s) 17.537 28.902 66.872
action of Fig. 10 are obtained. From the figures, although
Settling time before dist. (s) 52.198 101.431 141.697
similar pattern to the case with disturbance only is obtained,
Settling time after dist. (s) 256.838 268.362 297.849
the effect of measurement noise can be vividly seen especially
Overshoot (%) 0.000 4.523 0.000
on both the response and the control action of the PPI
controller. Table IV gives the numerical interpretation of Fig. Undershoot (%) 0.000 0.000 0.000
35
to improve overall control performance of the system. It is
better than both PI and Smith predictor controllers in most
30
cases. The significant improvement can be seen through the
25 reduction in IAE, the shortening of rise time for all cases,
and settling time for the case of no noise and disturbance.
Temperature(C)

20

15
However, the drawback of the PPI design like other deadtime
compensators is that it is being affected by noise due to the
10
predictive action involved. As part of future work, the use of
5
Setpoint
appropriate filter to curtail this effect will be attempted.
PPI control
0 PID control
Smith predictor ACKNOWLEDGMENT
-5
0 50 100 150 200 250 300 350 400 450 500 The authors acknowledged the support of Universiti
Time(s)
Teknologi PETRONAS for the award of Yayasan (Y-UTP)
Fig. 9: Response of the system with disturbance in the pres- grant No. 0153AA-A74 and Graduate Assistantship Scheme
ence of measurement noise for the first and fourth authors.
R EFERENCES
6
[1] T. Blevins, D. Chen, S. Han, M. Nixon, and W. Wojsznis, Process
5
control over real-time wireless sensor and actuator networks, in 2015
IEEE 17th International Conference on High Performance Computing
4
and Communications (HPCC). IEEE, 2015, pp. 11861191.
[2] T. Blevins, M. Nixon, and W. Wojsznis, Pid control using wireless
Control signal

3 measurements, in 2014 American Control Conference. IEEE, 2014,


pp. 790795.
2 [3] T. L. Blevins, Pid advances in industrial control, IFAC Proceedings
Volumes, vol. 45, no. 3, pp. 2328, 2012.
1 [4] T. D. Chung, R. B. Ibrahim, V. S. Asirvadam, N. B. Saad, and S. M.
Hassan, Adopting ewma filter on a fast sampling wired link contention
0 PPI control
PID control
in wirelesshart control system, IEEE Transactions on Instrumentation
Smith predictor and Measurement, vol. 65, no. 4, pp. 836845, 2016.
-1
0 50 100 150 200 250 300 350 400 450 500
[5] S. M. Hassan, R. Ibrahim, N. Saad, V. S. Asirvadam, and T. D. Chung,
Time(s) Setpoint weighted wirelesshart networked control of process plant, in
2016 IEEE International Instrumentation and Measurement Technology
Fig. 10: Control signal of plant with disturbance and measure- Conference Proceedings. IEEE, 2016, pp. 16.
ment noise [6] F. Seibert and T. Blevins, Wirelesshart successfully handles control,
Chemical Processing, 2011.
[7] D. Chen, M. Nixon, and A. Mok, Future of wireless and the wirelesshart
standard, in WirelessHART. Springer, 2010, pp. 227243.
will be required in the design of these controllers to fitter the [8] M. Nixon and T. Round Rock, A comparison of wirelesshart and isa100.
11a, Whitepaper, Emerson Process Management, pp. 136, 2012.
measurement noise. [9] D. Chen, M. Nixon, and A. Mok, Why wirelesshart, in WirelessHART.
Springer, 2010, pp. 195199.
V. C ONCLUSION [10] J. Song, S. Han, A. Mok, D. Chen, M. Lucas, M. Nixon, and W. Pratt,
Wirelesshart: Applying wireless technology in real-time industrial pro-
This paper has presented the application of the PPI control cess control, in Real-Time and Embedded Technology and Applications
strategy for wireless control systems characterized by both Symposium, 2008. RTAS08. IEEE. IEEE, 2008, pp. 377386.
process deadtime and network random delay. The performance [11] W. Ikram and N. F. Thornhill, Wireless communication in process
automation: a survey of opportunities, requirements, concerns and chal-
of the controller is compared to two other controllers, PI and lenges, in Control 2010, UKACC International Conference on. IET,
Smith Predictor. The abilities of the controllers is evaluated 2010, pp. 16.
for robustness to input disturbance and measurement noise. [12] M. Huba, Comparing 2dof pi and predictive disturbance observer based
filtered pi control, Journal of Process Control, vol. 23, no. 10, pp. 1379
As seen from the simulation results, the PPI controller helps 1400, 2013.
[13] B. Kristiansson and B. Lennartson, Robust tuning of pi and pid
controllers: using derivative action despite sensor noise, IEEE control
TABLE IV: C ONTROL P ERFORMANCE WITH I NPUT D ISTUR - systems, vol. 26, no. 1, pp. 5569, 2006.
BANCE AND N OISE [14] P. Larsson and T. Hagglund, Comparison between robust pid and pre-
dictive pi controllers with constrained control signal noise sensitivity,
Control Strategy IFAC Proceedings Volumes, vol. 45, no. 3, pp. 175180, 2012.
Performance Parameter [15] K.-K. Tan, K.-Z. Tang, Y. Su, T.-H. Lee, and C.-C. Hang, Dead-
PPI PI Smith Pred.
time compensation via setpoint variation, Journal of Process Control,
IAE 76,731 10,8810 14,4340 vol. 20, no. 7, pp. 848859, 2010.
Rise time (s) 17.415 29.707 66.872 om and T. Hagglund, Advanced PID control.
[16] K. J. Astr ISA-The
Instrumentation, Systems and Automation Society, 2006.
Settling time before dist. (s) 63.847 117.436 129.122
[17] K. J. strom and T. Hagglund, Pid controllers: theory, design, and
Settling time after dist. (s) 480.583 266.053 297.849 tuning, Instrument Society of America, Research Triangle Park, NC,
Overshoot (%) 2.843 6.545 1.150 vol. 10, 1995.
[18] T. Hagglund, A predictive pi controller for processes with long dead
Undershoot (%) 0.790 0.276 0.270 times, IEEE control systems, vol. 12, no. 1, pp. 5760, 1992.
[19] F. Shinskey, Pid-deadtime control of distributed processes, Control
engineering practice, vol. 9, no. 11, pp. 11771183, 2001.
[20] A. ODwyer, Handbook of PI and PID controller tuning rules. World
Scientific, 2009, vol. 57.

You might also like