You are on page 1of 21

89

CHAPTER 5

DEVELOPMENT OF GENERIC MODEL CONTROL


ALGORITHM FOR NONLINEAR
INTERACTING SYSTEMS

5.1 INTRODUCTION

Generic Model Control (GMC) was introduced by Lee and Sullivan


in 1988. GMC allows nonlinear process models to be directly embedded
within the control structure, providing a possibility of a better process control
because an accurate process model is employed (Wang et al 2003). In order to
calculate a control law that induces linear input/ output behavior in the MIMO
system GMC algorithm is used. The control law allows controlling process
control systems without imposing any structural constraints on the closed-
loop dynamics of the system. The main objective of GMC is to guide a
system from its initial condition to a desired setpoint by manipulating its input
so that the system follows the behavior of a pre defined reference model. This
algorithm is also used to eliminate the interaction of the nonlinear MIMO
process.

5.2 DEVELOPMENT OF GMC ALGORITHM

As industrial processes have become more integrated and more


flexible, the demands on their efficiency have increased, creating more
complex operational and control problems. In the last decades, model-based
controllers have been the most feasible alternative to tackle more complex
90

industrial control problems, where the traditional PID controllers in many


cases are not able to give satisfactory closed-loop performance (James and
Peter 1997). Linear model-based control techniques have demonstrated a
superior performance over the traditional techniques for the control of a wide
variety of chemical processes.

The output of the MIMO process can either be interacting or


non-interacting. If one output is affected by only one input, then it is called
non-interacting, otherwise it is called as an interacting system. In MIMO
system, the input variable affects more than one output variable which causes
interaction between the input/output loops. So, the control of multivariable
systems is difficult compared to the SISO system. Therefore, the degree of
interaction plays an important role to quantify the proper input/output pairings
that minimize the impact of the interaction. The interaction between inputs
and outputs can be eliminated by using decouplers, cross controllers or many
other methods (Marshall Duvall et al 2001).

GMC uses a model of the process in formulating the control law


(James Riggs and Peter Lee 1997). The design framework is similar to other
model based approaches such as Dahlins algorithm and IMC. However,
rather than adopting a classical approach of comparing the trajectory of the
process output against a desired trajectory, GMC defines the performance
objective in terms of the time derivatives of the process output, i.e minimizing
the difference between the desired derivative of the process output and the
actual derivative. GMC is a multivariable controller with interactions between
variables taken into account in the calculation of the values of the
manipulated variables (Marshal et al 2001). Nonlinearity and interactions are
two of the most important influences that make control a difficult task.
Figure 5.1 shows Multivariable Generic Model Structure (Suleyman Karacan
et al 2007).
91

The objective of the controller is to keep the reactor conversion and


the reactor temperature at their desired setting by manipulating the
constrained inputs. Successful application of GMC formulation requires the
nonlinear system with a relative degree of one. In the present research, GMC
is used to produce the manipulated input for controller so that a linear
response from the system in the sense of input/output is obtained. In the GMC
formulation, the desired response trajectory is not limited to a linear system
response alone but also to nonlinear responses. GMC formulation yields
outputs u1 and u2, which form the inputs to the controllers u1 and u2 which
force the system to produce a linear input/output behavior.

GMC is a multivariable controller with interactions between


variables taken into account in the calculation of the values of the
manipulated variables (Marshal et al 2001). Nonlinearity and interactions are
two of the most important influences that make control a difficult task.
Figure 5.1 shows Multivariable Generic Model Structure (Suleyman et al
2007).

y
y* GMC u Process

Multivariable
System
Identification
Parameters

Figure 5.1 Multivariable generic model control structure


92

Here current output is y, setpoint is y* and u is the control law.

Control input u(t) is as given below (Erik Weyer et al 1999)

t
u(t) k1(t) ysp (t) y (t) k 2 (t) (ysp ) y ) (5.1)
ref 0 ref

where k1and k2are positive definite diagonal matrices and Ysp is the set
point. Equation (5.1) resembles a PI feedback controller. The first term of
Equation (5.1) ensures the output to move towards the setpoint and the
integral term makes the system capable of tracking a setpoint (Scott Flathouse
and James Riggs 1996). Assume that k1 and k2 are constant with respect to
time. Equation (5.1) for the SISO case can be solved as follows

A standard nonlinear system representation is shown (Dunia and


Edgar 1996)

x f(x) g(x)u (5.2)

y h(x)

where x is the state vector of dimension n, u is an input vector of


dimension m, y is an output vector of dimension p and f(x) is an (n,1)
vector with each row fi(x) is a smooth function, h(x) is a (p,1) vector with a
row element hj(x) also a smooth function and g(x) is an (n,m) matrix with
element of each column gj(x).

A Lie derivative of h with respect to f is written as Lfh(x). It is


defined as (Wang et al 2003)

h(x)
L h(x) f(x) (5.3)
f x
93

h (x)
j
Note that is a (1,n) row vector and f(n,1) column vector.
x

The lie derivative of Lf h(x) with respect to f(x) is

L L h(x) L 2 h(x) ,
f f f

The repeated lie derivative notation is

L k h(x) L L k 1h(x) (5.4)


f f f

L 0 h(x) h(x) (5.5)


f

Considering the nonlinear system Equation (5.2) by taking the time


derivative of the each output component, yi = hi(x), this study denotes the
smallest order of each output derivative that explicitly depends on the input
u as the relative degree i. Initial behavior for step change is really dictated
by the relative order of the system. The relative order is the difference
between the orders of numerator and denominator polynomials in the transfer
function.

h h dx
y1 i i L h (x) (5.6)
t x dt f i

y L 2h (x) (5.7)
1 f i

y i 1 L i 1h (x) (5.8)
1 f i

y i L i h (x) L g L i 1h (x)u (5.9)


i f i f i
94

where Lg L i 1h (x) denotes that the (1,m) row vector with the jth
f i

component L L i 1h (x) . The purpose of the differentiation is to obtain an


gj f i
explicit expression for the control input u.

GMC technique has many advantages (Massimiliano Veronesi and


Antonio Visioli 2010). For example, the controller structure is very simple,
controller parameters can be easily tuned, measurable disturbances can be
compensated effectively and control strategy has strong robust stability.

Interesting features of GMC are (Yudi Samyudia and Peter Lee


2002):

(1) This controller consists of not only feedback but also feed-
forward structures in its implementation.

(2) Its simplicity in treating the nonlinearities of the process.

(3) The control law may be solved either explicitly or via


iteration, depending on the nature of the process model
(Yudi Samyudia and Peter Lee 2004). This technique proves
to be attractive for industrial process applications.

5.3 IMPLEMENTATION OF GMC ALGORITHM IN LIQUID


LEVEL-TEMPERATURE CASCADED PROCESS

GMC algorithm steps are implemented for a liquid Level-


Temperature cascaded process.

The model is explained in chapter 2. The model equations for the


process are
95

k 1 1
x x1 2 u (5.10)
1 S S 1

To x 1
x 2 u u (5.11)
2 Sx1 1 Cp Sx1 2

x1 is liquid level of the tank and x2 is liquid temperature in the tank.


The output equations are y1 = x1 and y2 = x2.

The parameters are as explained in chapter 2. According to GMC


formulation

y1 k 0 y y k 0 y y
11 1sp 1 21 1sp 1
dt (5.12)
0 k12 y 2sp y 2 0 k 22 y 2sp y 2
y
2

The value of k11 and k12 and k21 and k22 can be selected by using
tuning method to achieve a desired process response. The standard GMC
design approach is limited to a nonlinear process with relative degree one. For
process with relative degree Rd>1, the selection of GMC parameters becomes
more complicated. Further, the GMC design method is a special case of
feedback linearizing controller. y1sp and y2sp are the setpoints of the liquid
level of the tank and liquid temperature in the tank.

y k y y 0 k y y
1 11 1sp 1 21 1sp 1
(5.13)
y2 0 k y y k y y
12 2sp 2 22 2sp 2

y1 k11 y1sp y1 k 21 y1sp y1 x1 (5.14)


96

y2 k12 y 2sp y12 k 22 y12p y 2 x2 (5.15)

Substituting Equation (5.10) in Equation (5.14) gives

k 1 1
k11 y1sp y1 k 21 y1sp y1 x1 2 u (5.16)
S S 1

Substituting Equation (5.11) in Equation (5.15) gives

k12 y2sp y2 k 22 y2sp y 2

To x 1
2 u u (5.17)
Sx1 1 C p Sx1 2

Equation (5.16) gives

k 1 1
k11 y1sp y1 k 21 y1sp y1 x1 2 u (5.18)
S S 1

From the above equation, u1 is

k 1
u1 S k11 y1sp y1 k 21 y1sp y1 x1 2 (5.19)
S

From the Equation (5.17)

y k y y k y y x (5.20)
2 12 2sp 12 22 12p 2 2

Substitute the value of u1 and rearrange the Equation (5.17) gives as


97

y2 k12 y 2sp y12 k 22 y12p y 2

To x
2 Sv kx 12 1
u (5.21)
Sx1 1 1 Cp Sx1 2

There fore

To x
k v 2 Sv kx 12 1
u (5.22)
12 2 Sx1 1 1 Cp Sx1 2

The control input u2 is

To x 2 1
C p Sx k v Sv kx 2 u (5.23)
1 12 2 Sx 1 1 2
1

Control inputs u1 and u2 are

1
u SV kx 2 (5.24)
1 1 1

T x T x
u2 C p Sx 1 k12 V2 0 2 V 0 2 Kx 12 } (5.25)
x1 1 Sx 1 1

Substituting u1 and u2 in the state model, results in state equation in


decoupled form and linearized form. Then, the system is decoupled and
interaction eliminated and is given as

x v (5.26)
1 1
98

x v *k (5.27)
2 2 12

where v1 and v2 are new control inputs. x1 is liquid level x2 is liquid


temperature. 10 = 20 is constant coefficients.

5.4 CONTROLLER DESIGN OF LEVEL-TEMPERATURE


CASCADED PROCESS

Decentralized PI controller is one of the most widely used control


schemes for interacting MIMO plants in the chemical and process industries.
The main reason is its relatively simple structure, which is easy to understand
and implement (Halevi et al 1997).

Control of the interacting multivariable processes can be realized


either by centralized MIMO controllers or by a set of SISO local controllers
(Truong et al 2007). For large-scale industrial processes, decentralized control
is preferred from the viewpoints of implementation, requiring fewer
parameters to tune and loop failure tolerance of the resulting control system.
A common design procedure in decentralized control is the sequential design
strategy. Sequential design strategy involves closing and tuning one loop at a
time. PI, Setpoint Weighting PI (SPW-PI) and Fuzzy Logic Controller (FLC)
are designed and implemented.

Using Zeigler- Nichols method and further fine tuning, the values
for PI controllers are obtained.

For Level-Temperature cascaded process, PI controller is designed.


It has been determined that the PI parameters are as Kp = 1.2, i = 50 for
getting the desired response.
99

For Level-Temperature cascaded process, PI-SPW controller is


designed. It has been found that the parameter values are Kp = 1.2, i = 50,
b = 0.9 for getting desired response.

Level-Temperature process FLC controller is designed. Fuzzy rules


are framed and 7 linguistic variables are chosen for error e value for level
parameter and 7 linguistic variables are chosen for change in error ce values
for level parameter. Totally 49 rules are farmed for level parameter. Similarly
fuzzy rules are framed and 7 linguistic variables are chosen for error e value
for temperature parameter and 7 linguistic variables are chosen for change in
error ce values for temperature parameter. Totally 49 rules are farmed for
the temperature parameter

5.5 IMPLEMENTATION OF GMC ALGORITHM IN LIQUID


LEVEL-pH CASCADED PROCESS

The topic of pH control has attracted literature much. The problem


of controlling the pH, in a batch fed reactor where precipitation occurs, is
considered (Barraud et al 2009). GMC algorithm steps are implemented in a
liquid Level-pH cascaded process. Level-pH process model is explained in
chapter 2. The model equations for the process are

k 1 1
x x 2 u u (5.28)
1 S 1 S 1 2

1
x b CAO u b CBO u (5.29)
2 Sx log (a) 1 2
1 10

x1 is liquid level of a tank and x2 is liquid pH of a tank. The output


equations are y1 = x1 and y2 = x2. System Parameters are as explained in
chapter 2.
100

Here x1 and x2, S, k, 10, 20 possess the same values as mentioned


in chapter 2. The feed concentrations are chosen as CAO=CBO=0.03 molcm-3.
y1sp and y2sp are the setpoints of the liquid level and pH parameter.

y1 k11 y1sp y1 k 21 y1sp y1 x1 (5.30)

y2 k12 y 2sp y12 k 22 y 2sp y 2 x2 (5.31)

Substituting Equation (5.28) in Equation (5.30)

k11 y1sp y1 k 21 y1sp y1

k 1 1
x1 2 u u (5.32)
S S 1 2

Substituting Equation (5.29) in Equation (5.31)

k y y k y y
21 2sp 2 22 2sp 2

1
b CAO u b CBO u (5.33)
Sx log (a) 1 2
1 10

From the Equation (5.32)

k 1 1 1
v1 x1 2 u1 u (5.34)
S S S 2

Where v1 is control input 1 and v2 is control input 2.


101

Similarly from Equation (5.33)

1
v2 b CAO u1 b CBO u 2 (5.35)
Sx log (a)
1 10

From the Equation (5.34)

k 1 1
u1 S v1 x1 2 u (5.36)
S S 2

Substituting Equation (5.36) in Equation (5.35)

1
v2
Sx1log10 (a)

k 1 1
b CAO S v1 x1 2 u b CBO u 2 (5.37)
S S 2

By rearranging the above equation, so as to get u2

Sx log (a)v
1 10 2

1
b CAO Sv kx 2 u b CBO u (5.38)
1 1 2 2
1 1
Sx log (a)v b Sv kx 2 CAO Sv kx 2
1 10 2 1 1 1 1
u (5.39)
2 CAO CBO

Considering Equation (5.34)

k 1 1 1
v x 2 u u (5.40)
1 S 1 S 1 S 2
102

Substituting Equation (5.39) in the above equation

k 1 1
v x 2 u
1 S 1 S 1

1 1
Sx log (a)v b Sv kx 2 CAO Sv kx 2
1 1 10 2 1 1 1 1
(5.41)
S CAO CBO

Rearranging the above equation, gives

k 1
v x 2
1 S 1

1 1
Sx log (a)v b Sv kx 2 CAO Sv kx 2
1 1 10 2 1 1 1 1
S CAO CBO

1
u (5.42)
S 1

1
u Sv kx 2
1 1 1

1 1
Sx log (a)v b Sv kx 2 CAO Sv kx 2
1 10 2 1 1 1 1
(5.43)
CAO CBO

Control inputs u1 and u2 are


103

1
u Sv kx 2
1 1 1

1 1
Sx log (a)v b Sv kx 2 CAO Sv kx 2
1 10 2 1 1 1 1
(5.44)
CAO CBO

1 1
Sx log (a)v b Sv kx 2 CAO Sv kx 2
1 10 2 1 1 1 1
u (5.45)
2 CAO CBO

Substituting u1 and u2 in the state model, results in state equation


being in decoupled and linearized form.

k 1
x1 x1 2
S

1
Sv kx 2 (2CAO bCAO CBO)
1 1 1
(5.46)
S CAO CBO

1
x
2 CAO CBO

1
b(Sv1 kx1 2 ) (2CAO CBO bCAO - b) -
1
CAO(Sx1 kx1 2 )[CAO CBO b bCAO] (5.47)
1
CBO b Sx log (a)v (Sv kx 2 ) (CAO b)
1 10 2 1 1

But in this case, x 2 depends on both u1 and u2. So interaction exists


even after implementing GMC algorithm in the process. System could not be
fully decoupled.
104

5.6 CONTROLLER DESIGN OF LEVEL-pH CASCADED


PROCESS

For Level-pH cascaded process, PI controller is designed. It has


been found that the PI parameters are selected as Kp = 0.05, i = 51 for getting
desired response. Using Zeigler- Nichols and further finite tuning values, we
the values for PI controllers are obtained.

For Level-pH cascaded process, PI-SPW controller is designed It


has been found that the parameter values Kp = 0.05, i = 51, b = 0.9 for getting
desired response.

In Level-pH cascaded process, FLC is designed. Fuzzy rules are


framed and 8 linguistic variables are chosen for error e value for level
parameter and 8 linguistic variables are chosen for change in error ce values
for level parameter. Totally, 64 rules are framed for level parameter.
Similarly, fuzzy rules are framed and 8 linguistic variables are chosen for
error e value for pH parameter and 8 linguistic variables are chosen for
change in error ce values for pH parameter. Totally, 64 rules are framed for
pH parameter.

5.7 SIMULATION RESULTS AND DISCUSSION

The following simulation results prove that the desired objectives


of achieving decoupling and linearization were accomplished. This is to
illustrate the sudden change in level input does not affect the temperature.
Simulation results of the Level-Temperature process system using GMC
algorithm with external PI controller is shown in Figure 5.2. It shows the
output response of the level and temperature when the setpoint of the level is
raised from 1 to 60 cm and temperature is raised from 1 to 30C. The sudden
disturbance introduced at time of 200 secs in level does not affect the
105

temperature of the process. It can be seen from the figure that under the
influence of the controller, ISE is improved.

Figure 5.2 Output responses for the Level-Temperature process with


PI controllers

Next GMC algorithm with external PI-Setpoint Weighting


controller is tested and the results are shown in Figure 5.3. The sudden
disturbance introduced at time of 200 secs in level does not affect the
temperature process. It can be seen from the figure that the ISE is improved
than Figure 5.2. In PI-SPW controller settling time of level and temperature
improved both at 50 secs than both at 80 secs achieved by PI controller.

Figure 5.3 Output responses for the Level-Temperature process with


PI-SPW controllers
106

Next GMC algorithm with Fuzzy Logic Controller is tested and the
results are as shown in Figure 5.4. The sudden disturbance introduced at time
of 200 secs in level does not affect the temperature of the process. It can be
seen from the figure that under the influence of the controller, performance
has improved when compared to Figure 5.3. In FLC, controller settling time
of level improved at 40 secs than at 50 secs achieved by PI-SPW controller.

Figure 5.4 Output responses for the Level-Temperature process with


FLC controllers

For Level-pH process, GMC algorithm with external PI controller


is applied as shown in Figure 5.5. It shows the output response of the level
and pH when the setpoint of the level is raised from 1 to 30 cm and pH is
raised from 1 to 4. The sudden disturbance introduced at time of 100 secs in
level affects the pH process.
107

Figure 5.5 Output responses for the Level-pH process with PI controllers

Next GMC algorithm with external PI setpoint weighting controller


is applied as shown in Figure 5.6. The sudden disturbance introduced at time
of 100 secs in level also affects the pH process.

Figure 5.6 Output responses for the Level-pH process with PI-SPW
controllers

Next decoupling and linearization along with external Fuzzy logic


controller simulation is applied and the results are shown in Figure 5.7. The
sudden disturbance introduced at time of 100 secs in level affects the pH
process.
108

Figure 5.7 Output responses for the level-pH process with FLC controllers

Table 5.1 shows the performance of Level-Temperature cascaded


process and Level-pH cascaded process with controllers.

Table 5.1 Comparison of controller performance in GMC algorithm

Controller Process Ts ISE


PI Level 80 3.077e+004
Temp 80 1.988e+003
PI-SPW Level 50 2.03e+004
Temp 50 2.433e+003
FLC Level 40 1.888e+003
Temp 100 3.977e+004
PI Level 80 Interaction exists
pH 80 Interaction exists
PI-SPW Level 50 Interaction exists
pH 50 Interaction exists
FLC Level 30 Interaction exists
pH 30 Interaction exists
109

It is inferred that FLC controller gives less Settling Time (Ts) and
Integral Square Error (ISE) for the Level-Temperature cascaded process to
reach the desired setpoint. But in Level-pH cascaded process, interaction
exists.

5.8 SUMMARY

In this chapter, the detailed explanation of GMC algorithm is


presented. The control strategies are used to maintain the liquid level of the
tank and temperature of the liquid in a Level-Temperature cascaded process
and liquid level of the tank and pH of the liquid in a Level-pH cascaded
process. The coding is done in MATLAB environment. The simulations are
carried out for set point changes for the chosen models and the simulation
results are obtained. From the simulation results, it is observed that the Fuzzy
Logic Controller performs relatively better for the performances indices of
Settling Time (Ts) and Integral Square Error (ISE).

You might also like