F i. R. r. H. h. F. Example: Linearization of a nonlinear model involving a nonlinear function of two variables.
Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.
Fi
R
r
H
h
F
Example: Linearization of a nonlinear model involving a nonlinear function of two variables.
The dynamic behavior of the liquid level h in the conical storage tank system shown in the figure can be shown to be represented below, where now the cross section area of the tank is given by:
The tank model becomes
where α and β are parameters defined by:
Dr. Kalyana Veluvolu
This process model has two types of nonlinear functions: Fih2 , a product of two functions, and h3/2. We shall have to linearize each of these functions seperately around the steady state (hs, Fis).
The linearization of f(h, Fi) = Fih2 proceeds as follows:
whereupon carrying out the indicated operations now gives:
If we ignore the higher order terms.
The steps involved in linearizing the second nonlinear term are no different from those illustrated in the previous example; the result is
Dr. Kalyana Veluvolu
We may now introduce these expressions in place of corresponding nonlinear terms in
Recalling that under steadystate conditions αFis=βhs1/2, and introducing the deviation variables y=(hhs) and u = (FiFis), the approximate linear model is obtained upon further simplification as:
where the steadystate gain, and time constant associated with this approximate linear model, are given by:
and
Dr. Kalyana Veluvolu
If we desire an approximate transformdomain transfer function model, Laplace transformation again gives
Note that the approximate linear model for the conical tank has the same process gain, K, as for the cylindirical tank; but the time constant is a much stronger function of the liquid level hs.
Some important points to note about the results of these two illustrative examples are the following:
Dr. Kalyana Veluvolu
Let us conclude with a comparison of the responses of the nonlinear model and an approximate linear model for the conical tank of the Example. Suppose that α=2, β=1, c=0.5 so that the nonlinear model is:
while the approximate linear model has the transfer function
which in the time domain is
Dr. Kalyana Veluvolu
tank1.m
tank.m
clc;clear
tss=4;
[t,h]=ode45('tank1',0,tss,[0.5]);
plot(t,h(:,1))
function hdot=tank1(t,h)
hs=0.5;
dfi=0.15;
hdot=[1/(2*hs^2.5)*(hsh+4*hs^0.5*dfi)];
Nonlinear model
function hdot=tanknl1(t,h)
alfa=2;
beta=1;
Fs=0.3536;
dfi=0.05;
Fi=Fs+dfi;
hdot=[alfa*Fi/h^2beta*h^(1.5)];
clc;clear
tss=4;
[t,h]=ode45('tanknl1',0,tss,[0.5]);
plot(t,h(:,1))
However, in some situations this may be inadequate (e.g. for the control of highly nonlinear processes) so the development of nonlinear controllers has featured prominently in process control in the last decade.
Adaptive Control:
A controller designed for the systems in the level control example in the previous example using the approximate transfer function model will function acceptably as long as the level, h, is maintaned at, or close to, the steadystate value hs. Under these conditions (hhs) will be small enough to make the linear approximation adequate. However, if the level is required to change over a wide range, the farther away from hs the level gets, the poorer the linear approximation.
Observe also from the approximate transfer function model that the apparent steady statestate gain and time constant are dependent on the steadystate around which the linearization was carried out. This implies that in going from one steady state to the other, the approximate process parameters will change.
Dr. Kalyana Veluvolu
The main problem with applying the classical approach under these circumstances is that it ignores the fact that the characteristics of the approximate model must change as the process moves away from hs if the approximate model is to remain reasonably accurate. The immediate implication is that any controller designed on the basis of this changing approximate model must also have its parameters adjusted if it is to remain effective.
In the adaptive control scheme, the controller parameters are adjusted (in an automatic fashion) to keep up with the changes in the process characteristics. We know intuitively that, if properly designed, this procedure will be a significant improvement over the classical scheme.
There are various types of adaptive control methods, differing mainly in the way the controller parameters are adjusted. The three most popular methods are: scheduled adaptive control, model reference adaptive control, and selftuning controllers.
Dr. Kalyana Veluvolu
Parameter Adjustment
Figure. Scheduled adaptive controller.
Scheduled Adaptive Control
A scheduled adaptive control method is one in which, as a result of a priori knowledge and easy quantification of what is responsible for the changes in the process charactersitics, the commensurate changes required in the controller parameters are programmed (or scheduled) ahead of time. This type of adaptive controller, sometimes referred to as gain scheduling, is illustrated by the block diagram in the following figure.
Dr. Kalyana Veluvolu
Continuing with the example of liquid level system, the figure represents control of linearised approximation of the original nonlinear system. The approximation is in the sense that the process transfer function G(s) only holds good for a particular level of the liquid in the tank. If the level h changes drastically, the transfer function G(s) may fail to represent the original system even approximately. Then there is a need to change the transfer function G(s) to account for the change in level h. One way of doing this is through Scheduled adaptive controller shown in previous page. From the system transfer function
it is seen that both K and τ are affected by the liquid level h. The effect on K is more prominent since it affects the system gain. The effect on the time constant is less apparent and not so prominent. Thus we only compansate for the effect of liquid level h on K. This is done in a way that the overall system gain KKc is constant. Notice that
Dr. Kalyana Veluvolu
So, to keep KKc = K0 (a constant)
This means that the controller gain Kc has to be adjusted inversely as the square root of the liquid level h. This is what will be done by the block represented as “Parameter adjustment” in the figure.
More generally, if K = K(t), that is, the process gain changes with time, then to keep the product KKc constant it is necessary to adjust Kc inversely with respect to K(t), i.e.
K(t) itself can be determined periodically by inserting a signal u into the process and monitoring the process output y.
Dr. Kalyana Veluvolu
Table: Scheduled control of Valve
Valve opening Kp Ti Td
0.00.15 1.7 95 23
0.150.22 2.0 89 22
0.220.35 2.9 82 21
0.351.0 4.4 68 17
In practice, it is often possible to find measured variables that correlate well with changes in process dynamics. These variables can be used to change the controller parameters, using a precalculated schedule. The controller parameters are computed offline for several operating conditions and stored in memory.
It is difficult to give general rules. Each case must be treated individually. The key question is to determine the auxiliary variables to be used as scheduling variables. It is necessary to have good insight into the dynamics of the process if gain scheduling is going to be used. The controller can be automatically tuned at a different operating points and the resulting tuning parameters can be saved and a schedule created.
Example: The figure given below shows different valve characteristics, if the scheduling variable in a level control system is the opening of a control valve, then a schedule may look as in the Table:
Flow
Quick opening
Linear
Equal percentage
Dr. Kalyana Veluvolu
Position
Contol signal u
Command signal uc
Output y
+
PI
f1(c)
f(u)
G(s)

Nonlinear Valve
Process
Example: The nonlinearity of the valve is assumed to be:
Let f1 be an approximation of the inverse of the valve characteristics. To compensate for the nonlinearity, the output of the controller fed through this function before it is applied to the valve. This gives the relation:
where c is the output of the PI controller
Dr. Kalyana Veluvolu
15
10
0
2
Time
1
Figure. A crude approximation of valve characteristics.
Assume that f(u)=u4 is approximated by two lines as shown in the figure. One from (0,0) to (1.3,3) and the other from (1.3,3) to (2,16).
Then we have:
Dr. Kalyana Veluvolu
Dr. Kalyana Veluvolu
Model Reference Adaptive Controller (MRAC)
Dr. Kalyana Veluvolu
Tracking error:
Introduce the cost function J:
Where θ is a vector of controller parameters. Change the parameters in the direction of the negative gradient of e2.
is called the sensitivity derivative. It indicates how the error is influenced by the adjustable parameters θ.
MIT Rule
Example:
Process:
Model:
Controller:
Closed loop system:
Ideal controller parameters for perfect modelfollowing
Derivation of adaptive law
Error:
where
Sensitivity derivatives
Approximate
where
MRAC of Pendulum
d2
d1
dc
T
Keith Sevcik
MRAC of Pendulum
Model
ymodel
Adjustment
Mechanism
Controller Parameters
uc
Controller
u
yplant
Keith Sevcik
MRAC of Pendulum
sensitivity
derivative
Keith Sevcik
MRAC of Pendulum
MRAC of Pendulum
MRAC of Pendulum
MRAC of Pendulum
MRAC of Pendulum
Reference Model
ymodel

+
Π
+

uc
yplant
θ1
e
Plant
Π
θ2
Π
Π
Keith Sevcik
MRAC of Pendulum
Keith Sevcik
Keith Sevcik
MRAC of Pendulum
Keith Sevcik
MRAC of Pendulum
Keith Sevcik