You are on page 1of 32

AIChE Journal

Half Order Plus Time Delay (HOPTD) Models to Tune PI


Controllers

Journal: AIChE Journal


Fo
Manuscript ID AIChE-16-17775

Wiley - Manuscript type: Research Article

Date Submitted by the Author: 05-Feb-2016


rp
Complete List of Authors: Lee, Jietae; Kyungpook National University, Department of Chemical
Engineering
Lee, Yongjeh; Korea University, Chemical and Biological Engineering
Yang, Dae Ryook; Korea University, Chemical and Biological Engineering
ee

Edgar, Thomas; The university of Texas at Austin, Chemical engineering

Half-Order Plus Time Delay Model, First-Order Plus Time Delay Model,
Keywords:
Proportional Integral Control, Tuning, Step Response
rr
ev
ie
w
on
ly

AIChE Journal
Page 1 of 31 AIChE Journal

1
2
3
4
5 Half Order Plus Time Delay (HOPTD) Models to Tune PI Controllers
6
7
8 Jietae Leea*, Yongjeh Leeb, Dae Ryook Yangb, Thomas F. Edgarc
9
a
10 Department of Chemical Engineering, Kyungpook National University, Daegu 702-701, Korea
b
11 Department of Chemical and Biological Engineering, Korea University, Seoul 136-713, Korea
c
12 Department of Chemical Engineering, University of Texas at Austin, Austin, TX78712, U. S. A.
13
*Corresponding author. Tel: +82-53-950-5620, Fax: +82-53-950-6615, E-mail: jtlee@knu.ac.kr
14
15
Fo
16
17
18
19 Abstract: Methods based on the first-order plus time delay (FOPTD) model are very popular for tuning
rp
20
21 proportional-integral (PI) controllers. The FOPTD model-based methods are simple and their utility has
22
23 been proved with many successful applications to a wide range of processes in practice. However, even
ee

24
25 for some overdamped processes where the FOPTD model has been applied successfully, these empirical
26
FOPTD model-based methods can fail to provide stable tuning results. To remove these drawbacks, a PI
27
rr

28 controller tuning method based on half-order plus time delay (HOPTD) model is proposed. Because
29
30 FOPTD model-based methods can be applied to higher order processes, the proposed HOPTD model-
ev

31
32 based method can be applied to higher order processes as well. It does not require any additional process
33
34 information compared to the FOPTD model-based method and hence can be used in practice,
ie

35
36 complementing the traditional FOPTD model-based methods.
w

37
38
39
40
on

41 Keywords: Half-Order Plus Time Delay Model, First-Order Plus Time Delay Model, Proportional Integral
42
43 Control, Tuning, Step Response
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 1

AIChE Journal
AIChE Journal Page 2 of 31

1
2
3
4
5 Introduction
6
7 Proportional-integral (PI) controllers are still used widely in practice due to their simplicity and are
8
9 covered in detail in almost all process control textbooks.1 The PI controller gain and integral time can be
10
11 designed based on the first-order plus time delay (FOPTD) model.2 The FOPTD model has three
12
parameters (process gain, time constant, and time delay) and many PI controller tuning rules based on
13
14 FOPTD model parameters are available. For the successful PI control system, a key but time-consuming
15
Fo
16 step is to obtain the FOPTD model of process. There are many open-loop and closed-loop approaches for
17
18 this identification step.3
19
rp
20 For known high order process models, their FOPTD models can be obtained by applying the SIMC
21
22 model reduction rule.4-5 For this, full high order models are required and sometimes it may be
23
ee

24 complicated to obtain high order models from process responses. Traditional simple methods to obtain
25
FOPTD models are based on the several measurements of process step responses. Two point data6 and
26
27 two area data7 of process step responses can be used. Relay feedback responses can also be used to obtain
rr

28
29 the FOPTD models.8
30
ev

31 Due to the structural limitations, the FOPTD models cannot be used for underdamped processes
32
33 showing overshoots. In addition, it is not well known that FOPTD models do not fit responses of some
34
ie

35 overdamped processes whose transfer functions have negative zeros. For high order processes, the phase
36
shifts due to additional poles increase the time delays in the FOPTD models and PI controllers tuned
w

37
38 based on these FOPTD models will be more conservative, being acceptable in field. However, negative
39
40
on

zero terms give phase leads and decrease the time delays in the FOPTD models. Decreased time delays
41
42 make PI controllers more aggressive. For some processes with negative zeros, their performances can be
43
44 too oscillatory and not acceptable. To remove these drawbacks without increasing the model complexities,
ly

45
46 fractional order models have been available.9 However, computations for the fractional order derivatives
47
48 and integrators required for the fractional order models are rather complex.10
49
To tune PI controllers for overdamped processes including processes for which the FOPTD models
50
51 fail to provide acceptable PI controller parameters, a method based on a half-order plus time delay
52
53 (HOPTD) model is proposed. It provides more conservative tuning results compared to the FOPTD
54
55 model-based method. When slightly slower responses are acceptable, the proposed HOPTD model-based
56
57 method can replace the FOPTD model-based methods to tune PI controllers. Otherwise, the proposed
58
59
60 2

AIChE Journal
Page 3 of 31 AIChE Journal

1
2
3
4
5 method can be used selectively, complementing the existing FOPTD model-based methods. Several
6
7 representative examples are presented.
8
9
10
11
12
Motivations
13
14 Consider a linear time-invariant process;
15
Fo
16 Y (s) = G(s)U (s) (1)
17
18 Here G(s) is the process transfer function, and U(s) and Y(s) are the Laplace transforms of the process
19
rp
20 input and output variables, respectively. To design PI controllers, the process transfer function G(s) is
21
22 often approximated by the FOPTD model of
23
ee

24 k exp(−θ F s )
25 G (s ) ≈ GF ( s) = (2)
τ F s +1
26
27
rr

Three parameters of GF(s) can be identified from process responses such as step responses, pulse
28
29 responses and relay feedback responses. The FOPTD model of Eq. 2 can be applied successfully to
30
ev

31 overdamped processes, the main process type in practice. From step responses, we can determine clearly
32
33 whether processes are included in this type. Typical step responses of overdamped processes are shown in
34
ie

35 Figure 1. Three parameters in Eq. 2 are, from the new steady state value of the output,
36
k = yinf / M (3)
w

37
38
39 where M is the size of input step change. From two point measurements (two-point method6),
40
on

41 τ F = 1.5(t b − t a )
(4)
42 θ F = tb −τ F
43
44 or, from two area measurements (two-area method7),
ly

45
46 τ F = 2.7183 A1 / y inf
(5)
47 θ F = A0 / yinf −τ F
48
49 The above graphical identification methods for FOPTD models are very successful and are being used
50
51 in practice. However, for some overdamped processes with negative zeros, this FOPTD model-based
52
53 method can fail to provide stable PI controllers. For example, consider the last diagonal element of the
54
55 Ogunnaike and Ray column;11
56
57
58
59
60 3

AIChE Journal
AIChE Journal Page 4 of 31

1
2
3
4
5 (11.61s + 1) exp(−s) 0.4822exp(−s) 0.5178exp(−s)
6 G( s ) = = + (6)
(18.8s + 1)(3.89s + 1) 18.8s + 1 3.89s + 1
7
8
Identified FOPTD models from the step response are given in Table 1 and their fitting results are given in
9
10 Figure 2. The two-area method shows a negative time delay that cannot be implemented. The two-point
11
12 method shows a positive time delay that is less than the process time delay of 1. The integral of absolute
13
14 error (IAE) method is such that IAE is minimized under the constraints of positive time delay. Although
15
Fo
16 approximate FOPTD models seem not bad except for the two-area method, PI controllers tuned by the
17
18 SIMC tuning rules4 based on these three FOPTD models all fail to provide stable closed-loop systems.
19
rp
20 This is mainly due to the estimated time delays lower than the process time delay of 1. Low time delays
21
are caused by negative zeros in the process transfer functions. As shown in Eq. 6, negative zeros appear in
22
23 the parallel connection of two FOPTD processes and there can be many processes with negative zeros in
ee

24
25 practice. For such processes, an empirical model different from the FOPTD model will be required to
26
27 tune PI controllers.
rr

28
29
30
ev

31
32
33 HOPTD Model
34
ie

Consider a process transfer function;


35
36 1
G ( s) =
w

37 (7)
38 ( s / β + 1) β
39
40 It can describe the linear time-invariant process whose response in time-domain is
on

41
42 t β β exp( − βtˆ)tˆ β −1
43
y (t ) = ∫0 g (tˆ)u (t − tˆ)dtˆ, g (tˆ) =
Γ( β − 1)
(8)

44
ly

45 Figure 2 shows step responses (u(t)=1) for various β. For the first-order process (β=1), g (tˆ) = exp(−tˆ)
46
47 and y(t)=1-exp(-t). The half-order process (β=0.5) shows initial steep rise with long tails. Processes with
48
49 negative zeros and some diffusion processes12-14 show this half-order type of step response.
50
51 For some overdamped processes whose step responses show an initial steep rise and long tails,
52
53 FOPTD models suffer when tuning PI controllers as shown in the previous section. To remove these
54
55 difficulties, a half-order plus time delay (HOPTD) model is considered;
56
57
58
59
60 4

AIChE Journal
Page 5 of 31 AIChE Journal

1
2
3
4
5 k exp(−θ H s)
6 G ( s) ≈ GH ( s ) = (9)
τ H s +1
7
8
9 Properties of the half-order process are given in Appendix. Like the time delay term, Pade approximates
10
are available for the half-order model. Cheon et al.9 proposed a fractional-order model of
11
12
G ( s ) ≈ k exp( −θs ) /((τs ) β + 1) to overcome drawbacks of the FOPTD model. This fractional-order
13
14
model has no Pade approximates and is rather difficult to handle.10
15
Fo
16
17
18
19
rp
20 Graphical Identification from Open-loop Step Responses
21
22 The unit step response of HOPTD model (Eq. 9) is given as (see Appendix)
23
ee

24  0, t < θH
25
26
y (t ) = 
k erf ( (t − θ H ) /τ H )
, otherswise
(10)

27
rr

Here erf (⋅) is the error function which is found as a built-in function in recent scientific program
28
29
30 languages. Since erf ( 0.0659 ) = 0.283 and erf ( 0.4054 ) = 0.632 , for the measurements in Figure
ev

31
32 1(a), we have
33
34 t a = θ H + 0.0659 τ H
ie

(11)
35 t b = θ H + 0.4054 τ H
36
w

37 Hence the HOPTD model parameters are


38
39 τ H = 2.9455 (t b − t a )
40 (12)
θ H = t b − 0.4054 τ H
on

41
42
For two areas in Figure 1(b), we have
43
44
ly

A0 / y inf = θ H + 0.5 τ H
45 0.5 (13)
46 A1 / y inf =τ H ∫0 erf ( ξ ) dξ = 0.2420 τ H
47
48 and the HOPTD model parameters are
49
50 τ H = 4.1328 A1 / y inf
51 (14)
θ H = A0 / yinf − 0.5τ H
52
53
Relationship between Eq. 4 for the FOPTD model and Eq. 12 for the HOPTD model is
54
55
τ H = 1.9637τ F
56 (15)
57 θ H = θ F + 0.2039τ F
58
59
60 5

AIChE Journal
AIChE Journal Page 6 of 31

1
2
3
4
5 The time constant of HOPTD model is about 2 times that of FOPTD model and the time delay of HOPTD
6
7 model is greater than that of FOPTD model.
8
9
10
11 [Small Time Delay Process]: When the process time delay θH is very small compared to τH and the
12
process gain k is large, the process can be approximated by a half-order integral plus time delay (HOIPTD)
13
14
model as
15
Fo
16
exp(−θ H s )
17 G ( s ) ≈ G HI ( s ) = kˆ (16)
18 s
19
rp
20 For this model, it is not needed for the output to reach new steady state for the step input change. For the
21
step input change, the output becomes
22
23
 0, t < θH
ee

24 
y (t ) =  2 kˆ (17)
25  π t − θ H , otherswise
26 
27
rr

28 Hence, from the slope of y2(t), we can find the process gain;
29
30 π y b2 − y a2
kˆ = (18)
ev

31 2 tb − ta
32
33 The time delay is
34
ie

35 y b2 t a − y a2 t b
36 θH = (19)
y b2 − y a2
w

37
38
39 This model can be used when θH<τH/10.
40
on

41
42
43
44
ly

PI Controller for the HOPTD model


45
46 PI controller parameters are designed based on the gain margins.15 For the PI controller
47
48 C ( s) = k C + k I / s (20)
49
50 we use
51
52 k C = k Cu / 3
53 (21)
k I =k Iu / 2
54
55 Here kCu is the ultimate gain for the proportional only control and kIu is the ultimate gain for the integral
56
57
58
59
60 6

AIChE Journal
Page 7 of 31 AIChE Journal

1
2
3
4
5 only control. Controller gains of kC and kI are detuned by factors of 3 and 2, respectively. These detuning
6
7 factors can be adjusted for robustness and performances.
8
9 The characteristic equation for the PI control system is
10
11 1 + GH (s)(kC + k I / s ) = 0 (22)
12
13 Hence
14
15
k Cu =
1
=
(1 + (τ H ω180 ) 2 )
0 .25
Fo
16 | G H ( jω180 ) | k
17 (23)
18 k Iu =
ω 90 (
ω 1 + (τ H ω 90 ) 2
= 90
)
0.25

19 | G H ( jω 90 ) | k
rp
20
21 where ω90 and ω180 are the angular frequencies whose phase angles of GH(s) are -90o and -180o,
22
23 respectively. Specifically, ω180 can be computed from
ee

24
25 ∠G H ( jω180 ) = −θ H ω180 − 0.5 tan −1 (τ H ω180 ) = −π (24)
26
27 Its approximate solution is16
rr

28
29  3π 1 + 1 + 8θ H /(9πτ H ) θH
30  , < 2.7
 8 θ τH
ev

31 ω180 = H (25)
 8π
32 , otherwise
 8θ H + πτ H
33
34
ie

35 The frequency of ω90 can be computed from


36
∠G H ( jω90 ) = −θ H ω 90 − 0.5 tan −1 (τ H ω 90 ) = −π / 2
w

37 (26)
38
39 It can be rewritten as
40
on

41 − 2θ H ω 90 − tan −1 (τ H ω 90 ) = −π (27)
42
43 The angular frequency of Eq. 27 becomes the ultimate frequency of process
44
ly

45 G ( s ) = exp( −2θ H s ) /(τ H s + 1) and its approximate solution can be obtained as16
46
47
π 1 + 1 + 8θ H /(πτ H )
48 ω 90 = (28)
49 8 θH
50
51
52
[Small Time Delay Process]: For the HOIPTD model of Eq. 16, we use
53
54
55
56
57
58
59
60 7

AIChE Journal
AIChE Journal Page 8 of 31

1
2
3
4
5 ω180 3π
6 k C = k Cu / 2 = =
2kˆ 4kˆ θ H
7 (29)
8 ω 90 ω 90 π 1.5
9 k I = k Iu / 2 = =
2kˆ 16 kˆθ H1.5
10
11 Here the detuning factor for kC is decreased from 3 (in Eq. 21) to 2 for better transients.
12
13
14
15
Fo
16
17 PI Controller with Filter (PIF Controller)
18
19 The internal model control (IMC) method cancels stable poles of the process by the controller zeros.1,4
rp
20
21 Similarly, a controller based on this concept can be designed. Consider a controller that cancels the
22
process pole dynamics;
23
ee

24
25 kI τ H s +1
C (s) = (30)
26 s
27
rr

28 Then the closed loop transfer function becomes


29
30 Y (s) G H ( s )C ( s ) kk I exp(−θ H s )
= = (31)
ev

31 R( s ) 1 + G H ( s )C ( s ) s + kk I exp(−θ H s )
32
33 This is the same as that for the SIMC control system.4 When the controller gain is chosen as
34
ie

35 1
kI = (32)
36 2 kθ H
w

37
38 the set-point responses will be the same as those of SIMC control systems for FOPTD processes.
39
40 The controller can be implemented in the form of PI controller with the half-order filter as
on

41
42 k I (τ H s + 1) 1  k  1
C (s) = =  k Iτ H + I  (33)
43 s τ H s +1  s  τ H s +1
44
ly

45
46 In Eq. 33, the half-order filter part, 1 / τ H s + 1 , can be implemented as (see Appendix)
47
48 x& H (t ) = AH x H (t ) + bH e(t )
(34)
49 z (t ) = c H x H (t )
50
51 where, for a given filter order of nH,
52
53
54
55
56
57
58
59
60 8

AIChE Journal
Page 9 of 31 AIChE Journal

1
2
3
4
5 1 1 1 L 1 
6  
1 3 3 L 3 
7 2 
AH =− 1 3 5 L 5 
8 τH  
9 M M M O M 
 
10 1 3 5 L 2 n H − 1
11 1
12 bH = (1, 1, 1 L 1)T
τH
13
14 c H = 2 (1, 1, 1 L 1)
15
Fo
16 The filter output z(t) goes to the PI control part of C(s). Figure 4 shows this PI controller with the half-
17
order filter. As in the SIMC tuning, this PI controller with the half-order filter can show poor load
18
19
responses when τH is much large compared to the time delay θH. This controller will be effective for
rp
20
21
θH>τΗ/10.
22
23 Table 2 shows the proposed tuning rules for HOPTD model-based PI controllers. Figure 6 shows
ee

24
25 closed-loop responses of proposed PI controllers with and without filter when applied to HOPTD
26
27 processes.
rr

28
29
30
ev

31
32
33 Simulations
34
ie

Process 1 (Integer Order Plus Time Delay Process):


35
36 Consider a process with positive integer n,
w

37
38 exp( −0.3s )
39 G(s) = (35)
( s + 1) n
40
on

41
This process is a usual one for which the FOPTD model can be effectively applied. Figure 7 shows fitting
42
43 results and performances of PI controllers. For n=1, the FOPTD model describes the process exactly and
44
ly

45 performances of the FOPTD model-based PI controller are excellent. On the other hand, PI controllers
46
47 tuned based on the HOPTD model show slower closed-loop responses. When conservative tuning is
48
49 acceptable,17 the proposed HOPTD model-based method can be used still for n=1. For n=3, both PI
50
51 controllers tuned by the FOPTD and HOPTD models show similar closed-loop responses.
52
53
54
Process 2 (Ogunnaike and Ray Column11):
55
56 Consider the process of Eq. 6 which is the last diagonal element of the Ogunnaike and Ray column.11
57
58
59
60 9

AIChE Journal
AIChE Journal Page 10 of 31

1
2
3
4
5 Figure 7 shows fitting results and performances of PI controllers. FOPTD models obtained graphically
6
7 from the open-loop step response all provide unstable PI control systems. The proposed PI and PIF
8
9 controllers based on the HOPTD model show excellent closed-loop performances. For comparisons,
10
11 performances of FOPTD model obtained by applying the SIMC reduction method4 are given. The
12
FOPTD model by the SIMC reduction is
13
14
(11.61s + 1) exp(−s) 11.61 exp(−s) 0.6176exp(−s)
15 G(s) = ≈ = (36)
(18.8s + 1)(3.89s + 1) 18.8 3.89s + 1 3.89s + 1
Fo
16
17
18 This FOPTD model provides the PI controller with excellent closed-loop performances. However, this
19
rp
SIMC method requires the full high order process model.
20
21
22
23 Process 3:
ee

24
25 Consider a process,
26
27 (1.2 s + 1) 2 exp( −0.5s )
rr

28 G (s) = (37)
(5s + 1)( s + 1) 2
29
30
Figure 9 shows fitting results and performances of PI controllers. For this process, the accuracy of the
ev

31
32 fitted FOPTD model is excellent as shown in Table 1 (the integral of absolute error (IAE) of the FOPTD
33
34
ie

model is far lower than that of the HOPTD model). However, closed-loop performances of the PI control
35
36 system based on the FOPTD model is too oscillatory as indicated by the low gain and phase margins and
w

37
38 high peak amplitude ratio of the sensitivity function (Table 3). This is due to the time delay that is smaller
39
40 than the process time delay of 0.5. The proposed PI and PIF controllers tuned based on the HOPTD
on

41
42 model show excellent closed-loop responses.
43
44
ly

45 Process 4 (Process with Inverse Response):


46
47 Consider a process
48
49 (5s + 1)(−2s + 1)
50 G(s) = (38)
(10s + 1)(2s + 1)(s + 1)
51
52 This process has a right-half plane zero (positive zero) and the open-loop step response shows the inverse
53
54 response. Figure 10 shows fitting results and performances of PI controllers. Closed-loop performance of
55
56 the PI control system based on the FOPTD model is oscillatory, which is indicated by the low gain margin
57
58
59
60 10

AIChE Journal
Page 11 of 31 AIChE Journal

1
2
3
4
5 and high peak amplitude ratio of the sensitivity function (Table 3). On the other hand, the proposed PI and
6
7 PIF controllers tuned based on the HOPTD model show excellent closed-loop responses.
8
9
10
11 Process 5 (Packed Bed Reactor18):
12
Consider an isothermal, isobaric packed bed reactor in Kim.18 There are two partial differential
13
14 equations that represent the mass balance equations for the axial bulk flow and inside the porous catalyst
15
Fo
16 particles. The dimensionless mass balance equations are
17
18 ∂y b 1 ∂ 2 y b ∂y b 1 − ε b ∂y p
19 = − −3 α
∂t Pe ∂z 2
∂z εb ∂x
rp
20 x =1
(39)
21 ε p + K ∂y p 1 ∂  2 ∂y p 
= 2 x  −φ2 yp
22 α ∂t x ∂x  ∂x 

23
ee

24 with initial and boundary conditions;


25
26 y b ( z ,0) = 0, y p ( z , x,0) = 0,
27
rr

28 ∂y p
y p ( z ,1, t ) = y b ( z , t ), =0 (40)
29 ∂x x =0
30 ∂y b 1 ∂y b
ev

31 = 0, y b (0, t ) − = u (t )
∂z z =1 Pe ∂z z =0
32
33
34 Here t is time, z is the axial position in bed, x is the radial position in catalyst particle, yb is the
ie

35
36 concentration in the bulk flow, yp is the concentration in catalyst particle, Pe is the Peclet number, α is the
w

37
38 ratio of space time and diffusion time, εb is the bed void fraction and εp is the void fraction in the catalyst
39
particle, K is adsorption equilibrium constant and φ is the Thiele modulus. All variables are dimensionless.
40
on

41
Details can be found in Kim.18
42
43 Laplace domain solution is available and the transfer function between the concentration in bed outlet,
44
ly

45 yb(1,t), and the forcing function in the bed inlet, u(t), is


46
47 Yb (1, s ) 2bPe exp( Pe / 2)
48 G ( s) = =
U (s) exp(b) ( Pe 2 / 4 + ( Pe + b)b) − exp( −b) ( Pe 2 / 4 + (− Pe + b)b)
49
Pe
50 b= 1 + 4 B / Pe (41)
51 2
52 3(1 − ε b )α  2 ε p + K  εp +K  
53 B=s+  φ + s coth  φ 2 + s  − 1
εb  α  α  
54    
55
56 Figure 11 shows fitting results and performances of PI controllers tuned for Pe=1000, K=14, α=0.1,
57
58
59
60 11

AIChE Journal
AIChE Journal Page 12 of 31

1
2
3
4
5 εb=0.5, εp=0.4 and φ=0. Closed-loop performances of the PI control system based on the FOPTD model is
6
7 very oscillatory as indicated by the low gain and phase margins and high peak amplitude ratio of the
8
9 sensitivity function (Table 3). On the other hand, the proposed PI and PIF controllers tuned based on the
10
11 HOPTD model show excellent closed-loop responses.
12
13
14
Process 6 (Process with a Small Time Delay):
15
Fo
16 Consider a process
17
18 20(50 s + 1) exp( − s )
19 G (s) = (42)
1250 s 2 + 100 s + 1
rp
20
21 This process has a small time delay compared to time constants and a large process gain of 20. So an
22
23 integral plus time delay (IPTD) model and a half-order integral plus time delay (HOIPTD) model can be
ee

24
25 applied. It will not be necessary to wait for the output to reach a new steady state. Figure 11 shows fitting
26
27 results and performances of PI controllers. The IPTD model obtained from the initial part of open-loop
rr

28
29 step response is not appropriate to tune the PI controller because the time delay estimated is negative. On
30
the other hand, the proposed HOIPTD model has proper model parameters and the proposed PI controller
ev

31
32 tuned based on the HOIPTD model shows excellent closed-loop responses. For comparisons,
33
34 performances of IPTD model obtained by applying the SIMC reduction method4 are given. The IPTD
ie

35
36 model by the SIMC reduction is
w

37
38 20(50s + 1) exp(−s) 1000 exp(−s) 0.8 exp(−s)
G( s) = ≈ = (44)
39 2
1250s + 100s + 1 1250 s s
40
on

41 This IPTD model provides PI controller with excellent closed-loop performances. However, this SIMC
42
43 method requires the full high order process model.
44
ly

45
46
47
48
49 Conclusion
50
To tune PI controllers in field, it is proposed to use the HOPTD model
51
52
k exp(−θ H s )
53 GH ( s) =
54 τ H s +1
55
56 Field data for the three model parameters of (k, τH, θH) are just the same as those for the FOPTD model,
57
58
59
60 12

AIChE Journal
Page 13 of 31 AIChE Journal

1
2
3
4
5 i.e., only two point data or two areas with the final steady state are the field data required. Simple PI
6
7 controller tuning rule for the HOPTD model has been suggested. The Pade approximates are available for
8
9 this HOPTD model and its mathematical operations for open-loop and closed-loop simulations are simple.
10
11 For some overdamped processes with negative zeros for which the FOPTD model-based PI
12
controllers are oscillatory or even unstable, stable PI controllers can be obtained by the proposed HOPTD
13
14 model-based method. For first and high order processes with time delays, the HOPTD model-based
15
Fo
16 method provides PI control systems with slower performances compared to the FOPTD model-based
17
18 method, because the time delays estimated for the HOPTD model are higher than those for the FOPTD
19
rp
20 model. When slightly slower closed-loop responses are allowed, the HOPTD model-based method can
21
22 replace the FOPTD model-based method. Otherwise, the proposed HOPTD model-based method can be
23
ee

24 used selectively for processes where the FOPTD model shows smaller time delays than the actual process
25
time delays.
26
27
rr

28
29
30
ev

31 Appendix (Half-Order Process19-21)


32
33 Consider a half-order process;
34
ie

35 1
G( s) = (A1)
36 τs + 1
w

37
38
By applying successively the relationship for p = τs + 1
39
40
on

τs τs
41 p = 1+ = 1+ (A2)
42 1+ p τs
1+1+
43 1+ p
44
ly

45 we have the continued fraction of19


46
47 1 1
G (s) = = (A3)
48 τs + 1 τs
1+
49 τs
2+
50 τs
2+
51 2 +L
52
53 The continued fraction of Eq. A3 converges fast and, truncating the continued fraction, Pade
54
55 approximations of the half-order process can be obtained, for example,
56
57
58
59
60 13

AIChE Journal
AIChE Journal Page 14 of 31

1
2
3
4
5 1 1
G ( s ) ≈ P0 / 1 ( s ) = = (A4)
6 τs τ
7 1+ s +1
2 2
8
9 τ
10 s +1
1 2
G ( s ) ≈ P1/ 2 ( s) = = (A5)
11 τs τ2
1+ s + τs + 1
2
12 τs 8
13 2+
τs
14 2+
2
15
Fo
16 These Pade approximates can be realized in the state space form as20
17
18 x& (t ) = AH x(t ) + bH u (t )
19 (A6)
y(t ) = c H x(t )
rp
20
21
where
22
23
1 1 1 L 1 
ee

24  
25 1 3 3 L 3 
2
26 AH =− 1 3 5 L 5 
τ 
27 M M M O M 
rr

28  
1 3 5 L 2 n − 1
29
1
30 bH = (1, 1, 1 L 1)T
τ
ev

31
32 c H = 2 (1, 1, 1 L 1)
33
34 The unit step response of the half-order process (U(s)=1/s) is21
ie

35
t t /τ

∫ exp (− ζ )dζ
36  1  1 1  ξ 2
y (t ) = L−1   = ∫ exp −  dξ = 2
w

37  τ
 s τs + 1  π τ 0
ξ π 0
(A7)
38
39 = erf ( t /τ )
40
on

41 The amplitude ratio and phase angle are given as


42
43 1
G ( jω ) =
44 (1 + (τω ) 2 ) 0.25
ly

(A8)
45 ∠ G ( jω ) = −0.5 tan (τω ) −1
46
47
48 Hence the corner frequency is ω=1/τ and, at that frequency, the phase angle is –π/8. The phase angle goes
49
50 to –π/4 as ω increases.
51
52
53
54
55 Acknowledgments
56
57 This work (2014004928) was supported by Mid-career Research Program through NRF grant funded
58
59
60 14

AIChE Journal
Page 15 of 31 AIChE Journal

1
2
3
4
5 by the MEST.
6
7
8
9
10
11 Literature Cited
12
1. Seborg DE, Mellichamp DA, Edgar TF, Doyle FJ. Process Dynamics and Control, 3rd ed. New York:
13
14 Wiley, 2010.
15
Fo
16 2. O'Dwyer A. Handbook of PI and PID Controller Tuning Rules, 3rd ed. London, UK: Imperial College
17
18 Press, 2009.
19
rp
20 3. Liu T, Wang QG, Huang HP. A tutorial review on process identification from step or relay feedback test.
21
22 J Process Control. 2013;23:1597-1623.
23
ee

24 4. Skogestad S. Simple analytic rules for model reduction and PID controller tuning. J Process Control.
25
2003;13:291-309.
26
27
rr

5. Lee J, Cho W, Edgar TF. Simple analytic PID controller tuning rules revisited. Ind Eng Chem Res.
28
29 2014;53:5038-5047.
30
ev

31 6. Marlin TE, Process Control: Designing Processes and Control Systems for Dynamic Performance, 2nd
32
33 ed. New York: McGraw-Hill, 1995.
34
ie

35 7. Astrom KJ, Hagglund T. PID Controllers: Theory, Design, and Tuning, 2nd ed. Research Triangle Park,
36
NC: Instrument Society of America, 1995.
w

37
38 8. Lee J, Sung SW, Edgar TF. Integrals of relay feedback responses for extracting process information.
39
40
on

AIChE J. 2007;53:2329-2338.
41
42 9. Cheon YJ, Ryu KH, Sung SW, Lee J, Lee IB. PID auto-tuning using new model reduction method and
43
44 explicit PID tuning rule for a fractional order plus time delay model. J Process Control. 2014;24:113-
ly

45
46 128.
47
48 10. Monje CA, Chen Y, Vinagre BM, Xue D, Feliu V. Fractional-order Systems and Controls:
49
Fundamentals and Applications, New York: Springer, 2010.
50
51 11. Ogunnaike BA, Ray WH. Multivariable controller design for linear systems having multiple time
52
53 delays. AIChE J. 1979;25:1043-1057.
54
55 12. Lee J, Kim DH, High-order approximations for noncyclic and cyclic adsorption in a particle. Chem
56
57 Eng Sci. 1998;53:1209-1221.
58
59
60 15

AIChE Journal
AIChE Journal Page 16 of 31

1
2
3
4
5 13. Lee J, Kim DH. Global approximations of unsteady-state adsorption, diffusion, and reaction in a
6
7 porous catalyst. AIChE J. 2013;59:2540-2548.
8
9 14. Lee J, Kim DH. Simple high-order approximations for unsteady-state diffusion, adsorption and
10
11 reaction in a catalyst: a unified method by a continued fraction for slab, cylinder and sphere
12
geometries. Chem Eng J. 2011;173:644-650.
13
14 15. Lee J, Sung SW, Edgar TF. Stability margin interpretation of the SIMC tuning rule for PI controllers
15
Fo
16 and its applications. 9th International Symposium on Advanced Control of Chemical Processes
17
18 (ADCHEM). Whistler, British Columbia, Canada, June 7-10, 2015.
19
rp
20 16. Lee J, Park H, Sung SW. Analytic expressions of ultimate gains and ultimate periods with phase-
21
22 optimal approximations of time delays. Can J Chem Eng. 2005;83:990-995.
23
ee

24 17. Skogestad S. Tuning for smooth PID control with acceptable disturbance rejection. Ind Eng Chem Res.
25
2006;45:7817-7822.
26
27
rr

18. Kim DH. Approximations for unsteady-state diffusion and reaction in porous catalyst and their
28
29 application to packed-bed reactor. AIChE J. 2008;54:2423-2431.
30
ev

31 19. Abramowitz M, Stegun IA. Handbook of Mathematical Functions, New York: Dover Pub, 1972.
32
33 20. Chen CF, Shieh LS. Continued fraction inversion by Routh’s algorithm. IEEE Trans Circuit Theory.
34
ie

35 1969;16:197–202.
36
w

37 21. Zill DG, Cullen MR. Advanced Engineering Mathematics, Boston: PWS-KENT Pub, 1992.
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 16

AIChE Journal
Page 17 of 31 AIChE Journal

1
2
3
4
5 Table 1. Model parameters estimated and integral of absolute errors (IAE).
6
7 Process Method k τ θ IAE
8
exp(−0.3s ) n=1 FOPTD 1 1.0200 0.3000 0.02
9
n
10 ( s + 1) HOPTD 1 2.0029 0.5078 0.22
11 n=3 FOPTD 1 2.1000 1.4600 0.29
12
13 HOPTD 1 4.1237 1.8878 0.71
14 (11.61s + 1) exp(−s) FOPTD *1
1 9.2250 0.1950 2.78
15
(18.8s + 1)(3.89s + 1) *2
Fo
16 FOPTD 1 13.898 -1.864
*3
17 FOPTD 1 11.68 0 2.15
18
HOPTD 1 18.115 2.0744 1.16
19
rp
20 (1.2s + 1) 2 exp( −0.5s) FOPTD 1 4.7700 0.2300 0.13
21
(5s + 1)( s + 1) 2 HOPTD 1 9.3667 1.2018 0.85
22
23 (5s + 1)(−2s + 1) FOPTD 1 6.8250 1.8850 1.35
ee

24
(10s + 1)(2s + 1)(s + 1) HOPTD 1 13.4020 3.2755 0.50
25
26 Packed bed reactor (Eq. 40; FOPTD 1 12.255 1.0050 2.25
27 Pe=1000, K=14, α=0.1,
rr

HOPTD 1 24.153 3.5059 0.35


28 εb=0.5, εp=0.4, φ=0)
29 20 (50 s + 1) exp( − s ) IPTD 0.3819 -5.080
30 1250 s 2 + 100 s + 1 HOIPTD 2.0907 4.4337
ev

31
32
*1: Two-point method.
33 *2: Two-area method.
34
ie

*3: IAE method.


35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 17

AIChE Journal
AIChE Journal Page 18 of 31

1
2
3
4
5 Table 2. Tuning rules for HOPTD models
6
7 Process Controller
8
9
k exp(−θ H s) PI:
kC =
(1 + (τ H ω180 ) )
2 0.25

10 τ H s +1 k 3k
kC + I
11
12
s
kI =
(
ω 90 1 + (τ H ω90 ) 2 )
0.25

13 2k
14
 3π 1 + 1 + 8θ H /(9πτ H ) θH
15  , < 2.7
 8
Fo
16 ω180 = θ H τH
17  8π
, otherwise
18  8θ H + πτ H
19
rp
20 π 1 + 1 + 8θ H /(πτ H )
ω90 =
21 8 θH
22
23 PIF: 1
kI =
ee

24 2kθ H
kI τ H s +1
25
26 s
27 kˆ exp(−θ H s)
rr

PI: 3π 0.7675
28 kC = =
s kI ˆ
4k θ H kˆ θ H
29 kC +
30 s π 1.5 0.3480
kI = =
ev

31 16 kˆθ H1.5 kˆθ H θ H


32
33
34
ie

35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 18

AIChE Journal
Page 19 of 31 AIChE Journal

1
2
3
4
5 Table 3. Tuning results.
6
7 Process Method kC kI Gain Phase Ms*4
8 Margin Margin
9 exp(−0.3s ) n=1 FOPTD 1.7000 1.6667 3.1 61o 1.6
10 ( s + 1) n
HOPTD 1.0279 1.6920 4.5 53 o
1.5
11 o
12 n=3 FOPTD 0.7192 0.3425 3.8 62 1.6
13 HOPTD 0.7747 0.3936 3.3 56 o
1.7
14
(11.61s + 1) exp(−s) FOPTD*1 23.654 2.5641 Unstable
15
Fo
16 (18.8s + 1)(3.89s + 1) FOPTD *2
3.1459 0.8096 3.1 57 o 1.6
17 o
HOPTD 1.5189 0.5499 6.1 57 1.3
18
*3 o
19 HOPTD 18.115 0.2410 6.2 65 1.4
rp
20 (PIF)
21 (1.2s + 1) 2 exp(−0.5s) FOPTD 10.370 5.6356 1.1 5.2o 20.1
22 (5s + 1)( s + 1) 2 HOPTD 1.4359 0.9074 7.5 55 o
1.2
23 *3 o
ee

24 HOPTD 9.3667 0.4160 8.2 61 1.3


(PIF)
25
(5s + 1)(−2s + 1) FOPTD 1.8103 0.2653 1.5 51 o 3.4
26
27 (10s + 1)(2s + 1)(s + 1) HOPTD 1.0464 0.2653 2.3 54 o
1.9
rr

28 HOPTD 13.402 *3
0.1526 2.8 59 o
1.7
29 (PIF)
30 Packed bed reactor (Eq. 40; FOPTD 6.0970 0.4975 1.2 13 o 8.2
ev

31 Pe=1000, K=14, α=0.1, o


32 HOPTD 1.3494 0.2967 4.5 58 1.4
εb=0.5, εp=0.4, φ=0)
33 20 (50 s + 1) exp( − s ) IPTD *2
0.6250 0.0781 3.1 54 o
1.6
34
ie

1250 s 2 + 100 s + 1 HOIPTD 0.2430 0.0373 7.8 58 o


1.2
35
36 *1: Tuning based on the FOPTD model estimated by the two-point method.
w

37 *2: SIMC tuning based on the exact process transfer function.


38 *3: τˆ
39 *4: Peak amplitude ratio of the sensitivity function, S(s)=1/(1+G(s)C(s)).
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 19

AIChE Journal
AIChE Journal Page 20 of 31

1
2
3
4
5
6
7
8
9 yinf yinf
10 y(t) y(t)
11
12 0.632yinf A0
13
14 0.283yinf
A1
15
Fo
16
17 0 ta tb t 0 A0/yinf t
18 (a) Two-point method (b) Two-area method
19
rp
20
21 Figure 1. Two popular methods to identify FOPTD models from open-loop step responses.
22
23
ee

24
25
26
27
rr

28
29
30
ev

31
32
33
34
ie

35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 20

AIChE Journal
Page 21 of 31 AIChE Journal

1
2
3
4
5
6
7 1
8 y
9 0.8
10 Exact
0.6
11 Two-point method
12 0.4
13
Two-area method
14 0.2 IAE method
15
Fo
16 0
0 20 40 60 80 100
17 t
18
19
rp
20 Figure 2. Open-loop step responses of three FOPTD models for the process,
21 (11.61s + 1) exp(−s)
22 G(s) = .
23 (18.8s + 1)(3.89s + 1)
ee

24
25
26
27
rr

28
29
30
ev

31
32
33
34
ie

35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 21

AIChE Journal
AIChE Journal Page 22 of 31

1
2
3
4
5
6
1
7
8 y
9 β=0.5
10
11 β=3
12
13 β=1
14
15 0
Fo
16 0 1 2 3 4 t 5
17
18
19 Figure 3. Open-loop step responses of the process, G ( s) = 1 / (s / β + 1)β .
rp
20
21
22
23
ee

24
25
26
27
rr

28
29
30
ev

31
32
33
34
ie

35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 22

AIChE Journal
Page 23 of 31 AIChE Journal

1
2
3
4
5
6
7
8
9 y2(t)
10
11 yb2
12
ya2
13
14
15
0 θ ta tb t
Fo
16
17
18 Figure 4. Half-order integral plus time delay (HOIPTD) model identification.
19
rp
20
21
22
23
ee

24
25
26
27
rr

28
29
30
ev

31
32
33
34
ie

35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 23

AIChE Journal
AIChE Journal Page 24 of 31

1
2
3
4
5
6
7
8 R(s) U(s) Y(s)
9 HOF(s) PI(s) G(s)
10
11
12
13
14
15 Figure 5. PI controller with the half-order filter (HOF(s)= 1 / τ H s + 1 , PI(s)= k I τ H + k I / s ).
Fo
16
17
18
19
rp
20
21
22
23
ee

24
25
26
27
rr

28
29
30
ev

31
32
33
34
ie

35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 24

AIChE Journal
Page 25 of 31 AIChE Journal

1
2
3
4
5
6
7
8
9 1.2
10 θ=0.1
11
1
12
13 y
14 θ=1
0.8
15
Fo
16
17 0.6
18
19
rp
20 0.4
21 θ=1
22
23 0.2
ee

24
25
26 0 PI(HOPTD)
27 θ=0.1
rr

28 PIF(HOPTD)
29 -0.2
30 0 4 8 12 16 20
t
ev

31
32
33
34
ie

exp(−θs)
35 Figure 6. Closed-loop responses of the proposed PI control systems for G( s) = .
36 s +1
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 25

AIChE Journal
AIChE Journal Page 26 of 31

1
2
3
4
5
6
7
8
9 1
10 y
11 0.8
n=1 n=3
12
0.6
13
14
Exact
0.4
15 FOPTD
Fo
16 0.2
HOPTD
17
18 0
0 4 8 12 16 t 20
19
rp
20 (a) Open-loop response
21 1.5
22
23 n=1 n=3
ee

24 y 1
25
26
27 0.5 PI(FOPTD)
rr

28 PI(HOPTD)
29
30 0
0 4 8 12 16 t 20
ev

31
32 (b) Closed-loop response
33
34
ie

exp( −0.3s )
35 Figure 7. Open-loop and closed-loop responses for G ( s ) = .
36 ( s + 1) n
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 26

AIChE Journal
Page 27 of 31 AIChE Journal

1
2
3
4
5
6
7
8
9 1
10 y
0.8
11
12 0.6
13 Exact
14 0.4
15 FOPTD
Fo
16 0.2
HOPTD
17
0
18 0 20 40 60 80 t 100
19 (a) Open-loop response
rp
20
21 1.5
22
PI(SIMC)
23 y 1
ee

24
25 0.5 PI(FOPTD)
26
27 PI(HOPTD)
rr

0
28 PIF(HOPTD)
29
-0.5
30 0 40 80 120 160 t 200
ev

31 (b) Closed-loop response


32
33
34
ie

(11.61s + 1) exp(−s)
35 Figure 8. Open-loop and closed-loop responses for G(s) = .
36 (18.8s + 1)(3.89s + 1)
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 27

AIChE Journal
AIChE Journal Page 28 of 31

1
2
3
4
5
6
7 1
8 y
0.8
9
10 0.6
11 Exact
12 0.4
FOPTD
13
14 0.2 HOPTD
15
0
Fo
16 0 10 20 30 40 t 50
17 (a) Open-loop response
18
19 1.5
rp
20 y
21 1
22
23 0.5 PI(FOPTD)
ee

24
25 PI(HOPTD)
0
26 PIF(HOPTD)
27
rr

-0.5
28 0 20 40 60 80 t 100
29 (b) Closed-loop response
30
ev

31
(1.2 s + 1) 2 exp( −0.5s )
32 Figure 9. Open-loop and closed-loop responses for G ( s ) = .
33 (5s + 1)( s + 1) 2
34
ie

35
36
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 28

AIChE Journal
Page 29 of 31 AIChE Journal

1
2
3
4
5
6
7
8
9 1
10 y
11
0.5
12
13 Exact
14 FOPTD
0
15
Fo
16 HOPTD
17 -0.5
18 0 10 20 30 40 t 50
19 (a) Open-loop response
rp
20
21 1.5
22
23 y 1
ee

24
25 0.5 PI(FOPTD)
26
PI(HOPTD)
27
rr

0
28 PIF(HOPTD)
29 -0.5
30 0 20 40 60 80 t 100
ev

31 (b) Closed-loop response


32
33
34
ie

(5s + 1)(−2s + 1)
35 Figure 10. Open-loop and closed-loop responses for G(s) = .
36 (10s + 1)(2s + 1)(s + 1)
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 29

AIChE Journal
AIChE Journal Page 30 of 31

1
2
3
4
5
6
7
8
1
9
10
y
0.8
11
12 0.6
13 Exact
14 0.4
FOPTD
15
0.2
Fo
16 HOPTD
17 0
18 0 20 40 60 80 t 100
19 (a) Open-loop response
rp
20
1.5
21
22
y 1
23
ee

24
25 0.5
26 PI(FOPTD)
27 0
rr

PI(HOPTD)
28
29 -0.5
30 0 40 80 120 160 t 200
ev

31 (b) Closed-loop response


32
33
34 Figure 11. Open-loop and closed-loop responses for a packed bed reactor, Eq. 40 with Pe=1000, K=14,
ie

35
36 α=0.1, εb=0.5, εp=0.4 and φ=0.
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 30

AIChE Journal
Page 31 of 31 AIChE Journal

1
2
3
4
5
6
7
8
10
9
y 8 Exact IPTD(SIMC)
10
11 IPTD
12 6
13 HOIPTD
4 Data points
14
15 2
Fo
16
17 0
18 0 5 10 t 15
19 (a) Open-loop response
rp
20
1.5
21
22 y 1
23
ee

24
25 0.5
26 PI(SIMC)
27 0
rr

28 PI(HOIPTD)
29 -0.5
30 0 20 40 60 80 t 100
ev

31 (b) Closed-loop response


32
33
34
ie

20(50 s + 1) exp( − s )
35 Figure 12. Open-loop and closed-loop responses for G ( s ) = .
36 1250 s 2 + 100 s + 1
w

37
38
39
40
on

41
42
43
44
ly

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 31

AIChE Journal

You might also like