Professional Documents
Culture Documents
Jobrun Nandong1, 2
1
Department of Chemical Engineering; 2 Curtin Sarawak Research Institute
Curtin University Sarawak
98009 Miri, Sarawak, MALAYSIA
jobrun.n@curtin.edu.my
Abstract—A new method of centralized PI/PID control system simultaneous equation solving or optimization, and (5)
synthesis for TITO processes using the non-conventional Inter- independent methods. The detuning method based on the BLT
Communicative Decentralized Multi-Scale Control (ICD-MSC) [7] is probably among the most commonly used in process
scheme is presented. Explicit mathematical tuning equations are industry because of its simplicity. But the BLT method often
constructed which are used to systematically convert the inter- leads to under satisfactory performance, i.e., can be either too
communicative multi-scale controllers into an industrial sluggish or too oscillatory. Another popular design approach is
implementable form of centralized PID controllers. An based on the sequential tuning where the PID controllers are
illustrative example shows the effectiveness of the proposed designed sequentially one after another [8]. The main issue in
synthesis method which is easy to use and understand.
this approach is that one has to select the right order of the loop
Keywords—multivariable PID control; decentralized PID
closing sequence as the control performance is strongly
control; MIMO control design; multi-scale control influenced by this order. In addition to the sequential and
detuning approaches, several researchers have developed PID
tuning methods based on the independent approach, e.g., see
I. INTRODUCTION [9], 10].
Over the last few decades, various forms of advanced
In process control, one common approach to improve the
control techniques have been developed, such as the linear-
performance of decentralized PID control is to augment the
quadratic control [1], generalized model predictive control [2]
control system with decouplers, which are used to mitigate the
and model predictive control (MPC) [3]. In spite of all the
adverse effect of process interactions. There are several forms
advances in control theories, the PID control is still the most
of decoupling methods, where the basic techniques can be
widely used controller in industries [4]. In process industry,
divided into 3 categories [11]: (a) the ideal, (b) simplified, and
most of the systems of interest are multi-input and multi-output
(c) inverted decoupling methods. One of the major problems in
(MIMO) in nature where multi-loop (decentralized) PID
using the decoupling techniques is that the decouplers might
controllers are often adopted to control such systems. The main
not be physically realizable, especially those that are based on
challenge for designing effective decentralized PID controllers
the ideal decoupling method. There are several methods of
for a MIMO process arises from the presence of process
decentralized PID control design which include the decoupling
interactions, which fairly often impose a limitation on control
controllers, e.g., see [12], [13], [14].
performance. It has been recognized that even for single-input
and single-output (SISO) processes, the tuning of a PID Note that, Nandong and Zang [15] have recently proposed
controller is often difficult without a systematic procedure [5]. the Inter-Communicative Decentralized Multi-scale Control
The PID controller tuning for a MIMO process is even more (ICD-MSC) scheme as an alternative to traditional decoupling
challenging because the number of tuning parameters increases approach for overcoming the performance limitation imposed
with the system dimension, i.e., an n × n MIMO system will by the process interactions. The ICD-MSC is basically founded
leads to 2n, 3n and 4n numbers of tuning parameters for on the more basic Multi-scale Control (MSC) scheme for SISO
multi-loop PI, PID and PID with filter controllers, respectively. nonminimum-phase processes; see [16], [17]. Additionally, the
In the fully centralized PID control system, there are 3n 2 same basic MSC scheme has recently been adopted in the
number of controller parameters that need to be tuned. Hence, construction of several PID tuning relations; see [18], [19].
it would be much more difficult to design the multivariable In the present work, the ICD-MSC scheme [15] and the
PID control system than the decentralized one. procedure for constructing some PID tuning relations in [18]
The existing methods for decentralized PID controller are combined together in order to synthesize a multivariable
tuning can be broadly categorized as in [6]: (1) the detuning, PI/PID control system for the two-input and two-output (TITO)
(2) sequential loop closing, (3) iterative or trial-and-error, (4) processes. It is worth noting that the ICD-MSC scheme in [15]
R1 -
II. CONSTRUCTION OF TUNING RELATIONS Fr1 K11,0
C1
h11
+ U1
g11
+ E1 + Y1
The details about the ICD-MSC scheme can be found in + +
[15] while the construction procedure of PID controller tuning h12 g12
formulas can be found in [18]. The basic idea of the ICD-MSC
scheme is to reduce the effect of MIMO process interactions h21 g21
via the inner-layer of a given multi-scale controller. The multi-
scale controller has two loops – a fast inner feedback loop and R2 E2 +
+
+
Y2
a relatively slow outermost feedback loop. The effect of the Fr1
+
K22,0
C2
h22
+ U2
g11
process interactions is treated as an input disturbance, which is -
to be removed by the fast feedback inner loop. In contrast to
this idea, in the conventional decoupling techniques, the loop Fig. 2. Simplified block diagram of the Input-Input Structure of the
interactions are mitigated via the conventional feedforward ICD-MSC scheme.
cancellation approach. In the ICD-MSC, the loop interactions
are removed or reduced by the fast feedback control action of
-
the given multi-scale controller – a feedback approach. R1
Fr1 Gc11
+ U1
g11
+ E1 + Y1
Fig. 1 shows the block diagram of an Input-Input structure + +
of ICD-MSC scheme. Other ICD-MSC structures are called the g12
Gc21
Input-Output and Output-Output structures [15]. In Fig. 1, gii
denotes the diagonal transfer function, gij the off-diagonal Gc12 g21
transfer function, Wii the multi-scale predictor, I ij the inter- + +
R2 E2 + Y2
communicative controller, K ii ,1 the inner sub- controller, K ii ,0 Fr1
+
Gc22
+ U2
g11
-
the outermost sub-controller and Fri the setpoint pre-filter; Ri
and Yi denote the setpoint and controlled signals, respectively. Fig. 3. Equivalent multivariable control block diagram of the Input-
Input Structure of the ICD-MSC scheme.
In this work, the Input-Input structure of ICD-MSC scheme
(Fig. 1) is adopted to synthesize a centralized PI/PID control
system for TITO processes.
g ii ( s ) = K pii exp( −θ ii s ) (τ ii s + 1) (2)
Let us consider a TITO process given as follows
First, the dead-time in (2) is approximated using the 1/1
ª g11 ( s ) g12 ( s ) º Padé formula. Then, the approximated model is decomposed
P( s) = « » (1) into a sum of an outermost mode ( mii ,0 ) and an inner-layer
¬« g 21 ( s ) g 22 ( s )¼» mode ( mii ,1 ) as given by
mii ,0 mii ,1
where α ii = 0.5θii and it is assumed that α ii < τ ii . In (3), the β ij (τ cii,1s + 1)(τ cjj,1 s + 1)
I ij ( s ) = (14)
gains for the outermost and innermost modes are as (α ii s + 1)(α jj s + 1)
k ii ,0 = K pii (τ ii + 0.5θ ii ) ( τ ii − 0.5θ ii ) (4)
where βij denotes the IC controller gain.
After substituting (13) into (12), and then into (11)
k ii ,1 = K pii θ ii (0.5θ ii − τ ii ) (5)
hij ( s ) = β ijo k cjj ,1 (α ii s + 1) (τ cii ,1 s + 1)
o
(15)
Note that, Fig. 1 can be simplified to Fig. 2. Then, based on
Fig. 2 the inner-loop transfer function corresponding to each
Meanwhile, if the IC controller form in (14) is used, a static
diagonal element of (1) is written as in (6):
form of (11) is obtained as follows
hii ( s ) = k o
(α ii s + 1) (τ cii,1s + 1) (7) ȕijo = ȕij (1 + k cii ,1k ii ,1 ) (17)
cii ,1
In (7), the overall sub-controller gain and the closed-loop Now, let us define an MSC parameter (see reference [18]),
time constant are respectively given by (8) and (9): which is a ratio of the open-loop time constant to that of the
closed-loop of the inner mode given as follows
,1 = k cii ,1 (1 + k cii ,1 k ii ,1 )
o
k cii (8) λii,1 = α ii τ cii ,1 = 1 + k cii ,1k ii,1 (18)
The overall diagonal controller can now be written as follows τ f ij = τ cii ,1 = θ ii (2λii ,1 ) (36)
Gcii ( s ) = K ii,0 ( s )hii ( s) S (k cii,1 ) (26) If the IC controller is chosen to be in the form of (14), then
one obtains the following off-diagonal controller:
where S (k cii ,1 ) denotes the sign of the inner sub-controller,
included to obtain the correct overall controller gain. It has Gcij ( s ) = K jj,0 ( s )h ij (37)
been shown in [19], [20] that the multi-scale controller (26)
can be converted into an equivalent PID controller in (27): Since the outermost sub-controller is in the form of a PI
controller and the off-diagonal inner-layer transfer function
§ ·§ · takes a static form, the overall off-diagonal controller is simply
1 1
Gc ii ( s) = K C ii ¨¨1 + + τ D ii s ¸¸ ¨¨ ¸
¸ (27) equivalent to a PI controller overall, i.e.:
© τ I ii s ¹ © τ f ii s + 1 ¹
Gcij ( s ) = K C ij (1 + 1 (τ I ij s )) (38)
In (27), the PID controller parameters are given in terms of
the model parameters and the MSC tuning parameters:
The PI controller parameters in (38) are given as follows:
ª § ( 2τ ii − θ ii ) 2 · º
« ¨ ¸ » § β ij (λ jj ,0 − 1)(λ jj,1 − 1) ·§ ( 2τ jj − θ jj ) 2 ·
§ (λii ,0 − 1)(λii ,1 − 1) · « ¨© K pii K pii ¸¹ » K Cij = ¨¨ ¸¸¨¨ ¸¸ (39)
K Cii = ¨¨ ¸« (28) © 2λii,1λ jj ,1 ( 2τ jj + θ jj )θ jj ¹© K pjj K pjj ¹
© 4 γ ii λii ,1 ¸ § (2τ + θ )τ θ
¹« ¨ ·»
ii ii ii ii
¸»
« © 2γ iτ ii +θ ii ¹ »¼
¬
τ I ij = γ jj τ jj (40)
Yi ( s ) = [h jj ( s ) g ij ( s ) + hij ( s ) g ii ( s )]C j ( s ) (44) Remark 2: The proposed design procedure is quite general
(it can be further refined in future study). A more rigorous
tuning procedure using the constructed tuning relations can be
Let the overall interaction transfer function be in (45) performed via optimization techniques, e.g., genetic algorithm.
Remark 3: We propose robustness 14 dB ≤ GM * ≤ 16 dB
H ij ( s ) = h jj ( s ) g ij ( s ) + hij ( s ) g ii ( s ) (45)
for gain margin range and 60 0 ≤ PM * ≤ 80 0 for phase margin
range. It is crucial to ensure sufficient robustness margins to
A simple choice is to set (45) at the steady-state as follows accommodate the effect of process uncertainties.
(47).
[1] A. Bemporad, M. Morari, V. Dua and E. N. Pistikopoulos, "The explicit
2 linear quadratic regulator for constrained systems," Automatica, vol. 38,
pp. 3-20, 2002.
1
[2] D. W. Clarke, C. Mohtadi and P. S. Tuffs, "Generalized predictive
control - part 1. The basic algorithm," Automatica, vol. 23, pp. 137-148,
1987.
Y1