Int. J. Appl. Math. Comput. Sci., 2013, Vol. 23, No.

2, 291–307
DOI: 10.2478/amcs-2013-0022

A GENERAL TRANSFER FUNCTION REPRESENTATION FOR A CLASS

OF HYPERBOLIC DISTRIBUTED PARAMETER SYSTEMS

K RZYSZTOF BARTECKI

Institute of Control and Computer Engineering

Opole University of Technology, ul. Sosnkowskiego 31, 45-272 Opole, Poland
e-mail: k.bartecki@po.opole.pl

Results of transfer function analysis for a class of distributed parameter systems described by dissipative hyperbolic partial
differential equations defined on a one-dimensional spatial domain are presented. For the case of two boundary inputs, the
closed-form expressions for the individual elements of the 2×2 transfer function matrix are derived both in the exponential
and in the hyperbolic form, based on the decoupled canonical representation of the system. Some important properties of
the transfer functions considered are pointed out based on the existing results of semigroup theory. The influence of the
location of the boundary inputs on the transfer function representation is demonstrated. The pole-zero as well as frequency
response analyses are also performed. The discussion is illustrated with a practical example of a shell and tube heat
exchanger operating in parallel- and countercurrent-flow modes.

Keywords: distributed parameter system, hyperbolic system, partial differential equation, transfer function, heat exchanger.

1. Introduction conduction and diffusion problems), hyperbolic type

(representing convection, advection and wave propagation
Distributed Parameter Systems (DPSs) represent a large phenomena) or elliptic type (describing steady-state
class of dynamical systems in which process variables physical phenomena, e.g., electrostatic, magnetostatic
depend not only on time but also on the spatial variables, or gravitational fields). Mathematical models of DPSs
which express the geometry of the system. In practice, obtained on the basis of PDEs are described in an
almost all industrial processes fall into this category, infinite-dimensional state space, usually in a Hilbert
while the existence of the so-called Lumped Parameter space, and their transfer functions have the form of
Systems (LPSs) results mainly from the adoption of a irrational functions, as opposed to the rational ones
simplified model of the reality, in which spatial effects describing the dynamic properties of LPSs (Callier and
are neglected or averaged. Typical examples of a Winkin, 1993; Rabenstein, 1999; Grabowski and Callier,
DPS include heat transfer and fluid flow phenomena, 2001a; Zwart, 2004; Curtain and Morris, 2009).
as well as processes occurring in chemical reactors, Comparative analyses of rational and irrational
semiconductor manufacturing, polymer processing, air transfer functions show many essential differences
pollution monitoring and many others (Friedly, 1972; Wu between them, and one of the most important concerns
and Liou, 2001; Bounit, 2003; Bartecki, 2007; 2009; the number of their poles and zeros. In the case of
Zavala-Rio et al., 2009; Li and Qi 2010; Ucinski, 2012; rational transfer functions, the number of poles and zeros
Patan, 2012). is finite and corresponds to the number of polynomial
Based on the phenomenological models of the roots appearing in the numerator and the denominator
processes, established usually on the basis of the of the transfer function, respectively. On the other
mass or energy conservation balance laws, one obtains hand, irrational transfer functions, due to the presence of
their mathematical description, mostly in the form of functions which are periodic for the imaginary argument
Partial Differential Equations (PDEs) (Evans, 1998; (as, e.g., exponential or hyperbolic ones), may have an
Mattheij et al., 2005). Depending on the nature of infinite number of poles and/or zeros. However, some
the modeled phenomena, the equations can be of the of them have neither poles nor zeros, as is in the case of
parabolic type (which are typical for unsteady heat the well-known transfer function of a time delay system,
 292 K. Bartecki

G(s) = e−τ s . Another striking difference concerns the so-called Callier–Desoer algebra, for which the Nyquist
behavior of the transfer functions in the limit cases, e.g., theorem provides necessary and sufficient conditions for
when the argument s tends to infinity. In the case of proper the input-output closed-loop stability. The problems of
rational functions the limit is zero, while for irrational robust control of hyperbolic PDE systems have been
ones the limits at the infinity may depend on the chosen studied extensively, e.g., by Christofides and Daoutidis
direction on the complex plane. For example, in the case (1998a; 1998b) or Litrico et al. (2007). Interesting results
of the above-mentioned transfer function G(s) = e−τ s , the on the stability and well-posedness analysis of the class
limit at infinity calculated along the real axis equals zero, of systems considered were presented recently, e.g., by
while calculated along the imaginary one does not exist Xu and Sallet (2002), Bounit (2003), Coron et al. (2007),
due to the periodicity of the exponential function for the Dos Santos et al. (2008), Chentouf and Wang (2009),
imaginary argument (Curtain and Morris, 2009). Diagne et al. (2012), as well as Jacob and Zwart (2012).
The knowledge of transfer functions enables the The structure of the article is as follows. After
design of control systems using frequency response Introduction, Section 2 reviews the mathematical model
techniques, and this statement refers both to the LPS and of the discussed class of DPSs in the form of a set of
the DPS (Jovanović and Bamieh, 2006). However, due PDEs and formulates its hyperbolicity conditions. Next,
to the mathematical complexity and the above-mentioned the analysis is focused on the systems with two distributed
peculiarities of irrational transfer functions, their analysis state variables and boundary-type control. Based on
is more difficult, and possible applications more limited the semigroup approach, some known theoretical results
than in the case of finite-dimensional models. Therefore, of well-posedness and stability for the given class of
in order to enable the implementation of the techniques dissipative hyperbolic systems are recalled. Section 3
developed over the years and still commonly used for starts with a discussion of some known properties of
the synthesis of control systems, infinite-dimensional DPS the transfer functions of DPSs with boundary inputs.
models are usually replaced by their finite-dimensional The closed-form expressions for the individual elements
approximations (Filbet and Shu, 2005; Contou-Carrere of the 2×2 transfer function matrix are derived based
and Daoutidis, 2008; Ding et al., 2009; Jones and on the decoupled canonical system of two hyperbolic
Kerrigan, 2010). These approximations often require a PDEs. In Section 4, a shell and tube heat exchanger
further reduction using, for example, techniques based on operating in parallel- and countercurrent-flow modes is
the Principal Component Analysis (PCA) (Park and Cho, presented as a typical example of the hyperbolic DPS
1996; Li and Qi, 2010; Bartecki, 2011; 2012a; 2012b). with boundary inputs considered. Based on the derived
However, regardless of the approximation method used, transfer functions, selected frequency responses of the
the starting point for the synthesis of a control system heat exchanger are also presented here. The article
should be based on a possibly accurate description of the concludes with a summary of new results and directions
DPS, taking into account its infinite-dimensional nature, for further research.
e.g., a model in the form of an irrational transfer function.
The paper presents the results of transfer function 2. Hyperbolic systems
analysis for a certain class of DPSs in which the mass,
heat and energy transport phenomena occur. This class 2.1. General case. Many of the above-mentioned
of systems, among which one can mention, e.g., heat DPSs can be described, after possible linearization in a
exchangers, transport pipelines, irrigation channels or given operating point, by the following system of linear
electrical transmission lines, is usually described by a homogeneous PDEs of the first order (see Christofides
PDE of the hyperbolic type and known under the common and Daoutidis, 1998b; Evans, 1998; Strikwerda, 2004;
name of the hyperbolic systems of conservation laws Mattheij et al., 2005; Chentouf and Wang, 2009; Diagne
(Friedly, 1972; Delnero et al., 2004; Bounit, 2003; et al., 2012):
Kowalewski, 2009; Bartecki and Rojek, 2005; Bartecki,
∂w (l, t) ∂w (l, t)
2007; 2009). Their dynamical properties have been E +F = W w (l, t) , (1)
∂t ∂l
intensively studied in the literature in recent years. For
example, Zwart et al. (2010) prove that hyperbolic where w(l, t) : Q → Rn is a vector function representing
boundary control systems are well-posed if and only the spatio-temporal distribution of the n state variables
if the state operator generates a strongly continuous
 T
one-parameter semigroup. Furthermore, they show that w(l, t) = w1 (l, t) w2 (l, t) . . . wn (l, t) , (2)
the corresponding transfer function is regular, i.e., has a
limit for the argument s going to infinity. Litrico and defined on a set Q = Ω × Θ, where Ω = [0, L] ⊂ R is the
Fromion (2009a; 2009b) use a frequency domain method domain of the spatial variable l, Θ = [0, +∞) ⊂ R is the
for the boundary control of hyperbolic conservation laws domain of the time variable t, and E, F, W ∈ Rn×n are
and show that the resulting transfer functions belong to the matrices with constant entries.
A general transfer function representation for a class of hyperbolic distributed parameter systems 293

Assuming that for det(E) = 0 and det(F ) = 0 there Remark 2. If the matrix F E −1 is symmetric, then the
exists a non-singular transformation matrix S ∈ Rn×n system (1) is hyperbolic. In the case of hyperbolic PDEs
such that the following equation holds: describing physical phenomena mentioned in Section 1,
its eigenvalues usually represent the wave propagation
S −1 F E −1 S = Λ, (3)
velocities (as for the electrical transmission line) or mass
where Λ is a diagonal matrix, pre-multiplying both the and energy transport rates (as for the shell and tube heat
sides of (1) by S −1 and using the following identity exchanger).
E −1 SS −1 E ≡ I, (4)
2.2. Initial and boundary conditions. In order to
Eqn. (1) can be transformed into the form obtain a unique solution of (7), one must specify the
∂w (l, t) ∂w (l, t) appropriate initial and boundary conditions. The initial
S −1 E + S −1 F E −1 SS −1 E conditions represent the initial (i.e., determined for t = 0)
∂t ∂l (5)
distribution of the values of all n state variables for the
= S −1 W w (l, t) .
whole set Ω,
Then, taking into account (3) and (4) and introducing the x(l, 0) = x0 (l), (12)
transformed state variable vector x(l, t) defined as
where x0 (l) : Ω → Rn is a given vector function.
−1
x(l, t) = S Ew(l, t), (6) On the other hand, the boundary conditions represent
the requirements to be met by the solution x(l, t) at the
Eqn. (1) can be finally written as follows: boundary points of Ω. In general, these conditions may
∂x (l, t) ∂x (l, t) take the form of a linear combination of the Dirichlet and
+Λ = Kx (l, t) , (7)
∂t ∂l Neumann boundary conditions, as the so-called boundary
where conditions of the third kind (Dooge and Napiorkowski,
K = S −1 W E −1 S. (8) 1987; Christofides and Daoutidis, 1997; Sutherland and
Kennedy, 2003; Ancona and Coclite, 2005). For the
class of hyperbolic systems considered, we assume the
Remark 1. Owing to the diagonal form of the matrix
Dirichlet boundary conditions, which can be written in the
Λ, each equation of the system (7) contains both temporal
following compact way (see Xu and Sallet, 2002; Diagne
and spatial derivatives of the same state variable xi (l, t),
et al., 2012):
for i = 1, 2, . . . , n. Therefore, this system is commonly
referred to as decoupled or weakly coupled, i.e., coupled     +   
only through the terms that do not contain derivatives, as x+ (0, t) P00 P01 x (L, t) R0
opposed to the original system (1) which in general can be = + u (t)
x− (L, t) P10 P11 x− (0, t) R1
strongly coupled.
(13)
Definition 1. (Mattheij et al., 2005) The system (1) is with ⎡ ⎤ ⎡ ⎤
said to be hyperbolic if all the eigenvalues of the matrix x1 xp+1
F E −1 are real and different from zero. Additionally, if ⎢ ⎥ ⎢ ⎥
x+ = ⎣ ... ⎦ , x− = ⎣ ... ⎦ . (14)
all the eigenvalues are distinct, then Eqn. (1) is said to be
xp xn
strictly hyperbolic.
Therefore, the strict hyperbolicity of the system The vector function u(t) ∈ L2 ([0, +∞); Rr ) in (13)
means that the matrix Λ in (7) takes the following form: expresses the inhomogeneity of the boundary conditions
which can be identified with r external inputs to the
Λ = diag (λ1 , . . . , λp , λp+1 , . . . , λn ) , (9)
system, including control signals as well as external
where λi ∈ R \ {0} for i = 1, 2, . . . , n, represents the disturbances. The constant matrices P00 ∈ Rp×p , P01 ∈
eigenvalues of the matrix F E −1 arranged in descending Rp×(n−p) , P10 ∈ R(n−p)×n , P11 ∈ R(n−p)×(n−p)
order, express boundary feedbacks and reflections, whereas
R0 ∈ Rp×r and R1 ∈ R(n−p)×r represent the effect of the
λ1 > . . . > λp > 0 > λp+1 > . . . > λn , (10) external inputs u(t) on the boundary conditions x+ (0, t)
with p ≤ n representing the number of positive and x− (L, t), respectively.
eigenvalues.
The matrix S in (3) can be presented as follows: 2.3. Second-order systems. Among many different
  kinds of DPSs, an important class is constituted by
S = s1 . . . sp sp+1 . . . sn , (11)
the systems with two spatio-temporal state variables
where si ∈ Rn for i = 1, 2, . . . , n denotes the which can be described, after appropriate assumptions,
corresponding column eigenvectors of the matrix F E −1 . by a second-order hyperbolic PDE. The following typical
 294 K. Bartecki

examples can be mentioned here (Górecki et al., 1989; [1 0] and R1 = [0 1], which leads to the input vector of
Gvozdenac, 1990; Miano and Maffucci, 2001; Arbaoui the form
et al., 2007; Dos Santos et al., 2008; Bartecki, 2009):  T  T
u(t) = u1 (t) u2 (t) = x1 (0, t) x2 (L, t) . (18)
• the voltage u(l, t) and the current i(l, t) in the
Remark 3. Taking into account (10), it can be noticed
electrical transmission line,
that the congruent boundary inputs should be imposed for
• the pressure p(l, t) and the flow q(l, t) of the medium λ1 > 0 and λ2 > 0, while the incongruent ones for λ1 > 0
transported through the pipeline, and λ2 < 0.
The above assumptions about the form of the
• the temperatures ϑ1 (l, t) and ϑ2 (l, t) of the heating boundary conditions representing the external influences
and the heated fluid in the case of a shell and tube on the system have their practical reasons. For example,
heat exchanger. in the case of the above-mentioned shell and tube heat
In the case of the above-mentioned systems, exchanger operating in the so-called parallel-flow mode,
the temperatures of the heated and the heating fluid are
Eqn. (7) takes, after possible previous linearization and
diagonalization, the form of the following two PDEs: given for the same geometric point of the exchanger. On
the other hand, the temperatures of the fluids flowing into
∂x1 (l, t) ∂x1 (l, t) the exchanger operating in the countercurrent-flow mode
+λ1 = k11 x1 (l, t)+k12 x2 (l, t) , (15) are specified for its two opposite sides.
∂t ∂l
∂x2 (l, t) ∂x2 (l, t)
+λ2 = k21 x1 (l, t)+k22 x2 (l, t) , (16) 2.4. State and output equations. Ignoring for the
∂t ∂l moment the boundary inputs, Eqn. (1) can be written
where k11 , k12 , k21 , k22 are constant elements of the in the following homogeneous abstract Cauchy form
matrix K given by (8). (Curtain and Zwart, 1995):
dx(t)
It is assumed here that the only external influence = Ax(t), x(0) = x0 , (19)
dt
on the state variables is given by the boundary conditions
(13) and no boundary feedbacks nor reflections are present with the state variable x(t) belonging to the Hilbert space
in the system, i.e., P00 , P01 , P10 , P11 in (13) are all X = L2 (Ω, Rn ) and the operator A : X ⊃ D(A) → X
zero matrices. The second important assumption is that which can be expressed based on (7) as
the system is given directly by the two weekly coupled
dh
PDEs (15) and (16), or is considered in terms of the Ah = −Λ + Kh. (20)
transformed state variables (6). Two cases often occurring dl
in practice are studied here: in the first one, both boundary Considering the second-order system (15) and (16), the
conditions are given for the same edge (l = 0) of Ω, domain D(A) is given for h = [h1 , h2 ]T by
and in the second the input function u(t) acts on the two  
different edges, l = 0 and l = L, respectively. Such a D(A) = h ∈ H 1 (Ω)×H 1 (Ω) h1 (0) = h2 (0) = 0 ,
situation occurs, e.g., in the case of shell and tube heat (21)
exchangers, for which two different flow configurations for λ1 > 0, λ2 > 0, and by
are possible (Bartecki and Rojek, 2005; Bartecki, 2007;  
D(A) = h ∈ H 1 (Ω)×H 1 (Ω) h1 (0) = h2 (L) = 0 ,
Delnero et al., 2004; Gvozdenac, 1990; Zavala-Río et al.,
(22)
2009; Maidi et al., 2010). Therefore, two new definitions
for λ1 > 0, λ2 < 0, with H 1 (Ω) the Hilbert–Sobolev
are now introduced in order to discriminate between the
space of functions, whose distributional derivatives lie in
two above-mentioned classes of boundary inputs.
L2 (Ω).
Definition 2. The boundary inputs of the system (15) Assumption 1. The matrix K in (7) is symmetric and
and (16) will be referred to as congruent for the following negative-semidefinite, i.e., it fulfills ξ T Kξ ≤ 0 for any
parameter values of Eqn. (13): n = l = p = 2 and R0 = ξ ∈ Rn .
I2 , which leads to the input vector of the form
Remark 4. Assumption 1, together with the proper
 T  T
u(t) = u1 (t) u2 (t) = x1 (0, t) x2 (0, t) . (17) boundary conditions, ensures the so-called dissipativity
of the operator A in (20) (Phillips, 1957; Lumer and
Definition 3. The boundary inputs to the system (15) and Phillips, 1961; Xu and Sallet, 2002; Chentouf and Wang,
(16) will be referred to as incongruent for the following 2009; Tucsnak and Weiss, 2006). This means that the
parameter values of Eqn. (13): n = l = 2, p = 1, R0 = associated physical model has no internal energy sources
A general transfer function representation for a class of hyperbolic distributed parameter systems 295

and the boundary conditions are such that no energy enters or

the model through the boundaries. These conditions ⎡L ⎤

are satisfied by many chemical engineering systems like δ (L) x (ξ, t) dξ  
⎢0 1 ⎥ x1 (L, t)
tubular reactors and heat exchangers. ⎢
Cx(t) = ⎣ L ⎥
 ⎦ = x2 (0, t) , (31)
Theorem 1. (Rauch and Taylor, 1974; Lumer and δ (0) x2 (ξ, t) dξ
0
Phillips, 1961) Let the assumptions on the hyperbolicity
be satisfied for the system (7). Then for each x0 ∈ X the which can be treated as “opposite boundary” counterparts
system has a unique solution of the input operators (26) and (27), with y(t) ∈
L2 ([0, +∞); R2 ).
x(·, t) ∈ C ([0, +∞) ; X) . (23) Although the above formulation of input and output
operators seems to be soundly justified by practical
Moreover, if the system is dissipative, the semigroup of
considerations (e.g., inlet and outlet points of a shell and
bounded linear operators T (t) from X into itself such that
tube heat exchanger are usually located on its extreme
x(·, t) = T (t)x0 , generated by the operator A (20), is
opposite ends), it causes some mathematical difficulties
contractive, i.e., it fulfills
since B and C are unbounded, δ(0), δ(L) ∈ / L2 (Ω).

T (t)
L(X) ≤ 1 for all t ≥ 0. (24) However, as will be shown in the next section, the
notion of the transfer function can be preserved for some
Considering the second-order system (15) and (16), class of distributed parameter systems with a certain
the boundary inputs introduced in Definitions 2 and 3 can unboundedness of the control and observation operators.
be included as an inhomogeneity into (19), giving as a
result the following state equation (see Curtain and Weiss, 3. Transfer function description
2006):
3.1. Definition and properties.
dx(t)
= Ax(t) + Bu(t), x(0) = x0 , (25) Definition 4. (Curtain and Zwart, 1995) The trans-
dt fer function G(s) of the linear system described by
where u(t) ∈ L2 ([0, +∞); R2 ), and the state and output equations (25) and (28), is the
  L(R2 , R2 )-valued function of a complex variable given by
λ1 δ (0) u1 (t)
Bu(t) = (26) y(s) = G(s)u(s) for Re(s) > α,
λ2 δ (0) u2 (t) (32)

or   where y(s) and u(s) are the Laplace transforms of y(t)

λ1 δ (0) u1 (t) and u(t), respectively,
Bu(t) = , (27)  and α is2 a real parameter such
λ2 δ (L) u2 (t) that e−α u (·) ∈ L 2
[0, +∞) ; R produces e −α
y (·) ∈

where B is the control operator for the case of the L2 [0, +∞) ; R2 for every u.
congruent and incongruent boundary inputs, respectively.
In the above formulation, the input points are modeled as
Assuming temporarily that both 2 B and C are
bounded
 linear
 operators: B ∈ L R , X and C ∈
a Delta distribution at the extremities l = 0 and l = L (see
L X, R , the transfer function G(s) of the system is
2
Tucsnak and Weiss, 2006; 2009; Maidi et al., 2010).
given by (Callier and Winkin, 1993; Curtain and Zwart,
The state equation (25) can be completed by the
1995)
following output equation:
−1
G(s) = C (sI − A) B for Re(s) > ω0 , (33)
y(t) = Cx(t), (28)

where y is in the Hilbert space Y and C is a linear output where (sI − A)−1 ∈ L (X, D (A)) is the resolvent of
(observation) operator. This operator can be taken as A, i.e., the Laplace transform of the semigroup T (t)
identity, giving as a result a “distributed output”, i.e., generated by A, and ω0 is the growth constant of this
semigroup.
y(t) = x(t), (29) The above definition of the transfer function can
be extended to a more general class of systems, where
or as a pointwise output operator with some unboundedness of the input and output operators
⎡L ⎤ is possible, as shown by Salamon (1987), Curtain et al.
 (1992), Callier and Winkin (1993), Grabowski and Callier
δ (L) x (ξ, t) dξ  
⎢0 1 ⎥
Cx(t) = ⎢ ⎥ = x1 (L, t) (30)
(2001a; 2001b), or Cheng and Morris (2003). Such
⎣L ⎦ x2 (L, t) general systems are interesting since they allow, e.g.,
δ (L) x2 (ξ, t) dξ
0
the exact boundary control as well as the pointwise
 296 K. Bartecki

observation, as considered in Section 2.4. To this end, differential equation and solving the resulting boundary
the notion of the so-called admissibility of the input and value problem. This approach will be applied here to the
output operators has been introduced in the literature. The analysed class of hyperbolic DPSs.
mild solution of (25) and (28) can be written, assuming
t ≥ 0, in the following general form (Curtain and 3.2. General transfer function matrix. After Laplace
Zwart, 1995): transform Lt with respect to the variable t, the general
system (7) can be written as
t
x(t) = T (t)x0 + T (t − τ )Bu(τ ) dτ, (34) ∂x(l, s)
= Λ−1 (K − sI) x(l, s) + Λ−1 x(l, 0), (36)
0 ∂l
t where x(l, 0) represents the initial conditions (12).
y(t) = CT (t)x0 + C T (t − τ )Bu(τ ) dτ, (35) Introducing, for the sake of brevity, the matrix P (s) of
0 the following form:

where T (t) is the semigroup generated by A (20). Without P (s) = Λ−1 (K − sI) , (37)
going too far into details, it may be stated that the operator
B for which the integral term in (34) lies in X for every the general solution of (36) is given by
 
u ∈ L2 [0, t1 ] , R2 and some t1 > 0 is called an admis-
l
sible input operator. Similarly, the operator C for which −1
x(l, s) = Λ eP (s)(l−ξ) x(ξ, 0) dξ + eP (s)l x(0, s),
one can uniquely define a function y ∈ L2 [0, t1 ] , R2
for some t1 > 0 and all x0 ∈ X, is called an admis- 0
(38)
sible output operator (Curtain et al., 1992; Weiss, 1994;
where the first term on the right-hand side represents the
Staffans and Weiss, 2000).
response to the initial conditions at t = 0 and the second
If both B and C are admissible operators, the system
the response to the boundary conditions at l = 0.
(34) and (35) is called a Pritchard–Salamon system
Equation (38) can be identified with the state
(Pritchard and Salamon, 1984; Curtain et al., 1992). Since
transition equation (34), except that the former is in the
the admissibility of the studied unbounded operators
Laplace transform domain whereas the latter in the time
B and C for the semigroup T (t) generated by A in
domain. Moreover, the role of the input function u(t)
(20) has been already proved (see, e.g., Tucsnak and
in (34) is played by the Laplace-transformed boundary
Weiss, 2006) one may state that the dissipative hyperbolic
conditions x(0, s) in (38). Therefore, assuming the case
systems considered belong to the Pritchard–Salamon
of the distributed output (29), the matrix eP (s)l in (38),
class. Moreover, as proved by Curtain (1988), if A
which is the resolvent of the operator A,
generates a C0 -semigroup T (t), either μ-exponential
stabilizability or μ-exponential detectability of this R (s, A) = (sI −A)−1 = eP (s)l for Re (s) > 0, (39)
semigroup is sufficient for the given systems to belong
to the Callier–Desoer class B(μ), for which the Nyquist can be referred to as a general n × n transfer function
theorem provides necessary and sufficient conditions for matrix G(l, s) of the system (7) with n boundary inputs
input-output closed-loop stability (Callier and Winkin, imposed at l = 0 and the parameter l indicating the spatial
1993; Curtain and Zwart, 1995; Sasane, 2002). The above position for which the transfer function is to be evaluated.
statement holds true even if B and C are unbounded In order to calculate the matrix exponential eP (s)l in
operators, which is the case considered here. The same (39), we assume that P (s) can be diagonalized as follows:
has been proved by Litrico and Fromion (2009a) for a very
particular hyperbolic system of conservation laws, based P (s) = Ψ(s) Φ(s) Ψ−1 (s), (40)
on the transfer function decomposition into an unstable
where Φ(s) is a diagonal matrix of eigenvalues of
finite dimensional part and a stable infinite dimensional
P (s) and Ψ(s) consists of its (column) eigenvectors.
part.
Consequently, Eqn. (38) can be written in the following
From the semigroup analysis performed in
form:
Section 2.4 we have that the transfer functions of
the hyperbolic systems considered are in H∞ , i.e., they l
−1
are analytic and bounded in the open right half of the x(l, s) =Λ G(l−ξ, s)x(ξ, 0) dξ +G(l, s)x(0, s),
complex plane, Re(s) > 0. As shown by Callier and 0
Winkin (1993) as well as Curtain and Zwart (1995), (41)
it is possible for a boundary control system to obtain where
a closed form expression for the transfer function by
taking the Laplace transforms of the original partial G(l, s) = eP (s)l = Ψ(s)eΦ(s)l Ψ−1 (s). (42)
A general transfer function representation for a class of hyperbolic distributed parameter systems 297

3.3. Transfer function matrices of the second-

order system. In the case of the second-order system
described by (15) and (16), one can distinguish two
lumped input signals represented by the boundary
conditions (17) or (18), and two output signals
representing the distribution of the state variables x1 (l, t)
and x2 (l, t) (see Eqn. (29)). For the purpose of further
analysis, the following two definitions are introduced,
based on the well-known notion of the transfer function
as the ratio of the Laplace transform of the output signal
to the Laplace transform of the input signal under the
assumption that the other input signals as well as all initial
conditions are equal to zero.
Fig. 1. Block diagram of the transfer function model for the
Definition 5. The transfer function matrix of the system congruent boundary inputs.
described by Eqns. (15) and (16) is defined for the case
of the congruent boundary inputs (17) as the matrix of the
following form: can therefore be determined for the congruent boundary
  inputs, assuming zero initial conditions, based on the
G11 (l, s) G12 (l, s) following equation:
G(l, s) = , (43)
G21 (l, s) G22 (l, s)
x(l, s) = G(l, s)u(s) (50)
where
T
for u(s) = [x1 (0, s) x2 (0, s)] , which is obviously a
x1 (l, s) x2 (l, s)
G11 (l, s) = , G21 (l, s) = , (44) particular case of (41). A block diagram of the transfer
x1 (0, s) x1 (0, s) function model for the congruent boundary inputs is
for x2 (0, s) = 0, and presented in Fig. 1.
On the other hand, for the case of the incongruent
x1 (l, s) x2 (l, s) boundary inputs (see Fig. 2), the formula for the vector of
G12 (l, s) = , G22 (l, s) = , (45) the state variables takes the following form:
x2 (0, s) x2 (0, s)

for x1 (0, s) = 0, all for zero initial conditions, x1 (l, 0) = x(l, s) = Ḡ(l, s)u(s) (51)
x2 (l, 0) = 0. T
for u(s) = [x1 (0, s) x2 (L, s)] . Referring to the
Definition 6. The transfer function matrix of the system analysis presented in Section 3.2, the matrix P (s) in (37)
described by Eqns. (15) and (16) is defined for the case of takes for the case considered the following form:
the incongruent boundary inputs (18) as the matrix of the ⎡k − s k12 ⎤
  11
following form: p11 (s) p12 ⎢ λ1 ⎥ , (52)
P (s) = = ⎣ λ1

Ḡ11 (l, s) Ḡ12 (l, s)

p21 p22 (s) k 21 k22 − s⎦
Ḡ(l, s) = , (46) λ2 λ2
Ḡ21 (l, s) Ḡ22 (l, s)
and its eigendecomposition (40) results in the following
where matrix of eigenvalues:
x1 (l, s) x2 (l, s)
Ḡ11 (l, s) = , Ḡ21 (l, s) = , (47) Φ(s) = diag (φ1 (s), φ2 (s)) , (53)
x1 (0, s) x1 (0, s)
with
for x2 (L, s) = 0, and φ1,2 (s) = α(s) ± β(s), (54)
x1 (l, s) x2 (l, s) where
Ḡ12 (l, s) = , Ḡ22 (l, s) = , (48) 1
x2 (L, s) x2 (L, s) α(s) = (p11 (s) + p22 (s)) (55)
2
for x1 (0, s) = 0, all for zero initial conditions, x1 (l, 0) = and
x2 (l, 0) = 0. 
1 2
The vector of Laplace transforms of the state β(s) = (p11 (s) − p22 (s)) + 4p12 p21 . (56)
2
variables
 T Assuming as an example k11 < 0, k22 < 0 and
x(l, s) = x1 (l, s) x2 (l, s) (49) k12 = k21 = 0 in (15) and (16) for the case of the
 298 K. Bartecki

p21  
G21 (l, s) = eφ1 (s)l − eφ2 (s)l , (62)
φ1 (s) − φ2 (s)
φ1 (s) − p11 (s) φ1 (s)l
G22 (l, s) = e
φ1 (s) − φ2 (s)
(63)
φ2 (s) − p11 (s) φ2 (s)l
− e ,
φ1 (s) − φ2 (s)

where p11 (s), p12 , p21 and p22 (s) are elements of the
matrix P (s) in (52) and φ1 (s), φ2 (s) are its eigenvalues
given by (53)–(56).

Fig. 2. Block diagram of the transfer function model for the in- Proof. By applying the Laplace transform Lt to (15)
congruent boundary inputs. and (16), one obtains, assuming zero initial conditions
x1 (l, 0) = 0 and x2 (l, 0) = 0, the following equations:

∂x1 (l, s)
congruent boundary inputs (17), we immediately obtain, λ1 = (k11 −s) x1 (l, s) + k12 x2 (l, s) , (64)
∂l
based on (52) and (42), the transfer function matrix G(l, s)
∂x2 (l, s)
in the diagonal form, λ2 = k21 x1 (l, s) + (k22 −s) x2 (l, s) . (65)
∂l
 
κ1 (l)e−sτ1 (l) 0 Taking the Laplace transform again, now with respect to
G(l, s) = , (57)
0 κ2 (l)e−sτ2 (l) the spatial variable l,
 
where x (q, s) = Ll x (l, s) (66)
k11 k22
l l and taking into account that
κ1 (l) = e λ1 , κ2 (l) = e λ2 , (58)
l l  
τ1 (l) = , τ2 (l) = . ∂x(l, s)
λ1 λ2
(59) Ll = qx(q, s) − x(0, s) (67)
∂l
The above representation should be transforms Eqns. (64) and (65) into the following form:
considered a particular case of the two separate
k22 −s k12
proportional-time-delay systems. Furthermore, assuming q− λ2 λ1
also k11 = k22 = 0, we obtain two pure time-delay x1 (q, s) = x1 (0, s) + x2 (0, s) ,
M (q, s) M (q, s)
systems, G11 (l, s) = e−sτ1 (l) and G22 (l, s) = e−sτ2 (l) . (68)
In the next subsection the analytical closed form
k21 k11 −s
expressions for the individual elements of the transfer λ2 q− λ1
x2 (q, s) = x1 (0, s) + x2 (0, s) ,
function matrices G(l, s) and Ḡ(l, s) are derived for M (q, s) M (q, s)
the general case of the non-zero parameters, taking (69)
into account the two different forms of the boundary
conditions introduced in Section 2.3. where M (q, s) is the characteristic polynomial of the
matrix P (s)
3.4. Analytical expressions for the transfer functions.
M (q, s) = det (qI − P (s)) =
(70)
Proposition 1. The transfer functions included in the = (q − φ1 (s)) (q − φ2 (s)) .
matrix G(l, s) in (43) take for the case of the congruent
boundary inputs (17) the following form: Finding the inverse Laplace transform of (68) and
(69) with respect to q by taking advantage of the following
φ1 (s) − p22 (s) φ1 (s)l property (Friedly, 1972):
G11 (l, s) = e
φ1 (s) − φ2 (s) ⎧ ⎫
(60) ⎪
⎪ ⎪
⎪
φ2 (s) − p22 (s) φ2 (s)l ⎪
⎨ S(q) ⎪
⎬ " N
− e , −1 S(q) S(λj )
φ1 (s) − φ2 (s) Lq = N = eλj t ,
⎪ T (q)  ⎪ dT (q)
⎪ (q − λj ) ⎪
p12   ⎪
⎩ ⎭ j=1 dq q=λj
⎪
G12 (l, s) = eφ1 (s)l − eφ2 (s)l , (61) j=1
φ1 (s) − φ2 (s) (71)
A general transfer function representation for a class of hyperbolic distributed parameter systems 299

where S(q) and T (q) represent polynomials in q of degree Proof. The transfer function G11 (l, s) described by
M and N > M , respectively, and λj is a single root of (60) can be transformed using (54) and Lemma 1 into the
T (q), yields the following form of the equations: following form:
x1 (l, s) = G11 (l, s) x1 (0, s)+G12 (l, s) x2 (0, s) , (72) G11 (l, s)
x2 (l, s) = G21 (l, s) x1 (0, s)+G22 (l, s) x2 (0, s) , (73) φ1 (s)
= eα(s)l eβ(s)l
φ1 (s) − φ2 (s)
where the expressions on G11 (l, s), G12 (l, s), G21 (l, s)
p22 (s)
and G22 (l, s) are given by (60)–(63).  − eα(s)l eβ(s)l
φ1 (s)−φ2 (s)
Lemma 1. For any x, y, z ∈ C such that z = 0 and φ2 (s)
− eα(s)l e−β(s)l
z = y, the following identity holds: φ1 (s) − φ2 (s)
# $ p22 (s)
x y −x z − y z+y + eα(s)l e−β(s)l
e − e = cosh x + sinh x . (74) φ1 (s)−φ2 (s)
z z z−y # $
φ1 (s) α(s)l β(s)l φ2 (s) −β(s)l
Proof. By using the well-known identities = e e − (s)e
φ1 (s) − φ2 (s) φ1 (s)
ex − e−x p22 (s)  
sinh x = (75) − eα(s)l eβ(s)l − e−β(s)l
2 φ1 (s) − φ2 (s)
# $
and φ1 (s) + φ2 (s)
ex + e−x = eα(s)l coshβ(s)l + sinhβ(s)l
cosh x = , (76) φ1 (s) − φ2 (s)
2
p22 (s)
the right-hand side of (74) can be transformed in the −2 eα(s)l sinhβ(s)l.
φ1 (s) − φ2 (s)
following way:
(82)
# $
z−y z+y
cosh x + sinh x
z z−y Then, by making the substitutions φ1 (s) − φ2 (s) =
# $ 2β(s) and φ1 (s) + φ2 (s) = 2α(s), one obtains the
z−y x −x z+y  x −x

= e +e + e −e hyperbolic form (78) of G11 (l, s). Due to the obvious
2z z−y (77)
# $ symmetry, one can similarly obtain the hyperbolic version
z−y 2z x 2y −x (81) of G22 (l, s). On the other hand, for the transfer
= e − e
2z z−y z−y function G12 (l, s) we have
y
= ex − e−x . p12  
z G12 (l, s) = eα(s)l eβ(s)l − e−β(s)l
 φ1 (s) − φ2 (s)
p12
Proposition 2. The transfer functions (60)–(63) can be =2 eα(s)l sinh β(s)l
φ1 (s) − φ2 (s)
expressed in the following equivalent form using the hy- p12 α(s)l
perbolic functions: = e sinh β(s)l,
β(s)
(83)

G11 (l, s) = eα(s)l cosh β(s)l
and, in a similar fashion, we come to the hyperbolic form
α(s) − p22 (s)  (78) (80) of G21 (l, s). 
+ sinh β(s)l ,
β(s)
Remark 5. By making in the transfer functions (60)–(63)
p12 α(s)l the substitution l = 0, one obtains eφ1 (s)l = eφ2 (s)l = 1
G12 (l, s) = e sinh β(s)l, (79)
β(s) and finally G11 (0, s) = 1, G12 (0, s) = 0, G21 (0, s) = 0,
G22 (0, s) = 1. This result means that for l = 0, a given
p21 α(s)l
G21 (l, s) = e sinh β(s)l, (80) output variable xi (l, t) is identically equal to the relevant
β(s)
boundary input ui (t), without any cross-interactions from
 the second boundary input (see Fig. 1). Analogous results
G22 (l, s) = eα(s)l cosh β(s)l
can be obtained based on the analysis of the hyperbolic
α(s) − p11 (s)  (81) form (78)–(81) of the transfer functions. In this case one
+ sinh β(s)l , obtains, for l = 0, eα(s)l = 1, sinh β(s)l = 0 and
β(s)
cosh β(s)l = 1, which leads to the same values of the
where α(s) and β(s) are given by (55) and (56). transfer functions.
 300 K. Bartecki

Proposition 3. The transfer functions included in the ma-

trix Ḡ(l, s) in (46) for the case of the incongruent bound- Ḡ12 (l, s)
ary inputs (18) take the following form:
p12 eα(s)(l−L) sinh β(s)(l)
= ,
Ḡ11 (l, s) β(s) cosh β(s)L + (α(s) − p11 (s)) sinh β(s)L
(91)
eφ2 (s)L eφ1 (s)l (φ1 (s) − p22 (s))
=
eφ2 (s)L (φ1 (s) − p22 (s)) − eφ1 (s)L (φ2 (s) − p22 (s))
Ḡ21 (l, s)
eφ1 (s)L eφ2 (s)l (φ2 (s) − p22 (s))
− φ (s)L , p21 eα(s)l sinh β(s)(l − L)
e 2 (φ1 (s) − p22 (s)) − eφ1 (s)L (φ2 (s) − p22 (s)) = ,
(84) β(s) cosh β(s)L − (α(s) − p22 (s)) sinh β(s)L
(92)
Ḡ12 (l, s)
  Ḡ22 (l, s)
p12 eφ2 (s)l − eφ1 (s)l
= φ (s)L ,
e 2 (φ2 (s) − p11 (s)) − eφ1 (s)L (φ1 (s) − p11 (s)) eα(s)(l−L) β(s) cosh β(s)l
=
(85) β(s) cosh β(s)L + (α(s) − p11 (s)) sinh β(s)L

Ḡ21 (l, s) eα(s)(l−L) (α(s) − p11 (s)) sinh β(s)l

+ .
  β(s) cosh β(s)L + (α(s) − p11 (s)) sinh β(s)L
p21 eφ2 (s)L eφ1 (s)l − eφ1 (s)L eφ2 (s)l (93)
= φ (s)L ,
e 2 (φ1 (s) − p22 (s)) − eφ1 (s)L (φ2 (s) − p22 (s))
(86) Proof. The result is derived as in the case of the proof of
Result 2, i.e., using (54) and Lemma 1. 

Ḡ22 (l, s) Remark 6. In the case of the incongruent boundary

eφ2 (s)l (φ2 (s) − p11 (s)) − eφ1 (s)l (φ1 (s) − p11 (s)) inputs one obtains Ḡ11 (0, s) = 1, Ḡ12 (0, s) = 0,
= φ (s)L ,
e 2 (φ2 (s) − p11 (s)) − eφ1 (s)L (φ1 (s) − p11 (s)) Ḡ21 (L, s) = 0 and Ḡ22 (L, s) = 1 (see Fig. 2).
(87)
3.5. Pole-zero analysis. The pole-zero analysis can be
where p11 (s), p12 , p21 and p22 (s) are elements of the
started with the simplified form of the analysed transfer
matrix P (s) in (52) and φ1 (s), φ2 (s) are its eigenvalues
functions, discussed at the end of Section 3.3, i.e., for
given by (53)–(56).
k12 = k21 = 0. In this case we obtain φ1 (s) = p11 (s),
φ2 (s) = p22 (s) and the transfer functions are given by
Proof. The solution of the set of the differential equations the matrix (57) with G12 (l, s) = G21 (l, s) = 0 and
(64) and (65), now with the incongruent boundary G11 (l, s), G22 (l, s) in the pure exponential form having
conditions (see Definition 3), is given in the following neither poles nor zeros. Similar results can be obtained
form: for the case of the incongruent inputs.
x1 (l, s) = Ḡ11 (l, s) x1 (0, s)+ Ḡ12 (l, s) x2 (L, s) , (88) Considering the more general case of the non-zero
parameters, one can suppose that the poles of the transfer
x2 (l, s) = Ḡ21 (l, s) x1 (0, s)+ Ḡ22 (l, s) x2 (L, s) , (89) functions (60)–(63) and (84)–(87) occur when

where the expressions for Ḡ11 (l, s), Ḡ12 (l, s), Ḡ21 (l, s) φ1 (s) = φ2 (s) (94)
and Ḡ22 (l, s) are given by (84)–(87). 
with
Proposition 4. The transfer functions (84)–(87) can be √
k22 λ1 − k11 λ2 2 −k12 k21 λ1 λ2
expressed in the following equivalent form using the hy- s1,2 = ± . (95)
λ1 − λ2 λ1 − λ2
perbolic functions:
However, Eqn. (94) expresses also the condition for
Ḡ11 (l, s)
“zeroing” all transfer function numerators. Thus, since
eα(s)l β(s) cosh β(s)(l − L) the poles and zeros are defined for the reduced form of
=
β(s) cosh β(s)L − (α(s) − p22 (s)) sinh β(s)L the transfer function, we can state that all the transfer
functions considered have no poles. Furthermore, transfer
eα(s)l (α(s) − p22 (s)) sinh β(s)(l − L)
+ , functions G11 (l, s), G22 (l, s) and their incongruent
β(s) cosh β(s)L − (α(s) − p22 (s)) sinh β(s)L counterparts have no zeros either. The above results can
(90) be regarded as a generalization of the results obtained for
A general transfer function representation for a class of hyperbolic distributed parameter systems 301

the simplified form of the transfer functions considered in the operator variable s with the expression iω, where
Section 3.3. ω is the angular frequency. As a result, one obtains
Slightly different is the case of the “crossover” the corresponding frequency responses Gij (l, iω) of the
channels G12 (l, s) and G21 (l, s), for which the condition channel connecting the i-th output to the j-th input, i, j =
for the occurrence of zeros is 1, 2.
The graphical representation of these responses can
eφ1 (s)l = eφ2 (s)l . (96) take the form of three-dimensional graphs, taking into
account the dependence of the frequency response on
Thus, due to the periodicity of the exponential function both the angular frequency ω and the spatial variable l.
along the imaginary axis, this condition means that the Another possibility is the representation in the form of
transfer functions considered have infinitely many zeros the classical two-dimensional plots, determined for the
as stated in the following result. fixed value of the spatial variable (Jovanović and Bamieh,
Proposition 5. The transfer functions G12 (l, s), 2006; Bartecki, 2007; 2009; Litrico and Fromion, 2009a).
G21 (l, s) and Ḡ12 (l, s) have infinitely many zeros of the Considering as an example the Bode plot of the frequency
following form: response, the expressions for the logarithmic gain and
phase take the following form:
(k) k22 λ1 − k11 λ2
z1,2 (l) = ± Lij (l, ω) = 20 log Gij (l, iω) (101)
(λ1 − λ2 )
% (97)
2 −λ21 λ22 k 2 π 2 − λ1 λ2 k12 k21 l2 and
±  
(λ1 − λ2 ) l ϕij (l, ω) = arg Gij (l, iω) , (102)
for k ∈ N. where the expressions for the modulus and argument of
the frequency response are as follows:
Proof. Based on (79), (80) and (91), one can observe that
    
the zeros of these transfer functions are given as the roots Gij (l,iω) = Re2 Gij (l,iω) +Im2 Gij (l,iω)
of the following equation:
(103)
sinh β (s) l = 0, (98) and
 
  Im Gij (l, iω)
which is satisfied for arg Gij (l, iω) = atan  . (104)
Re Gij (l, iω)
β (s) l = kπi, k ∈ Z. (99)
4. Examples
Taking into account (56), Eqn. (99) can be transformed to
For a practical illustration of the above discussed issues,
2 4k 2 π 2 this section performs a transfer function analysis of a
(p11 (s) − p22 (s)) + 4p12 p21 =− 2 . (100)
l shell and tube heat exchanger, which can be considered a
typical DPS whose mathematical description, after some
Using the relationships in (52) and finding the solution of assumptions, takes the form of Eqns. (15) and (16). The
(100) in terms of s gives as a result the expression (97). analysis is performed for both the exchanger operating in
 the parallel-flow mode, for which the boundary conditions
Remark 7.
(0)
The zeros z1,2 (l) for k = 0 in (97) are have the form specified in Definition 2, and for the
canceled by the poles s1,2 in (95), which is easy to show countercurrent-flow mode, with boundary conditions of
by inserting k = 0 into (97) and comparing to (95). the form adopted in Definition 3. Having derived,
based on Eqns. (60)–(63) and (84)–(87), the transfer
Result 1. Taking into account (92), the zeros of the functions of the exchanger, the pole-zero analysis is
transfer function Ḡ21 (l, s) should satisfy the condition performed. Selected frequency responses for the parallel
sinh β (s) (l − L) = 0. Thus, in order to determine their and countercurrent flow modes are also presented, both
values, one should replace in (97) the variable l with l−L. in the form of three-dimensional graphs as well as the
classical two-dimensional Bode and Nyquist plots.
3.6. Frequency responses. Based on the knowledge
of the transfer functions it is possible to determine the 4.1. Parallel-flow heat exchanger. Under some
frequency responses of the individual channels of the simplifying assumptions, the dynamic properties of a
system under consideration. For this purpose, one should shell and tube heat exchanger can be described, based on
replace in the relationships (60)–(63) and (84)–(87), or the thermal energy balance equations, by the following
in their hyperbolic equivalents (78)–(81) and (90)–(93), PDE system (Bartecki and Rojek, 2005; Bartecki, 2007;
 302 K. Bartecki

Delnero et al., 2004; Gvozdenac, 1990; Zavala-Río et al.,

2009; Maidi et al., 2010):

∂ϑ1 (l,t) ∂ϑ1 (l,t)  

+v1 = α1 ϑ2 (l,t)−ϑ1 (l,t) , (105)
∂t ∂l
∂ϑ2 (l,t) ∂ϑ2 (l,t)  
+v2 = α2 ϑ1 (l,t)−ϑ2 (l,t) , (106)
∂t ∂l
where the 1- and 2- sub-indexed figures represent
the tube-side and shell-side fluid variables/coefficients,
respectively; specifically, ϑ1 (l, t) and ϑ2 (l, t) are the
temperature variations, v1 and v2 are the velocities, α1
and α2 are the heat transfer coefficients.
Referring Eqns. (105) and (106) to Eqn. (1), one
obtains the following vector of the state variables: Fig. 3. Real part of the frequency response function G11 (l, iω)
 T for the parallel-flow heat exchanger.
w(l, t) = ϑ1 (l, t) ϑ2 (l, t) , (107)
and the following matrices of constant coefficients:
     
1 0 v 0 −α1 α1
E= , F= 1 , W= .
0 1 0 v2 α2 −α2
(108)

Since F E −1 is diagonal, Eqns. (105) and (106) are

decoupled in the differential part and the diagonalization
procedure described by Eqns. (3)–(8) is not necessary.
Assuming v1 = 1 m · s−1 , v2 = 0.2 m · s−1 , α1 = α2 =
0.05 s−1 in (105) and (106), one obtains the following
matrices of the system (7):
 
1 0
Λ=F = , (109)
0 0.2
Fig. 4. Imaginary part of the frequency response function
 
−0.05 0.05 G11 (l, iω) for the parallel-flow heat exchanger.
K=W = , (110)
0.05 −0.05
with λ1 = v1 = 1, λ2 = v2 = 0.2 being the
eigenvalues of the system, and K symmetric and negative where ϑi (l, s) = Lt {ϑi (l, t)} is the Laplace transform
semi-definite, which makes the system dissipative. of the function representing the distribution of the
The fluid inlet temperatures ϑ1i , ϑ2i can be taken temperature of the i-th fluid, i = 1, 2.
as the input signals, which in the given case of According to the results presented in Section 3.5, all
the parallel-flow corresponds the following congruent the transfer functions of the heat exchanger considered
boundary conditions: have no poles. Additionally, transfer functions G11 (l, s)
and G22 (l, s) have no zeros either. On the other hand, for
ϑ1 (0, t) = ϑ1i (t) , (111) the transfer functions G12 (L, s), G21 (L, s) and parameter
ϑ2 (0, t) = ϑ2i (t) . (112) values (109)–(110) we obtain, based on Proposition 5,
infinitely many pairs of zeros of the following form:
The transfer functions of the heat exchanger are given
by Eqns. (60)–(63) or, equivalently, by (78)–(81) and (1)
z1,2 ≈ −0.0289 ± 0.3187i, (115)
represent the ratio of the Laplace transform of the fluid
(2)
temperature along the spatial axis to the Laplace transform z1,2 ≈ −0.0289 ± 0.6306i, (116)
of the fluid temperature in the inlet section of the
exchanger, i.e., for l = 0: etc.
Figures 3 and 4 show three-dimensional graphs of
ϑ1 (l, s) = G11 (l, s)ϑ1 (0, s) + G12 (l, s)ϑ2 (0, s), (113) the real and imaginary parts of the frequency response
G11 (l, iω) of the exchanger operating in the parallel-flow
ϑ2 (l, s) = G21 (l, s)ϑ1 (0, s) + G22 (l, s)ϑ2 (0, s), (114) mode, determined based on (60) for the assumed
A general transfer function representation for a class of hyperbolic distributed parameter systems 303

zeros of the transfer functions G12 (l, s) and G21 (l, s) and
the shape of the corresponding frequency responses. For
example, the imaginary parts of the zeros z (k) = α + ωk i
calculated for k = 1, 2 in (115) and (116) correspond to
the frequency values ωk for which the local minima occur
on the amplitude plot in Fig. 5.

4.2. Countercurrent-flow heat exchanger. In the

countercurrent mode of operation, the fluids involved
in the heat exchange enter the exchanger from its
opposite ends. The PDEs describing the dynamics of
the heat exchanger have the same form (105)–(106) as
for the parallel-flow mode, and the difference in the
mathematical description consists in the opposite signs of
Fig. 5. Bode plot of the frequency response G12 (L, iω) for the fluid velocities (v1 > 0, v2 < 0) as well as in the different
parallel-flow heat exchanger. boundary conditions

ϑ1 (0, t) = ϑ1i (t) , (117)

ϑ2 (L, t) = ϑ2i (t) . (118)

This situation represents the case of the incongruent

boundary inputs (18). Assuming the parameter values
(110) and the following values of fluid velocities: v1 =
1 m · s−1 , v2 = −0.2 m · s−1 , one obtains, based on (97),
the following values of the zeros of the transfer function
Ḡ12 (L, s):
(1)
z1,2 ≈ −0.0406 ± 0.2064i, (119)
(2)
z1,2 ≈ −0.0406 ± 0.4174i, (120)

etc., and the same values for the zeros of the transfer
Fig. 6. Nyquist plot of the frequency response G12 (L, iω) for function Ḡ21 (0, s).
the parallel-flow heat exchanger. Figures 7 and 8 show the three-dimensional graphs
of the real and the imaginary parts of the frequency
response Ḡ21 (l, iω) of the exchanger operating in the
countercurrent-flow mode. Unlike in the case of the
parameter values (109) and (110). One can observe parallel mode, due to the reversal of the flow direction for
characteristic oscillations caused by the aforementioned the second fluid, the largest amplitude of the oscillations
periodicity of the exponential function for the imaginary occurs now at l = 0. Figure 9 shows classical Bode
argument. plots of the heat exchanger frequency responses Ḡ22 (l, s),
Next, Fig. 5 shows classical, two-dimensional Bode determined based on Eqns. (87) and (101)–(104) for the
plots of the heat exchanger frequency responses G12 (l, s), outlet point of the the second fluid, i.e., for l = 0. The
determined for l = L based on Eqns. (61) and Nyquist plot of the same frequency response is shown in
(101)–(104). The Nyquist plot for the same transfer Fig. 10.
function channel is presented in Fig. 6. As seen An analysis of the frequency responses of the heat
from the Bode plot, the increase in the frequency of exchanger exhibits typical characteristics of systems with
the sinusoidal input signal initially causes a decrease distributed delay (Górecki et al., 1989). In particular,
in the amplitude of the output signal, and then it in the case of the “straightforward” transfer functions
gives rise to a local maximum. To a lesser extent G11 (l, s), G22 (l, s) and their incongruent counterparts,
it affects the phase characteristics. As the frequency one can notice the dominant influence of the transport
increases, the phenomenon repeats itself, which can be delay in the fluid flow. The amplitude damping of the
also observed as characteristic “loops” on the Nyquist sinusoidal oscillations in the real and imaginary parts
plot. These oscillations are closely associated with the of the frequency response is relatively small, which is
wave phenomena taking place inside the exchanger pipes reflected in the circular-shaped Nyquist plot, similar to
(Friedly, 1972). On can notice a relationship between the the one of the pure delay system (see Figs. 9 and 10).
 304 K. Bartecki

Fig. 7. Real part of the frequency response function Ḡ21 (l, iω) Fig. 9. Bode plot of the frequency response Ḡ22 (0, iω) for the
for the countercurrent-flow heat exchanger. countercurrent-flow heat exchanger.

Fig. 8. Imaginary part of the frequency response function Fig. 10. Nyquist plot of the frequency response Ḡ22 (0, iω) for
Ḡ21 (l, iω) for the countercurrent-flow heat exchanger. the countercurrent-flow heat exchanger.

On the other hand, in the case of the “crossover” transfer and/or zeros. Also the spatial location of the input signals,
functions G12 (l, s) and G21 (l, s), the damping of the represented by the appropriate boundary conditions,
input signal with increasing frequency is much greater, as significantly influences the form of the transfer functions.
for the “straight-forward” channels (see Figs. 5 and 6). Two kinds of Dirichlet boundary conditions have been
examined: congruent, where both inputs are given at the
same point of the spatial domain, and incongruent, where
5. Conclusion each input is located in a different point of the domain.
In the article, a general analytic form of the transfer The results of the transfer function analysis have
functions for a class of DPSs of hyperbolic type been illustrated on the example of a shell and tube heat
with two boundary inputs and two distributed outputs exchanger operating in parallel- and countercurrent-flow
has been derived and analyzed. As shown here, modes, considered to be a typical example of a
taking into account the spatio-temporal dynamics of hyperbolic DPS with congruent and incongruent boundary
these systems significantly affects their transfer function inputs, respectively. Based on the transfer function
representation. Unlike in the case of lumped systems, representation, the zero-pole analysis has been performed
the transfer functions derived for the DPS contain and the selected frequency responses of the exchanger
irrational functions, e.g., exponential and/or hyperbolic have been shown. They have been presented both in
ones. The “non-standard” form of these transfer functions the form of three-dimensional graphs, taking into account
causes some peculiarities which do not appear in the the spatio-temporal specificity of the system, and as the
rational-form case, such as an infinite number of poles classical two-dimensional Bode and Nyquist plots derived
A general transfer function representation for a class of hyperbolic distributed parameter systems 305

for a fixed value of the spatial variable. A specific feature heat exchanger: Modeling and adaptive predictive
of the obtained frequency responses are the oscillations, functional control, Applied Thermal Engineering
closely associated with the presence of the exponential 27(13): 2332–2338.
and hyperbolic functions which are periodic along the Bartecki, K. (2007). Comparison of frequency responses of
imaginary axis. parallel- and counter-flow type of heat exchanger, Pro-
The transfer function representation discussed here ceedings of the 13th IEEE IFAC International Confer-
has been obtained based on the system of hyperbolic ence on Methods and Models in Automation and Robotics,
equations in the decoupled canonical form. Thus, the use Szczecin, Poland, pp. 411–416.
of the word “general” in the paper title might appear to Bartecki, K. (2009). Frequency- and time-domain analysis of
be somewhat excessive. However, it can be shown that a simple pipeline system, Proceedings of the 14th IEEE
the transfer function matrix for the case of the strongly IFAC International Conference on Methods and Mod-
coupled system can be obtained from its decoupled els in Automation and Robotics, Mi˛edzyzdroje, Poland,
counterpart via relatively simple transformations, giving pp. 366–371.
as a result transfer functions formulated directly in their Bartecki, K. (2010). On some peculiarities of neural network
physical context. The problem will be addressed in one of approximation applied to the inverse kinematics problem,
the future articles together with some practical examples Proceedings of the Conference on Control and Fault-
such as, e.g., electrical transmission line or transport Tolerant Systems, Nice, France, pp. 317–322.
pipeline. Bartecki, K. (2011). Approximation of a class of distributed
The future works could also include, e.g., parameter systems using proper orthogonal decomposition,
consideration of the effects of the boundary feedback Proceedings of the 16th IEEE International Conference
and/or reflection on the transfer function description, on Methods and Models in Automation and Robotics,
as well as determination of a general form of the Mi˛edzyzdroje, Poland, pp. 351–356.
spatio-temporal responses for the individual transfer Bartecki, K. (2012a). Neural network-based PCA: An
function channels, both for the congruent and incongruent application to approximation of a distributed parameter
boundary inputs. Another issue to be thoroughly system, in L., Rutkowski, M., Korytkowski, R., Scherer,
examined, which is very important from the control R., Tadeusiewicz, L. Zadeh and J. Zurada (Eds.), Artifi-
synthesis point of view, is selecting an appropriate cial Intelligence and Soft Computing, Lecture Notes in
method for the transfer functions and spatio-temporal Computer Science, Vol. 7267, Springer, Berlin/Heidelberg,
responses approximation using finite-dimensional pp. 3–11.
models. The analysis of the approximated models can be Bartecki, K. (2012b). PCA-based approximation of a class
difficult since the irrational transfer functions presented of distributed parameter systems: Classical vs. neural
here are not nuclear in general, i.e., they do not have a network approach, Bulletin of the Polish Academy of Sci-
nuclear Hankel operator. It stems from the fact that the ences: Technical Sciences 60(3): 651–660.
semigroup T (t) considered is stable but not analytic, as Bartecki, K. and Rojek, R. (2005). Instantaneous linearization
shown, e.g., by Curtain and Sasane (2001). Therefore, no of neural network model in adaptive control of heat
theoretical upper bound is known for the H∞ errors in the exchange process, Proceedings of the 11th IEEE Interna-
rational approximants produced by truncated balanced tional Conference on Methods and Models in Automation
realizations and optimal Hankel-norm approximation and Robotics, Mi˛edzyzdroje, Poland, pp. 967–972.
(Sasane, 2002). In the non-linear case, the approximated Bounit, H. (2003). The stability of an irrigation canal system, In-
models could be based, e.g., on the approximation ternational Journal of Applied Mathematics and Computer
properties of artificial neural networks (Bartecki, 2010; Science 13(4): 453–468.
2012a). Callier, F.M. and Winkin, J. (1993). Infinite dimensional
system transfer functions, in R.F, Curtain, A. Bensoussan
Acknowledgment and J.L. Lions (Eds.), Analysis and Optimization of
Systems: State and Frequency Domain Approaches
The author gratefully acknowledges helpful comments for Infinite-Dimensional Systems, Lecture Notes in
from the anonymous reviewers. Control and Information Sciences, Vol. 185, Springer,
Berlin/Heidelberg, pp. 75–101.
References Cheng, A. and Morris, K. (2003). Well-posedness of boundary
Ancona, F. and Coclite, G.M. (2005). On the boundary control systems, SIAM Journal on Control and Optimiza-
controllability of first-order hyperbolic systems, Non- tion 42(4): 1244–1265.
linear Analysis: Theory, Methods & Applications Chentouf, B. and Wang, J.M. (2009). Boundary feedback
63(5–7): e1955–e1966. stabilization and Riesz basis property of a 1-D first order
Arbaoui, M.A., Vernieres-Hassimi, L., Seguin, D. and hyperbolic linear system with L∞ -coefficients, Journal of
Abdelghani-Idrissi, M.A. (2007). Counter-current tubular Differential Equations 246(3): 1119–1138.
 306 K. Bartecki

Christofides, P.D. and Daoutidis, P. (1997). Finite-dimensional Dos Santos, V., Bastin, G., Coron, J.-M. and d’Andréa Novel,
control of parabolic PDE systems using approximate B. (2008). Boundary control with integral action for
inertial manifolds, Journal of Mathematical Analysis and hyperbolic systems of conservation laws: Stability and
Applications 216(2): 398–420. experiments, Automatica 44(5): 1310–1318.
Christofides, P.D. and Daoutidis, P. (1998a). Distributed Evans, L.C. (1998). Partial Differential Equations, American
output feedback control of two-time-scale hyperbolic PDE Mathematical Society, Providence, RI.
systems, Applied Mathematics and Computer Science Filbet, F. and Shu, C.-W. (2005). Approximation of hyperbolic
8(4): 713–732. models for chemosensitive movement, SIAM Journal on
Christofides, P.D. and Daoutidis, P. (1998b). Robust control of Scientific Computing 27(3): 850–872.
hyperbolic PDE systems, Chemical Engineering Science
Friedly, J.C. (1972). Dynamic Behaviour of Processes, Prentice
53(1): 85–105.
Hall, New York, NY.
Contou-Carrere, M.N. and Daoutidis, P. (2008).
Górecki, H., Fuksa, S., Grabowski, P. and Korytowski, A.
Model reduction and control of multi-scale
(1989). Analysis and Synthesis of Time Delay Systems,
reaction–convection processes, Chemical Engineering
Wiley, New York, NY.
Science 63(15): 4012–4025.
Grabowski, P. and Callier, F.M. (2001a). Boundary
Coron, J., d’Andrea Novel, B. and Bastin, G. (2007). A strict
control systems in factor form: Transfer functions and
Lyapunov function for boundary control of hyperbolic
input-output maps, Integral Equations and Operator The-
systems of conservation laws, IEEE Transactions on Au-
ory 41(1): 1–37.
tomatic Control 52(1): 2–11.
Grabowski, P. and Callier, F.M. (2001b). Circle criterion and
Curtain, R.F. (1988). Equivalence of input-output stability
boundary control systems in factor form: Input-output
and exponential stability for infinite-dimensional systems,
approach, International Journal of Applied Mathematics
Mathematical Systems Theory 21(1): 19–48.
and Computer Science 11(6): 1387–1403.
Curtain, R.F., Logemann, H., Townley, S. and Zwart, H.
Gvozdenac, D.D. (1990). Transient response of the parallel flow
(1992). Well-posedness, stabilizability, and admissibility
heat exchanger with finite wall capacitance, Archive of Ap-
for Pritchard–Salamon systems, Journal of Mathematical
plied Mechanics 60(7): 481–490.
Systems, Estimation, and Control 4(4): 1–38.
Jacob, B. and Zwart, H.J. (2012). Linear Port-Hamiltonian
Curtain, R.F. and Sasane, A.J. (2001). Compactness and
Systems on Infinite-dimensional Spaces, Operator Theory:
nuclearity of the Hankel operator and internal stability
Advances and Applications, Vol. 223, Springer-Verlag,
of infinite-dimensional state linear systems, International
Basel.
Journal of Control 74(12): 1260–1270.
Jones, B.L. and Kerrigan, E.C. (2010). When is the discretization
Curtain, R.F. and Weiss, G. (2006). Exponential stabilization of
of a spatially distributed system good enough for control?,
well-posed systems by colocated feedback, SIAM Journal
Automatica 46(9): 1462–1468.
on Control and Optimization 45(1): 273–297.
Jovanović, M.R. and Bamieh, B. (2006). Computation of
Curtain, R.F. and Zwart, H. (1995). An Introduction to Infinite-
the frequency responses for distributed systems with one
Dimensional Linear Systems Theory, Springer-Verlag,
spatial variable, Systems Control Letters 55(1): 27–37.
New York, NY.
Kowalewski, A. (2009). Time-optimal control of infinite
Curtain, R. and Morris, K. (2009). Transfer functions of
order hyperbolic systems with time delays, International
distributed parameters systems: A tutorial, Automatica
Journal of Applied Mathematics and Computer Science
45(5): 1101–1116.
19(4): 597–608, DOI: 10.2478/v10006-009-0047-x.
Delnero, C.C., Dreisigmeyer, D., Hittle, D.C., Young, P.M.,
Li, H.-X. and Qi, C. (2010). Modeling of distributed
Anderson, C.W. and Anderson, M.L. (2004). Exact
parameter systems for applications: A synthesized review
solution to the governing PDE of a hot water-to-air
from time–space separation, Journal of Process Control
finned tube cross-flow heat exchanger, HVAC&R Research
20(8): 891–901.
10(1): 21–31.
Litrico, X. and Fromion, V. (2009a). Boundary control of
Diagne, A., Bastin, G. and Coron, J.-M. (2012). Lyapunov
hyperbolic conservation laws using a frequency domain
exponential stability of 1-D linear hyperbolic systems of
approach, Automatica 45(3): 647–656.
balance laws, Automatica 48(1): 109–114.
Ding, L., Johansson, A. and Gustafsson, T. (2009). Application Litrico, X. and Fromion, V. (2009b). Modeling and Control of
of reduced models for robust control and state estimation of Hydrosystems, Springer, London.
a distributed parameter system, Journal of Process Control Litrico, X., Fromion, V. and Scorletti, G. (2007). Robust
19(3): 539–549. feedforward boundary control of hyperbolic conservation
Dooge, J.C.I. and Napiorkowski, J.J. (1987). The effect of laws, Networks and Heterogeneous Media 2(4): 717–731.
the downstream boundary conditions in the linearized St Lumer, G. and Phillips, R.S. (1961). Dissipative operators
Venant equations, The Quarterly Journal of Mechanics in a Banach space, Pacific Journal of Mathematics
and Applied Mathematics 40(2): 245–256. 11(2): 679–698.
A general transfer function representation for a class of hyperbolic distributed parameter systems 307

Maidi, A., Diaf, M. and Corriou, J.-P. (2010). Boundary Tucsnak, M. and Weiss, G. (2006). Passive and Conservative
control of a parallel-flow heat exchanger by Linear Systems, Nancy University, Nancy.
input-output linearization, Journal of Process Control Tucsnak, M. and Weiss, G. (2009). Observation and Control for
20(10): 1161–1174. Operator Semigroups, Birkhäuser, Basel.
Mattheij, R.M.M., Rienstra, S.W. and ten Thije Boonkkamp, Uciński, D. (2012). Sensor network scheduling for identification
J.H.M. (2005). Partial Differential Equations: Modeling, of spatially distributed processes, International Journal of
Analysis, Computation, Society for Industrial and Applied Applied Mathematics and Computer Science 22(1): 25–40,
Mathematics, Philadelphia, PA. DOI: 10.2478/v10006-012-0002-0.
Miano, G. and Maffucci, A. (2001). Transmission Lines and Weiss, G. (1994). Transfer functions of regular linear systems,
Lumped Circuits, Academic Press, San Diego, CA. Part I: Characterizations of regularity, Transactions of the
Park, H.M. and Cho, D.H. (1996). The use of the American Mathematical Society 342(2): 827–854.
Karhunen–Loève decomposition for the modeling of Wu, W. and Liou, C.-T. (2001). Output regulation of
distributed parameter systems, Chemical Engineering Sci- two-time-scale hyperbolic PDE systems, Journal of Pro-
ence 51(1): 81–98. cess Control 11(6): 637–647.
Patan, M. (2012). Distributed scheduling of sensor networks for Xu, C.-Z. and Sallet, G. (2002). Exponential stability and
identification of spatio-temporal processes, International transfer functions of processes governed by symmetric
Journal of Applied Mathematics and Computer Science hyperbolic systems, ESAIM: Control, Optimisation and
22(2): 299–311, DOI: 10.2478/v10006-012-0022-9. Calculus of Variations 7: 421–442.
Phillips, R.S. (1957). Dissipative hyperbolic systems, Trans- Zavala-Río, A., Astorga-Zaragoza, C.M. and
actions of the American Mathematical Society 86(1): Hernández-González, O. (2009). Bounded positive
109–173. control for double-pipe heat exchangers, Control Engi-
Pritchard, A.J. and Salamon, D. (1984). The linear quadratic neering Practice 17(1): 136–145.
optimal control problem for infinite dimensional systems Zwart, H. (2004). Transfer functions for infinite-dimensional
with unbounded input and output operators, in F. Kappel systems, Systems and Control Letters 52(3–4): 247–255.
and W. Schappacher (Eds.), Infinite-Dimensional Sys-
tems, Lecture Notes in Mathematics, Vol. 1076, Springer, Zwart, H., Gorrec, Y.L., Maschke, B. and Villegas, J. (2010).
Berlin/Heidelberg, pp. 187–202. Well-posedness and regularity of hyperbolic boundary
control systems on a one-dimensional spatial domain,
Rabenstein, R. (1999). Transfer function models for ESAIM: Control, Optimisation and Calculus of Variations
multidimensional systems with bounded spatial domains, 16(04): 1077–1093.
Mathematical and Computer Modelling of Dynamical Sys-
tems 5(3): 259–278.
Rauch, J. and Taylor, M. (1974). Exponential decay of solutions Krzysztof Bartecki received his M.Sc. and
Ph.D. degrees in electrical engineering from the
to hyperbolic equations in bounded domains, Indiana Uni- Faculty of Electrical Engineering, Automatic
versity Mathematical Journal 24(1): 79–86. Control and Computer Science, Opole University
Salamon, D. (1987). Infinite dimensional linear systems with of Technology, in 1996 and 2004, respectively.
unbounded control and observation: A functional analytic He is an assistant professor at the Institute of
Control and Computer Engineering. He has au-
approach, Transactions of the American Mathematical So- thored or co-authored some 40 papers, concern-
ciety 300(2): 383–431. ing mainly the modeling of distributed parameter
Sasane, A. (2002). Hankel Norm Approximation for Infinite- systems as well as application of artificial neural
networks in identification and control of dynamic objects.
dimensional Systems, Springer-Verlag, Berlin.
Staffans, O.J. and Weiss, G. (2000). Transfer functions of Received: 25 April 2012
regular linear systems, Part II: The system operator and Revised: 9 January 2013
the Lax–Phillips semigroup, Transactions of the American
Mathematical Society 354(8): 3229–3262.
Strikwerda, J. (2004). Finite Difference Schemes and Partial
Differential Equations, Society for Industrial and Applied
Mathematics, Philadelphia, PA.
Sutherland, J.C. and Kennedy, C.A. (2003). Improved boundary
conditions for viscous, reacting, compressible flows, Jour-
nal of Computational Physics 191(2): 502–524.