PRINCIPLES OF

THERMAL CONTROL
Mihir Sen
Department of Aerospace and Mechanical Engineering
University of Notre Dame
Notre Dame, IN 46556, U.S.A.
January 27, 2004
Contents
Preface 4
1 Introduction 5
2 Systems and control theory 8
2.1 Systems without feedback control . . . . . . . . . . . . . . . . . . . . 8
2.2 Linear systems theory . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Ordinary dierential equations . . . . . . . . . . . . . . . . . . 9
2.2.2 Algebraic-dierential equations . . . . . . . . . . . . . . . . . 11
2.2.3 Partial dierential equations . . . . . . . . . . . . . . . . . . . 11
2.2.4 Other models . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Nonlinear eects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 Bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 Controllability and reachability . . . . . . . . . . . . . . . . . 13
2.3.3 Bounded variables . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.4 Relay and hysteresis . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 System identication . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Estimation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 Open-loop control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.7 Feedback control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.8 Control strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.9 Soft computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.10 Nondimensionalization and scaling . . . . . . . . . . . . . . . . . . . 17
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Lumped parameter approximation 19
3.1 Heating and cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 PID control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.1 Proportional control . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.2 Integral control . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1
3.2.3 Integral-derivative control . . . . . . . . . . . . . . . . . . . . 22
3.2.4 PID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.5 Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.6 Variable ambient temperature . . . . . . . . . . . . . . . . . . 24
3.3 On-o control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.1 Variable ambient temperature . . . . . . . . . . . . . . . . . . 28
3.3.2 Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Two-uid problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Two-body problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4 Conductive systems 32
4.1 Systems with one spatial dimension . . . . . . . . . . . . . . . . . . . 32
4.1.1 Fins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 Stefan problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5 Natural convective systems 35
5.1 Natural convection loop . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 Deterministic chaos . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 Two-dimensional cavity . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.4 Porous media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6 Forced convective systems 36
6.1 One-dimensional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.1.1 Long duct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.1.2 Heat exchanger model . . . . . . . . . . . . . . . . . . . . . . 37
6.2 Potential ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.3 Viscous ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.4 Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7 Radiative systems 41
7.1 Radiative n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2 Radiative plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.3 Furnace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2
8 Hardware 42
8.1 Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.1.1 Thermocouples . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.1.2 Thermistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.1.3 MEMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.2 Actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.2.1 Furnace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.2.2 Fans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.3 Microprocessors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
9 Concluding remarks 43
References 44
Index 49
3
Preface
Even though the thermal sciences constutute an important application of control
theory, there are few basic texts dealing exclusively with the subject. On the one
hand there are many on the mathematical theory of control and its broad applications,
and on the other there are specialized books for the design engineer on the devices
and tools to use. The present work is intended to ll the gap. It is written for
thermal engineers; the writing is at the level of a college senior or beginning graduate
student who has taken courses in linear algebra, ordinary and partial dierential
equations, uid mechanics and heat transfer. A working knowledge of computation
is also assumed.
These pages are at present in the process of being written. I will be glad to
receive comments and suggestions, or have mistakes brought to my attention.
Mihir Sen
Department of Aerospace and Mechanical Engineering
University of Notre Dame
Notre Dame, IN 46556
U.S.A.
Copyright c _ by M. Sen, 2004
4
Chapter 1
Introduction
This is a brief introduction to the control of thermal systems and some of its applica-
tions. Fundamentals of the linear theory applied to ordinary and partial dierential
equations are discussed from the point of view of system and output controllability.
Some nonlinear eects that are characteristic of thermal systems are pointed out,
and commonly used system identication models presented. Details of the control
of lumped convective systems with on-o and PID control are given. Conductive,
convective and advective phenomena are studied using one- dimensional models in
the context of ns, long ducts and heat exchangers. Finite-dierence approximations
are employed to enable the use of nite-dimensional theory. Applications to complex
thermal systems like heat exchangers and networks are also described and represen-
tative experimental results shown. An extensive list of references for further study is
included.
There are many kinds of thermal systems in common industrial, transportation
and domestic use that need to be controlled in some manner, and there are many
ways in which that can be done. One can give the example of heat exchangers
[1, 2], environmental control in buildings [38], satellites [912], thermal packaging
of electronic components [13, 14], manufacturing [15], rapid thermal processing of
computer chips [1618], and many others. If precise control is not required, or if the
process is very slow, control may simply be manual; otherwise some sort of mechanical
or electrical feedback system has to be put in place for it to be automatic.
Most thermal systems are generally complex involving diverse physical processes.
These include natural and forced convection, radiation, complex geometries, property
variation with temperature, nonlinearities and bifurcations, hydrodynamic instability,
turbulence, multi-phase ows, or chemical reaction. It is common to have large
uncertainties in the values of heat transfer coecients, approximations due to using
lumped parameters instead of distributed temperature elds, or material properties
that may not be accurately known. In this context, a complex system can be dened as
5
one that is made up of sub-systems, each one of which can be analyzed and computed,
but when put together presents such a massive computational problem so as to be
practically intractable. For this reason large, commonly used engineering systems are
hard to model exactly from rst principles, and even when this is possible the dynamic
responses of the models are impossible to determine computationally in real time.
Most often some degree of approximation has to be made to the mathematical model.
Approximate correlations from empirical data are also heavily used in practice. The
two major reasons for which control systems are needed to enable a thermal system
to function as desired are the approximations used during design and the existence of
unpredictable external and internal disturbances which were not taken into account.
There are many aspects of thermal control that will not be treated in this brief
review. The most important of these are hardware considerations; all kinds of sensors
and actuators [19,20] developed for measurement and actuation are used in the control
of thermal systems. Many controllers are also computer based, and the use of micro-
processors [21, 22] and PCs in machines, devices and plants is commonplace. Flow
control, which is closely related to and is often an integral part of thermal control,
has its own extensive literature [23]. Discrete-time (as opposed to continuous-time)
systems will not be described. The present paper will, however, concentrate only on
the basic principles relating to the theory of control as applied to thermal problems,
but even then it will be impossible to go into any depth within the space available.
This is only an introduction, and the interested reader should look at the literature
that is cited for further details. There are good texts and monographs available on
the basics of control theory [2427], process control [2831], nonlinear control [32],
innite-dimensional systems [33, 34], and mathematics of control [35, 36] that can be
consulted. These are all topics that include and are included within thermal control.
There are many kinds of thermal systems in common use in industry and house-
holds that need to be controlled. Examples include heating and cooling systems,
spacecraft environment control, industrial processes, rapid thermal processing of com-
puter chips, etc. Control is an important aspect of these systems, usually of the
temperature.
If strict control is not required, or if the process is very slow, control may be
manual. Otherwise some sort of mechanical or electrical system is designed to provide
automatic control. The term classical control generally refers to the body of theory
derived on the basis of Laplace transforms for linear systems. On the other hand
we will be using the so-called modern control which is based on state-space analysis
mainly because it can be extended to nonlinear systems.
Thermal systems are generally complex involving diverse physical processes, some
of which are nonlinear. The study of these systems thus uses approaches that are
analytical, experimental or computational.
Undergraduate-level textbooks in the area of heat transfer include [37], and in
6
control [38]. More advanced books such as [39, 40] are very useful.
7
Chapter 2
Systems and control theory
A thermal system (not to be confused with a thermodynamic system) is an identiable
part of a building or machine that exchanges uid and heat with the surroundings.
A block diagram of a system (alternatively it may be shown as a signal ow graph) is
shown in Fig. 2.1. The input is u(t) and the output y(t). The main component is the
plant (also called the process) that is to be controlled. Some basic ideas of systems
will be dened here even though, because of the generality involved, it is hard to be
specic at this stage.
2.1 Systems without feedback control
The dynamic behavior of a system such as that in Fig. 2.1 (often called the open-loop
system or plant to distinguish it from the system with controller described below),
may be mathematically represented as
L
s
(x, u, w, ) = 0, (2.1)
where L
s
is a system operator, t is time, x(t) is the state of the system, u(t) is its
input, w(t) is some external or internal disturbance, and is a parameter that denes
the system. Each one of these quantities belongs to a suitable set or vector space and
there are a large number of possibilities. For example L
s
may be an algebraic, integral,
ordinary or partial dierential operator, while x may be a nite-dimensional vector
or a function. u(t) is usually a low-dimensional vector.
In general, the output of the system y(t) is dierent from its state x(t). For
example, x may be a spatial temperature distribution, while y are the readings of
one or more temperature measurement devices at a nite number of locations. The
relation between the two may be written as
L
o
(y, x, u, w, ) = 0, (2.2)
8
E
p E
c c
u(t)
x(t)
plant
y(t)
w(t)
Figure 2.1: System without controller.
where L
o
is the output operator.
The system is single-input single-output (SISO) if both u and y are scalars. A
system is said to be controllable if it can be taken from one specic state to another
within a prescribed time with the help of a suitable input. It is output controllable if
the same can be done to the output. It is important to point out that output control-
lability does not imply system controllability. In fact, in practice for many thermal
systems the former is all that is required; it has been reported that most industrial
plants are controlled quite satisfactorily though they are not system controllable [41].
All possible values of the output constitute the reachable set. A system is said to be
observable if its state x can be uniquely determined from the input u and output y
alone. The stability of a system is a property that leads to a bounded output if the
input is also bounded.
2.2 Linear systems theory
The system may be continuous in which case it is modeled mathematically thermal as
dierential equations. It may be lumped giving rise to ordinary dierential equations,
or distributed with partial dierential equations. Discrete-time systems are generally
those that are sampled periodically. System theory can be developed for dierent
linear operators, and some of these are outlined below.
2.2.1 Ordinary dierential equations
Much is known about a linear dierential system in which Eqs. (2.1) and (2.2) take
the form
dx
dt
= Ax + Bu, (2.3)
y = Cx + Du, (2.4)
where x R
n
, u R
m
, y R
p
, A R
nn
, B R
nm
, C R
pn
, D R
pm
. x, u
and y are vectors of dierent lengths and A, B, C, and D are matrices of suitable
9
sizes. Though A, B, C, and D can be functions of time in general, here they will be
treated as constants.
The solution of Eq. (2.3) is
x(t) = e
A(tt
0
)
x(t
0
) +
_
t
t
0
e
A(ts)
Bu(s) ds. (2.5)
where the exponential matrix is dened as
e
At
= I + At +
1
2!
A
2
t
2
+
1
3!
A
3
t
3
+ . . . ,
with I being the identity matrix. From Eq. (2.4), we get
y(t) = C
_
e
At
x(t
0
) +
_
t
t
0
e
A(ts)
Bu(s) ds
_
+ Du. (2.6)
Eqs. (2.5) and (2.6) dene the state x(t) and output y(t) if the input u(t) is given.
A system is said to be observable if its state x can be uniquely determined from
the input u and output y. It can be shown to be the case for Eqs. (2.3) and (2.4) if
the observability matrix
M
o
=
_

_
C
CA
CA
2
.
.
.
CA
n1
_

_
R
pnn
(2.7)
is of rank n.
A linear system is said to be controllable if it can be taken from any state x
0
at
time t
0
to any other state x in a nite time. It can be shown that for the system
governed by Eq. (2.3), a u(t) can be found to take x(t) from x(t
0
) at t = t
0
to x(t
f
) = 0
at t = t
f
if and only if the matrix
M =
_
B
.
.
. AB
.
.
. A
2
B
.
.
. . . .
.
.
. A
n1
B
_
R
nnm
(2.8)
is of rank n. The system is then controllable. For a linear system, controllability
from one state to another implies that the system can be taken from any state to any
other. It must be emphasized that the u(t) that does this is not unique.
Similarly, it can be shown that the output y(t) is controllable if and only if
N =
_
D
.
.
. CB
.
.
. CAB
.
.
. CA
2
B
.
.
. . . .
.
.
. CA
n1
B
_
R
p(n+1)m
(2.9)
10
is of rank p.
The Laplace transform of f(t) dened as
L(f) =
_

0
f(t)e
st
dt (2.10)
can be used to dene a transfer function relating the output y(t) and input u(t) as
G(s) =
L(y)
L(u)
(2.11)
2.2.2 Algebraic-dierential equations
This is a system of equations of the form
E
dx
dt
= A x + B u, (2.12)
where the matrix E R
nn
is singular [42]. This is equivalent to a set of equations,
some of which are ordinary dierential and the rest are algebraic. As a result of this,
Eq. (2.12) cannot be converted into (2.3) by substitution. The index of the system
is the least number of dierentiations of the algebraic equations that is needed to
get the form of Eq. (2.3). The system may not be completely controllable since
some of the components of x are algebraically related, but it may have restricted or
R-controllability [43].
2.2.3 Partial dierential equations
Partial dierential equations (PDEs) are an example of innite- dimensional systems
that are very common in thermal applications [44, 45]. Exact controllability exists if
the function representing the state can be taken from an initial to a nal target state,
and is approximate if it can be taken to a neighborhood of the target [46]. Determi-
nation of approximate controllability is usually sucient for practical purposes.
Consider a system governed by
X
t
= /X +Bu, (2.13)
with homogeneous boundary and suitable initial conditions, where / is a bounded
semi-group operator [44], and B is a another linear operator. The state X(, t) is a
function of spatial coordinates and time t. If / is self-adjoint, then it has real eigen-
values and a complete orthonormal set of eigenfunctions
m
(), with m = 0, 1, 2 . . .,
which forms a complete spatial basis for X. It is known [45] that the system is
approximately state controllable if and only if all the inner products
B,
m
) , = 0. (2.14)
11
2.2.4 Other models
x +
_
t
0
x() d = f(x, u, w) (2.15)
2.3 Nonlinear eects
A lumped system may be modeled by two equations
x = f(x, u, w
s
) (2.16)
y = g(x, u, w
m
) (2.17)
where x(t) 1
n
is the state of the system, u(t) 1
m
is the input, w
s
(t) 1
r
is a disturbance to the system, w
m
(t) 1
r
is a disturbance to the measurements,
y(t) 1
s
is the output, and t 1
+
is time. Notice that in principle (and in fact for
simple systems) x can be eliminated between Eqs. (2.16) and (2.17) to get a single
relation between u and y . The separation is usually made, however, to distinguish
between x which completely describes the system, and y which is an output that can
be observed, i.e. measured and used for feedback control or other purposes.
The system is nonlinear if either of the functions f or g in Eqs. (2.16) and (2.17)
are nonlinear. It is hard to make general statements about such systems, so that
special cases are often treated. For w constant, the model may be linearized about
an equilibrium point E with x = x, u = u where f(x, u, w) = 0 and y = g(x, u, w) to
give
x

=
f
x

E
x

+
f
u

E
u

= Ax

+ Bu

(2.18)
u

=
g
x

E
x

+
g
u

E
u

= Cx

+ Du

(2.19)
where x = x + x

and u = u + u

. The Jacobian matrices f/x, f/u, g/x and

g/u are evaluated at the equilibrium point. The primes in Eqns. (2.18) and (2.19)
are usually dropped. If x, u and y are written as column vectors of size n, m and
r, respectively, then A, B, C and D are n n, n m, s n and s m matrices,
respectively. Eqs. (2.18) and (2.19) have the same form as the linear equations (2.3)
and (2.4).
12
2.3.1 Bifurcations
2.3.2 Controllability and reachability
General theorems for the controllability of nonlinear systems are not available at this
point in time. Results obtained from the linearized equations generally do not hold
for the nonlinear equations. The reason is that in the nonlinear case one can take a
path in state space that travels far from the equilibrium point and then returns close
to it. Thus regions of state space that are unreachable with the linearized equations
may actually be reachable. In a thermal convection loop it is possible to go from one
branch of a bifurcation solution to another in this fashion [47].
Observability.
2.3.3 Bounded variables
It is common to have the variables in Eqs. (2.16) and (2.17) conned to certain
ranges depending on hardware constraints. In this case x X 1
n
, u U1
m
and
y Y 1
r
. The system is now nonlinear since, for instance, the sum of solutions
x
1
and x
2
may not be a valid solution. In practice, due to hardware constraints it is
common to have the physical variables conned to certain ranges, so that variables
such as x and u in Eqs. (2.1) and (2.2), being temperatures, heat rates, ow rates
and the like, are bounded. If this happens, even systems locally governed by Eqs.
(2.3) and (2.4) are now nonlinear since the sum of solutions may fall outside the range
in which x exists and thus may not be a valid solution. On the other hand, for a
controllable system in which only u is bounded in a neighborhood of zero, x can reach
any point in R
n
if the eigenvalues of A have zero or positive real parts, and the origin
is reachable if the eigenvalues of A have zero or negative real parts [35].
2.3.4 Relay and hysteresis
A relay is an element of a system that has an input-output relationship that is not
smooth; it may be discontinuous or not possess rst or higher-order derivatives. This
may be accompanied by hysteresis where the relationship also depends on whether
the input is increasing or decreasing. Valves are typical elements in ow systems that
have this kind of behavior.
Describing function
13
2.4 System identication
To be able to design appropriate control systems, one needs to have some idea of
the dynamic behavior of the thermal system that is being controlled. Mathematical
models of these systems can be obtained in two entirely dierent ways: from rst
principles using known physical laws, and empirically from the analysis of experi-
mental information (though combinations of the two paths are not only possible but
common). The latter is the process of system identication, by which a complex
system is reduced to mathematical form using experimental data [31, 48, 49]. There
are many dierent ways in which this can be done, the most common being the t-
ting of parameters to proposed models [50]. In this method, a form of L
s
is assumed
with unknown parameter values. Through optimization routines the values of the un-
knowns are chosen to obtain the best t of the results of the model with experimental
information. Apart from the linear Eq. (2.3), other models that are used include the
following.
There are many forms based on Eq. (2.16), one of which is the closed-ane
model
dx
dt
= F
1
(x) +F
2
(x)u (2.20)
The bilinear equation for which F
1
(x) = Ax and F
2
(x) = Nx + b is a special
case of this.
Volterra models, like
y(t) = y
0
(t) +

i=1
_

. . .
_

k
i
(t; t
1
, t
2
, t
3
, . . . , t
i
)u(t
1
) . . . u(t
i
) dt
1
. . . dt
i
(2.21)
for a SISO system, are also used.
Functional [51], dierence [52] or delay [53] equations such as
dx
dt
= A x(t s) + B u (2.22)
also appear in the modeling of thermal systems.
Fractional-order derivatives, of which there are several dierent possible deni-
tions [5458] can be used in dierential models. As an example, the Riemann-
Liouville denition of the nth derivative of f(t) is
a
T
n
t
f(t) =
1
(mn + 1)
d
m+1
dt
m+1
_
t
a
(t s)
mn
f(s) ds, (2.23)
where a and n are real numbers and m is the largest integer smaller than n.
14
2.5 Estimation theory
Stochastic disturbance.
Stochastic control: Kalman and Extended Kalman lters
2.6 Open-loop control
If the system is controllable, the objective of open-loop control is to nd u such that
y = y
s
(t), where y
s
, known as a reference value, is prescribed. The problem is one of
regulation if y
s
is a constant, and tracking if it is function of time.
The objective of control is to have a given output y = y
r
(t), where the reference
or set value y
r
is prescribed. The problem is called regulation if y
r
is a constant, and
tracking if it is function of time.
In open-loop control the input is selected to give the desired output without using
any information from the output side; that is one would have to determine u(t) such
that y = y
r
(t) using the mathematical model of the system alone. This is a passive
method of control that is used in many thermal systems. It will work if the behavior
of the system is exactly predictable, if precise output control is not required, or if
the output of the system is not very sensitive to the input. If the changes desired
in the output are very slow then manual control can be carried out, and that is also
frequently done. A self-controlling approach that is sometimes useful is to design the
system in such a way that any disturbance will bring the output back to the desired
value; the output in eect is then insensitive to changes in input or disturbances.
2.7 Feedback control
Open-loop control is not usually eective for two reasons: (a) The mathematical
model of the plant may not be correctly known due to uncertainty in the heat trans-
fer coecients, for instance, the model may have been simplied by using lumped
parameters instead of distributed, material properties may not be accurately known,
etc. (b) In addition there may be an unknown external disturbance such as a change
in environmental temperature, surface characteristics due to fouling, etc.
Open-loop control is not usually eective for many systems. For thermal systems
contributing factors are the uncertainties in the mathematical model of the plant and
the presence of unpredictable disturbances. Internal disturbances may be noise in the
measuring or actuating devices or a change in surface heat transfer characteristics due
to fouling, while external ones may be a change in the environmental temperature. For
these cases closed-loop control is an appropriate alternative. For closed-loop control,
there is a feedback from the output to the input of the system. This is done using
15
E
e(t)
controller
E
p E
c c
u(t)
x(t)
plant
y(t)
w(t)
j
C
c
E y
r
(t)
Figure 2.2: System with comparator C and controller.
feedback from the output, as measured by a sensor, to the input side of the system,
as shown in Fig. 2.2; the gure actually shows unit feedback. Some physical quantity
is measured by a sensor, the signal is processed by a controller, and then used to
move an actuator such as a fan or heater. There is a comparator which determines
the error signal e(t) = e(y
r
, y), which is usually taken to be
e = y
r
y. (2.24)
The key role is played by the controller which puts out a signal that is used to move
an actuator in the plant.
Sensors that are commonly used are temperature-measuring devices such as ther-
mocouples, resistance thermometers or thermistors. The actuator can be a fan or a
pump if the ow rate is to be changed, or a heater if the heating rate is the appropriate
variable. The controller itself is either entirely mechanical if the system is not very
complex, or is a digital processor with appropriate software. In any case, it receives
the error in the output e(t) and puts out an appropriate control input u(t) that leads
to the desired operation of the plant.
The control process can be written as
L
c
(u, e, ) = 0, (2.25)
where L
c
is a control operator. The controller designer has to propose a suitable L
c
,
and then Eqs. (2.1) (2.25) form a set of equations in the unknowns x(t), y(t) and
u(t). Choice of a control strategy denes L
o
and many dierent methodologies are
used in thermal systems. The process can be represented mathematically by
x = f(x, u, w) (2.26)
y = g(x, u, w) (2.27)
u = h(u, u
s
) (2.28)
The sensor may be used to determine the error
e = y
s
y (2.29)
through a comparator.
16
2.8 Control strategies
The term classical control is often used to refer to theory derived on the basis of
Laplace transforms. Since this is exclusively for linear systems, we will be using the
so-called modern control or state-space analysis which is based on dynamical systems,
mainly because it can be extended to nonlinear systems. Where they overlap, the
issue is only one of preference since the results are identical.
It is common to use on-o (or bang-bang, relay, etc.) control. This is usually used
in heating or cooling systems in which the heat coming in or going out is reduced
to zero when a predetermined temperature is reached and set at a constant value
at another temperature. Another method is Proportional-Integral-Derivative (PID)
control [59] in which
u = K
p
e(t) + K
i
_
t
0
e(s) ds + K
d
de
dt
. (2.30)
An example of a nonlinear form of this method is when
u = K
1
e + K
2
e
3
(2.31)
In feedback linearization, the nonlinearity is included in the feedback loop and
the system itself is linear. Sliding-mode control can also be employed. Control theory
itself is a vast subject, with specialized branches like optimal [60], robust [61], and
stochastic control [62] that are well developed. Many of the tools in these areas nd
applications in thermal systems.
2.9 Soft computing
Articial neural networks and fuzzy logic control can also be used for control [63].
2.10 Nondimensionalization and scaling
It is common in studies of uid and thermal systems to work with nondimensionalized
versions of the variables and parameters. The advantages are frugality (least number
of parameters invoked), generality (study not conned to a specic uid or system, but
more general), and normalization (variables usually between zero and unity, though
not necessarily the parameters). This distinguishes between quantities that are xed
during the study, called parameters, and those that are variables.
The same can be done for specic control applications. However,the method of
nondimensionalization may vary according to the problem; some generality is conse-
quently lost: a quantity that is constant in one control strategy may be a manipulated
17
variable in others. For this reason it will be convenient to leave some quantities in
their dimensional form while scaling others.
Amplitude and time
Problems
18
Chapter 3
Lumped parameter approximation
3.1 Heating and cooling
We can use internal heat generation to control the temperature of a body losing heat
to the environment . Consider a body that is cooled from its surface by convection to
the environment with a constant ambient temperature T

. It also has an internal heat

source Q(t) to compensate for this heat loss, and the control objective is to maintain
the temperature of the body at a given level by manipulating the heat source. The
Biot number for the body is Bi = hL/k, where h is the convective heat transfer
coecient, L is a characteristics length dimension of the body, and k is its thermal
conductivity. If Bi < 0.1, the body can be considered to have a uniform temperature
T(t). Under this lumped approximation the energy balance is given by
Mc
dT
dt
+ hA
s
(T T

) = Q

(t), (3.1)
where M is the mass of the body, c is its specic heat, and A
s
is the surface area for
convection.
With the nondimensional variables
=
t
Mc/hA
s
, (3.2)
Q =
Q

hA
s
(T
i
T

)
, (3.3)
this equation becomes
d
dt
+ = Q(t), (3.4)
where
=
T T

T
i
T

. (3.5)
19
T(0) = T
i
, so that (0) = 1. With x = , u = Q, n = m = 1 in Eq. (2.3), we nd
from Eq. (2.8) that rank(M)=1, so the system is controllable.
Open-loop operation to maintain a given non-dimensional temperature
r
is easily
calculated. Choosing Q =
r
, it can be shown from the solution of Eq. (3.4), that is
(t) = (1
r
)e
t
+
r
, (3.6)
that
r
as t . In practice, to do this the dimensional parameters hA
s
and
T

must be exactly known. Since this is usually not the case some form of feedback
control is required.
We can design a feedback controller for the lumped system as shown in Fig. 2.2
to attempt to maintain the temperature of the mass at a set temperature
s
.
3.2 PID control
We will rst investigate proportional-integral-derivative (PID) control. The control
input is taken to be
q = K
p
e + K
i
_

0
e(

) d

+ K
d
de
d
(3.7)
where the error is
e() =
s
(3.8)
3.2.1 Proportional control
Let
q = K
p
(
s
) (3.9)
from which we get
d
d
+ = K
p
(
s
) (3.10)
The solution is
= Ce
(1+Kp)
+
K
p

s
1 +K
p
(3.11)
From the initial condition (0) = 1, we have C = 1 K
p

s
/(1 +K
p
), so that
=
_
1
K
p

s
1 +K
p
_
e
(1+Kp)
+
K
p

s
1 +K
p
(3.12)
If 1 + K
p
> 0, we nd that as
() =
K
p
1 +K
p

s
(3.13)
20
The set point
s
is thus never achieved, and there is an oset
oset
dened by
() =
s

oset
(3.14)
where

oset
=
s
_
1
K
p
1 +K
p
_
(3.15)
the oset can be reduced by choosing a large K
p
. For large K
p
, the oset is approxi-
mated by

oset
=

s
K
p
(3.16)
3.2.2 Integral control
In this case
q = K
i
_

0
(
s
(

)) d

(3.17)
so that
d
d
+ = K
i
_

0
(
s
(

)) d

(3.18)
Dierentiating with respect to , we get
d
2

d
2
+
d
d
+ K
i
= K
i

s
(3.19)
The complementary function is e
m
, where
m
2
+ m + K
i
= 0 (3.20)
so that
m =
1
2
_
1
_
1 4K
i
_
(3.21)
The complete solution is
= C
1
e
m
1

+ C
2
e
m
2

+
s
(3.22)
where m
1
and m
2
are the roots from equation (3.21).
There is no oset in integral control. There are damped oscillations for K
i
> 1/4,
and no oscillations if 0 < K
i
< 1/4.
21
3.2.3 Integral-derivative control
Derivative control is usually not used alone since any constant would trigger no
response from the controller. Combining with integral control, we have
q = K
i
_

0
(
s
(

)) d

K
d
d
d
(3.23)
from which
d
d
+ = K
i
_

0
(
s
(

)) d

K
d
d
d
(3.24)
Dierentiating with respect to , we get
d
2

d
2
+
1
1 +K
d
d
d
+
K
i
1 +K
d
=
K
i
1 +K
d

s
(3.25)
This can show overdamped or underdamped oscillatory behavior, depending on the
values of the constants K
i
and K
d
.
3.2.4 PID
If all three control actions are included, we have
q = K
p
(
s
) + K
i
_

0
(
s
(

)) d

+ K
d
d
d
(3.26)
from which
d
d
+ = K
p
(
s
) + K
i
_

0
(
s
(

)) d

+ K
d
d
d
(3.27)
This can be solved as in the above examples.
The error e =
r
and control input u = Q can be used in Eq. (2.30), so that
the derivative of Eq. (3.4) gives
(K
d
+ 1)
d
2

dt
2
+ (K
p
+ 1)
d
dt
+ K
i
= K
i

r
, (3.28)
with the initial conditions
=
0
at t = 0, (3.29)
d
dt
=
K
p
+ 1
K
d
+ 1

0
+
K
p

r
K
d
+ 1
at t = 0. (3.30)
The response of the closed-loop system can be obtained as a solution. The steady-
state for K
i
,= 0 is given by =
r
. It can be appreciated that dierent choices
of the controller constants K
p
, K
i
and K
d
will give overdamped or underdamped
oscillatory or unstable behavior of the system. Fig. 3.1 shows two examples of closed-
loop behavior with dierent parameter values.
22
0 10 20 30 40 50 60 70
1.5
1
0.5
0
0.5
1
1.5

t
(a)
(b)
Figure 3.1: Lumped approximation with PID control; K
i
= K
d
= 0.1,
r
= 0.5, (a)
K
p
= 0.1, (b) K
p
= 0.9.
3.2.5 Delay
Let us apply proportional control with a time delay of , so that
q() = K
p
(
s
( )) (3.31)
The governing equation is
d
d
+ + K
p
( ) = K
p

s
(3.32)
with the solution
= Ce
m
+
K
p

s
1 +K
p
(3.33)
where m satises the transcendental equation
m+ 1 +K
p
e
m
= 0 (3.34)
For small , we nd the approximate value
m =
1 +K
p
1 K
p

(3.35)
23
m>0
m<0
K
p

Unstable
Stable
Figure 3.2: Convective cooling.
The control system is unstable if m > 0. If stable, the long time solution is
=
K
p
1 +K
p

s
(3.36)
which is the same as equation (3.13). In Fig. 3.2 the region of instability is wh ere
m > 0.
3.2.6 Variable ambient temperature
Let the ambient temperature vary sinusoidally as in
T

= T

+ T sin t (3.37)
The governing equation (3.1) can be nondimensionalized using equations (3.5), (3.2)
and (3.3), where we use T

instead of T

. The nondimensional equation is

d
d
+ = T sin + q() (3.38)
where
=
T
T
i
T

(3.39)
=
Mc
hA
(3.40)
24
Using PID control, we take the heat source to be given by equation (3.26). Equa-
tion (3.38) becomes
d
d
+ = sin + K
p
(
s
) + K
i
_

0
(
s
(

)) d

K
d
d
d
(3.41)
Dierentiating
d
2

d
2
+
1 +K
p
1 +K
d
d
d
+
K
i
1 +K
d
=

1 +K
d
cos +
K
i
1 +K
d

s
(3.42)
The complementary functions are e
m
1

and e
m
2

, where m
1
and m
2
are obtained from
m =
1
2
_
_

1 +K
p
1 +K
d

_
1 +K
p
1 +K
d
_
2

4K
i
1 +K
d
_
_
(3.43)
Appropriate values of K
p
, K
i
and K
d
will produce m
1
and m
2
with negative real parts
for which the transient disappears with time and the control system is stable. The
particular solution is
=
s
+ a cos( + ) (3.44)
where
a =

(1 +K
p
)sin + [(1 +K
d
)
2
K
i
] cos
(3.45)
tan =
(1 +K
p
)
(1 +K
p
)
2
K
i
(3.46)
For optimal control the values of the parameters can be found to satisfy certain a
priori criteria for optimality.
3.3 On-o control
This is a common form of thermal control in which the heating or cooling is turned
o at a predetermined temperature and turned on at another. Equation (3.1) for this
case is
Mc
dT
dt
+ hA(T T

) =
_
Q
0
on
0 o
(3.47)
where for the moment we take the ambient temperature, T

to be constant. The heat

rate is Q = Q
0
when the system is on and Q = 0 when it is o. With the system in its
o mode, as t , T T
min
= T

and in its on mode T T

max
= T

+ Q
0
/hA.
25
Note that with the nomenclature T
min
and T
max
, we have implicitly assumed that we
are dealing with a heater, though the analysis is also applicable to cooling; in this
case Q
0
would be negative and T
min
would be higher than T
max
.
Taking the nondimensional temperature to be
=
T T
min
T
max
T
min
(3.48)
(3.49)
and the time as in equation (3.2), the governing equation is
d
d
+ =
_
1 on
0 o
(3.50)
The solution is
=
_
1 +C
1
e

on
C
2
e

o
(3.51)
We will assume that the heat source comes on when temperature falls below
a value T
L
, and goes o when it is above T
U
. These lower and upper bounds are
nondimensionally

L
=
T
L
T
min
T
max
T
min
(3.52)

U
=
T
U
T
min
T
max
T
min
(3.53)
The result of applying this form of control is an oscillatory temperature that
looks like that in Fig. 3.3, the period and amplitude of which can be chosen using
suitable parameters. The heat source is on for a time interval
on
during which time
the system temperature goes from
L
to
U
. The heater source is then switched o
and the system goes from
U
back to
L
in time
o
. These temperature conditions
can be applied to the solution, equation (3.51).
During the on interval

L
= 1 + C
1
(3.54)

U
= 1 + C
1
e
on
(3.55)
which gives

on
= ln
1
L
1
U
(3.56)
Similarly, in the o interval

L
= C
2
(3.57)

U
= C
2
e

o
(3.58)
26

=1

U
L

on off

Figure 3.3: Response to on-o control.

from which

o
= ln

U

L
(3.59)
The total period of the oscillation is

p
=
on
+
o
(3.60)
= ln

U
(1
L
)

L
(1
U
)
(3.61)
If we make a small dead-band assumption, we can write

L
=
s
(3.62)

U
=
s
+ (3.63)
where 1. A Taylor-series expansion gives

on
= ln
1 +/(1
s
)
1 /(1
s
)
(3.64)
=
2
1
s
+ . . . (3.65)
and

o
= ln
1 +/
s
1 /
(3.66)
=
2

s
+ . . . (3.67)
27
so that

p
= 2
_
1

s
+
1
1
s
_
+ . . . (3.68)
The period is proportional to the width of the dead band.
Alternatively, the small dead-band period can be calculated by assuming a saw-
toothed (piecewise linear) shape of the temperature-time curve. In equation (3.50),
if we take =
s
, the equation approximates to
2

on
+
s
= 1 (3.69)

o
+
s
= 0 (3.70)
for the on and o periods, respectively. From this, we get

p
=
on
+
o
(3.71)
= 2
_
1
1
s
+
1

s
_
(3.72)
The complete problem can also be looked at as one with variables coecients.
Writing

=
_
/
on
on
1 + (
on
)/
o
o
(3.73)
equation (3.50) becomes
d
d

+
_

on

o
_
=
_

on
0

< 1
0 1

< 2
(3.74)
The solution is
=
_
1 (1
L
)e
on

< 1

U
e

o
(

1)
1

< 2
(3.75)
3.3.1 Variable ambient temperature
3.3.2 Delay
It is possible in real systems that the on and o operations are delayed by a certain
time interval . This is shown in Fig. 3.4 where the temperature is seen to vary
between the limits
1
and
2
instead of between
L
and
U
. The frequency and
amplitude of the oscillation is thus aected by the delay.
28

U
L

on
off

1
2

2
1
Figure 3.4: On-o control with time delay .
To determine the time period of the oscillation, let us begin with the last part of
the o period when the system goes from temperature
L
to
1
in time . From the
exponential solution for this time interval, equation (3.51), we get

L
= C
2
(3.76)

1
= C
2
e

(3.77)
from which

1
=
L
e

(3.78)
When the system is on, the change in temperature from
1
to
L
occurs in time
1
,
so that

1
= 1 +C
1
(3.79)

L
= 1 +C
1
e

1
(3.80)
From this

1
= ln
1
1
1
L
(3.81)
= ln
1
L
e

1
L
(3.82)
Again with the system on, the temperature goes from
L
to
U
in time
on
that is
given by equation (3.56). Next, goes from
U
to
2
in time . This gives

U
= 1 +C
1
(3.83)

2
= 1 +C
1
e

(3.84)
29
from which

2
= 1 (1
U
)e

(3.85)
The next time interval to consider is that when changes from
2
to
U
in time
2
.
This gives

2
= C
2
(3.86)

U
= C
2
e

2
(3.87)
so that

2
= ln

2

U
(3.88)
= ln
1 (1
U
)e

U
(3.89)
Finally, changes from
U
to
L
with the system o. This happens in time
o
that
is given by equation (3.59).
Summing up all these time intervals gives us the total period

p
=
on
+
o
+ 2 +
1
+
2
(3.90)
= ln
_

U
(1
L
)

L
(1
U
1
L
e

1
L
1 (1
U
)e

U
_
+ 2 (3.91)
which is larger due to the delay.
On the other hand the temperature excursion is given by

d
=
2

1
(3.92)
= 1 (1
U
+
L
)e

(3.93)
which has also increased.
For a small delay, i.e. with 1, we get the approximations

d
P
=
p
+
_
2 +

L
1
L
+
1
U

U
_
+ . . . (3.94)

d
= (
U

L
) + (1
U
+
L
) + . . . (3.95)
3.4 Two-uid problem
3.5 Two-body problem
Problems
1. Study the PID control system for the two-body problem in thermal contact, each of which has
a heat source that can be independently controlled. Show numerical results for the dierent
30
types of responses possible (damped, oscillatory, unstable, stable, etc.). Take the ambient
temperature, T

, to be (a) constant and (b) oscillatory.

2. Consider on-o control for the two-body problem. Show analytical or numerical results for
the temperature responses of the two bodies. If you do the problem analytically, take the
ambient temperature, T

, to be constant, but if you do it numerically, then you can take it

to be (a) constant, and (b) oscillatory.
31
Chapter 4
Conductive systems
4.1 Systems with one spatial dimension
The lumped approximation in the previous section, valid for Bi 1, is frequently
not good enough for thermal systems, and the spatial variation of the temperature
must be taken into account. The system is then described by PDEs that represent
a formidable challenge for control analysis. The simplest examples occur when only
one spatial dimension is present.
4.1.1 Fins
Fig. 4.1 shows a n of length L with convection to the surroundings [64]. It is thin
and long enough such that the transverse temperature distribution may be neglected.
The temperature eld is governed by
T
t
=

2
T

2
(T T

), (4.1)
where T(, t) is the temperature distribution that represents the state of the system,
T

is the temperature of the surroundings, t is time, and is the longitudinal coor-

dinate measured from one end. The thermal diusivity is , and = hP/cA
c
where
h is the convective heat transfer coecient, A
c
is the constant cross- sectional area of
the bar, P is the perimeter of the cross section, is the density, and c is the specic
heat. For simplicity it will be assumed that is independent of . The end = 0 will
be assumed to be adiabatic so that (T/)(0, t) = 0.
Since a linear system that is controllable can be taken from any state to any other,
we can arbitrarily assume the n to be initially at a uniform temperature. There
are two ways in which the temperature distribution on the bar can be controlled:
32

E
= 0 = L
T
L
T

Figure 4.1: One-dimensional n with convection.

in distributed control
1
the surrounding temperature T

is the control input and in

boundary control it is the temperature of the other end T(L, t) of the n.
(a) Distributed control: The boundary temperature T(L, t) = T
L
is xed. Using it as
a reference temperature and dening = T T
L
, Eq. (4.1) becomes,

t
=

2
+

(t) (4.2)
with the homogeneous boundary and initial conditions (/)(0, t) = 0, (L, t) = 0,
and (, 0) = 0.
The operators in Eq. (2.13) are / =
2
/
2
, B = , and u =

. / is a
self-adjoint operator with the eigenvalues and eigenfunctions

m
=
(2m+ 1)
2

2
4L
2
,

m
=
_
2
L
cos
(2m+ 1)
2L
,
respectively. Inequality (2.14) is satised for all m, so the system is indeed state
controllable. It can be shown that the same problem can also be analyzed using a
nite-dierence approximation [65].
(b) Boundary control: Using the constant outside temperature T

as reference and
dening = T T

, Eq. (4.1) becomes

t
=

2
, (4.3)
with the initial and boundary conditions (/)(0, t) = 0, (L, t) = T
L
(t) T

, and
(, 0) = 0.
To enable a nite-dierence approximation [65], the domain [0, L] is divided into
n equal parts of size , so that Eq. (4.3) becomes
d
i
dt
=
i1
(2 + )
i
+
i1
, (4.4)
1
This term is also used in other senses in control theory.
33
where = /
2
. The nodes are i = 1, 2, . . . , n + 1, where i = 1 is at the left and
i = n+1 at the right end of the n in Fig. 4.1. With this Eq. (4.3) can be discretized
to take the form of Eq. (2.3), where x is the vector of unknown
i
. Thus we nd
A =
_

_
(2 + ) 2 0 0
(2 + )
.
.
.
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 (2 + )
_

_
R
nn
, (4.5)
B = [0, , ]
T
R
n
. (4.6)
The boundary conditions have been applied to make A non-singular: at the left end
the n is adiabatic, and at the right end
n+1
is the control input u.
The controllability matrix M is
M =
_

_
0 0
n
0 0
n1

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0
3

0
2
2
2
(2 + )
(2 + )
3
+ (2 + )
2

_

_
(4.7)
The rank of M is n, indicating that the state of the system is also boundary control-
lable.
4.2 Stefan problem
Problems
34
Chapter 5
Natural convective systems
5.1 Natural convection loop
Much of the work in this area has been done by Bau and co-workers [6670].
5.2 Deterministic chaos
5.3 Two-dimensional cavity
Stabilization and control of convection in horizontal uid layers has been investigated
in [7180].
5.4 Porous media
Porous media have been studied in [81].
Problems
35
Chapter 6
Forced convective systems
6.1 One-dimensional
6.1.1 Long duct
The diusion problem of the previous section does not have advection. Transport of
uids in ducts introduces a delay between the instant the particles of uid go into the
duct and when they come out, which creates a diculty for outlet temperature control.
The literature includes applications to hot-water systems [82,83] and buildings [84,85];
transport [86] and heater [87, 88] delay and the eect of the length of a duct on
delay [89] have also been looked at.
A long duct of constant cross section, schematically shown in Fig. 6.1 where the
ow is driven by a variable-speed pump, illustrates the basic issues [64, 90]. The uid
inlet temperature T
in
is kept constant, and there is heat loss to the constant ambient
temperature T

through the surface of the duct.

With a one-dimensional approximation, energy conservation gives
T
t
+ v
T

+
4h
cD
(T T

) = 0, (6.1)
with the boundary condition T(0, t) = T
in
, where T(, t) is the uid temperature, t
is time, is the distance along the duct measured from the entrance, v(t) is the ow
E
= ow
velocity = v(t)
temperature = T(, t)

= 0 = L
T
in
T
out
(t)
T

Figure 6.1: Schematic of duct.

36
velocity, h is the coecient of heat transfer to the exterior, is the uid density, c
is its specic heat, and D is the hydraulic diameter of the duct. The ow velocity is
taken to be always positive, so that the = 0 end is always the inlet and = L the
outlet, where L is the length of the duct. The temperature of the uid coming out of
the duct is T
out
(t).
Using the characteristic quantities of L for length, cD/4h for time, and hL/cD
for velocity, the non-dimensional version of Eq. (6.1) is

t
+ v

+ = 0, (6.2)
where = (T T

)/(T
in
T

), with (0, t) = 1. The other variables are now

non-dimensional. Knowing v(t), this can be solved to give
(, t) = e
t
f
_

_
t
0
v(s) ds
_
, (6.3)
where the initial startup interval in which the uid within the duct is ushed out has
been ignored; f is an arbitrary function. Applying the boundary condition at = 0
gives
1 = e
t
f
_

_
t
0
v(s) ds
_
. (6.4)
The temperature at the outlet of the duct, i.e. at = 1, is

out
(t) = e
t
f
_
1
_
t
0
v(s). ds
_
(6.5)
Eqs. (6.4) and (6.5) must be simultaneously solved to get the outlet temperature

out
(t) in terms of the ow velocity v.
The problem is non-linear if the outlet temperature T
out
(t) is used to control the
ow velocity v(t). The delay between the velocity change and its eect on the outlet
temperature can often lead instability, as it does in other applications [9194]. Fig.
6.2 shows a typical result using PID control in which the system is unstable. Shown
are the outlet temperature, ow velocity and residence time of the uid in the duct,
all of which ultimately achieve constant amplitude oscillations.
6.1.2 Heat exchanger model
A cross-ow heat exchanger model, schematically shown in Fig. 6.3, has been studied
using nite dierences [64, 65, 95]. Water is the in-tube and air the over-tube uid in
37
0 50 100 150
0.2
0.3
0.4
0.5
T
o
u
t
0 50 100 150
0.5
1
1.5
v
0 50 100 150
0.6
0.8
1
1.2
1.4
t

Figure 6.2: Outlet temperature, velocity and residence time for K

i
= 5 and K
p
= 2.5
[64].
the heat exchanger. This example includes all the conductive, advective and convec-
tive eects discussed before. The governing equations on the outside of the tube, in
the water, and in the wall of the tube are
m
a
L
c
a
(T
a
in
T
a
out
) = h
o
2r
o
(T
a
T
t
), (6.6)

w
c
w
r
2
i
T
w
t
+ m
w
c
w
T
w

= h
i
2r
i
(T
t
T
w
), (6.7)

t
c
t
(r
2
o
r
2
i
)
T
t
t
= k
t
(r
2
o
r
2
i
)

2
T
t

2
+ 2r
o
h
o
(T
a
T
t
) 2r
i
h
i
(T
t
T
w
), (6.8)
respectively. L is the length of the tube; m
a
(t) and m
w
(t) are the mass ow rates
of air and water; T
a
in
and T
a
out
(t) are the inlet and outlet air temperatures; T
a
(t)
is the air temperature surrounding the tube; T
t
(, t) and T
w
(, t) are the tube-wall
and water temperatures; h
i
and h
o
the heat transfer coecients in the inner and
outer surfaces of the tube; r
i
and r
o
are the inner and outer radii of the tube; c
a
,
c
w
and c
t
are the specic heats of the air, water and tube material;
w
and
t
are
the water and tube material densities; and k
t
is the thermal conductivity of the tube
material. In addition, the air temperature is assumed to be T
a
= (T
a
in
+T
a
out
)/2. The
boundary and initial conditions are T
t
(0, t) = T
w
(0, t) = T
w
in
, T
t
(L, t) = T
w
(L.t), and
T
t
(, 0) = T
w
(, 0). Suitable numerical values were assumed for the computations.
38
Water in
Air in

L
Water out
Air out
Figure 6.3: Schematic of single-tube cross-ow heat exchanger.
The inlet temperatures T
a
in
and T
w
in
, and the ow rates m
a
and m
w
can all be used
as control inputs to obtain a desired outlet temperature, T
a
out
or T
w
out
. The ow rates
present a special diculty; they appear in nonlinear form in Eqs. (6.6) and (6.7), and
the outlet temperature is bounded. Fig. 6.4 shows the steady-state range of values
of T
w
out
that can be achieved on varying m
w
; temperatures outside this range cannot
be obtained. It is also seen that the outlet water temperature is hard to control for
large water ow rates. As an example of control dynamics, Fig. 6.5 shows the results
of applying PI control on m
w
to obtain a given reference temperature T
w
out
= 23

C.
6.2 Potential ow
6.3 Viscous ow
6.4 Turbulence
Problems
39
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
22.9
23
23.1
23.2
23.3
23.4
23.5
23.6
T




[

C
]
a
o
u
t
[kg/s] m
a
.
= 0.23 kg/s
m
a
.
0.17 kg/s
Figure 6.4: The relation between T
a
out
and m
w
for dierent m
a
[64].
0 200 400 600 800 1000 1200 1400 1600 1800 2000
21.5
22
22.5
23
23.5
24
24.5
25
t [s]

5
T




[

C
]
a
o
u
t
K = 100,000
p
10,000
1,000
100
Figure 6.5: Behavior of T
a
out
as a function of time for K
i
= 50 and dierent K
p
[64].
40
Chapter 7
Radiative systems
One example is in [96].
7.1 Radiative n
7.2 Radiative plate
7.3 Furnace
Problems
41
Chapter 8
Hardware
8.1 Sensors
8.1.1 Thermocouples
8.1.2 Thermistors
8.1.3 MEMS
8.2 Actuators
8.2.1 Furnace
8.2.2 Fans
8.3 Microprocessors
Problems
42
Chapter 9
Concluding remarks
There are a large number of other thermal problems in which control theory has been
applied. Agent-based controls have been proposed by complex thermal systems such
as in buildings [97], microwave heating [98], and materials processing and manufac-
turing [99, 100].
This has been a very brief introduction to the theory of thermal control. The
fundamental ideas in this subject are rmly grounded on the mathematics of systems
and control theory which should be the starting point. There are, however, a few
aspects that are particularly characteristic of thermal systems. Phenomena such as
diusion, convection and advection are common and the systems are usually complex,
nonlinear and poorly predicted dynamically. The governing equations cover a wide
range of possibilities, from ordinary and partial dierential equations to functional
and dierential-algebraic systems.
The study of thermal control will continue to grow from the point of view of
fundamentals as well as engineering applications. There are many outstanding prob-
lems and issues that need to be addressed. To cite one specic example, networking
between a large number of coupled components will become increasingly important; it
is known that unexpected synchronization may result even when multiple dynamical
systems are coupled weakly [101]. It is hoped that the reader will use this as a starting
point for further study and apply control theory in other thermal applications.
43
References
[1] G. Huang, L. Nie, Y. Zhao, W. Yang, Q. Wu, and L. Liu. Temperature control
system of heat exchangers, an application of DPS theory. Lecture Notes in
Control Information Sciences, 159:6876, 1991.
[2] J.R. Leigh. Temperature Measurement and Control. Peregrinus, London, 1987.
[3] Z. Zhang and R. M. Nelson. Parametric analysis of a building space conditioned
by a VAV system. ASHRAE Transactions, 98(1):4348, 1992.
[4] P. Riederer, D. Marchio, and J.C. Visier. Inuence of sensor position in building
thermal control: criteria for zone models. Energy and Buildings, 34(6):785798,
2002.
[5] E.G. Hansen. Hydronic System Design and Operation. McGraw-Hill, New York,
1985.
[6] M.F. Hordeski. HVAC Control in the New Millennium. Fairmont Press, Liburn,
GA, 2001.
[7] J.I. Levenhagen and D.H. Spethmann. HVAC Controls and Systems. Mc-Graw-
Hill, New York, 1993.
[8] L. Hach and Y. Katoh. Thermal responses in control loop in indirect control of
indoor environment of non-air-conditioned space with quasi-steady-state model.
JSME International Journal Series C-Mechanical Systems Machine Elements
And Manufacturing, 46(1):197211, 2003.
[9] S. Zumbo, J. Leofanti, S. Corradi, G. Allegri, and M. Marchetti. Design of a
small deployable satellite. Acta Astronautica, 53(4-10):533540, 2003.
[10] K. Shimazaki, A. Ohnishi, and Y. Nagasaka. Development of spectral selec-
tive multilayer lm for a variable emittance device and its radiation properties
measurements. International Journal of Thermophysics, 24(3):757769, 2003.
[11] T.D. Swanson and G.C. Birur. NASA thermal control technologies for robotic
spacecraft. Applied Thermal Engineering, 23(9):10551065, 2003.
[12] R.D. Karam. Satellite Thermal Control for Systems Engineers. AIAA, Reston,
VA, 1998.
[13] D.C. Price. A review of selected thermal management solutions for military
electronic systems. IEEE Transactions on Components and Packaging Tech-
44
nologies, 26(1):2639, 2003.
[14] M. Sweetland and J.H. Lienhard. Active thermal control of distributed param-
eter systems with application to testing of packaged IC devices. ASME Journal
of Heat Transfer, 125(1):164174, 2003.
[15] C. Doumanidis and N. Fourligkas. Distributed-parameter control of the heat
source trajectory in thermal materials processing. ASME Journal of Manufac-
turing Science and Engineering, 118(4):571578, 1996.
[16] V. Trivedi and S.J. Pearton. Evaluation of rapid thermal processing systems
for use in CMOS fabrication. Solid-State Electronics, 46(5):777783, 2002.
[17] P.K. Roy, S.M. Merchant, and S. Kaushal. A review: thermal processing in fast
ramp furnaces. Journal of Electronic Materials, 30(12):15781583, 2001.
[18] C.J. Huang, C.C. Yu, and S.H. Shen. Selection of measurement locations for
the control of rapid thermal processor. Automatica, 36(5):705715, 2000.
[19] P. Elgar. Sensors for Measurement and Control. Longman, Essex, U.K., 1998.
[20] M. Tabib-Azar. Microactuators: Electrical, Magnetic, Thermal, Optical, Me-
chanical, Chemical & Smart Structures. Kluwer Academic, Boston, 1998.
[21] W.F. Stoecker and P.A. Stoecker. Microcomputer Control of Mechanical and
Thermal Systems. Van Nostrand Reinhold, New York, 1989.
[22] D. Ibrahim. Microcontroller-Based Temperature Monitoring and Control.
Newnes, Oxford, 2002.
[23] M. Gad-el-Hak. Flow Control: Passive, Active and Reactive Flow Management.
Cambridge University Press, Cambridge, U.K., 2000.
[24] W.S. Levine, editor. Control System Fundamentals. CRC Press, Boca Raton,
FL, 2000.
[25] E.O. Roxin. Control Theory and its Applications. Gordon and Breach, Ams-
terdam, 1997.
[26] K. Ogata. Modern Control Engineering. Prentice-Hall, Englewood Clis, N.J.,
second edition, 1990.
[27] P.N. Paraskevopoulos. Modern Control Engineering. Marcel Dekker, New York,
2002.
[28] E.F. Camacho and C. Bordons. Model Predictive Control in the Process Indus-
try. Springer, London, 2001.
[29] O.L.R. Jacobs. Introduction to Control Theory. 2nd Ed., Oxford University
Press, Oxford, 1993.
[30] W. H. Ray. Advanced Process Control. McGraw-Hill, New York, 1981.
[31] M.A. Henson and D.E. Seborg, editors. Nonlinear Process Control. Prentice
Hall, Upper Saddle River, NJ, 1997.
[32] A. Isidori. Nonlinear Control Systems. Springer, London, 1995.
[33] P.D. Christodes and Dautidis P. Feed-back control of hyperbolic PDE systems.
AIChE Journal, 412:3063, 1978.
45
[34] L. Irena and R. Triggiani. Deterministic Control Theory for Innite Dimen-
sional Systems, Vols. I and II Encyclopedia of Mathematics. Cambridge Uni-
versity Press, 1999.
[35] E.D. Sontag. Mathematical Control Theory. Springer, New York, 1998.
[36] P.J. Antsaklis and A.N. Michel. Linear Systems. McGraw-Hill, New York, 1997.
[37] F.P. Incropera and D.P. DeWitt. Fundamentals of Heat and Mass Transfer, 4th
Ed. Wiley, New York, 1996.
[38] G.F. Franklin, J.D. Powell, and A. Emami-Naeini. Feedback Control of Dynamic
Systems, 3rd Ed. Addison-Wesley, Reading, MA, 1994.
[39] M.A. Henson and D.E. Seborg. Nonlinear Process Control. Prentice Hall, Upper
Saddle River, NJ, 1997.
[40] O.L.R. Jacobs. Introduction to Control Theory. Oxford University Press, Ox-
ford, U.K.,, 1993.
[41] H.H. Rosenbrock. State-Space and Multivariable Theory. John Wiley, New
York, 1970.
[42] A. Kumar and P. Daoutidis. Control of Nonlinear Dierential Algebraic Equa-
tion Systems. Chapman and Hall/CRC, Boca Raton, FL, 1999.
[43] L. Dai. Singular Control Systems. Springer-Verlag, Berlin, 1989.
[44] N.U. Ahmed. Semigroup Theory with Applications to Systems and Control.
Longman, Harlow, U.K., 1991.
[45] J. Klamka. Controllability of Dynamical Systems. Kluwer, Dordrecht, Nether-
lands, 1991.
[46] J.L. Lions. On the controllability of distributed systems. Natl. Acad. Sci. USA,
94:48284835, 1997.
[47] R. Acosta, M. Sen, and E. Ramos. Single-phase natural circulation in a tilted
square loop. W arme- und Sto ubertragung, 21(5):269275, 1987.
[48] L. Ljung. System Identication: Theory for the User. Prentice-Hall, Upper
Saddle River, N.J., second edition, 1999.
[49] O. Nelles. Nonlinear System Identication. Springer, Berlin, 2001.
[50] A. Pacheco-Vega, M. Sen, K.T. Yang, and R.L. McClain. Genetic-algorithm
based prediction of a n-tube heat exchanger performance. Proceedings of the
11th International Heat Transfer Conference, 6:137142, 1998.
[51] J.K. Hale and S.M. Lunel. Introduction to Functional Dierential Equations.
Springer-Verlag, New York, 1993.
[52] R. Bellman and K.L. Cooke. Dierential-Dierence Equations. Academic Press,
New York, 1963.
[53] R.D. Driver. Ordinary and Delay Dierential Equations. Springer-Verlag, New
York, 1977.
[54] A. Oustaloup. Syst`emes Asservis Lineaires dOrdre Fractionnaire: Theorie et
Practique. Masson, Paris, 1983.
46
[55] A. Oustaloup. La Derivation Non Enti`ere. Herm`es, Paris, 1995.
[56] R. Arena, R. Caponetto, L. Fortuna, and D. Porto. Nonlinear Noninteger Order
Circuits and SystemsAn Introduction. World Scientic, Singapore, 2000.
[57] I. Podlubny. Fractional Dierential Equations. Academic Press, San Diego,
CA, 1999.
[58] J.L. Battaglia, L. Le Lay, J.C. Batsale, A. Oustaloup, and O. Cois. Heat ux
estimation through inverted non-integer identication models. International
Journal of Thermal Sciences, 39(3):374389, 2000.
[59] C.-C. Yu. Autotuning of PID Controllers. Springer, London, 1999.
[60] K.-H. Homan and W. Krabs. Optimal Control of Partial Dierential Equa-
tions. Springer-Verlag, Berlin, 1991.
[61] P.D. Christodes. Nonlinear and Robust Control of PDE Systems. Birkhauser,
Boston, 2002.
[62] G. Chen and G. Chen. Linear Stochastic Control Systems. CRC Press, Boca
Raton, 1995.
[63] M. Sen. Lecture Notes on Intelligent Systems. Department of Aerospace and
Mechanical Engineering, University of Notre Dame, Notre Dame, IN 46556,
2004.
[64] S. Alotaibi. Temperature Controllability in Cross-Flow Heat Exchangers and
Long Ducts. PhD thesis, Department of Aerospace and Mechanical Engineering,
University of Notre Dame, Notre Dame, IN, 2003.
[65] S. Alotaibi, W.J. Goodwine M. Sen, and K.T. Yang. Controllability of cross-
ow heat exchangers. To be published in International Journal of Heat and
Mass Transfer, 2004.
[66] P.K. Yuen and H.H. Bau. Optimal and adaptive control of chaotic convection
- theory and experiments. Physics of Fluids, 11(6):14351448, 1999.
[67] P.K. Yuen and H.H. Bau. Rendering a subcritical Hopf bifurcation supercritical.
Journal of Fluid Mechanics, 317:91109, 1996.
[68] Y.Z. Wang, J. Singer, and H.H. Bau. Controlling chaos in a thermal-convection
loop. Journal of Fluid Mechanics, 237:479498, 1992.
[69] J. Singer and H.H. Bau. Active control of convection. Physics of Fluids A-Fluid
Dynamics, 3(12):28592865, 1991.
[70] J. Singer, Y.Z. Wang, and H.H. Bau. Controlling a chaotic system. Physical
Review Letters, 66(9):11231125, 1991.
[71] H.H. Bau. Control of Marangoni-Benard convection. International Journal of
Heat and Mass Transfer, 42(7):13271341, 1999.
[72] J. Tang and H.H. Bau. Numerical investigation of the stabilization of the no-
motion state of a uid layer heated from below and cooled from above. Physics
of Fluids, 10(7):15971610, 1998.
[73] J. Tang and H.H. Bau. Experiments on the stabilization of the no-motion state
47
of a uid layer heated from below and cooled from above. Journal of Fluid
Mechanics, 363:153171, 1998.
[74] J. Tang and H.H. Bau. Stabilization of the no-motion state of a horizontal uid
layer heated from below with Joule heating. ASME Journal of Heat Transfer,
117(2):329333, 1995.
[75] J. Tang and H.H. Bau. Stabilization of the no-motion state in the Rayleigh-
Benard problem. Proceedings of the Royal Society of London Series A-
Mathematical and Physical Sciences, 447(1931):587607, 1994.
[76] J. Tang and H.H. Bau. Stabilization of the no-motion state in Rayleigh-
Benard convection through the use of feedback-control. Physical Review Letters,
70(12):17951798, 1993.
[77] J. Tang and H.H. Bau. Numerical investigation of the stabilization of the no-
motion state of a uid layer heated from below and cooled from above. Physics
of Fluids, 10(7):15971610, 1998.
[78] H.M. Park and W.J. Lee. Feedback control of the Rayleigh-Benard convection
by means of mode reduction. International Journal for Numerical Methods in
Fluids, 40(7):927949, 2002.
[79] T. Marimbordes, A.O. El Moctar, and H. Peerhossaini. Active control of natural
convection in a uid layer with volume heat dissipation. International Journal
of Heat and Mass Transfer, 45(3):667678, 2002.
[80] L.E. Howle. Control of Rayleigh-Benard convection in a small aspect ratio
container. International Journal of Heat and Mass Transfer, 40(4):817822,
1997.
[81] J. Tang and H.H. Bau. Feedback-control stabilization of the no-motion state
of a uid conned in a horizontal porous layer heated from below. Journal of
Fluid Mechanics, 257:485505, 1993.
[82] W. Munk. The delayed hot-water problem, brief notes. ASME J. Applied
Mechanics, 21(10), 1954.
[83] C. Comstock. On the delayed hot water problem. Heat and Mass Transfer,
96:166171, 1974.
[84] N. Saman and H. Mahdi. Analysis of the delay hot/cold water problem. Energy,
21(5):395400, 1996.
[85] K.A. Antonopoulos and C. Tzivanidis. A correlation for the thermal delay of
building. Renewable Energy, 6(7):687699, 1995.
[86] C. Tin-Tai. Numerical modeling of thermal behavior of uid conduit ow with
transport delay. ASHRAE Transactions, 102(2), 1996.
[87] J. Chu, Su H., and x. Hu. A time-delay control algorithm for an industrial
electrical heater. J. Proc. Cont., 3(4):220224, 1993.
[88] J. Chu. Application of a discrete optimal tracking controller to an industrial
electrical heater with pure delay. J. Proc. Cont., 5(1):38, 1995.
48
[89] D. Daz, M. Sen, and K.T. Yang. Eect of delay in thermal systems with long
ducts. To be published in International Journal of Thermal Sciences, 2004.
[90] S. Alotaibi, M. Sen, W.J. Goodwine, and K.T. Yang. Flow-based control of
temperature in long ducts. In preparation.
[91] C Baruch and S. Darrell. On stability of systems of delay dierential equations.
J. Comp. Appl. Math., 117:137158, 2000.
[92] J. S. Guillermo, D. Aniruddha, and S.P. Bhattacharyya. PI stabilization of
rst-order systems with time delay. Automatica, 37:20202031, 2001.
[93] N. Silviu-lulian. Delay Eects on Stability. Lecture Notes in Control and Infor-
mation Sciences, Springer, London,, 2001.
[94] G. Henryk, G. Piotr, and K. Adam. Analysis and Synthesis of Time Delay
Systems. John Wiley Sons, Chichester, 1989.
[95] S. Alotaibi, M. Sen, J.W. Goodwine, and K.T. Yang. Numerical simulation of
the thermal control of heat exchangers. Numerical Heat Transfer A, 41:229244,
2002.
[96] M.V. Papalexandris and M.H. Milman. Active control and parameter updating
techniques for nonlinear thermal network models. Computational Mechanics,
27(1):1122, 2001.
[97] K.T. Yang and M. Sen. Agent networks for intelligent dynamic control of com-
plex hydronic HVAC building systemsPart 1: Framework for agent network
development. International Journal on Architectural Science, 3(1):4350, 2002.
[98] B. Liu and T.R. Marchant. The occurrence of limit-cycles during feedback
control of microwave heating. Mathematical and Computer Modelling, 35(9-
10):10951118, 2002.
[99] C. Doumanidis and N. Fourligkas. Distributed-parameter control of the heat
source trajectory in thermal materials processing. ASME Journal of Manufac-
turing Science and Engineering, 118(4):571578, 1996.
[100] C.D. Schaper, M.M. Moslehi, K.C. Saraswat, and T. Kailath. Modeling, identi-
cation, and control of rapid thermal-processing systems. Journal of the Elec-
trochemical Society, 141(11):32003209, 1994.
[101] S.H. Strogatz. Sync: The Emerging Science of Spontaneous Order. Theia, New
York, 2003.
49
Index
actuators, 6, 42
advective, 38
algebraic-dierential equations, 11
ambient temperature, variable, 24, 28
approximation, one-dimensional, 36
articial neural networks, 17
bifurcations, 13
boundary control, 33
bounded variables, 13
cavity, two-dimensional, 35
conductive, 38
control strategies, 17
control theory, 8
control, discrete-time, 6
control, manual, 5
control, agent-based, 43
control, automatic, 5
control, ow, 6
control, integral, 21
control, on-o, 25
control, PI, 22
control, PID, 20, 22
controllability, 13
controllability matrix, 34
convection, natural, 35
convective, 38
cooling, 19
delay, 14, 23, 28, 36
deterministic chaos, 35
dierence equation, 14
distributed control, 33
estimation theory, 15
fans, 42
feedback control, 15
n, 41
ow, potential, 39
ow, viscous, 39
fractional-order derivaives, 14
functional equation, 14
furnace, 41, 42
fuzzy logic control, 17
heat exchanger, 37
heating, 19
hysteresis, 13
Kalman lters, 15
Laplace transform, 11
linear systems theory, 9
long duct, 36
loop, natural convective, 35
lumped parameters, 19
MEMS, 42
microprocessors, 6, 42
nondimensionalization, 17
nonlinear eects, 12
open loop control, 8
open-loop control, 15
ordinary dierential equations, 9
50
partial dierential equations, 11
porous media, 35
radiation, 41
reachability, 13
relay, 13
scaling, 17
sensors, 6, 42
soft computing, 17
system identication, 14
systems theory, 8
systems, complex, 5
thermal system, 8
thermistors, 42
thermocouples, 42
thermosyphon, 35
turbulence, 39
two-body, 30
two-uid, 30
Volterra model, 14
51