IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 36, NO.

4, JULY 2006 543

Differential Petri Net Models for Industrial
Automation and Supervisory Control
Isabel Demongodin, Member, IEEE, and Nick T. Koussoulas, Member, IEEE
AbstractSupervisory control systems play a central role in
modern industrial automation. However, control theory has re-
cently made signicant advances in modeling mixed continu-
ous/discrete event systems (hybrid control systems), whose typi-
cal instantiations include the industrial supervisory controller. This
article shows howdifferential Petri nets, a model for hybrid control
systems, can be used to represent industrial supervisory systems
in a unied way. Typical industrial automation tests can be mod-
eled, whereas the effect of communication protocols and software
can be straightforwardly included using conventional Petri nets.
Therefore, a global model for the operation of an industrial control
system can be formed and its behavior analyzed.
Index TermsAutomation, hybrid systems, Petri nets.
I. INTRODUCTION
A
MORE recent point of convergence between industrial
automation and control theory has been the objective to
deal effectively with supervisory control systems. The benets
from this convergence are mutual: industry can certainly benet
from the theoretical support in terms of better and more reliable
design and increased power of analysis, whereas control theory
can have a vast and important domain of application.
From the part of control theory, the effort has been to create a
modeling framework that can include not only the technological
process, but also its supervisory scheme(s) (Fig. 1). These two
parts are totally different in nature. The technological process
has continuous dynamics that are typically modeled by ordinary
differential equations. The supervisory mechanism responsible
for deciding on commands (e.g., set points) and control poli-
cies (e.g., PID, self-tuning) typically consists of logical oper-
ations executed via programmable logic controllers, and the
corresponding models belong to the class of discrete event dy-
namic systems (DEDS). Moreover, networking can introduce
additional complexity. The resulting overall supervisory control
system, being a mixture of continuous time and discrete event
dynamic processes, has been termed hybrid or discontinu-
ous. Modeling, analysis, control, and synthesis of such systems
pose a number of challenging problems. Different approaches to
modeling have been proposed and, at present, there are already
Manuscript received November 14, 2003; revised December 2, 2004. This
work was partially supported by the European Commission of the European
Union through the ESPRIT-8924 program SESDIP-Structural Evaluation and
Synthesis of Distributed Industrial Processes. This paper was recommended by
Associate Editor P. Borne.
I. Demongodin is with the Ecole des Mines de Nantes, France. She is
now with LISA (Laboratory of Automation and System Engineering), Uni-
versity. AngersCNRS FRE 2656, 49000 Angers, France (e-mail: isabel.
[email protected]).
N. T. Koussoulas is with the Electrical and Computer Engineering Depart-
ment, University of Patras, Patras, Greece (e-mail: [email protected]. gr).
Digital Object Identier 10.1109/TSMCC.2005.848154
a large number of models for hybrid control systems. The basis
of the modeling effort is to start from a model for either part of
the hybrid system and extend it so it will be able to account for
the dynamics of the other part. Most of the proposed models are
based on a DEDS model with a suitable extension for the con-
tinuous dynamics representation. An informative overview can
be found in [1] and [2], whereas current and past developments
and applications are well represented in the series [3][7].
Among the proposed DEDS models, Petri nets have a spe-
cial position because they seem to be more accepted in industry
and the applications domain, in general. Extensions of the basic
Petri model have already been used to model systems where
some kind of continuity or full continuity as rst described in
hybrid Petri nets is present [8]. In the domain of hybrid systems,
more than 15 Petri net-based models have been developed so
far, and they are summarized in an excellent survey [9]. Among
those models, differential Petri nets (DPNs) [10] constitute an
extension of hybrid Petri nets [11], [12] and can represent simul-
taneously continuous dynamic systems, modeled as systems of
ordinary differential equations, and DEDS. This article demon-
strates the application of DPN in modeling common industrial
supervision tasks.
Section II presents an extensive introduction to the DPN
model, including the construction of the evolution graph. In
section III, we provide DPN models for typical industrial au-
tomation tests and monitoring functions. Section IV contains a
detailed example of a systemwhose model consists of switching
between two alternative representations dependent on current
conditions, and is followed by the Conclusion.
II. BACKGROUND ON DIFFERENTIAL PETRI NETS
A. Model Denition
DPNs extend and combine the advantages of continuous-
type Petri net models and those of the discrete Petri net. The
novel features of DPNs are the negative real values accepted
for place markings and the use of an integration mechanism for
the approximate representation of the continuous system. Under
the assumption that the continuous systemcan be represented by
a nite number of linear rst-order difference state equations,
DPNs provide enough power to model a hybrid system in a
single graph. Full development of the DPN model can be found
in [10]. However, for the convenience of the reader, we present
a succinct description.
A DPN is composed of two kinds of places and two kinds of
transitions:
1) discrete place and discrete transition
2) differential place and differential transition
as shown in Fig. 2.
1094-6977/$20.00 2006 IEEE
544 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 36, NO. 4, JULY 2006
Fig. 1. Structure of a typical industrial automation system.
Fig. 2. Nodes of a differential Petri net.
Fig. 3. Explicit representation of a discrete implicit differential transition.
Marking of places is similar in principle to that of the hybrid
Petri nets, except that in DPNs the marking is allowed to also
take negative real values. Another interesting aspect is that to
every differential transition we associate a ring speed repre-
senting either a variable proportional to a state variable (or a
marking of a differential place) or an independent variable. Be-
cause a differential transition is always enabled, to discretize
the continuous system, we introduce to every differential tran-
sition a ring frequency representing the integration step that
would be used when carrying out a simulation. According to
the Petri net theory, this delay is associated with the implicit
discrete transition linked by a discrete place to this differential
transition, as shown in Fig. 3.
The intrinsic characteristics associated with a differential
transition do not allow the denition of autonomous DPNs. In
fact, inherent in the differential transition is the notion of time,
permitting the discretized view of continuous systems with
a certain period (namely, the integration step). Thus, a DPN is
actually a timed DPN whose denition follows:
Denition: A timed differential Petri net is dened by B =
R, f, M
0
, T ) verifying the following conditions:
1) R is a Petri net dened by R = P, T, Pre, Post) with
a) P: a nite set of places with dimP = n < ;
b) T: a nite set of transitions with dimT = m < ;
c) P T = and P T ,= ;
d) Pre(P
i
, T
j
) is a function that denes arcs from a
place to a transition with Pre(P
i
, T
j
) R;
e) Post(P
i
, T
j
) is a function that denes arcs from a
transition to a place with Post(P
i
, T
j
) R.
2) f: P T D, DF, named differential function, in-
dicates for every node whether it is a discrete or differential
node;
3) M
0
is the initial marking;
4) T is a map, to be called timing map, that associates a
real number to every transition that can evolve in time,
and also a delay for each differential transition.
The delay in each differential transition corresponds directly
to the integration step that would be used when a simulated
evolution of the continuous system was carried out. The choice
of the numerical value of this delay is constrained by numerical
stability considerations and is a tradeoff between calculation
speed and accuracy. More on the selection of integration steps
can be found even in elementary books on numerical methods
covering the integration of ordinary differential equations. It
must be noted that it is possible to use a different delay for
each differential transition. This may prove quite helpful in the
representation of distributed systems, for instance.
The marking vector is composed of two components. The rst
one, named M
D
, is a discrete vector where the integer value
corresponds to the number of tokens inside discrete places. The
second one, named M
DF
, contains real values corresponding
to the continuous markings of differential places. From this
decomposition, we can dene the global marking vector of a
DEMONGODIN AND KOUSSOULAS: DIFFERENTIAL PETRI NET MODELS FOR INDUSTRIAL AUTOMATION AND SUPERVISORY CONTROL 545
Fig. 4. Node of the evolution graph.
DPN as follows:
M(t) =
_
M
D
(t)
.............
M
DF
(t)
_
=
_
_
_
_
_
_
_
_
_
_
_

_
m
D
1
(t)
.
.
.
m
D
n
d
(t)
_

_
_

_
m
DF
1
(t)
.
.
.
m
DF
n
c
(t)
_

_
_
_
_
_
_
_
_
_
_
_
with
_
_
_
_
_
_
_
_
_
_
_

_
m
D
1
(t) N
.
.
.
m
D
n
d
(t) N
_

_
_

_
m
DF
1
(t) R
.
.
.
m
DF
n
c
(t) R
_

_
_
_
_
_
_
_
_
_
_
_
where n
d
and n
c
are the number of discrete places and of dif-
ferential places, respectively. It should be noted that a num-
ber of discrete places and discrete transitions are implicit (see
Fig. 3). For these nodes we have that dimP
DI
= dimT
DI
=
dimT
DF
, i.e., the number of discrete implicit places is equal to
the number of discrete implicit transitions, and both are equal
to the number of differential transitions because they serve as
support for the latter. The marking of the discrete implicit
places is always equal to 1
m
DI
i
(t) = 1, t, i P
DI
where P
DI
is the set of discrete implicit places, with P
DI
P.
It should also be noted that when a discrete transition is enabled
at a date t, the tokens required to re this transition are reserved
during the associated delay. When the delay is over, the transi-
tion is red, and the tokens (or marking) reserved for ring are
removed from the preplaces of this discrete transition, whereas
the nonreserved tokens (or marks) are added to the postplaces.
Thus, at any time, the present marking for discrete places is
the sum of the reserved and the unreserved markings. Reserved
marks can be dened in a similar way for the differential places.
The corresponding markings are denoted by r
D
and r
DF
, re-
spectively. At the time of expiration of the delay in the discrete
transition, the marking of the differential place is reduced by
the value of the reserved marking. The previous markings are
explained more extensively in [10]. Finally, invariants can be
introduced in the same way as for general Petri nets, under the
constraint that the values corresponding to the discrete part must
be integer.
B. Dynamics
For applications, it is more important to emphasize the more
applied aspects of DPNs such as the evolution graph used to
analyze the dynamic behavior. The evolution graph of a DPN
can be formed by nodes corresponding to interevent states (also
called stable states) and transitions with delay between these
nodes corresponding to events. A node is divided in two parts,
the rst one representing the markings of the discrete places,
and the second representing the ring speed vector of the dif-
ferential transitions and the reserved markings, as shown in
Fig. 4. These nodes are linked by arcs and transitions that deter-
mine the occurrence of events and the expended delay between
two consecutive interevent states. The transition associated to
this arc is noted as follows:
occurrence of events [ d
k
, where d
k
is the elapsed time.
We can also dene an incidence matrix W for every DPN
W = [W
ij
]
nm
where
W
ij
= Post(P
i
, T
j
) Pre(P
i
, T
j
).
Furthermore, a fundamental relation can be formulated
from which it is possible to deduce from a given marking at
time t
i
the reachable marking at the date t
k
M(t
k
) = M(t
i
) +W
_
(t
k
) +
_
t
k
t
i
v(u) du
_
where the characteristic vector (t) represents the ring se-
quence for the discrete transitions and the speeds vector
v(t) contains the instantaneous ring speeds of differential
transitions.
546 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 36, NO. 4, JULY 2006
Fig. 5. DPN representation of the system.
C. Example
We consider a system whose behavior can be represented by
an alternation between the following two models:
x(t) = 3
x(t) = 0.5x(t).
The switching between these two continuous models is gov-
erned by the following simple rule. For three time units, the
system admits model I, and for the following ve time units,
model II is valid; then the same cycle repeats. At the switching
instant, there occurs a jump of constant and known magnitude
in the state of the system. Therefore, the rules are
for kT t < kT +
3T
8
, k = 0, 1, . . . use model x(t) =3
for kT +
3T
8
t (k + 1)T, k = 0, 1, . . . set
x
_
kT +
3T
8
_
x
_
kT +
3T
8
_
+ 4
and use model x(t) = 0.5x(t).
We choose the integration step h = 0.5, which is crude but
effective for demonstration purposes. Considering an initial con-
dition of 3 (marking of P
5
in Fig. 5), the DPNmodel is as shown
in Fig. 5:
We dene the following sets:
P
D
= P
1
, P
2
, discrete places
T
D
= T
1
, T
2
, discrete transitions
P
DI
= P
3
, P
4
, discrete implicit places
T
DI
= T
3
, T
4
, discrete implicit transitions
P
DF
= P
5
, differential place
T
DF
= T
5
, T
6
, differential transitions
Then, the Pre-, Post-, and incidence matrices become
Pre =
_
_
_
_
_
0 1 0 0 1 0
1 0 0 0 0 1
0 0 1 0 1 0
0 0 0 1 0 1
0 0 0 0 0 0
_
_
_
_
_
,
Post =
_
_
_
_
_
1 0 0 0 1 0
0 1 0 0 0 1
0 0 1 0 1 0
0 0 0 1 0 1
4 0 0 0 0.5 3
_
_
_
_
_
so we obtain
W =
_
_
_
_
_
1 1 0 0 0 0
1 1 0 1 0 0
0 0 0 0 0 0
0 0 0 0 0 0
4 0 0 0 0.5 3
_
_
_
_
_
_
P
D
_
P
DI
P
DF
..
T
D
..
T
DI
..
T
DF
.
Places P
1
and P
2
, along with T
1
and T
2
, realize the model
selection rules. Place P
5
represents the state variable x(t), and
its marking indicates the value of that state. The combinations
T
6
, P
5
and T
5
, P
5
realize the two continuous dynamic
models I and II, respectively. Each of these continuous mod-
els is supported by the corresponding discrete implicit nodes
T
4
, P
4
and T
3
, P
3
, respectively. Place P
1
activates model
II, whereas place P
2
activates model I.
The dynamics of the net are determined by the instantaneous
ring speeds [12] of the differential transitions. Thus, for model
I, we notice that the dynamics are described by the following
relations:
v
6
(t) = 1, when T
6
is enabled
v
6
(t) = 0, otherwise.
The dynamics of the DPN are basically the dynamics of
P
5
, which are characterized by the previous relations and the
DEMONGODIN AND KOUSSOULAS: DIFFERENTIAL PETRI NET MODELS FOR INDUSTRIAL AUTOMATION AND SUPERVISORY CONTROL 547
(a)
Fig. 6. (a) Evolution graph (complete) of the example DPN.
following:
v
5
(t) = m
5
(t)
dm
5
(t)
dt
= W v(t)
dm
5
(t)
dt
= 0.5v
5
(t) 3v
6
(t).
(b)
(c)
Fig. 6. (Continued.) (b) Compressed evolution graph of the example DPN. (c)
System state trajectory (marking of place P
5
) vs. time of the example DPN.
To fully understand the behavior of a particular DPN, its evo-
lution graph must be constructed. Alternatively, a summarized
version of the behavior can be given by the compressed evolu-
tion graph, when possible. The full and compressed evolution
graphs for the DPN of the previous example appear in Fig. 6(a)
and (b), respectively. Fig. 6(c) shows the trajectory of the state
(actually, the marking of place P
5
).
III. DIFFERENTIAL PETRI NETS IN SUPERVISORY CONTROL
DPNs constitute a natural model for supervisory control
systems, where discrete logic (usually in software form or pro-
grams for programmable logic controllers) and continuous sys-
tems intermix. Within supervisory control, there are a number
of functionalities necessary for the supervisor to carry out its
548 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 36, NO. 4, JULY 2006
task(s). An important class of functionalities is the monitoring
of the systems operation and the issuance of alarms or reactions
to specic events. Such events may be the simple crossing of
signal levels, the crossing of signal levels coming froma specic
direction (e.g., up or down, in a scalar signal), and entering or
leaving a band-like region (dened by two signal levels). Such
events may be the result of normal deviations from set points,
disturbances, or failures, and usually there are corresponding
prescribed reactions, such as changing of plant model for the
computation of control policies, changing of the control mode,
and similar counteractions. This information can be useful for
decision making in the context of modeling. For example, the
overshooting of a set point may signify that a rst-order (per-
haps dominant pole approximation) model may be inadequate
to describe the process and a higher order model must be used.
Modeling such actions can be a substantial task in the study
of the supervisory systems behavior and performance. It is
thus important to validate the power of DPNs in this setting.
We provide below a number of typical testing schemes that
are widespread in industrial automation environments. A more
formal model-based example can be found in section IV.
Before proceeding to the test models, it is useful to introduce
a concept that is closely related to invariance. This construct,
which we call forced invariant, proved useful for modeling
purposes. In the forced invariant setting, we introduce a new
place that will be the complement of the place containing the
value of the state variable that needs to be tested. Thus, if m(t) is
the marking of the original place, the marking of the new place,
will be m(t). Each input (respectively, output) transition will
be duplicated as output (respectively, input) transition of the
new place, whereas the new arcs will carry equal weights with
the corresponding original arcs. More formally, the complement
of a differential place is dened as follows (T
DF
denotes the
set of differential transitions):
Denition: To each differential place P
j
, with f(P
j
) = DF,
a complementary differential place P
/
j
can be associated, with
respect to pre- and posttransitions, for which the following are
true:
f(P
/
j
) = DF
Pre(P
/
j
, T
k
) = Post(T
k
, P
j
) and
Post(P
/
j
, T
k
) = Pre(T
k
, P
j
), T
k
T
DF
m
P
/
j
(t) = m
P
j
(t).
This construct is shown in Fig. 7.
A. Crossing a Signal Level Upward
This test refers to the event of a positive signal exceeding a
higher positive signal level or of a negative signal exceeding a
higher negative signal level (Fig. 8). The corresponding test can
be modeled via DPN in the following way.
Output in Fig. 9 and subsequent gures, refers to the answer
of the test: if the place is marked after an afrmative answer to
the test, then a transition can be suitably enabled. Finally, note
the different time constant h
/
for the timed discrete place in the
test section of the DPN. This means that the time constants of
Fig. 7. The complement of a differential place with respect to differential
transitions.
Fig. 8. Level crossing.
Fig. 9. DPN model of an upward-level crossing.
Fig. 10. Downward-level crossing.
the test(s) can be different (equal or slower, as necessary) than
those of the continuous system.
B. Crossing a Signal Level Downward
This condition refers to the event of a positive signal crossing
fromabove a lower signal level or of a negative signal exceeding
a lower negative signal level (Fig. 10). Because we want to know
whether a marking is inferior to a value, we use the notion of
complementary differential place previously dened. This test
can be modeled via DPN in the way shown in Fig. 11.
DEMONGODIN AND KOUSSOULAS: DIFFERENTIAL PETRI NET MODELS FOR INDUSTRIAL AUTOMATION AND SUPERVISORY CONTROL 549
Fig. 11. DPN model of a downward-level crossing.
Fig. 12. Variation within a band-like region.
Fig. 13. DPN model of testing variation within a band.
C. Variation Within a Region
This test refers to the event of a signal taking values within
a band-like region, dened by two levels A
1
and A
2
(signs are
not important as long as A
1
> A
2
) as in Fig. 12.
A
2
m(t) A
1
, t, A
1,2
R
or, equivalently
m(t) A
2
and m(t) A
1
, t, A
1,2
R.
The last relation leads to the test that can be modeled via DPN
in the way shown in Fig. 13.
D. Variation Outside a Region
This test refers to the event of a signal taking values outside
a band-like region, dened by two levels A
1
and A
2
(signs are
not important, again, as long as A
1
> A
2
), as in Fig. 14.
m(t) A
2
or m(t) A
1
, t, A
1,2
R.
The corresponding test can be modeled via DPN as shown in
Fig. 15:
Fig. 14. Variation outside a band-like region.
Fig. 15. DPN model of testing variation outside a band.
Fig. 16. Using the test result in many locations simultaneously.
E. Remarks
1) When the result of any of the previous test is to be used in
many locations simultaneously, it is possible to do so in
a straightforward and elegant way through the construct
shown in Fig. 16.
2) The time constant h is related to the integration step used
for the continuous model. It is possible, however, and
in fact quite probable in the context of hybrid control
systems that various continuous models may be used for
the same system (see also the example in the next section)
or within the same plant. Each of these models can have
its own integration constant. In such a case where more
than one time constant is present, we can either duplicate
the test construct for each constant, or, more elegantly,
duplicate only the timed transition and enable it when the
corresponding model is active.
3) As noted earlier, each test can be carried out at its own
pace through the introduction of appropriate time con-
stants (marked h
/
in the gures). Thus, the exibility of
the model is not hampered in any way.
4) It is worth noting that it is possible to construct large
models by combining the individual tests. Typically, a set
550 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 36, NO. 4, JULY 2006
of tests will have as target a set of specic continuous
quantities. Then, a model can be constructed for the con-
tinuous part, and each test can be realized around the same
set of differential places. The example in the next section
demonstrates this.
5) For the time being, no tool has been developed specif-
ically supporting DPN models. However, OMOLA, an
object-oriented modeling language for representing dy-
namic systems, has been enhanced successfully to de-
scribe and simulate DEDS and continuous time dynamic
systems operating concurrently. DPNs have been imple-
mented, validated, and compared with other mixed Petri
nets within OMOLAs companion modeling environment
OmSim [13].
IV. EXAMPLE: STATE-DEPENDENT MODEL SWITCHING
Consider a continuous dynamic system for which more de-
tailed modeling is required, depending on the value and behavior
of the output. The two models are the following:
MODEL I: x(t) = 0.5x(t)
MODEL II:
_
x
1
(t) = x
2
(t)
x
2
(t) = 0.3x
1
(t) 0.3x
2
(t).
For model I, the output is x(t), whereas for model II the
output is x
2
(t), and both quantities correspond to the same
physical entity. The rules for selecting the appropriate model
are the following:
RULE 1: When the output is greater than 1.0 in absolute value,
use model I.
RULE 2: When the output is less than 0.5 in absolute value, use
model II.
RULE 3: When the output enters the zone between 0.5 and 1.0
(in absolute value) although it was less than 0.5, (in
absolute value) the moment before, continue to use
model II.
RULE 4: When the output enters the zone between 0.5 and
1.0 (in absolute value), although it was larger than
one (in absolute value) the moment before, continue
using model I.
Fig. 17 presents a typical scenario with two disturbances (not
represented in the DPN model) at approximately 2000 and 5500
time units (events B and C). Initially, the system is supposed to
follow model I with initial condition x(0) = 1.5.
To understand more clearly the behavior of the model, note
that the oscillatory behavior corresponds to model II. The pre-
vious system, consisting of a set of models and a set of rules
on the regions of validity of these models, can be represented
via a DPN as shown in the following steps. The two alter-
nate models have the DPN representation as shown in Fig. 18.
Place P
1
represents the continuous state variable x
1
of model
II, whereas place P
2
represents the continuous state variable x
2
of model II, which is identical to the continuous state variable x
of model I. Table I collects the information regarding necessary
tests.
Figs. 1921 showthe implementation of the tests described in
Section III for the system of the example. In building the overall
Fig. 17. Simulation of the example systemA, B, C, Dare the times of events
(crossing of levels; disturbances), whereas on the upper side the corresponding
valid model is indicated.
Fig. 18. DPN representation of the alternate models.
TABLE I
RULES FOR MODEL SELECTION
DEMONGODIN AND KOUSSOULAS: DIFFERENTIAL PETRI NET MODELS FOR INDUSTRIAL AUTOMATION AND SUPERVISORY CONTROL 551
Fig. 19. Implementation of tests 1 and 2 of Table I.
Fig. 20. Implementation for positive values of tests 3 and 4 of Table I.
Fig. 21. Implementation for negative values of tests 3 and 4 of Table I.
model, we follow the composition method. We rst construct
the continuous models. Then, we construct each test according
to Table I. Finally, we add the logic that ensures the selection
of the correct model for each setting. To fuse the four tests, we
must bring together the results of the tests and drive a suitable
logic. This is done in Fig. 22 in a summarized fashion because
the complexity of the global model does not permit a satisfactory
depiction. Note that the differential places and transitions in
552 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 36, NO. 4, JULY 2006
Fig. 22. Model selection based on the results of the tests.
Fig. 22 are of course identical with the corresponding entities in
the tests of Figs. 1921
0.5 x
2
(t) 0.5 and x
2
(t

) < 0.5 Output A

0.5 x
2
(t) 0.5 and x
2
(t

) > 1 Output C
1 x
2
(t) 0.5 and x
2
(t

) > 0.5 Output B

1 x
2
(t) 0.5 and x
2
(t

) < 1 Output D.
V. CONCLUSION
The objective of this work was to illustrate the use of the DPN
model for hybrid systems, in the context of industrial supervi-
sory control. The DPN model introduces several novel features
to the class of hybrid Petri nets to which it belongs and provides
a unied representation for hybrid control systems. In terms of
modeling capability, it can bring together models for decision
making enacted by the supervision, which is based on discrete
events, and models for continuous systems whose local controls
are affected by the decisions of the supervisor. The Petri net basis
endows DPNwith the capacity for detailed representation of the
supervisory logic, whereas its rules of evolution can capture the
nest synchronization properties between the discrete event and
the continuous part. Overall, this unied model can help evalu-
ate the integrated system from the control point of view, based
on the test structures presented in this article. Furthermore, it is
also possible to incorporate the communications point of view
in any industrial context, as soon as a suitable Petri net model is
available for the protocols (and networking, in general). The el-
ementary models presented here can be combined to formlarger
models to correspond to a real system of any level of complex-
ity. In general, the effect of supervisors, especially of complex
ones, and their sensitivity to the operating logic and commu-
nications intricacies, can be assessed for the integrated system.
Thus, a global analysis of integrated industrial automation sys-
tems can be put on a new basis and modications can be tested
out.
REFERENCES
[1] M. S. Branicky, Hybrid systems, Ph.D. dissertation, Mass. Inst. Tech-
nol., Cambridge, 1991.
[2] M. S. Branicky, V. S. Borkar, and S. K. Mitter, A unied framework for
hybrid control: Model and optimal control theory, IEEE Trans. Autom.
Cont., vol. 43, no. 1, pp. 3145, Jan. 1998.
[3] R. L. Grossman et al., Hybrid Systems, ser. Lecture Notes in Computer
Science. New York: Springer, 1993, vol. 736.
[4] P. Antsaklis, W. Kohn, A. Nerode, and S. Sastry, Eds., Hybrid Systems
II. ser. Lecture Notes in Computer Science. New York: Springer, 1995,
vol. 999.
[5] R. Alur, T. A. Henzinger, and E. D. Sontag, Eds., Hybrid Systems III. ser.
Lecture Notes in Computer Science. NewYork: Springer, 1996, vol. 1066.
[6] P. Antsaklis, W. Kohn, A. Nerode, and S. Sastry, Eds., Hybrid Systems
IV. ser. Lecture Notes in Computer Science. New York: Springer, 1997,
vol. 1273.
[7] P. J. Antsaklis, W. Kohn, M. Lemmon, A. Nerode, and S. Sastry, Eds.,
Hybrid Systems V. ser. Lecture Notes in Computer Science. New York:
Springer, 1999, vol. 1567.
[8] J. Le Bail, H. Alla, and R. David, Hybrid Petri nets, Proc. 1st European
Control Conference, Grenoble, France, July 1991, pp. 187191.
[9] P. Antsaklis, X. D. Koutsoukos, and J. Zaytoon, On hybrid control
of complex systems: A survey, Eur. J. Automat., vol. 32, nos. 910,
pp. 10231045, Dec. 1998.
[10] I. Demongodin and N. T. Koussoulas, Differential Petri nets: Represent-
ing continuous systems in a discrete event world, IEEE Trans. Autom.
Control, vol. 43, no. 4, pp. 573579, Apr. 1998.
[11] H. Alla and R. David, Continuous and hybrid Petri nets, J. Circ. Syst.
Comput., vol. 8, no. 1, pp. 159188, 1998.
[12] , Petri Nets & GrafcetTools for Modeling Discrete-Event Systems.
Englewood Cliffs, NJ: Prentice Hall, 1992.
[13] N. Pariset, Representation and Simulation of Hybrid Dynamic Systems
Implementation of Hybrid Petri Nets in Omola. Nantes, France: Univ.
Nantes, Jun. 1996, Rapport D.E.A..
Isabel Demongodin (M94) received the D.E.A. de-
gree in control engineering and the Ph.D. degree in
computer and systems engineering from the Univer-
sity of Montpellier, France, in 1990 and 1994, respec-
tively.
She held a postdoctorate position at the Laboratory
of Automation and Robotics, University of Patras,
Greece, in 1994. From 1995 to 2004, she was an As-
sistant Professor at the Ecole des Mines de Nantes,
France, and in 1999 joined the Research Institute
of Communications and Cybernetic of Nantes in the
Discrete Event Systems group. Currently she is a full Professor at the University
of Angers, France, with the Laboratory of Automation and SystemEngineering.
Her research interests are in the areas of discrete event systems, hybrid systems,
and Petri nets.
DEMONGODIN AND KOUSSOULAS: DIFFERENTIAL PETRI NET MODELS FOR INDUSTRIAL AUTOMATION AND SUPERVISORY CONTROL 553
Nick T. Koussoulas (S76M84) holds ve engi-
neering degrees from Greece, France, and the United
States, including the Ph.D. degree in applied dynamic
systems control from University of California, Los
Angeles, in 1984.
Since 1990, he has been with the Laboratory for
Automation and Robotics of the Electrical and Com-
puter Engineering Department at the University of
Patras, Greece, where he is now Professor and Head
of the Division of Systems & Automatic Control. He
was formerly with Bell Communications Research
(Bellcore), Red Bank, NJ, where he worked on modeling and designing com-
munication networks with dynamic, feedback-based routing. His main research
interests include modeling and control of hybrid systems, path planning and
control of mobile robots and transportation vehicles, and simulation.