SSRN Id4716404

ed
Velocity control design of hyperbolic distributed

parameter systems using zeroing dynamics and
iew
zeroing-gradient dynamics methods
Ahmed MAIDIa, , Radoslav PAULENb∗, and Jean-Pierre CORRIOUc
ev
a
Laboratoire de Conception et Conduite des Systèmes de Production,
Université Mouloud MAMMERI, 15 000 Tizi-Ouzou, Algérie.
r
b
Faculty of Chemical and and Food Technology,
c
er
Slovak University of Technology in Bratislava, Bratislava, Slovakia.
Reaction and Process Engineering Laboratory,

pe
UMR 7274-CNRS, Lorraine University, ENSIC
1, rue Grandville, BP 20451, 54001 Nancy Cedex, France.
February 2, 2024
ot
tn
Abstract: Velocity control proves to be an effective and a more easily implementable actuation than
boundary and distributed actuations for hyperbolic distributed parameter systems. However, the design
of velocity control for these systems, following the late lumping approach, poses a challenging problem
in control engineering. Noticeably, the velocity controller faces a control singularity issue, resulting in
rin
a loss of controllability that renders the controller impractical. In this paper, we demonstrate that the
zeroing dynamics method is an easy and interesting alternative design approach for velocity control of
hyperbolic distributed parameter systems following the late lumping approach. Furthermore, the control
ep
singularity problem is addressed using the zeroing gradient method. The effectiveness of these design
approaches is clearly demonstrated in the case of a steam-jacketed heat exchanger and a non-isothermal
plug flow reactor.
Pr
Keywords: hyperbolic distributed parameter system, velocity control, late lumping approach, zeroing
dynamics, zeroing gradient dynamics.
∗
Corresponding author: radoslav.paulen@stuba.sk
1
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4716404
1 Introduction
The dynamical behavior of an important class of distributed parameter systems (DPSs) is described
ed
by hyperbolic partial differential equations (PDEs). Examples of such systems include heat exchangers,
non-isothermal reactors, packed bed columns, adsorbers, absorbers, and crystallizers (Bastin and Coron,
2016; Bühler and Franke, 1980; Christofides, 2001a; Ray, 1989; Zuyev and Benner, 2021). Various control
iew
actuators can be applied to these systems, such as boundary control (Maidi et al., 2009, 2010), interior
control (Christofides and Daoutidis, 1996), and velocity control (Gundepudi and Friedly, 1998). From a
practical standpoint, velocity control, a type of bilinear control, proves to be a natural choice, an effective
and a more easily implementable than boundary and distributed actuations (Bühler and Franke, 1980;
Gundepudi and Friedly, 1998).
ev
It is well known that for control design of hyperbolic DPS, the late lumping approach is more effective
than the early lumping approach (Christofides, 2001a,b; Ray, 1989; Singh and Hahn, 2007). This can be
explained by the fact that the modes of the spatial differential operator of a hyperbolic DPS have the
r
same energy and cannot be captured by a lumped parameter model corresponding to a finite number of
modes (Christofides and Daoutidis, 1996; Gundepudi and Friedly, 1998; Shang et al., 2005).
er
Boundary and distributed control of hyperbolic DPSs, following the late lumping approach, have
been extensively studied in the literature (Bastin and Coron, 2016; Christofides and Daoutidis, 1996;
Gugat, 2015; Maidi et al., 2009, 2010; Prieur et al., 2012; Wu and Liou, 2001; Zuyev and Benner, 2021).
pe
Velocity control has been less investigated, and few contributions are reported in the literature. Sira-
Ramirez (1989) extended the theory of variable structure systems to hyperbolic DPS and developed
an infinite dimensional variable structure state feedback. In Bühler and Franke (1980), the Lyapunov
direct method is applied to design both linear and nonlinear controllers, while Butkovskiy’s maximum
ot
principle is employed to design optimal control laws for hyperbolic DPSs. The method of characteristics
and sliding mode control theory have been combined by Hanczyc and Palazoglu (1995) to design a
state feedback controller for a steam heater and non-isothermal plug flow reactor. The input-output
tn
linearization approach has been extended by Gundepudi and Friedly (1998) to hyperbolic DPS with a
single characteristic variable (single flow DPSs) with application to a non-isothermal plug flow reactor.
The same approach has been applied by Maidi et al. (2009) to velocity control of a dual flow hyperbolic
rin
DPS, i.e., a counter-current heat exchanger. Using the method of characteristics, Shang et al. (2005)
proposed a control law that enforces the output tracking successfully.
Input-output linearization approach has been successfully applied for control design of DPSs following
the late lumping approach (Christofides and Daoutidis, 1996; Gundepudi and Friedly, 1998; Maidi and
ep
Corriou, 2014; Wu and Liou, 2001). Nevertheless, its application is limited for DPS with boundary or
interior actuations. For hyperbolic DPS, with velocity control, even the input-output linearization can be
applied, but the controller may be impractical due to the control singularity (loss of controllability) that
Pr
may occur. For instance, the controller fails to kick off with a uniform spatial initial profile (Gundepudi
and Friedly, 1998; Maidi et al., 2009).
Recently, an interesting control design approach based on the Zeroing Dynamics (ZD) method has
been developed (Ding et al., 2023; Zhang et al., 2018; Zheng and Zeng, 2023). The design of the controller
2
is carried out using the ZD method and the activation function (Zhang et al., 2020). Nevertheless, even
though the controller design is a straightforward task, the resulting controller, akin to the input-output
linearization approach, suffers from control singularity (Li et al., 2017; Zhang et al., 2020) when the initial
ed
spatial profile is uniform (constant). Therefore, to implement the control, it is necessary to initiate the
system in open loop to ensure a non-uniform initial profile before switching to closed loop. However, this
solution is economically unattractive. Thus, to address the issue of a uniform initial profile, a Zeroing
iew
Gradient Dynamics (ZGD) method, which combines the ZD method and the Gradient Dynamics (GD)
method, has been proposed to design a controller without control singularity (Ding et al., 2023; Zhang
et al., 2020; Zheng and Zeng, 2023). The controller is designed by minimizing an energy function using
the Gradient Dynamics method (Zhang et al., 2020).
Many successful tracking control applications of the ZD and ZGD methods have been reported in
ev
the literature. Li et al. (2017) used these methods for tracking control of knee exoskeleton system.
Tracking control of a varactor system has been studied by Hu et al. (2017). Nonlinear and robust
control have been developed by Zhang et al. (2018) for a stirred tank system. Zheng and Zeng (2023)
r
investigated the tracking control for robot manipulator both with linear and nonlinear outputs. Both
tracking control and disturbance rejection have been investigated in the case of a double-holding water
er
tank by Ding et al. (2023). Applications of Zero Dynamics (ZD) and Zero-Gradient Dynamics (ZGD) in
solving control problems for various systems, including a simple pendulum system, a double-integrator
pe
system, an inverted pendulum on a cart, and a Van der Pol oscillator, are discussed in detail in Zhang
et al. (2020).
Inspired and motivated by these interesting applications, an extension of these design methods for
DPSs is investigated in this work. To the best of our knowledge, the applications of the ZD and ZGD
methods are limited only to lumped parameter systems (LPSs), i.e., systems described by Ordinary
ot
Differential Equations (ODEs). This paper extends the application of these methods to Distributed
Parameter Systems (DPSs), marking the initial exploration in this direction. Thus, in this paper, the
tn
ZD and ZGD methods are applied to design a velocity controller for single flow DPSs described by
hyperbolic PDEs, and the stability of the resulting closed loop system is investigated. The developed
state feedback control strategies are applied to solve the tracking control in the case of a steam-jacketed
heat exchanger and a non-isothermal plug flow reactor. The objective consists in enforcing the outlet
rin
temperature of the heat exchanger, and the concentration of a species, at the outlet reactor, to track
their desired references. Simulation results are given to demonstrate the performances of the designed
controller using the ZD and ZGD methods.
ep
The contributions of the paper are summarized by the following points:
• The ZD and ZGD design methods are employed to develop a velocity controller for hyperbolic
DPSs, representing the pioneering effort in this field.
Pr
• The use of ZGD controller allows us to overcome the challenging control singularity especially in
the case of a uniform spatial profile,
• Stability analysis is provided for the ZD and ZGD state feedback control strategies.
3
• Simulation results are provided to demonstrate the effectiveness of the ZD and ZGD controllers in
the case of two processes (steam-jacketed heat exchanger and non-isothermal plug flow reactor).
ed
The paper is outlined as follows: Section 2 is dedicated to the formulation of the velocity control
problem for hyperbolic DPSs. Section 3 focuses on the design of the ZD and ZGD state feedbacks,
following the late lumping approach, and includes stability analysis for the resulting closed loop system.
iew
In Section 4, the practical applicability of the developed controllers is demonstrated through their
application to two processes. Finally, Section 5 concludes the paper.
2 Control problem formulation
ev
In this work, let us consider the class of DPSs whose dynamical behavior is described by the following
mathematical model
∂x(t, z) ∂x(t, z)
r
=A + f (x(t, z)), z∈Ω (1)
∂t ∂z
x(t, 0) = x0 (2)
x(0, z) = x∗ (z) er
y(t) = hc(z), xi (t, z)i, 1≤i≤n
(3)
(4)
pe
where A ∈ ℜn×n is the following diagonal matrix
A = diag(−v(t), . . . , −v(t)), (5)

ot
t ∈ [0, ∞) and z ∈ Ω̄ are independent variables that represent time and space position. x(t, z) =
n
[x1 (t, z) . . . xn (t, z)]T ∈ L2 Ω̄ is the vector of state variables, v(t) ∈ ℜ+ the manipulated variable

(flow velocity) and y(t) ∈ ℜ the output variable. f f (x) = [f1 (x) . . . fn (x)]T is a smooth vector
tn
function. c(z) is a continuous function defined on the bounded and closed interval Ω̄ that defines
n
the geometric configuration of the sensor. x∗ ∈ L2 Ω̄

is the initial profile and x0 ∈ ℜn is the
boundary input. Ω = (0, L] and ∂Ω = {0} are the interior and boundary of the entire spatial domain
rin
Ω̄ (Ω̄ = Ω∪∂Ω = [0, L]), respectively. L2 Ω̄ denotes the space of real-valued square-integrable functions

defined on Ω̄ with the standard scalar product, i.e., for g, h ∈ L2 Ω̄ defined as

Z L
ep
hg(z), h(z)i = g(z) h(z) dz (6)

0
The aim is to design a state feedback that forces the output variable defined by Equation (4) to track
the desired reference y d (t).
Pr
Remark 1 Based on the linearized model, Christofides and Daoutidis (1996) investigated the stability of
the hyperbolic DPS (1–4). It is shown that this system is stable since the eigenvalues of the matrix A
are negative real, given that the fluid flow velocity v(t) is positive.
4
3 State feedback design
In this section, the ZD and ZGD methods are used to design, following the late lumping approach, the
ed
state feedback that enforces the controlled output defined by Equation (4) to track a desired trajectory
y d (t).
Assumption 1 The desired reference y d (t) is differentiable.
iew
3.1 State feedback design using the ZD method
To solve the tracking problem of the hyperbolic DPS (1–4), the tracking error is defined as
ev
e(t) = y d (t) − y(t) (7)
The control design is based on the following first-order ordinary differential equation which forces
r
the error to decrease exponentially
er
ė(t) + λ e(t) = 0
where λ is a strictly positive real number (λ > 0). This corresponds to the simplest expression of the
zeroing dynamics design formula according to (Zhang et al., 2020)
(8)
pe
ė(t) + λ φ(e(t)) = 0 (9)
where the activation function φ is linear, i.e., φ(e) = e.

ot
Taking into account Assumption 1 and employing Equation (8), it follows that
ẏ d (t) − ẏ(t) + λ e(t) = 0 (10)

tn
and using Equation (4), it gives

d ∂xi (t, z)
ẏ (t) − c(z), + λ e(t) = 0 (11)
∂t
rin
and considering the state equation (1), Equation (11) takes the following form

d ∂xi (t, z)
ẏ (t) − c(z), −v(t) + fi (x(t, z)) + λ e(t) = 0 (12)
ep
∂z
or equivalently

∂xi (t, z)
c(z), v(t) + ẏ d (t) − hc(z), fi (x(t, z))i + λ e(t) = 0 (13)
Pr
∂z
consequently, the state feedback is derived from Equation (13) as follows
5
hc(z), fi (x(t, z))i − λ e(t) − ẏ d (t)
v(t) = (14)
∂xi (t, z)
c(z),
ed
∂z
Remark 2 λ is a tuning parameter of the controller (14) that can be used, according to Equation (8), to
adjust the convergence dynamics rate of the tracking error e(t).
iew

∂xi (t, z)
Proposition 1 If c(z), 6= 0, the ZD state feedback (14) enforces the output tracking for the
∂z
hyperbolic DPS (1–4).
Proof 1 The design of the ZD controller (14) is based on Equation (8). Hence, in closed loop, the
ev
tracking error is given by
e(t) = e(0) e−λ t (15)
r
thus since λ > 0, at steady-state,
t→∞ er
lim e(t) = 0,
indicating that the output tracking is successfully achieved, that is, y(t) globally exponentially converges
(16)
pe
to y d (t).

∂xi (t, z)
Note that when c(z), = 0, the manipulated variable v(t) is infinite, which means that
∂z
the state feedback (14) is inapplicable. For example, in many DPSs, in the case where the initial profile
ot
x∗ (z) was chosen uniform (constant), the controller (14) would fail as the denominator of state feedback
(14) be zero. Of course, practically, this issue can be easily avoided for example by choosing a non
constant strictly linear initial profile. This control singularity problem is also encountered with the
tn
input-output linearization approach (Maidi et al., 2009). In the following subsection, the ZGD approach
is used to design a state feedback that solves the singularity problem.
rin
3.2 State feedback design using the ZGD method

To overcome the control singularity problem of the state feedback (14), the GD method is used to
design a valid state feedback. Thus, in accordance with the ZD design method, the function J(xi (t, , z))
ep
is defined as the left-hand side of Equation (10), i.e.,

∂xi (t, z)
J(xi (t, z)) = c(z), v(t) + ẏ d (t) − hc(z), fi (x(t, z))i + λ e(t) (17)
∂z
Pr
Now, by considering the following energy function
[J(xi (t, z))]2

ε(t) = (18)
2
6
and using the gradient of energy with respect to the manipulated input similarly to Zhang et al. (2020)
as a descent technique
ed
∂ε(t)
v̇(t) = −γ (19)
∂v(t)
the following state feedback results
iew
" #
∂xi (t, z) ∂xi (t, z)
v̇(t) = −γ c(z), c(z), v(t) + ẏ d (t) − hc(z), fi (x(t, z))i + λ e(t) (20)
∂z ∂z
where γ is a tuning parameter of the controller (20).
ev
Proposition 2 For a uniformly distributed initial profile x∗ (z) and an initial flow velocity control v(0) ∈
ℜ+ , the ZGD state feedback (14) ensures a bounded steady state tracking error in closed loop for the
r
hyperbolic DPS (1–4).
er
Proof 2 The ZGD state feedback (20) can be expressed in the following form
v̇(t) = −γ α(t) (α(t) v(t) + Φ(x)) (21)

pe
with

∂xi (t, z)
α(t) = c(z), , Φ(x) = ẏ d (t) − hc(z), fi (x(t, z))i + λ e(t) (22)
∂z
ot
Let v̄(t) be the theoretical flow velocity control that achieves the output tracking, i.e., v̄(t) is the
solution of (13) written under the form
tn
α(t) v̄(t) + Φ(x) = 0 (23)
In the same manner as in (Zheng and Zeng, 2023) and (Li et al., 2017), two cases are possible,
according to the value of α
rin
• Case α(t) 6= 0: Let E be the control error defined by

ep
E = v − v̄(t) (24)
hence
˙
Ė = v̇ − v̄(t) (25)
Pr
Substituting v̇ according to Equation (21) gives
˙
Ė = −γ α(t) (α(t) v(t) + Φ(x)) − v̄(t) (26)
7
Combining Equation (26) with Equation (23) yields
ed
˙
Ė(t) = −γ α(t) (α(t) v(t) − α(t) v̄(t)) − v̄(t) (27)
and considering (24), Equation (27) simplifies as
iew
˙
Ė(t) = A(t) E(t) − v̄(t) (28)
with A(t) = −γ α2 (t).

The operator A(t) generates a two parameter semigroup U (t, s) given by
ev
U (t, s) = eI(t)−I(s) , 0≤s≤t (29)
Rτ
with I(τ ) = 0
A(ξ) dξ.
r
Given that γ > 0, the semigroup U (t, s) is stable, indicating the existence of two constants M ≥ 1
and ω > 0 such that (Pazy, 1983)
er
kU (t, s)k ≤ M e−ω (t−s) (30)
pe
The solution of Equation (28) is given by (Pazy, 1983)
Z t
E(t) = U (t, 0) E(0) + ˙ ds
U (t, s) v̄(s) (31)
0
ot
hence, by Gronwall’s inequality (Robinson, 2001)
Z t
tn
kE(t)k ≤ kU (t, 0)k kE(0)k + ˙

kU (t, s)k kv̄(s)k ds (32)
0
Since γ > 0 and Φ(x) is bounded, it follows from Equation (21) that |v̇(t)| ≤ η ≤ ∞. Equations
(30) and (32) yield
rin
Z t
kE(t)k ≤ M e −ω t
kE(0)k + M e−ω (t−s) η ds (33)
ep
0
η M η
≤ M e−ω t kE(0)k − + (34)
ω ω
hence, it can be concluded that

Pr
Mη
lim kE(t)k ≤ (35)
t→∞ ω
Now, using Equations (13) and (22), it can be written
8
ė(t) + λ e(t) = α(t) v(t) + Φ(x) (36)
ed
and combining Equations (23), (24) and (36), yields
ė(t) + λ e(t) = α(t) E(t) (37)
iew
and since
−kα(t)k kE(t)k ≤ α(t) E(t) ≤ kα(t)k kE(t)k (38)
ev
therefore, as α(t) 6= 0, 0 < α2 (t) ≤ β 2 ≤ ∞ (β > 0), hence
−β kE(t)k ≤ −kα(t)k kE(t)k ≤ ė(t) + λ e(t) ≤ kα(t)k kE(t)k ≤ β kE(t)k (39)
r
Considering Equation (35), it gives
βMη
ω
− er
≤ ė(t) + λ e(t) ≤
and using Cronwall’s inequality (Robinson, 2001)

βMη
ω
(40)
pe
βMη βMη
−C e−λ t − ≤ e(t) ≤ C e−λ t + (41)
λω λω
βMη
and since is positive, Equation (41) simplifies to
ot
λω
βMη
|e(t)| ≤ |C| e−λ t + (42)
λω
tn
where C is an integration constant. Consequently,
βMη
lim |e(t)| ≤ (43)
rin
t→∞ λω
• Case α(t) = 0: Let t0 be the singularity control instant, i.e, limt→t0 α(t) = 0. Equation (20) yields
ep
lim v̇(t) = lim v̇(t) = 0, (44)

t→t0 α(t)→0
+
which means that v(t) is continuous at t0 , i.e., v(t− 0 ) = v(t0 ) = v(t0 ). Considering Remark 1, the
+ +
bounded control sequence v(t− −
0 ), v(t0 ), and v(t0 ) yields bounded outputs y(t0 ), y(t0 ), and y(t0 ).
Pr
+
As a result, the tracking errors e(t− 0 ), e(t0 ), and e(t0 ) remain bounded, given the bounded nature
of the desired reference y d (t).
On account of the above convergence analysis of the error control E(t), it can be inferred that the
9
control v(t) remains within finite bounds throughout the tracking operation for both singularity and non-
singularity cases.
ed
Remark 3 The ZGD state feedback given by (20) does not suffer from the control singularity problem
in comparison to the controller (14) and the controller designed using the input-output linearization
approach (Maidi et al., 2009).
iew
Remark 4 For simulation and implementation of the controller (14), an initial value of the velocity,
i.e., v(0) must be specified. This can be determined through trial and error or selected as a reasonable
practical value.
Remark 5 It is noteworthy that the parameter γ significantly impacts the value of ω, hence in the case
ev
of no control singularity (α(t) 6= 0), from Equation (43), it can be seen that the tracking error can be
rendered small by increasing the tuning parameters λ and γ.
r
4 Application examples
er
In this section, the output tracking performances of the controllers (14) and (20) resulting from the ZD
and ZDG methods respectively, are evaluated in the case of a steam-jacketed tubular heat exchanger and
plug flow reactor. The simulations are carried out using the method of lines based on finite difference
pe
schemes (Vande Wouwer et al., 2004). For both processes, the controlled output is defined at outlet
point z = L, hence the shaping function c(z) is modeled as a Dirac delta function at z = L, i.e.,
c(z) = δL (z) = δ(z − L). The desired reference y d (t) is obtained by filtering the set point y sp (t) with a
first order filter of time constant τ as
ot
Y d (s) 1
sp
= (45)
Y (s) τs+1
tn
where s is the Laplace variable. Y d (s) and Y sp are the Laplace transforms of y d (t) and y sp (s), respectively.
4.1 Steam-jacketed tubular heat exchanger

rin
The evolution of the temperature T (t, z) within the tube of a steam-jacketed tubular heat exchanger
(Figure 1), with a length L, is described by the following dimensionless model (Shang et al., 2005)
ep
∂T (t, z) ∂T (t, z)
= −v(t) + H (Tj (t) − T (t, z)) (46)
∂t ∂z
T (t, 0) = Tin (47)
T (0, z) = T ∗ (z) (48)
Pr
where all variables are dimensionless. v is the fluid velocity, Tin the inlet fluid temperature, Tj the
steam jacket temperature which is assumed to be constant, T ∗ (z) the initial temperature profile and H
a positive (thermal) constant. The output to be controlled is the outlet temperature, that is,
10
ed
Figure 1: Steam-jacketed heat exchanger.
Heat exchanger ZD state feedback ZGD state feedback
iew
Parameter Value Parameter Value Parameter Value
H 2 λ 1 λ 10
Tj 5 γ 0.1
Tin 0 v(0) 2
ev
L 1
v 2
Table 1: Steam-jacketed heat exchanger simulation parameters.
r
er
y(t) = T (t, L)
Using Equations (14) and (20), the following ZD and ZGD controllers result
(49)
pe

H hc(z), (Tj (t) − T (t, z))i − λ y d (t) − y(t) − ẏ d (t)
v(t) = (50)
∂T (t, z)
∂z z=L
ot
"
∂T (t, z) ∂T (t, z)
v(t) + λ y d (t) − y(t)

v̇(t) = −γ
∂z z=L ∂z z=L
tn
#
+ ẏ d (t) − H hc(z), (Tj (t) − T (t, z))i (51)
The simulations are carried out with parameters given in Table 1.

rin
4.1.1 ZD state feedback
In the time interval [0, 10[, it is assumed that the heat exchanger is at open loop steady state obtained
ep
with parameters given in Table 1. Then, in the time [10, 40], the ZD feedback controller is applied. The
ZD state feedback performance is evaluated by imposing two set points y sp (t) = 3 and y sp (t) = 3.5 at
t = 10 and t = 30 with the filter time constant τ = 2, respectively. The simulation results are depicted
Pr
in Figures 2–3. It can be observed that the ZD state feedback achieves the output tracking (Fig. 2).
The flow rate v(t) exhibits physically mild moves (Fig. 3).
11
3.6
ed
3.4
iew
3.2
ev
3
Open loop Closed loop
r
2.8
0 10 20 30 40 50
er
Figure 2: Steam-jacketed heat exchanger: output evolution with ZD state feedback.
pe
2.4
ot
2.2
tn
2
rin
1.8

ep
1.6
0 10 20 30 40 50
Pr
Figure 3: Steam-jacketed heat exchanger: fluid flow velocity evolution with ZD state feedback.
12
3.8
ed
3.6
iew
3.4
3.2
ev
3
r
0 20 40
er
pe
Figure 4: Steam-jacketed heat exchanger: output evolution with ZGD state feedback.
4.1.2 ZGD state feedback
In the case of a constant initial temperature profile, the ZGD state feedback should be used. However,
ot
a heat exchanger in open loop does not possess a constant temperature profile. Thus, in the first time
interval [0, 10[, the heat exchanger is operated in open loop with a velocity equal to 2. Then, at t = 10,
assuming that the actual spatial temperature profile is unknown, a constant initial temperature profile
tn
is guessed for the heat exchanger
T ∗ (z) = 3.0783 ∀z (52)

rin
equal to the previous outlet temperature in open loop. Thus, a control singularity occurs at t = 10.
This is due to the derivative of the temperature at the outlet z = L being null. Consequently, the
ZD state feedback (50) fails to track the desired reference. Thus, the performance of the ZGD state
ep
feedback (51) can be evaluated in the time interval [10, 40] where the same set points as for the ZD
control are imposed. The parameters of the ZGD controller are given in Table 1. After a very short
transient, the controlled temperature perfectly tracks the reference trajectory (Fig. 4). The flow rate
shows also a rapid variation before getting smoother (Fig. 5). These rapid variations are explained by
Pr
the fluctuations of the spatial derivative of the outlet temperature (Fig. 6) which is present in the ZGD
control law.
13
2.2
ed
2
iew
1.8
ev
1.6
r
0 20 40
er
pe
Figure 5: Steam-jacketed heat exchanger: evolution of the fluid flow velocity v(t) with ZGD state
feedback.
2
ot
1.95
tn
1.9
rin
1.85
ep
1.8
0 50
Pr
Figure 6: Steam-jacketed heat exchanger: evolution of the temperature derivative at z = L with ZGD
state feedback.
14
ed
Figure 7: Non-isothermal plug flow reactor.
iew
4.2 Non-isothermal plug flow reactor
The mathematical model of a plug flow reactor operating under non-isothermal conditions and some
reasonable assumptions (Christofides and Daoutidis, 1996), involving the occurrence of two first order
reactions (Fig. 7)
ev
1k 2 k
A −→ B −→ C (53)
is given by the following PDEs
r
∂CA (t, z)
∂t
∂CB (t, z)
∂t
= − v(t)
= − v(t)
∂CA (t, z)
∂z
∂CB (t, z)
∂z
er
− k1 CA (t, z)
+ k1 CA (t, z) − k2 CB (t, z)
(54)
(55)
pe
∂Tr (t, z) ∂Tr (t, z) ∆H1 ∆H2
= − v(t) − k1 CA (t, z) − k2 CB (t, z)
∂t ∂z ρm cpm ρm cpm
Uw
+ (Tj (t) − Tr (t, z)) (56)
ρm cpm Vr
ot
with the following boundary conditions

tn
CA (t, 0) = CA,in , CB (t, 0) = CB,in , Tr (t, 0) = Tr,in (57)
and the initial conditions

rin
CA (0, z) = CA∗ (z), CB (0, z) = CB∗ (z), Tr (0, z) = Tr∗ (z) (58)
with the kinetic constants following Arrhenius law
Ei
ep
−
ki = ki0 e R Tr (t, z) i = {1, 2} (59)
The variables CA and CB are the concentrations of the species A and B in the reactor, respectively,
Pr
and Tr the reactor temperature. The various model parameters and their corresponding values are
provided in Table 2 (Wu and Liou, 2001).
The control objective involves designing a state feedback (flow velocity) to control the concentration
of the species B at the reactor outlet z = L. Therefore, the output to be controlled is defined as follows
15
Parameter Designation Value Unit
v Fluid flow velocity 1 m.min−1
ed
L Reactor length 2 m
Vr Reactor volume 2 lt
4
E1 Activation energy for the first reaction 2 × 10 kcal.kmol−1
E2 Activation energy for the second reaction 2 × 104 kcal.kmol−1
iew
k10 Constant 3 × 1012 min−1
k20 Constant 8 × 1011 min−1
R Gas constant 1.987 kcal.kmol−1 .K−1
∆Hr1 Enthalpy of the first reaction −4 × 104 kcal.kmol−1
ev
∆Hr2 Enthalpy of the second reaction −2 × 105 kcal.kmol−1
ρm Fluid density 1000 kg.lt−1
cpm Fluid heat capacity 0.231 kcal.kg.K−1
r
Uw Heat transfer coefficient 2 × 104 kcal.min−1 .K−1
Tj Jacket temperature 350 K
CA,in
CB,in
Tr,in
Reactant A inlet concentration
Reactant B inlet concentration
Inlet stream temperature
er 1
0
350
mol.l−1
mol.l−1
K
pe
Table 2: Plug flow reactor parameters (Wu and Liou, 2001).
ot
y(t) = CB (t, L) (60)
By defining x1 = CA , x2 = CB , and x3 = Tr , the subscript i in Equation (4) is equal to 2. Therefore,

using Equations (14) and (20), the following ZD and ZGD controllers result
tn

hc(z), k1 CA (t, z) − k2 CB (t, z)i − λ y d (t) − y(t) − ẏ d (t)
v(t) = (61)
∂CB (t, z)
rin
∂z z=L
"
∂CB (t, z) ∂CB (t, z)
v̇(t) = −γ v(t) + ẏ d (t)
∂z ∂z
ep
z=L z=L
#
− hc(z), k1 CA (t, z) − k2 CB (t, z))i + λ (y d (t) − y(t)) (62)
Pr
The used controller parameters are provided in Table 3.
16
Designation ZD state feedback ZGD state feedback
λ 20 102
ed
γ 104
v(0) 1
Table 3: Non-isothermal plug flow reactor: controllers and reference parameters.
iew
1
0.5
0.5
ev
0 0
0 1 2 0 1 2
r
360
355 er
pe
350
0 1 2
Figure 8: Reactor steady state profile.

ot
4.2.1 ZD state feedback
The plug flow reactor is operated in open loop at steady state in the time interval [0, 10[ min, then in
tn
closed loop in [10, 40] min. The steady state profiles are given in Figure 8, obtained with parameters of
Table 2.
To evaluate the tracking performance of the ZD state controller, two set points y sp (t) = 0.6 mol.l−1
rin
and y sp (t) = 0.5 mol.l−1 at t = 10 min and t = 25 min with the filter time constant τ = 2 min, respectively.
Figures 9–10 show the obtained results. It can be clearly seen that the ZD state feedback performs well
to achieve perfect tracking with smooth fluctuations in the fluid flow velocity (Fig. 10). This is expected
ep
because, with the flow velocity as the control variable, and considering the delay between the inlet and
the outlet of the reactor, the closed loop system can be understood as a time-varying delay system,
leading to oscillations.
Pr
4.2.2 ZGD state feedback
To assess the performance tracking of the ZGD state feedback, it is assumed that the fluid crossing the
plug flow reactor contains no reactant A in the time interval [0, 10[ min so that CA,in = 0, then suddenly
17
0.65
ed
0.6
iew
0.55
ev
0.5
r
0.45
0 10 20 30 40
er
Figure 9: Non-isothermal plug flow reactor: output evolution with ZD state feedback.
pe
2
ot
tn
1.5
rin
1
ep
0.5 Open loop Closed loop
0 10 20 30 40
Pr
Figure 10: Non-isothermal plug flow reactor: fluid flow velocity evolution with ZD state feedback.
18
0.7
0.6
ed
0.5
0.4
0.3
iew
0.2
0.1
ev
0 5 10 15 20 25 30 35 40
Figure 11: Non-isothermal plug flow reactor: output evolution with ZGD state feedback.
r
the component A is introduced at t = 10 min at a concentration CA,in = 1 mol.l−1 . The reactor is
er
operated in open loop in the time interval [0, 10[ min, then in closed loop in [10, 40] min. Due to this
step change of concentration, it results that the spatial derivative of the outlet temperature is zero at
t = 10 min creating a control singularity with the ZD controller. Thus, the ZGD control is applied.
pe
The same set points as in the ZD case are imposed. Even, in this situation, the ZGD state feedback
enforces the outlet concentration CB (L, t) to track its desired reference (Fig. 11). It can be clearly seen
that the controlled output exhibits a delay and starts with a jump. This is due to the control singularity
observed at t = 10 min, which leads to a fluid flow velocity (Fig. 12) that exhibits subdued fluctuations
ot
in the transient response, particularly at the beginning, but the moves remain physically applicable. It
is noteworthy that, after overcoming the singularity issue, the control oscillations are reduced compared
tn
to the ZD controller case. This is attributed to the ZGD controller, which can be perceived as applying
a filtered control. Figure 13 displaying the spatial derivative of the outlet concentration shows the effect
of the control singularity at t = 10 min.
rin
5 Conclusion
In this paper, the Zeroing Dynamics and Zeroing Gradient Dynamics methods are applied to solve the
ep
velocity control problem of hyperbolic DPSs. It is shown that these methods easily allow the design,
following the late lumping approach, of a stable infinite-dimensional state feedback that enforces output
tracking. The design consists solely of evaluating the derivatives of the tracking error.
The ZD method yields a state feedback that is impractical when the initial profile is uniform, leading
Pr
to a control singularity issue. Therefore, by combining the ZD method with the Gradient Dynamics
method, a ZGD state feedback can be designed to address the tracking problem even when the initial
profile is uniform. The ZD state feedback involves one tuning parameter that fixes the tracking error
19
ed
0.15
0.1
iew
0.05
ev
0
-0.05
r
0 5 10 15 20 25 30 35 40
er
Figure 12: Non-isothermal plug flow reactor: fluid flow velocity evolution with ZGD state feedback.
pe
ot
0
tn
-0.2
-0.4
rin
-0.6
-0.8
ep
-1
0 5 10 15 20 25 30 35 40
Figure 13: Non-isothermal plug flow reactor: evolution of the CB concentration derivative at z = L with
Pr
ZGD state feedback.
20
convergence rate, whereas ZGD state feedback involves a second tuning parameter that can be used to
enhance the transient response.
The tracking capabilities of both ZD and ZGD state feedbacks are demonstrated through simulation
ed
in the case of a steam-jacketed tubular heat exchanger and a non-isothermal plug flow reactor. The
conducted study demonstrates that the ZGD controller not only successfully addresses the singularity
problem but also enhances the performance.
iew
The developments of the present study can serve as a starting point or catalyst, inspiring the extension
of ZD and ZGD methods to other classes of DPS problems and fostering the development of theoretical
results.
ev
Acknowledgement
Radoslav Paulen acknowledges the funding from the Slovak Research and Development Agency under the
project APVV-21-0019, the Scientific Grant Agency of the Slovak Republic under the grant 1/0691/21,
r
by the European Commission under the grant no. 101079342 (Fostering Opportunities Towards Slovak
Excellence in Advanced Control for Smart Industries).
References
er
pe
Georges Bastin and Jean-Michel Coron. Stability and Boundary Stabilization of 1-D Hyperbolic Systems.
Birkhäuser, Cham, 2016.
Erhard Bühler and Dieter Franke. Topics in Identification and Distributed Parameter Systems.
ot
Vieweg+Teubner Verlag, Wiesbaden, 1980.
Panagiotis D. Christofides. Nonlinear and robust control of PDE systems: methods and applications to
tn
transport-reaction processes. Birkhaüser, Boston, 2001a.
Panagiotis D. Christofides. Control of nonlinear distributed process systems: Recent developments and
challenges. AIChE Journal, 47(3):514–518, 2001b.
rin
Panagiotis D. Christofides and Prodromos Daoutidis. Feedback control of hyperbolic PDE systems.
AIChE Journal, 42(11):3063–3086, 1996.
ep
Yaqiong Ding, Hanguang Jia, Yunong Zhang, and Binbin Qiu. High-order modeling, zeroing dynamics
control, and perturbations rejection for non-linear double-holding water tank. Mathematics, 11(13),
2023.
Pr
Martin Gugat. Optimal Boundary Control and Boundary Stabilization of Hyperbolic Systems.
Birkhäuser, Cham, 2015.
Pavan Kumar Gundepudi and John C. Friedly. Velocity control of hyperbolic partial differential equation
systems with single characteristic variable. Chemical Engineering Science, 53(24):4055–4072, 1998.
21
Eric M. Hanczyc and Ahmet Palazoglu. Sliding mode control of nonlinear distributed parameter chemical
processes. Industrial & Engineering Chemistry Research, 34(2):557–566, 1995.
ed
Chaowei Hu, Dongsheng Guo, Xiangui Kang, and Yunong Zhang. Zhang dynamics tracking control of
varactor system with stability analysis. In 2017 13th International Conference on Natural Computa-
tion, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD), pages 166–171, 2017.
iew
Zhan Li, Ziguang Yin, and Hong Cheng. Tracking control of knee exoskeleton system with time-
dependent inertial and viscous parameters. IFAC-PapersOnLine, 50(1):1322–1327, 2017. 20th IFAC
World Congress.
Ahmed Maidi and Jean-Pierre Corriou. Boundary geometric control of a linear Stefan problem. Journal
ev
of Process Control, 24(6):939–946, 2014.
Ahmed Maidi, Moussa Diaf, and Jean-Pierre Corriou. Boundary geometric control of a counter-current
heat exchanger. Journal of Process Control, 19(2):297–313, 2009.
r
Ahmed Maidi, Moussa Diaf, and Jean-Pierre Corriou. Boundary control of a parallel-flow heat exchanger
er
by input–output linearization. Journal of Process Control, 20(10):1161–1174, 2010.
Amnon Pazy. Semigroups of Linear Operators and Applications to Partial Differential Equations.
pe
Springer, New York, 1983.
Christophe Prieur, Antoine Girard, and Emmanuel Witrant. Lyapunov functions for switched linear
hyperbolic systems. IFAC Proceedings Volumes, 45(9):382–387, 2012.
ot
Willis Harmon Ray. Advanced Process Control. Butterworth, Boston, 1989.
James C. Robinson. Infinite-dimensional dynamical systems. An introduction to dissipative parabolic

tn
PDEs and the theory of global attractors. Cambridge University Press & Assessment, Cambridge,
2001.
H. Shang, J. F. Forbes, and M. Guay. Feedback control of hyperbolic distributed parameter systems.
rin
Chemical Engineering Science, 60(4):969–980, 2005.
Abhay Singh and Juergen Hahn. Effect of finite-dimensional approximations on observability analysis of
distributed parameter models. IFAC Proceedings Volumes, 40(5):197–202, 2007. 8th IFAC Symposium
ep
on Dynamics and Control of Process Systems.
Hebertt Sira-Ramirez. Distributed sliding mode control in systems described by quasilinear partial
differential equations. Systems & Control Letters, 13(2):177–181, 1989.
Pr
Alain Vande Wouwer, P. Saucez, and William E. Schiesser. Simulation of distributed parameter systems.
Industrial & Engineering Chemistry Research, 43(14):3469–3477, 2004.
22
Wei Wu and Ching-Tien Liou. Output regulation of nonisothermal plug-flow reactors with inlet pertur-
bations. Computers & Chemical Engineering, 25(2):433–443, 2001.
ed
Yunong Zhang, Yaqiong Ding, Binbin Qiu, Jianfeng Wen, and Xiaodong Li. ZD method based nonlinear
and robust control of agitator tank. Asian Journal of Control, 20(5):1–16, 2018.
Yunong Zhang, Binbin Qiu, and Xiaodong Li. Zhang-Gradient Control. Springer, Singapore, 2020.
iew
Zheng Zheng and Delu Zeng. From zeroing dynamics to zeroing-gradient dynamics for solving track-
ing control problem of robot manipulator dynamic system with linear output or nonlinear output.
Mathematics, 11(7), 2023.
ev
Alexander Zuyev and Peter Benner. Stabilization of crystallization models governed by hyperbolic
systems. In Grigory Sklyar and Alexander Zuyev, editors, Stabilization of Distributed Parameter
Systems: Design Methods and Applications, pages 123–135, Cham, 2021. Springer.
r
er
pe
ot
tn
rin
ep
Pr
23

SSRN Id4716404

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id4716404

Uploaded by

Copyright:

Available Formats

ed

Velocity control design of hyperbolic distributed

Ahmed MAIDIa, , Radoslav PAULENb∗, and Jean-Pierre CORRIOUc

Université Mouloud MAMMERI, 15 000 Tizi-Ouzou, Algérie.

Reaction and Process Engineering Laboratory,

1, rue Grandville, BP 20451, 54001 Nancy Cedex, France.

The contributions of the paper are summarized by the following points:

2 Control problem formulation

A = diag(−v(t), . . . , −v(t)), (5)

defined on Ω̄ with the standard scalar product, i.e., for g, h ∈ L2 Ω̄ defined as

hg(z), h(z)i = g(z) h(z) dz (6)

Assumption 1 The desired reference y d (t) is differentiable.

where the activation function φ is linear, i.e., φ(e) = e.

ẏ d (t) − ẏ(t) + λ e(t) = 0 (10)

and using Equation (4), it gives

e(t) = e(0) e−λ t (15)

3.2 State feedback design using the ZGD method

is defined as the left-hand side of Equation (10), i.e.,

Now, by considering the following energy function

[J(xi (t, z))]2

where γ is a tuning parameter of the controller (20).

v̇(t) = −γ α(t) (α(t) v(t) + Φ(x)) (21)

α(t) v̄(t) + Φ(x) = 0 (23)

• Case α(t) 6= 0: Let E be the control error defined by

Substituting v̇ according to Equation (21) gives

and considering (24), Equation (27) simplifies as

with A(t) = −γ α2 (t).

hence, by Gronwall’s inequality (Robinson, 2001)

kE(t)k ≤ kU (t, 0)k kE(0)k + ˙

hence, it can be concluded that

ė(t) + λ e(t) = α(t) E(t) (37)

−kα(t)k kE(t)k ≤ α(t) E(t) ≤ kα(t)k kE(t)k (38)

−β kE(t)k ≤ −kα(t)k kE(t)k ≤ ė(t) + λ e(t) ≤ kα(t)k kE(t)k ≤ β kE(t)k (39)

and using Cronwall’s inequality (Robinson, 2001)

where C is an integration constant. Consequently,

lim v̇(t) = lim v̇(t) = 0, (44)

4.1 Steam-jacketed tubular heat exchanger

Heat exchanger ZD state feedback ZGD state feedback

Table 1: Steam-jacketed heat exchanger simulation parameters.

The simulations are carried out with parameters given in Table 1.

4.1.1 ZD state feedback

Open loop Closed loop

4.1.2 ZGD state feedback

is guessed for the heat exchanger

T ∗ (z) = 3.0783 ∀z (52)

is given by the following PDEs

with the following boundary conditions

CA (t, 0) = CA,in , CB (t, 0) = CB,in , Tr (t, 0) = Tr,in (57)

and the initial conditions

with the kinetic constants following Arrhenius law

y(t) = CB (t, L) (60)

By defining x1 = CA , x2 = CB , and x3 = Tr , the subscript i in Equation (4) is equal to 2. Therefore,

The used controller parameters are provided in Table 3.

Table 3: Non-isothermal plug flow reactor: controllers and reference parameters.

Figure 8: Reactor steady state profile.

4.2.1 ZD state feedback

4.2.2 ZGD state feedback

0.5 Open loop Closed loop

ZGD state feedback.

Vieweg+Teubner Verlag, Wiesbaden, 1980.

transport-reaction processes. Birkhaüser, Boston, 2001a.

Willis Harmon Ray. Advanced Process Control. Butterworth, Boston, 1989.

James C. Robinson. Infinite-dimensional dynamical systems. An introduction to dissipative parabolic