Professional Documents
Culture Documents
furnace
R. T. Bui, A. Meghlaoui and R. Ouellet
A general structure of the spectjic optimal control has been previously formulated and used to solve the fuel
optimal problem of an aluminum casting furnace. Proportional, integral, and derivative (P, I, D) closed-loop
control were applied to a IO-order nonlinear model of the furnace. This paper analyzes the resulting control
actions and the dynamic response to a step change in the target temperature of the liquid metal. It is shown
that P and PD schemes are stable but bring about a steady-state error, whereas PI and PID schemes cause
no steady-state error but involve considerable oscillations in the transient response and longer settling times.
In view ofthe systems high thermal inertia and the need to impose limits on fuelfrow rate, it is found that a
PD scheme is the most appropriate due to the absence of overshoot and a short settling time. The method is
also applied to another optimization criteria, the minimization of temperature oscillations. This shows the
applicability of the scheme to practical industrial problems.
2. Previous work
In the truly (i.e., open-loop) optimal control problem, the
optimal solution, u(t), is to be found as an open-loop
control variable. The drawback of open-loop control is
W that if a perturbation occurs, the behavior of the process
will deviate from its intended optimal trajectory. In
Figure 1. Longitudinal and transversal sections of the casting
furnace: (1, 2) refractory roof, (3) stack, (4) burner, (5) loading practice a closed-loop scheme is often preferred.
doors, (6, 7,8) floor, (9, 10) metal, (11) syphon, (12) spout, and With the exception of the special case of linear
(13) thermocouples. Metal, roof, and floor are each treated as a quadratic control, the explicit implementation of an
one-dimensional heat conducting medium optimal control in closed-loop form is a difficult problem
due to the large number of parameters to be determined.
A viable alternative is a suboptimal scheme, in which a
specified structure is chosen prior to implementation as
the control law.5 In this sense it cannot be seen as truly
optimal, this is why it is referred to as specific optimal
CVll control or S.O.C.
A general structure for the specific optimal control
was formulated recently. Solved by COMMIN, it de-
termined the optimal parameter values for a closed-loop
cvv ROOF
(zon. 18) scheme. The formulation was applied to the closed-loop
lx6 control of an aluminum casting furnace.6 Various s.0.c.
____________________~~~~~ laws were studied and the corresponding solutions were
CV?
obtained and compared. The control laws used were
t-*_______-k-XY!-_--__ based on a combination of the three basic actions,
* Tcm - 1,s PI CV6
l_a!!!fL__ -- ____r ____ _ __----- namely proportional (P), integral (I), and derivative (D)
A CH =a,T,, (l)*P,l,, (2 i. cvs
actions. In other words, the control variable, u(t), had the
a,-q(1) 1, (O- 76
general form:
GAS
-Qh Q,Q,.O, l L, - 11 CVl
(x011* 7)
(4 dAy(r)
I G(O-JJ Q, .
u(t) = k,Ay(t) + ki r *A,,(t)dt + k, ~ (1)
Jo (CL
cvz
where Ay(t) is the difference between the target value and
cv3 the actual value of the mid-depth liquid metal
____--_______________ temperature.
Tcw = 0, &PI + 6,T.W CV4
The problem was solved with and without inequality
+ o,- O.(4) lW-Tc constraint imposed on u(t). For the problem without
constraint, the control laws P, PD, PI, and PID were
Figure 2. Discretization into 10 control volumes (CV) used in the applied. For the problem with constraint, the control
10th.order simplified model. T are temperatures (state variables);
laws PD, PI, and PID were considered. In both cases
Q are heat flow rates between control volumes; and subscripts C,
D, E, and F indicate interfaces between zones or with the environ- the results were compared with those of the truly
ment (open-loop) optimal scheme.
+-----
.
500 m3/hr. In the solution with constraint (Figure 4), a
.3 @DC l-i&L -~~.._
_
!
much smaller violation of the limit occurred with the PD
and PI laws, while the PID law caused no violation. XC, -
....
Further analysis showed that in the presence of .:.y-
constraint, the control law used must be simple but must Q.$_.
contain at least two adjustable parameters to allow the -..
required parameter adjustments without serious con-
straint violation. The higher the number of parameters p--------------------------------q
to be adjusted, the more easily a convergence can be 01 . . . 1
reached. That previous work was intented to prove that 0 10 20 40 50 60
3. Review of S.O.C.formulation
W)
~ = fry(t), W), tl
dt
In this article the formulation of the S.O.C. problem,
adapted to solution by COMMIN, is applied to the WY(a), al = 0 (3)
control and dynamic performance analysis of the NCy(b), bl = 0 (4)
aluminum casting furnace under closed-loop control.
Relevance of the work comes from the fact that due to The system is subject to m inequality constraints on the
the complexity and the nonlinearity of the process, there control variables:
is no way for process engineers to know a priori which g[u(t), t] 2 0 for a I t I b (5)
control law would be best suited to their problem, both
in terms of fuel optimal control and in terms of dynamic In closed-loop s.o.c., the control law has the general
response (response time, overshoot, steady-state error, form:
oscillations). In the following, the control law and the u(t) = hCx(Q, kl (6)
dynamic behavior of the casting furnace in response to
a change in target temperature will be studied for each where x(t) is a subvector of y(t), and k is a constant
form of feedback, then results are compared and choices vector.
suggested. We wish to find k to minimize the cost function:
To begin with, this section gives a brief review of the
problem formulation, focusing on the S.O.C. part as J(k) = 0b dQ(t), u(t), tl dt (7)
opposed to the truly (open-loop) optimal alternative. For s
a more detailed presentation, the interested reader may Substituting equation (6) into equations (2), (5), and
(7) and noting that x(t) is a subvector of y(t) we obtain:
dy(Q
_ 700 y..............,....
~ = k, 4
f-L-y(t),
'...._
........
!T!!!%! dt
3 m- S.0.C.P
g i ___-__J..,,
e.o.c.Pcl
----__
k tl 2 0
gCy@), for a 5 t < b (9)
9 ? *-.. ..- 8.O.C.Pl b
_.-_-.
tl dt (10)
.,.....,, II
E 400- S.O.C.PID
-,.- ..-.._
J(k) = 4Cy(t), k,
s (1
.. A.-,
5 -_ 5..
x.., .
0 m- ..
6: Y.,
....
.
-we
Another equation must be added to ensure that k is
z 200- ,.
.....,,
_=. a constant vector:
g loo- ....__
% dk
I .I.,.,.,,%
.... ..._._. -_= 0 (11)
0
0 10 23 40 Xl 60
dt
Tiie;Pmin) The constraint in equation (9) now involves the state
Figure 3. Fuel flow rates for the optimal (open-loop) and for the
variable instead of the control variables as was the case
four specific optimal (closed-loop) control schemes, without con- in equation (5). We use the penalty method7 to take care
straint on the control variable of equation (9). A new variable z(t) is defined as the
integral from a to b of dz(t)/dt. The latter term, dz(t)/dt, We then use COMMIN4 to solve the minimum fuel
is taken as a weighted average of the squares of the scalar problem taking as cost function the following expression:
components of equation (9) and thus, with the integrals f
initial condition set to be z(a) = 0, z(b) gives a direct J(k) = 1 U(t) dt (18)
measure of the extent of violation of inequality constraint s0
(9). where u(t) is taken from equation (1).
This new variable, z(t), is added to the cost function
J to obtain the modified cost function which we call Jmod,
and the problem now is to find k that minimizes Jmod. 4. Closed-loop control and dynamic analysis
This is done through the usual formulation using the The modelling of the aluminum casting furnace and the
technique of Lagrange multipliers in which a function development of its simplified control model have been
called Hamiltonian is defined as published previously.3 Figures I and 2 describe the
furnace and the discretization used in developing the
dz(t)
H=4+LT.f+-- (12) lOth-order simplified control model. Specific optimal
dt control was applied to the furnace and optimal fuel flow
and a nonlinear operator called the Lagrangian is rate, u(t), was determined.j Figures 3 and 4 present the
defined as fuel flow rates obtained by solving the problems without
dk(t) and with constraint on u(t), respectively, and for different
@,=H-~T.f-flT~ (13) optimal (open-loop) and specific optimal (closed-loop)
dt
control schemes. The limits imposed on fuel flow rate
where h and q are time functions called Lagrange (50 I u(t) I 500 m3/hr) exclude the P control law from
multipliers, associated with equality constraints (8) and the solution with constraint, because at the end of the
(1 l), respectively. After substituting the integrand of Jmod batch, when metal temperature tends toward its target
into equation (12), the Euler-Lagrange equations can be value, as shown by equation (1) a P control law would
applied to obtain the conditions of optimality. These require zero fuel flow rate, which violates the lower limit.
conditions are: The results of Figures 3 and 4 have been briefly reviewed
at the end of Section 2 of this article.
&_H To show the benefits of S.O.C. and the resulting
(14)
ay closed-loop structure, we now study the dynamic
behavior of the furnace under the various control laws,
aH
il= -x3 db) - W = 0 (15) in respons to a step change in the target metal
temperature. Starting from a steady state with a metal
The mixed boundary conditions of equation (15) temperature of 704C a 10% step change is imposed on
complicate the solution, and to avoid this we replace it the target metal temperature. In practice, this happens
with the following reformulation to make the problem with a batch of liquid metal must be brought from an
amenable to COMMIN: initial to a final temperature then maintained there while
awaiting the next operation.
The closed-loop systems that are about to be analyzed
o(t) = - g , w(a) = o(b) = 0 (16) are the ones obtained by solving the S.O.C. problems
where: without and with constraints on u(t), the results of which
were discussed previously6 and again presented here for
f aH
o(t) = q(t) - q(a) = - ~ dt (17) reference in Table 1. An improvement has been
s (1 ak introduced in Table 1, however. In the previous work,j
Table 1. Solutions of the minimum fuel problem, solved using optimal (open-loop) and specific optimal (closed-loop) control schemes
without and with constraint imposed on the control variable
cost
u(t) Optimal parameters function
constraint Total fuel calculated
Constraint Control violation required kd as per
on u(t) law (%) (m3) ( x k-3, ( x 1%7, equation (20)
2x5
(19)
@= (1 _ [).
Table 2. Characteristics of transient response to a 10% input step change for the various specific optimal
control laws without and with constraint on the control variable u(t)
4 = integrand of the cost function Sage, P. S. and White, C. C. Optimal Systems Control, 2nd ed.,
Prentice-Hall, Englewood Clitfs, NJ, 1977
m = Lagrangian Shearer, J. L., Murphy, A. T. and Richardson, H. H. Introduction
Aq = Lagrange multipliers to System Dynamics. Addison-Wesley, Reading, MA, 1971