Documente Academic
Documente Profesional
Documente Cultură
Department of Electronics
and Instrumentation Engineering,
NIT Agartala, Tripura 799046, India
e-mail: pghorai@gmail.com
Somanath Majhi
Department of Electronics
and Electrical Engineering,
IIT Guwahati, Assam 781039, India
e-mail: smajhi@iitg.ernet.in
Saurabh Pandey
Department of Electronics
and Electrical Engineering,
IIT Guwahati, Assam 781039, India
e-mail: p.saurabh@iitg.ernet.in
Introduction
ke
T1 s 1T2 s 1
(1)
_ Axt Brt d
xt
(2)
yt Cxt
(3)
ka1 a2 ka1 a2
1
a1 0
;B
, and
where A
;C
0 a2
1
a1 a2 a1 a2
a1 1=T1 and a2 1=T2 are the eigenvalues of the assumed
process transfer function model.
(4)
Gs
(5)
(6)
(7)
(8)
(9)
Similarly, at time t2, xt2 is derived from Eqs. (4), (5), and (9) as
xt2 I eAT 1 eAsd l 1 hH l 2 h2 eATsd l 3 hL eAs l 4 hH
(10)
Considering an ideal relay and at limit cycle condition yt0
yt2 setpoint (R), therefore
Cxt0 R
(11)
Cxt2 R
(12)
Now, from the relay feedback test, the period of the oscillation T,
the time for which the plant output remains positive s, the upper
peak amplitude of the oscillation Aup, the lower peak amplitude of
the oscillation Adp, the area of the plant output signal Ay, and the
area of the plant input signal Ar over the last stable period of the
(14)
1 ea2 T
a2
"
#
Ra1 a2
1 ea1 Td ea1 sd hH hL
hL
1 ea1 T
ka1 a2
a1
"
#
1 ea2 Td ea2 sd hH hL
hL
1 ea2 T
a2
Cxtdt khH hL s hH T
(24)
rt ddt hH hL s hH T
(25)
Ay
k
Ar
(15)
(16)
From Eq. (19), the following expression can be derived for upper
peak time tup:
"
#
1 ea2 Ts 1 ea1 T
a1 a2 tup d
e
(20)
1 ea1 Ts 1 ea2 T
Subsequently, from Eq. (16), the expression for upper peak amplitude is derived as
"
a a
Aup khL 1 2
khH hL
Ay
The steady-state gain can be written from Eqs. (24) and (25) as
where xd1 and xd2 are found by putting A, B, and I in Eq. (4) as
2
3
hH
a1 d
a1 d
e
x
e
1
01
6
x
a1 7
7
(18)
xd d1 6
4 ad
h
xd2
H5
e 2 x02 ea2 d 1
a2
#a1
1 ea2 Ts
1 ea2 T
#a2
"
1 ea1 Ts
where tup can be obtained by taking the first derivative of Eq. (16)
and equating the result to zero. After simplification, it becomes
a ea1 tup d x ea1 tup d hL
0
(17)
C 1 a2 tup d d1
a2 e
xd2 ea2 tup d hL
(23)
1 ea1 T
1 ea2 T
khH hL
Ar
At time tup, the upper peak output of the plant occurs, so the
expression for Aup from Eq. (5) can be written as
Aup CeAtup d xt1 A1 eAtup d IBhL
Similar to the steps followed to derive Eq. (20), the expression for
tdp the time instant at which the downward peak output of the
plant occurs can be derived from Eq. (7) with the help of Eqs. (6)
and (10) as
"
#
1 ea2 s 1 ea1 T
a1 a2 tdp sd
(22)
e
1 ea1 s 1 ea2 T
(26)
!#1
Aup khH khL Adp
T1 T ln
Aup khL khH Adp
dT
Aup Adp
ln 1
khH hL
!
Aup khH khL Adp
ln
Aup khL khH Adp
(28)
(29)
(21)
3.2 Expressions for Overdamped SOPDT Processes. Consider a general transfer function model for SOPDT process
(T1 > T2) as in Eq. (1) with four unknown parameters. The steadystate gain (k) can be estimated from the measurements of Ay and
1 ea1 T
keds
T1 s 12
(30)
The derived expressions (14), (15), (21), and (23) are extended for
the development of generalized expressions for critically damped
SOPDT processes. Assuming a1 a and a2 a k and letting
k ! 0 , Eqs. (21) and (23) become
"
#
1 eaTs
Aup khH hL
1 eaT
"
#
(31)
aT seaTs
aTeaT
khL
1 eaT
e 1 ea Ts
"
Adp khH khH hL
1 e
1 eaT
as
"
aseas
aTeaT
as
aT
e1 e 1 e
(32)
Substituting a1 and a2 in Eq. (20), the expression for process time
delay d can be written as
"
#"
#
1 eas
eaTs
T
s
(33)
tup d
1 eaTs
1 eaT
One of the three unknown parameters of critically damped
SOPDT plant k can be estimated with the help of Eq. (26) from
the measurements of Ay and Ar. After getting k, another unknown
process time constant T1 can be derived from Eq. (31) or Eq. (32),
and thereafter, Eq. (33) can be used to estimate d. Alternatively,
Eqs. (26), (31), and (33) can be solved simultaneously to yield the
unknown process model parameters with a proper choice of initial
conditions for the measured quantities.
Simulation Study
In this section, the three well-known examples from the literature are considered for the illustration of proposed identification
scheme. Thereafter, the obtained models from each example are
validated through Nyquist frequency response plots. The accuracy
of the proposed method is estimated using integral of absolute
error (IAE) criterion as
xgm
Gm jx G jx
dx
(34)
IAE
G jx
0
where Gm jx is the identified process model, Gjx is the actual
process, and xgm is the frequency of the process, respectively.
Further, the effect of measurement is studied through the addition
of Gaussian distributed random noise which is quite common in
real-time environment [24]. To overcome this, an offline method
for reconstruction of original limit cycle output using Fourier
series-based curve fitting method [25] is utilized for obtaining a
best-fitted limit cycle output. The technique searches the best
021010-4 / Vol. 139, FEBRUARY 2017
where y^t and x, namely, the process output and ultimate frequency, are to be obtained through curve fitting method.
The Fourier series coefficients (a0, aj, and bj) 8 j 1 can be
T
T
obtained as a0 1=2T T y^tdxt; aj 1=T T y^t
T
cosjxtdxt, and bj 1=T T y^tsinjxtdxt.
4.1 Example 1. Consider a stable FOPDT plant widely studied in the literature [3,17] as Gm1 s e2s =10s 1. An asymmetrical relay settings (hH 65:0 and hL 45:0) is fed back with
a step input of R 50:0% yielding sustained asymmetrical limit
cycle output around the setpoint. The limit cycle parameters
Aup 52.7182, Adp 49.0942, T 8.9292, s 6:3425; Ay
453:5476, and Ar 453:5477 are substituted in the derived
mathematical expressions for estimation of unknown process
model parameters. From Eq. (26), the process steady-state gain is
estimated as k 0.9999, and further solving Eqs. (28) and (29),
the remaining plant model parameters are obtained as T1
10:0038 and d 1:9999. Now, the identified model is represented
in transfer function form as tabulated in Table 1. The Nyquist frequency response plots for estimated and actual FOPDT process
transfer function models are compared with the literature and
shown in Fig. 3. To consider the measurement noise, a 30 dB noise
is added at the process output, and thereafter, a noise-free output
is derived by Fourier series-based curve fitting method. Again
substituting the recovered limit cycle information in derived
mathematical expressions, the FOPDT process is identified in
comparison with the actual model through IAE in Table 1. The
robustness of the proposed scheme and variation of the process
input over the model parameter estimation are shown in Table 2
with an improved accuracy in Table 3.
4.2 Example 2. Let us consider an SOPDT plant studied by
Bajarangbali et al. [17] and Liu et al. [18] as
Gm2 s fe2s =10s 1s 1g. Using an asymmetrical relay
feedback test with relay settings (hH 70:0 and hL 40:0) and
step input of R 50:0% yields sustained asymmetrical limit cycle
output around the setpoint. The measured limit cycle parameters
Aup 53.8686, Adp 48.0809, tup 2.7545, s 7:0577; T
11:1339; Ay 567:6468, and Ar 567.6415 are substituted in the
derived mathematical expressions for estimation of unknown process model parameters. From Eq. (26), the process steady-state
gain is derived as k 1.0, and the remaining unknown plant model
parameters are estimated from the simultaneous solution of Eqs.
(21), (23), and (20) or (22) as T1 10:0115; T2 0:9962, and
d 2:0023. The identified model can be written in SOPDT
transfer function form in Table 1. The three different levels of
measurement noise are added at process output to yield noisy
outputs with 10 dB, 20 dB, and 30 dB signal to noise ratio (SNR)
values, respectively. Thereafter, the noise-free outputs are recovered using Fourier series-based curve fitting method as shown in
Fig. 4. Substituting the recovered limit cycle data in the derived
expressions, an overdamped SOPDT plant model is identified.
Comparison of identified models in terms of IAE is given in
Table 1 and in terms of percentage error in Table 3. Finally, the
actual and identified transfer function models are compared
through Nyquist frequency response plots as shown in Figs. 5
and 6.
4.3 Example 3. Consider the critically damped SOPDT
plant studied by Vivek and Chidambaram [26] as Gm3 s
fe0:5s =20s 12 g. Using an asymmetrical relay feedback
Transactions of the ASME
Methods
Identified models
IAE
Proposed model
0:9999e1:9999s
10:0038s 1
3:6850 104
1:0e1:997374s
9:9415s 1
38:1188 104
1:0048e2:0024s
10:0490s 1
2:2186 103
0:9467e2:0s
9:5028s 1
1:6065 102
Proposed model
1:0e2:0023s
10:0115s 10:9962s 1
3:3663 104
1:0014e2:1391s
10:2491s 11:0164s 1
30:55 103
1:0019e2:0769s
10:0691s 11:0751s 1
24:35 103
0:9999e1:9718s
9:9038s 10:9966s 1
79:449 104
0:9923e2:0s
9:8997s 11:0105s 1
4:3141 103
1:0122e2:0037s
10:1178s 10:9920s 1
5:3081 103
1:0e0:4983s
1:8589 104
Proposed model
19:9991s 12
64:5469 104
1:0004e0:4368s
19:9187s 1
1:6820 102
1:058e0:551s
20:81s 1
test with relay settings (hH 70:0 and hL 40:0) and step input
of R 50:0% yields sustained asymmetrical limit cycle output
around the setpoint. The measured limit cycle parameters
Aup 50.3504, Adp 49.8231, s 10:5783, T 15.9567, tup
4:8580s; Ay 799:6553, and Ar 799:6194 are substituted in the
derived mathematical expressions for estimation of unknown
Table 3
Ex.
2:1944 101
0:3753e3:8613s
64:7950s 1
Relay settings
T1
hH 80%, hL 45%
hH 65%, hL 45%
hH 70%, hL 30%
hH 50%, hL 55%
1.0000
0.9999
1.0000
1.0000
2.0005
1.9999
2.0001
2.0008
9.9999
10.0038
9.9999
9.9999
Methods
T1
T2
Proposed
As per Ref. [16]
As per Ref. [25]
0.01
0.48
5.33
0.0036
0.12
0
0.0387
0.49
4.972
Proposed
As per Ref. [6]
As per Ref. [15]
0
0.77
1.22
0.1171
0
0.185
0.1151
1.003
1.178
0.3712
1.05
0.8
Proposed
As per Ref. [26]
0.0044
5.8
0.3364
10.2
0.0043
4.05
Fig. 4 Limit cycle output for example 2: (a) noisy output with
20 dB SNR and (b) reconstructed output
5
Fig. 5 Nyquist frequency response plots for example 2: (a)
actual plant, (b) proposed model, (c) model by Bajarangbali
et al. [17], and (d) model by Liu and Gao [18]
Variation (%)
K%
T1 %
d%
0.5
1
2
0.5
1
2
0.5
1
2
0.5
1
2
0.5084
1.0088
2.0088
0.4870
0.9981
1.9874
0.4847
0.9841
1.9834
0.5237
1.0036
2.0225
Fig. 9 Response of level control system using an asymmetrical relay: (a) limit cycle around setpoint and (b) relay output
Identified models
0:9507e6:9442s
265:4280s 1
0:9388e7:0455s
273:5958s 1
0:9507e3:6964s
176:3120s 13:4908s 1
0:9507e8:2085s
29:9624s 12
0:9790e8:4127s
169:6874s 1
0:9589e8:4919s
167:6980s 1
Fig. 11 Nyquist frequency response plots for level control system: (a) proposed FOPDT process model, (b) proposed overdamped SOPDT model, (c) proposed critically damped SOPDT
model, and (d) model by ZN test
The limit cycle plant output with relay output signals is shown in
Fig. 9. A single asymmetrical relay feedback tests are conducted
by considering the same plant in two different orders of dynamics.
First, the plant is applied an asymmetrical relay feedback test of
relay heights hH 65:0 and hL 25:0 with a step input of R
40:0% and produces a sustained asymmetrical system output at
steady state. From the experimental results, the limit cycle parameters Aup 40.544, Adp 39.562, T 28.0, Ay 1121:84, and
Ar 1180:0 are measured. From Eq. (26), the process steadystate gain is derived as k 0.9507, and further solving Eqs. (28)
and (29), the unknown plant model is found in terms of stable
FOPDT process model parameters as T1 265:4280 and d
6:9442 which can be seen in Table 5. Second, the same plant is
considered as stable SOPDT model, and thereafter, from additional limit cycle information of s 16:0, tup 8.0, and simultaneously solution of Eqs. (20)(22), the unknown plant model
parameters are estimated as T1 176:3120; T2 3:4908, and
d 3:6964, i.e., as shown in Table 5. Similar to the above relay
settings, the real-time level control system is estimated from the
simultaneous solution of Eqs. (26), (31), and (33) in terms of critically damped SOPDT process model as k 0.9507,
T1 29:9624, and d 8:2085, and can be seen as transfer function form in Table 5 with the model obtained from
ZieglerNichols (ZN) test. For validation purpose, the obtained
dynamics of the real-time plant in transfer function forms of three
different types such as FOPDT, overdamped SOPDT, and ZN
model are simulated with the same relay setting and same set
value which was considered during the experimental study. Thereafter, three model outputs are plotted in Fig. 10 in comparison with
the actual experimental output of the real-time level control system.
To discard measurement noise, the noisy process output signal is
processed by MATLAB Fourier series-based curve fitting method [25].
Again substituting the recovered limit cycle information in derived
mathematical expressions, the stable FOPDT plant model is identified and tabulated in Table 5. Thereafter, the amount of measurement noise present at the process output as shown in Fig. 9 is
calculated, and thereafter, the SNR of the real-time level control is
found to be as 56.7144 dB. As the derived transfer functions are
obtained for single plant dynamics with three different forms, proper
validation is required to satisfy the experimental outcomes. The
Nyquist frequency response plots of three identified models and the
dynamic model obtained from ZN test are demonstrated in Fig. 11
for a wide range of frequency responses. Interestingly, it has been
observed that the identified transfer function models in terms of stable FOPDT and overdamped SOPDT are close to model obtained
from the ZN test as compared to critically damped SOPDT model
due to model mismatch.
Conclusions
References
[1] Ahmed, S., Huang, B., and Shah, S. L., 2007, Novel Identification Method
From Step Response, Control Eng. Pract., 15(5), pp. 545556.
[2] Ahmed, S., Huang, B., and Shah, S. L., 2008, Identification From Step
Responses With Transient Initial Conditions, J. Process Control, 18(2), pp.
121130.
[3] Majhi, S., and Atherton, D. P., 1999, Autotuning and Controller Design for
Processes With Small Time Delays, IEE Proc.-Control Theory Appl., 146(5),
pp. 415425.
[4] Bi, Q., Cai, W. J., Lee, E. L., Wang, Q. G., Hang, C. C., and Zhang, Y., 1999,
Robust Identification of First-Order Plus Dead-Time Model From Step
Response, Control Eng. Pract., 7(1), pp. 7177.
[5] Chang, R. C., Shen, S. H., and Yu, C. C., 1992, Derivation of Transfer Function From Relay Feedback Systems, Ind. Eng. Chem. Res., 31(3),
pp. 855860.
str
[6] A
om, K. J., and Hagglund, T., 1984, Automatic Tuning of Simple Regulators With Specifications on Phase and Amplitude Margins, Automatica, 20(5),
pp. 645651.
[7] Sung, S., and Lee, I. B., 1999, On-Line Process Identification and PID Controller Autotuning, Korean J. Chem. Eng., 16(1), pp. 4555.
[8] Scali, C., Marchetti, G., and Semino, D., 1999, Relay With Additional Delay
for Identification and Autotuning of Completely Unknown Processes, Ind.
Eng. Chem. Res., 38(5), pp. 19871997.
[9] Panda, R., and Yu, C. C., 2003, Analytical Expressions for Relay Feed Back
Responses, J. Process Control, 13(6), pp. 489501.
[10] Atherton, D. P., 2006, Relay Autotuning: An Overview and Alternative
Approach, Ind. Eng. Chem. Res., 45(12), pp. 40754080.
[11] Lin, C., Wang, Q. G., and Lee, T. H., 2004, Relay Feedback: A Complete
Analysis for First-Order Systems, Ind. Eng. Chem. Res., 43(26),
pp. 84008402.
[12] Kaya, I., and Atherton, D. P., 2001, Parameter Estimation From Relay Autotuning With Asymmetric Limit Cycle Data, J. Process Control, 11(4),
pp. 429439.
[13] Srinivasan, S. K., and Chidambaram, M., 2004, An Improved Autotune Identification Method, Chem. Biochem. Eng. Q., 18(3), pp. 249256.
[14] Vivek, S., and Chidambaram, M., 2005, Identification Using Single Symmetrical Relay Feedback Test, Comput. Chem. Eng., 29(7), pp. 16251630.
[15] Majhi, S., 2007, Relay Based Identification of Processes With Time Delay, J.
Process Control, 17(2), pp. 93101.
[16] Majhi, S., 2007, Relay-Based Identification of a Class of Nonminimum Phase
SISO Processes, IEEE Trans. Autom. Control, 52(1), pp. 134139.
[17] Bajarangbali, R., Majhi, S., and Pandey, S., 2014, Identification of FOPDT
and SOPDT Process Dynamics Using Closed Loop Test, ISA Trans., 53(4),
pp. 12231231.
[18] Liu, T., Gao, F., and Wang, Y., 2008, A Systematic Approach for Online Identification of Second Order Process Model From Relay Feedback Test, AIChE
J., 54(6), pp. 15601578.
[19] Liu, T., and Gao, F., 2008, Alternative Identification Algorithms for Obtaining
a First-Order Stable/Unstable Process Model From a Single Relay Feedback
Test, Ind. Eng. Chem. Res., 47(4), pp. 11401149.
[20] Fedele, G., 2009, A New Method to Estimate a First-Order Plus Time Delay
Model From Step Response, J. Franklin Inst., 346(1), pp. 19.
str
[21] Hang, C., A
om, K. J., and Wang, Q., 2002, Relay Feedback Auto-Tuning of
Process Controllers-A Tutorial Review, J. Process Control, 12(1),
pp. 143162.
[22] Liu, T., Wang, Q. G., and Huang, H. P., 2013, A Tutorial Review on Process
Identification From Step or Relay Feedback Test, J. Process Control, 23(10),
pp. 15971623.
[23] Bhanot, S., 2008, Process Control Principles and Applications, Oxford University Press, New Delhi, India.
[24] Ljung, L., 1999, System Identification: Theory for the User, 2nd ed., Prentice
Hall, Englewood Cliffs, NJ.
[25] MATLAB, 2007, Curve Fitting Toolbox, The Mathworks Inc., Natick, MA.
[26] Vivek, S., and Chidambaram, M., 2012, An Improved Relay Auto Tuning of
PID Controllers for Critically Damped SOPTD Systems, Chem. Eng. Commun., 199(11), pp. 14371462.
[27] Yokogawa Electric, 2002, Technical Information: CENTUM CS3000 Integrated Production Control System (IPCS)TI 33Q01B10-01E, 6th ed., Yokogawa Electric Corp., Tokyo, Japan.