International Journal of Control, Automation, and Systems (2015) 13(2):346-352 ISSN:1598-6446 eISSN:2005-4092

DOI 10.1007/s12555-013-0418-9 [Link]

Equivalent Transfer Function based Multi-loop PI Control for

High Dimensional Multivariable Systems
Xiaoli Luan*, Qiang Chen, and Fei Liu

Abstract: A new equivalent transfer function (ETF) parameterization algorithm to incorporate the loop
interaction effect into the design of multi-loop PI controllers for high dimensional multivariable
processes is presented in this paper. The design scheme consists of two stages. In the first, by exploit-
ing the relationship between the equivalent closed-loop transfer function and the inverse of open-loop
transfer function, the analytical expression of ETF is derived. In the second stage, based on the ETF,
controller parameters for each loop are determined by utilizing the existing PI tuning rules and the
simple internal model control method. The proposed ETF parameterization algorithm is more accurate
and reasonable compared to the conventional ETF model approximation methods. Furthermore, the
advantage of the multi-loop PI controller designed by the proposed ETF is more significant when ap-
plied to higher dimensional processes with complicated interaction modes. Several typical industrial
process examples show the well-balanced and robust response with the minimum integral absolute er-
ror.

Keywords: Equivalent transfer function, high dimensional processes, multi-loop PI control, multi-
variable systems.

1. INTRODUCTION and different multi-loop control design techniques have

been proposed, such as detuning method [5,6], sequential
Many systems encountered in practice consist of loop closing method [7,8], independent design method
several interconnected loops. The interactions between [9,10], and so on. Since controllers interact with each
loops are a common phenomenon and the main obstacle other in a decentralized control system, the performance
in the design of multi-loop controllers for multi-input of one loop cannot be evaluated without the information
multi-output (MIMO) systems. Adjusting controller of the controllers in other loops.
parameters of one loop affects the performance of others, To overcome this difficulty, several novel independent
sometimes to the extent of destabilizing the entire system. design approaches by constructing equivalent individual
Therefore, one of the main issues in the control of loops have recently been proposed [11-16]. The effective
MIMO processes is the effects generated by interaction open-loop process (EOP) is formulated without the prior
terms. Although a full multivariable controller is an knowledge of controller dynamics in other loops and the
option, it is a matter of fact that the most common controllers are independently designed for the
solution found in industry is still the decentralized corresponding equivalent single loops in [11]. Similarly,
solutions based on PID controllers [1-4]. The the effective open-loop transfer function (EOTF) is
decentralized option obviously requires fewer parameters derived to decompose multi-loop control system into a
to tune than the multivariable controller. Besides its set of equivalent independent single loops in [12], then
simple structure, decentralized PID control has a decentralized controllers are designed independently
potential advantage of failure tolerance of the overall based on the EOTF model. To obtain effective transfer
control systems and has a long record of satisfactory function (ETF) model of independent single loops by
performance. It is therefore of considerable interest to decoupling and reducing the model simultaneously,
apply well established single-loop PI/PID tuning reference [13] suggested the dynamic relative interaction
principles to MIMO processes on a decentralized setting, to derive the multiplicative model factor (MMF) for an
__________ individual control loop. The ETF is then obtained by
Manuscript received September 5, 2013; revised April 10, multiplying the original loop transfer function with the
2014; accepted August 6, 2014. Recommended by Associate approximated MMF within the neighborhood of the
Editor Sung-Kwun Oh under the direction of Editor Zengqi Sun. individual control loop critical frequency. In an alternate
This work was supported by the National Natural Science manner, the ETF in terms of effective relative gain array
Foundation of China under Grant 61473137. (ERGA), relative normalized gain array (RNGA) and
Xiaoli Luan, Qiang Chen, and Fei Liu are with the Institute of
Automation, Jiangnan University, Wuxi 214122, P. R. China (e-
effective relative energy array (EREA) for the
mails: xlluan@[Link], [Link]@[Link], fliu@ multivariable system is derived in [14-16], where ERGA
[Link]). utilizes both gain and phase information, RNGA includes
* Corresponding author. both the steady state and the transient information of the

© ICROS, KIEE and Springer 2015

 Equivalent Transfer Function based Multi-loop PI Control for High Dimensional Multivariable Systems 347

transfer function matrix, and EREA is utilized to propose ⎡ gc1 ( s ) 0 " 0 ⎤

a new loop pairing method for the MIMO systems. ⎢ 0 gc 2 (s) " 0 ⎥⎥
Methods of MMF/ERGA/RNGA/EREA calculate ETF Gc ( s ) = ⎢ . (2)
by combining the original loop transfer function with ⎢ # # % # ⎥
⎢ ⎥
relative factors. Therefore, these ETFs have the same ⎣⎢ 0 0 " g cn ( s ) ⎦⎥
structure as the corresponding open-loop transfer
function. What is more, this approximation way to obtain As depicted in Fig. 1, designing a multi-loop con-
ETF inevitably has large modeling errors, which lead to troller is decomposed into tasks of design for controllers
more conservativeness for higher dimensional processes. in a number of equivalent and independent single loops.
In the present work, a new and more accurate ETF When overall system is closed, there are interactions
parameterization algorithm to incorporate the loop among control loops as a result of the existence of off-
interaction effect into the design of multi-loop PI diagonal elements in the transfer function. Considering
controllers is proposed. By making use of the the interactions from other loops, the ETFs can be
relationship between ETF and the inverse of open-loop compactly expressed in terms of DRGA, as follows [12]:
transfer function, the analytical expression of ETF model
is calculated directly. Based on the corresponding ETF, ⎡ 1 1 ⎤
⎢ gˆ ( s ) " gˆ1n ( s ) ⎥ ⎡ g11 ( s ) " g1n ( s ) ⎤
−T
multi-loop PI controllers are designed using the simple ⎢ 11 ⎥
internal model control method. The main contribution of ⎢ # % # ⎥ = ⎢⎢ # % # ⎥⎥ , (3)
the work lies in that the proposed ETF parameterization ⎢ ⎥
⎢ 1 1 ⎥ ⎢⎣ g n1 ( s ) % g nn ( s ) ⎥⎦
algorithm is more accurate compared to the conventional "
ETF model approximation methods. Especially, it is ⎢⎣ gˆ n1 ( s ) gˆ nn ( s ) ⎥⎦
more advantageous for higher dimensional processes
with complicated interaction modes than other existing where gˆ ij ( s ) is described as ETF.
approaches. The feasibility of proposed scheme is Note that the ETFs of each loop consist of a process
verified by several illustrated examples on purpose to dynamics term only and not include knowledge of other
demonstrate the effectiveness of the proposed design controller parameters under the assumption of a perfect
method. control approximation.
As seen from (3), the fundamental relationship
2. EQUIVALENT TRANSFER FUNCTION between ETF and G–T(s) is obtained. By considering the
steady and transient information of processes, the
Consider an open-loop stable multivariable system derivation of ETF is illustrated as follows based on (3).
with n inputs and n outputs as shown in Fig. 1, where ri, Firstly, the process transfer function models are
ei, ui and yi , i = 1, 2,..., n are the reference inputs, errors, expressed as first order plus time delay (FOPTD) models.
manipulated variables and system outputs, respectively;
ke −θ s
Gc ( s ) is the decentralized controller to be determined g ( s) = . (4)
and G ( s ) is the process transfer function matrix, both τ s +1
of which are expressed by From (3) and (4), the ETF for each loop is derived as
⎡ g11 ( s ) g12 ( s ) " g1n ( s ) ⎤ τˆi s + 1 ˆ adjGii
⎢ g ( s) g ( s) eθi s = , (5)
" g 2 n ( s ) ⎥⎥ kˆi G
G ( s ) = ⎢ 21 22
, (1)
⎢ # # % # ⎥
⎢ ⎥ where G is the determinant of G(s), and adjGii ( s ) is
⎣⎢ n1
g ( s ) g n 2 ( s) " g nn ( s ) ⎦⎥ the adjugate matrix corresponding to Gii ( s ).
Taking the derivative to both sides of (5), it yields
ˆ ˆ
τˆi eθi s + (τˆi s + 1) θˆi eθi s
kˆi
1 ⎛ n n ⎞
2 ∑⎜∑
=− ′ ⎟ adjG pi ,
⎜ (adjGiq ) g qp (6)
G p =1 ⎝ q =1 ⎟
⎠
′ is the first derivative of g qp .
where g qp
Differentiating both sides of (6) again, we have
ˆ ˆ
2τˆiθˆi eθi s + (τˆi s + 1) θˆi2 eθi s
kˆ i

1 ⎛ n n ⎞
Fig. 1. Multi-loop control system and equivalent inde- = − 2 ∑ ⎜ ∑ (adjGiq ) g qp
′′ ⎟ adjG pi
pendent SISO systems. G p =1 ⎜⎝ q =1 ⎟
⎠
348 Xiaoli Luan, Qiang Chen, and Fei Liu

2 ⎛ n⎛ n ⎛ n
n ⎞ ⎞ ⎞ When the interactions between loops are so significant
3 ∑ ⎜ ∑⎜ ∑ ⎜ ∑
+ ⎜ ⎜ ⎜ (adjGiq )gqp ′ ⎟ adjGmi ,
′ ⎟ adjGpl ⎟ glm that the ETF model can not be properly expressed by
G m=1⎝ l =1 ⎝ p=1⎝ q=1 ⎟ ⎟ ⎟
⎠ ⎠ ⎠ FOPTD, more general dynamics such as second order
(7) plus time delay (SOPTD) ETF model should be
′′ is the second derivative of g qi . considered.
where g qi
Letting s = 0, and solving (5), (6) and (7), the ˆ
kˆ(τˆa s + 1)e −θ s
FOPTD model parameters of ETF can be calculated as: gˆ( s) = . (12)
ˆ ˆs +1
τˆ 2 s + 2τζ
1
kˆi = , (8) Similar with the derivation of FOPTD model, SOPTD
aii ETF of each loop can be deduced as
2
⎛b ⎞ ⎛c ⎞ 1
τˆi = ⎜ ii ⎟ − ⎜ ii ⎟ , (9) kˆi = , (13)
⎝ aii ⎠ ⎝ aii ⎠ aii
2 bii
b ⎛b ⎞ ⎛c ⎞ b X 0 − τˆai = − , (14)
θˆi = ii − ⎜ ii ⎟ − ⎜ ii ⎟ = ii − τˆi , (10) aii
aii ⎝ aii ⎠ ⎝ aii ⎠ aii cii
X1 − X 0τˆai = , (15)
where 2aii
adjK ii dii
aii = , X 2 − X1τˆai = − , (16)
K 6aii
s =0
eii
1 n ⎛ n ⎞ X 3 − X 2τˆai = , (17)
bii = − 2 ∑ ⎜⎜ ∑ (adjKiq ) gqp′ ⎟⎟ adjK pi , 24aii
K p =1 ⎝ q =1 ⎠ s =0
where
n
2
cii = 3 ∑ dii = {[G −1 (0)]}′′′ , eii = {[G −1 (0)]}(4) , X 0 = θˆi + 2θˆi ζˆi ,
K m =1
1
⎛ n ⎛ n ⎛ n ⎞ ⎞ ⎞ X 1 = θˆi 2 + 2θˆiζˆi X 0 − τˆi ,
⎜ ∑ ⎜ ∑ ⎜ ∑ adjK iq g qp
⎜ l =1 ⎜ p =1 ⎜⎝ q =1
(
′ ⎟ adjK pl ⎟ glm
⎟ ⎟ ⎟
)
′ ⎟ adjK mi 2
1
⎝ ⎝ ⎠ ⎠ ⎠ X 2 = θˆi 3 + 2θˆi ζˆi X 1 − X 0τˆi ,
6
1 n ⎛ n ⎞
− 2 ∑ ⎜⎜ ∑ (adjKiq ) gqp′′ ⎟⎟ adjK pi , 1
X 3 = θˆi4 + 2θˆi ζˆi X 2 − X 1τˆi .
K p =1 ⎝ q =1 ⎠ 24
s =0
k11  k1n
Remark: As seen from the ETF derived above, both
K =    , kinds of ETF model, FOPTD and SOPTD, can be used to
kn1  knn represent the open-loop transfer function due to the
various interactions between each loop. Therefore, the
adjKiq , adjK pi , adjK mi and adjK pl are simplified as ETF parameterization algorithm developed in this paper
a single formula using subscripts v and w, defining: is more reasonable compared to the ETF of
MMF/ERGA/RNGA/EREA methods, which have the
adjK vw =
limitation of same structures as the corresponding
k1,1  k1,v −1 k1,v +1  k1, n original transfer functions. Furthermore, it can be seen
      from (8)-(17) that we obtain a general form for the
parameters of ETF, which makes the proposed ETF
k w−1,1  k w−1,v −1 kw−1,v +1  kw−1, n
(−1)v + w . suitable for higher dimensional processes.
kw+1,1  kw+1,v −1 ki +1,v +1  kw+1,n
      3. MULTI-LOOP PI CONTROLLER DESIGN
kn,1  kn,v −1 kn,v +1  k n, n
Since ETFs have incorporated the information of loop
In order for the resulting FOPTD ETF model to be interactions, the multivariable processes can be decom-
feasible, τˆi and θˆi should be real and positive. It is posed into a set of single-input single-output (SISO)
clear from (8)-(10) that the following condition should processes and then multi-loop PI controllers can be
be satisfied for finding feasible τˆi and θˆi values.
designed to stabilize these SISO loops independently.
Here, the IMC-PID [17] tuning rules were introduced for
2 its many advantages, including simplicity, robust
⎛ bii ⎞ cii
⎜ ⎟ > > 0. (11) performance, and analytical form.
⎝ aii ⎠ aii The obtained ETF model contains two parts.
 Equivalent Transfer Function based Multi-loop PI Control for High Dimensional Multivariable Systems 349

gˆ ii = p Ai pMi , (18) 1 ⎡ f ′′′(0) ⎤

τ Fi = ⎢ i ⎥. (29)
where pAi and pMi are the non-minimum portion with all- 3 ⎣ fi′′(0) ⎦
pass form and the minimum phase portion, respectively.
The desired closed-loop transfer function is 4. PERFORMANCE AND ROBUSTNESS
y p Ai ( s )
= , (19) To analyze the performance of control system,
r ( λi s + 1)mi different evaluation criteria are used as follows. The
optimum settings of controller for systems are necessary
where λi is an adjustable time constant, mi is an order, to minimize the integral error criterion and integral time-
both of which are chosen to make the controller weighted absolute error as small as possible. In addition,
realizable and proper. robustness of the control system should be considered for
The ideal controller that yields the desired closed-loop ensuring a fair comparison.
response is given as follows:
−1 4.1. Performance index
qi pMi ( s)
gci = = m
, (21) Integral absolute error (IAE)
(1 − gˆ ii qi ) (λi s + 1) i − p Ai ( s )
∞
IAE = ∫ e(t ) dt ,
−1
where qi is the IMC controller, qi = pMi /(λi s + 1)mi . 0
Although the above resulting controller is physically where e(t ) = r (t ) − y (t ).
realizable, it doesn’t have the standard PID form. In Integral time-weighted absolute error (ITAE)
frequency and time domain, approximation of the ideal
controller by Maclaurin series leads to the PID controller ∞
ITAE = ∫ t e(t ) dt.
more accurately than that of existing methods [17]. 0
Expanding gci in a Maclaurin series in s presents
4.2. Robustness index
f ( s) 1 ⎛ f ′′(0) fi′′′(0) ⎞
gci ≡ i = ⎜ fi (0) + fi′(0) s + i + , (22) A well-known method [12,18] is utilized to analyze
s s⎝ 2! 3! ⎟⎠ robust stability. Since the output uncertainty is often less
−1
restrictive than input uncertainty on the basis of control
s ⋅ pMi ( s) performance, the robust stability of multi-loop control
where fi ( s ) = .
(λi s + 1)mi − p Ai ( s ) systems is evaluated under output multiplicative uncer-
The first three terms of (22) can be interpreted as the tainty. When a system with an output uncertainty as
standard PI controller given by [ I + Δ o ( s )]G ( s ), the closed-loop system is stable if

⎛ 1 ⎞ γ = Δ o ( jω )
gci = kCi ⎜ 1 + ⎟. (23)
⎝ τ Ii s ⎠ < 1/ σ {[ I + G ( jω )GC ( jω )]−1 G ( jω )GC ( jω )},
Making the coefficients in (23) coincide with those in where γ represents the degree of robust stability, Δ o is
(22), the controller parameters are expresses by multiplicative output perturbation, and σ is maximum
kCi = fi′(0), (24) singular value. It should be noted that a control system
with a larger γ means more robust stability. Therefore,
f ′(0)
τ Ii = i . (25) for a fair comparison, γ value should be the same as or
f i (0) larger than that of the other methods in the simulation.
If the ETF has a strong lead term, the derivative and
integral constants calculated from (25) could have 5. EXAMPLES
5.1. Example 1
negative values. Under this circumstance, a PI controller
The distillation column of industrial scale polymeri-
with the first-order lag filter structure is recommended as
zation (ISP) was quoted to verify the benefits of the new
follows:
proposed ETF parameterization technique over the ETF
⎛ 1 ⎞ 1 in terms of ERGA [14] and RNGA [19].
gci = kCi ⎜ 1 + ⎟ . (26)
⎝ τ Ii s ⎠ τ Fi s + 1 ⎡ 22.89e −0.2 s −11.64e −0.4 s ⎤
⎢ ⎥
Analogously, the controller parameters are calculated 4.572 s + 1 1.807 s + 1 ⎥
G( s ) = ⎢
as: ⎢ 4.689e −0.2 s 5.8e −0.4 s ⎥
⎢ ⎥
⎡ f ′′′(0) ⎤ ⎣ 2.174 s + 1 1.801s + 1 ⎦
kCi = − fi (0) ⎢ i ⎥ + fi′(0), (27)
⎣ 3 fi′′(0) ⎦ In this example, due to the strong interactions between
f ′′′(0) fi′(0) loops, it is infeasible to use a simple FOPDT model to
τ Ii = − i + , (28) express ETF. So the SOPDT model with a lead term is
3 fi′′(0) fi (0) considered and calculated as follows:
350 Xiaoli Luan, Qiang Chen, and Fei Liu

1.5 0.6
Proposed
0
Model Error
0.4 ERGA
1 Vu et al.

0.2

u1
y1
−0.1 Proposed (g11)
ERGA 0.5 Proposed
ERGA 0
RNGA Vu et al.
−0.2 0 −0.2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
Time [hour] Time [hour]
Time
0.05 1.5 0.3
Proposed
Model Error

1 0.2 ERGA
0 Vu et al.

0.5 0.1

u2
y2
Proposed (g22)
−0.05 ERGA 0
Proposed
ERGA 0
RNGA Vu et al.

−0.1 −0.5 −0.1

0 10 20 30 40 0 10 20 30 40
0 10 20 30 40 Time [hour] Time [hour]
Time

Fig. 2. The ETF model errors of the proposed, ERGA Fig. 3. Outputs and control signals of the step response
and RNGA for ISP reactor. of the ISP reactor.

approach given by Vu et al. [20] are shown in the same

32.3(2.878s + 1)e −0.2 s figure. In tuning the controller parameters, γ of proposed
gˆ11 = ,
9.972s 2 + 6.753s + 1 method is adjusted to obtain γ = 0.59, in order to have a
8.18(2.876s + 1)e −0.4 s higher robustness level than those of ERGA [14] and Vu
gˆ 22 = . et al. [20]. The final control parameters, together with
3.925s 2 + 3.98s + 1
the performance indices, are summarized in Table 1. It is
To evaluate how closely the proposed ETF approxi- obvious from Fig. 3 that the proposed PI controller
mates the original process transfer function, the time provides a good performance with fast and well-balance
responses of proposed ETF, RNGA based ETF, ERGA responses in comparison with the other two methods.
based ETF and original model are drawn in Fig. 2. It can The effectiveness of the proposed PI controller is also
be seen from Fig. 2 that the proposed ETFs are closely confirmed by its smallest IAE and ITAE value in Table 1.
approximated to the original model, which essentially The robustness of these controllers is evaluated by
leads to satisfactory control performance of the multi- inserting a perturbation uncertainty of +50% in the
loop controller designed based on the proposed ETFs. process gain. As shown in Table 1, the proposed
Fig. 3 shows the closed-loop responses of several controller affords a good robust performance consistently.
different tuning methods for ISP model, where the set-
point changes are sequentially introduced into the 5.2. Example 2
individual loop. For comparison, the result of ERGA To show the proposed method can be easily extend to
based decentralized tuning approach given by Xiong et higher dimensional process without any complication,
al. [14] and direct multi-loop PI controller design the dynamics of the temperature control of four-room

Table 1. Controller parameters and performance index for ISP reactor.

IAEs ITAEs
Tuning method Loop K Ci τ Ii λi γ
Nominal ISP(+50%) Nominal ISP(+50%)
Proposed PI 1 0.4244 3.9750 0.19 0.59 3.5187 2.8892 4.8475 3.3311
2 0.1328 1.3040 1.00
ERGA 1 0.2190 4.5720 - 0.54 4.4316 3.5642 7.9667 5.4953
2 0.1730 1.8010 -
Vu et al. 1 0.4300 3.9500 0.09 0.57 3.6326 2.9611 5.2112 3.6181
2 0.1300 1.1300 0.69
IAEs: total sum of each loop’s IAE, ITAEs: total sum of each loop’s ITAE, ISP(+50%) represents the plant-model mismatch cases under +50% gain
uncertainty.

Table 2. Controller parameters and performance index for Example 2.

IAEs ITAEs
Tuning method Loop K Ci τ Ii λi γ
Nominal ISP(+50%) Nominal ISP(+50%)
Proposed PI 1 -31.8380 110.4447 24 0.83 260.4318 244.7808 14399.316 12669.860
2 -38.6410 119.1050 24
3 -31.7032 113.9600 24
4 -29.0807 125.0009 24
RNGA 1 -44.6855 113.8299 - 0.50 274.7409 298.5346 19693.059 23130.421
2 -53.8792 119.5064 -
3 -40.1251 113.9022 -
4 -36.6175 118.6099 -
MMF 1 -33.3610 122.0000 - 0.67 273.3006 250.9985 17824.393 14301.600
2 -36.1948 130.0000 -
3 -32.7582 118.0000 -
4 -29.6219 128.0000 -
 Equivalent Transfer Function based Multi-loop PI Control for High Dimensional Multivariable Systems 351

process was utilized [24]. 6. CONCLUSIONS

G(s) = In this paper, a new ETF parameterization algorithm to
⎡ −0.098e−17s −0.036e−27s −0.014e−32s −0.017e−30s ⎤ design multi-loop PI controller for high dimensional
⎢ ⎥ multivariable processes is presented. Differing from
⎢ 122s + 1 149s + 1 158s + 1 155s + 1 ⎥
conventional ETF model approximation methods, the
⎢ −0.043e−25s −0.092e−16s −0.011e−33s −0.012e−34s ⎥
⎢ ⎥ proposed ETF is derived directly by exploiting the
⎢ 147s + 1 130s + 1 156s + 1 157s + 1 ⎥ relationship between the equivalent closed-loop transfer
⎢ −0.012e−31s −0.016e−34s −0.102e−16s −0.033e−26s ⎥ function and the inverse of open-loop transfer function.
⎢ ⎥ Based on the obtained ETF, the multi-loop controller is
⎢ 153s + 1 151s + 1 118s +1 146s + 1 ⎥
designed utilizing the existing PI tuning rules and the
⎢ −0.013e−32s −0.015e−31s −0.029e−25s −0.108e−18s ⎥
⎢ ⎥ simple internal model control method. The objective of
⎢⎣ 156s + 1 159s + 1 144s + 1 128s + 1 ⎥⎦ the present work is to show that the new proposed ETF
based controller design scheme provides better or
Though this process has higher dimension, ETFs of each compatible time domain performance and frequency
loop can be easily exploited by FOPDT as follows: domain robustness compared to conventional ETF based
design approaches. In addition, the proposed method can
−0.0803 −19.2 s −0.0754 −16.88 s
gˆ11 = e , gˆ 22 = e , be extended to the integrator plus time-delay (IPTD)
106.178s + 1 115.62 s + 1 processes and the first order delay integrating (FODI)
−0.0919 −16.81s −0.0971 −18.18 s processes.
gˆ 33 = e , gˆ 44 = e .
110.475s + 1 121.083s + 1
REFERENCES
According to the obtained ETFs, multi-loop PI [1] Z. Y. Nie, Q. G. Wang, M. Wu, and Y. He, “Tun-
controller is calculated with its parameters in Table 2, ing of multi-loop PI controllers based on gain and
where γ values are chosen larger than that of MMF [13] phase margin specifications,” Journal of Process
(γ = 0.67) and RNGA [24] (γ = 0.50). Control, vol. 21, no. 9, pp. 1287-1295, 2011.
Fig. 4 displays the closed-loop time response of the [2] B. M. Chung, Y. K. Lim, and G. D. Buckner, “A
proposed method in comparison with RNGA and MMF. learning method for multivariable PID control syn-
In MMF method, the ETF is determined by combining thesis based on estimated plant Jacobian,” Interna-
the original loop transfer function with the approximated tional Journal of Control, Automation and Systems,
MMF. Literature [24] used RNGA to obtain the ETF, vol. 7, no. 5, pp. 737-744, 2009.
both methods utilized the SIMC-PID [25] tuning rule by [3] T. N. L. Vu, J. Lee, and M. Lee, “Design of multi-
setting λii equal to θii, thus their respective tuning values loop PID controllers based on the generalized IMC-
are employed without adjusting the γ value. The unit set- PID method with Mp criterion,” International
points change in r1 at t = 0, r2 at t = 1000, at t = 2000 and Journal of Control Automation and Systems, vol. 5,
r4 at t = 3000, respectively. It is clear from Fig. 4 that the no. 2, pp. 212-217, 2007.
proposed PI controller yields fast and well-balance [4] D. Chen and D. E. Seborg, “Design of decentra-
response. Table 2 also demonstrates the good robust lized PI control systems based on Nyquist stability
performance of the proposed method. analysis,” Journal of Process Control, vol. 13, no.
1, pp. 27-29, 2003.
1.5 20
[5] T. J. Monica, C. C. Yu, and W. L. Luyben, “Im-
0
1
−20 proved multiloop single-input/single-output (SISO)
y1

u1

Proposed Proposed
0.5 RNGA
MMF
−40 RNGA
MMF
controllers for multivariable processes,” Ind. Eng.
0 −60
0 1000 2000 3000 4000 0 1000 2000 3000 4000 Chem. Res., vol. 27, no. 6, pp. 969-973, 1988.
time [sec] time [sec]
1.5
20
[6] W. L. Luyben, “Simple method for tuning SISO
1 0 controllers in multivariable systems,” Ind. Eng.
0.5 −20
u2
y2

0
Proposed
RNGA
−40 Proposed
RNGA
Chem. Process Des. Dev., vol. 25, no. 3, pp. 654-
MMF −60 MMF
−0.5
0 1000 2000 3000 4000
−80
0 1000 2000 3000 4000
660, 1986
time [sec] time [sec] [7] M. Hovd and S. Skogestad, “Sequential design of
1.5 20
1 0
decentralized controllers,” Automatica, vol. 30, no.
0.5 10, pp. 1601-1607, 1994.
u3
y3

Proposed −20 Proposed

0 RNGA RNGA
MMF −40 MMF [8] S. J. Shiu and S. H. Hwang, “Sequential design
−0.5
0 1000 2000
time [sec]
3000 4000 0 1000 2000
time [sec]
3000 4000 method for multivariable decoupling and multiloop
1.5
Proposed
20 PID controllers,” Ind. Eng. Chem. Res., vol. 37, no.
1
0.5
RNGA
MMF
0 1, pp. 107-119, 1998.
u4
y4

Proposed
0
−20 RNGA
MMF
[9] M. Hovd and S. Skogestad, “Improved independent
−0.5
0 1000 2000 3000 4000
−40
0 1000 2000 3000 4000
design of robust decentralized controllers,” Journal
time [sec] time [sec]
of Process Control, vol. 3, no. 1, pp. 43-51, 1993.
Fig. 4. Outputs and control signals of the step response [10] S. Skogestad and M. Morari, “Robust performance
of the Example 2. of decentralized control systems by independent
352 Xiaoli Luan, Qiang Chen, and Fei Liu

designs,” Automatica, vol. 25, no. 1, pp. 119-125, duction and PID controller tuning,” Journal of
1989. Process Control, vol. 13, no. 4, pp. 291-309, 2003.
[11] H. P. Huang, J. C. Jeng, C. H. Chiang, and W. Pan,
“A direct method for multi-loop PI/PID controller
design,” Journal of Process Control, vol. 13, no. 8, Xiaoli Luan was born in 1979. She re-
pp. 769-786, 2003. ceived her [Link]. degree in Industrial
Automation and her Ph.D. degree in Con-
[12] T. N. L. Vu and M. Lee, “Independent design of trol Theory and Control Engineering
multi-loop PI/PID controllers for interacting multi- from Jiangnan University, Wuxi, China,
variable processes,” Journal of Process Control, in 2002 and 2010, respectively. She is
vol. 20, no. 8, pp. 922-933, 2010. currently an Assistant Professor of the
[13] M. J. He, W. J. Cai, B. F Wu, and M. He, “Simple Institute of Automation, Jiangnan Uni-
Decentralized PID controller design method based versity. Her research interests include
on dynamic relative interaction analysis,” Ind. Eng. robust control and optimization of complex nonlinear systems.
Chem. Res., vol. 44, no. 2, pp. 8334-8344, 2005.
[14] Q. Xiong and W. J. Cai, “Effective transfer func-
Qiang Chen was born in 1989. He is a
tion method for decentralized control system design doctoral student of the Institute of Auto-
of multi-input multi-output processes,” Journal of mation, Southern Yangtze University.
Process Control, vol. 16, no. 8, pp. 773-784, 2006. His research interests include robust con-
[15] W. Hu, W. J. Cai, and G. Xiao, “Decentralized con- trol and optimization of complex nonli-
trol system design for MIMO processes with inte- near systems.
grators/differentiators,” Ind. Eng. Chem. Res., vol.
49, no. 24, pp. 12521-2528, 2010.
[16] N. Naini, M. Fatehi, and A. Sedigh, “Input-Output
pairing using effective relative energy array,” Ind.
Eng. Chem. Res., vol. 48, no. 15, pp. 7137-7144, Fei Liu was born in 1965. He received
2009. his [Link]. degree in Electrical Technology
[17] Y. Lee, S. Park, M. Lee, and C. Brosilow, “PID and his [Link]. degree in Industrial
controller tuning for desired closed-loop responses Automation from Wuxi Institute of Light
for SI/SO systems,” AIChE Journal, vol. 44, no. 1, Industry, Wuxi, China, in 1987 and 1990,
pp. 106-115, 1998. respectively, and his Ph.D. degree in
[18] S. Skogestad and I. Poslethwaite, Multivariable Control Science and Control Engineering
Feedback Control, John Wiley and Sons, New from Zhejiang University, Hangzhou,
China, in 2002. From 1990 to 1999, he
York, 1996.
was with Wuxi Institute of Light Industry, first as an Assistant
[19] C. Rajapandiyan and M. Chidambaram, “Controller Professor, next as a Lecturer, and then as an Associate
design for MIMO processes based on simple de- Professor. Since 2003, he has been a Professor with the
coupled equivalent transfer functions and simpli- Institute of Automation, Jiangnan University, Wuxi. In 2006,
fied decoupler,” Ind. Eng. Chem. Res., vol. 51, no. he was a Visiting Professor with The University of Manchester,
38, pp. 12398-12410, 2012. Manchester, U.K. His research interests include advanced
[20] T. N. L. Vu and M. Lee, “Multi-loop PI controller control theory and applications, batch process control
design based on the direct synthesis for interacting engineering, statistical monitoring and diagnosis in industrial
multi-time delay processes,” ISA Transactions, vol. process, and intelligent technique with emphasis on fuzzy and
neural systems.
49, no. 1, pp. 79-86, 2010.
[21] R. K. Wood and M. W. Berry, “Terminal composi-
tion control of binary distillation column,” Chemi-
cal Engineering Science, vol. 28, no. 9, pp. 1707-
17, 1973.
[22] Y. Shen, W. J. Cai, and S. Li, “Multivariable
process control: decentralized, decoupling or
sparse?” Ind. Eng. Chem. Res., vol. 49, no. 2, pp.
761-771, 2010.
[23] M. Lee, K. Lee, C. Kim, and J. Lee, “Analytical
design of multi-loop PID controllers for desired
closed-loop responses,” AIChE Journal, vol. 50, no.
7, pp. 1631-1635, 2004.
[24] Y. Shen, W. J. Cai, and M. J. He, “Normalized de-
coupling control for high-dimensional MIMO
processes for application in room temperature con-
trol HVAC systems,” Control Engineering Practice,
vol. 18, no. 6, pp. 652-694, 2010.
[25] S. Skogestad, “Simple analytic rules for model re-