Supremica A Tool for Verication and Synthesis

of Discrete Event Supervisors

Knut

Akesson, Martin Fabian, Hugo Flordal, and Arash Vahidi
Department of Signals and Systems
Chalmers University of Technology
SE-412 96 G oteborg, Sweden
Email: [email protected]
AbstractA tool, Supremica, for automatic verication and
synthesis of controllers for discrete event systems is presented. In
addition to algorithms for verication and synthesis, Supremica
can automatically generate code for a number of languages
including ANSI C and IEC 61131 languages like Structured
Text and Instruction List. Supremica has built in support for
executing the supervisors against either a simulated environment
or a physical environment. To handle verication and synthesis
of industrial size problems Supremica implements algorithms
that exploits the modular structure of the problem together with
symbolic methods to efciently represent large state-spaces.
I. INTRODUCTION
The supervisory control theory (SCT) [1] has potential to
solve some of the problems that are related to high require-
ments on exibility in many new production systems. Many
new production plants must be easy to recongure by adding
and removing resources and should also be able to produce a
variety of products. For example, the number of options that a
car customer can choose between implies that there are almost
all cars are different. That virtually all products are different
will put a lot of requirements on the control system used in
the production process. The SCT may be used to analyze and
also synthesize control code for the production system. When
both the production system and products changes often it may
be advantageous or even necessary to automatically synthesize
the control code for the current setup of production resources
and products. The SCT handles this challenge by using well
dened models of both the plant, i.e. the production system,
and the specications that model the products to be produced
within the plant. Much theory has been developed within
the SCT-framework, however very few industrial applications
of the theory have been presented so far. Several tools for
experimenting with the concepts introduced in the SCT have
been developed but most of them focus academic usage instead
of industrial. In this paper a new tool Supremica that
tries to solve some of the problems with the previous tools
is presented. Supremica implements the standard supervisory
control algorithms. In general, the number of states in a system
with concurrent subsystems is proportional to the product of
the number of states in the subsystems. This implies that it
is only possible to explicitly represent all states for small
systems. To better handle large systems modular algorithms
has been developed, see [2]. The modular algorithms try to
divide the main problem into multiple, hopefully smaller, sub-
problems that can be efciently solved. An efcient data-
structure, Binary Decision Diagram (BDD) [3], may also be
used to store states. In addition to verication and synthesis
of supervisors it is possible to automatically generate stan-
dard Programmable Logic Controller (PLC) programs that
implement the behavior of the veried/synthesized supervi-
sor. Supremica also has built in support for executing the
supervisor against a graphical simulation of the environment
or against a physical environment. A design goal was to
make Supremica user-friendly. As a result of this Supremica
has successfully been used to teach the concepts of SCT to
undergraduate automation engineering students at Chalmers
University of Technology. To show the usability of Suprem-
ica and the SCT, some example applications are presented.
Supremica is free for education and research and can be
downloaded from http://www.supremica.org.
A. Existing Tools
A number of tools implementing the main concepts in the
supervisory control theory have emerged.
TCT TCT is the original SCT tool developed at Uni-
versity of Toronto, Canada. TCT has a primitive
text-based interface. A new tool called STCT, [4],
with algorithms based on IDDs [5] has recently
been released.
UMDES Developed at University of Michigan, USA,
UMDES is a library of C-routines for the SCT.
UKDES Developed at University of Kentucky, USA,
UKDES, [6], contains the basic algorithms and
has a simple graphical user interface.
J-DES Developed at Pennsylvania State University, USA,
J-DES has a graphical user interface and provides
similar functionality as TCT and UKDES.
BSP Developed at University of LAquila, Italy, BSP
uses BDDs to solve large scale synthesis prob-
lems. The approach is presented in [7].
Ver In [8], [9] a tool for supervisory control that uses
BDDs to solve large scale synthesis problems
is presented. The BDD algorithms are presented
in [10].
Valid Valid is the only known commercial tool for
supervisory control theory. Valid is developed
by Siemens Corporate Research, Germany. Very
little information is released about the details of
this implementation. This seems to be the most
complete tool among the existing supervisory
tools, but the focus is on verication and not on
synthesis. Valid includes efcient algorithms for
verication and a recent version includes support
for synthesis but experiments indicates that it has
problems to handle large problems.
All tools suffer in at least one of the following areas; ease
of use; verication; synthesis; ability to handle large systems;
simulation; code generation; code execution.
II. SUPREMICA
In this section Supremica is presented but rst the assump-
tions that Supremica relies on are described. Supremica is a
tool that changes over time and the most up to date information
is available at http://www.supremica.org. Supremica
was initially built to evaluate the ideas presented in [11]. When
designing Supremica we used experiences gained from the
development of a prior SCT tool called Desco [12].
The basic data structure in Supremica is a deterministic
nite automaton. Each automaton has a nite set of states,
Q, one of these states is the initial state q
i
. The states can be
marked or non-marked and forbidden or non-forbidden. The
set of marked states are denoted Q
m
and the set of forbidden
states Q
x
. Associated with the transition between states are
events where each automaton has a nite set of events, the
alphabet . The transition from state q
k
to state q
l
on the
event is denoted as (q
k
, ) = q
l
. is a partial function,
i.e. it is not dened for all states and all events. Each event
is controllable or uncontrollable. All events are assumed to be
observable.
A Supremica project consists of multiple automata that
together dene the main problem. The behavior of all the
automata is dened by the prioritized composition operator
(PCO). It is important to note that the alphabets do not have
to be the same but they are of course allowed to be equal. If the
same event is present in multiple automata then that event must
be either controllable or uncontrollable in all of the automata.
The PCO is useful because it can express both full syn-
chronous composition as dened by Hoare [13], and broadcast
composition used, for example, by State diagrams in Unied
Modeling Language (UML). In prioritized composition is also
possible to mix full and broadcast composition. To support
the PCO each event is assumed to be either prioritized or not.
Note that the same event may be prioritized in one automaton
and not prioritized in another. The set of prioritized event in
an automaton is denoted by
p
. The prioritized synchronous
composition of n 2 nite automata is denoted by

p
(A
1
, A
2
, . . . , A
n
)
where
Q = Q
1
Q
2
. . . Q
n
,
=
n

i=1

i
,
q
i
= q
1
i
, q
2
i
, . . . , q
n
i
,
Q
m
= Q
1
m
Q
2
m
. . . Q
n
m
Q
x
= {q
1
i
, q
2
i
, . . . , q
n
i
| q
j
i
Q
j
x
}
and the transition function as
(

q
1
, q
2
, . . . , q
n

, ) =

1
(q
1
, ),

2
(q
2
, ), . . . ,

n
(q
n
, )

{i |
i
p
}

i
(q
i
, )!
undened otherwise.
where

i
(q
i
, ) =

i
(q
i
, ) if
i
(q
i
, )!
q
i
otherwise.
Informally the above denition implies that an event is
present from a state in the synchronous composition if all
automata that have the event as prioritized are in a state from
which they can execute the event. If the event can be executed
in the synchronous composition then the next state is given by
executing the same event in all automata that can execute the
event, i.e. both those automata that have the event as prioritized
and those that do not.
Each automaton does also have type, it is one of either plant,
specication, or supervisor. The prioritized synchronous com-
position of the plant automata denes the behavior of the plant.
The term system used frequently in the sequel is used to denote
the prioritized synchronous composition of all the automata in
the project. In Supremica the user typically starts with a set of
plant and specication automata. The user can then verify if
the plant is controllable with respect to the specication and
if the plant and specication is non-blocking when executing
under the prioritized synchronous composition. If the system
is not controllable or not non-blocking then the user can chose
to synthesize a set of supervisors (or a single supervisor) that
when executing together with existing plant and specications
guarantees that the system will be both controllable and non-
blocking.
Supremica has a graphical user interface where it is easy to
do supervisory synthesis and verication. Supremica does not
yet contain its own graphical editor for state automata; this is
instead handled by another tool - DescoGUI. DescoGUI was
rst intended as an interface to Desco but today supports also
Supremica. Supremica is able to graphically layout and display
an automaton using Graphviz, from AT&T Research. This is
very useful, at least for small automata, since it allows the
user to synchronize or synthesize a new automaton and then
graphically display the result.
Supremica is able to synthesize maximally permissive non-
blocking and controllable supervisors. Efcient algorithms that
exploit the modularity of the system, for verication and
synthesis of controllable supervisors, have been developed and
implemented. Efcient modular algorithms for non-blocking
verication and synthesis are under development by our re-
search group. Supremica also contains an efcient algorithm
for language inclusion checking. Language inclusion checks
are useful for verifying properties of a system. Supremica
also implements state minimization of an automaton. State
minimization computes the minimal automaton, in number of
states, with the same language as the original one. Supremica
also allows the user to restrict, or project out, events from
the alphabet in an automaton; this implies that all the events
that do not survive the restriction are removed from the
strings in the language generated by the original automaton.
Furthermore, Supremica implements algorithms for bounded
controllability that are useful in hybrid computer-human su-
pervision of discrete event systems.
When the number of states is small, a graphical presentation
of the automaton is advantageous to the user. However, if the
automaton contains many states or contains many transitions
a graphical presentation easily becomes cluttered. Supremica
has an explorer for an automaton or a set of automata. The
explorer displays a single state and the set of enabled events
in that state. The user might click on an event to execute that
event and as a result change state.
A. Synthesis
Supremica implements monolithic synthesis algorithms for
solving non-blocking, controllability, and combined non-
blocking and controllability problems. Supremica also im-
plements modular algorithms for verication and synthesis
of the controllability problems. The modular algorithms are
presented in [2]. The modular algorithms have been evaluated
on several large problems and have been able to synthesize
a set of interacting supervisors with very limited time and
memory requirements. For example, the modular algorithms
are able to verify and synthesize supervisors for a central-
locking system with 750 million reachable states in a couple
of seconds. There are ongoing work on adding BDD based
algorithms to Supremica, some of this work is reported in [14].
B. Verication
With verication we refer to proving properties of the
system by help of formal methods. A survey of verication
of PLC programs is presented in [15]. Supremica contains its
own algorithms for verication. Currently there are efcient
algorithms for verication of controllability. If the system is
sufciently modular, Supremica is able to verify problems of
almost arbitrary size. There are pathological cases where the
developed algorithms cannot take advantage of the modularity
and the only thing that Supremica can do is to compute
the complete reachability graph of the system. Experiments,
on different applications, show that many applications have
enough modular structure to make the modular verication
and synthesis algorithms efcient. To verify non-blocking
properties, Supremica uses a brute-force approach and is
therefore currently unable to handle systems with more than
a few million states.
C. Simulation
An interface against the Scenebeans framework, [16], for
graphical simulation has been developed. This makes it possi-
ble to write graphical animations and load them together with
the state automata. Supremica can then compute supervisors
and execute them on-line against the graphical animation.
D. Code generation
When a set of supervisors has been synthesized it is
desirable to be able to generate code that implements these
supervisors. Supremica can generate code in a number of
formats including IEC 61131Instruction List and Structured
Text; ABB Control BuilderInstruction List, Structured Text,
and Sequential Function Chart; ANSI C; and Java bytecode.
Programmable Logic Controllers (PLCs) are used in in-
dustry to control very different kinds of devices. A standard
for PLC languages has been dened in [17]. This standard
is widely adopted by industry and most tools claim compli-
ance with the standard. The standard denes ve different
programming languages each with its own strengths and weak-
nesses. The ve languages are Structured Text, Function Block
Diagram, Ladder Diagram, Instruction List, and Sequential
Function Chart.
It is important to note that in general it is not desirable
to generate a single automaton that will dene the allowed
behavior of the supervisor since the state-space might be
huge even for rather small examples. Instead, we want to
generate a more compact representation of the supervisor. The
approach in Supremica is to take advantage of the possibility
for modular supervisors and thus compute a set of (sub-)
supervisors where the PLC is responsible for computing the
set of enabled events in each state of the total supervisor.
Fortunately, on-line computation of the set of enabled events
in the current state is a relatively simple operation that can be
computed efciently by a PLC.
In [18] it is shown how it is possible to, given a set
of automata interacting through events, compute a set of
SFCs interacting through signals, where the SFCs will behave
very similar to the automata. Different PLC vendors have
interpreted the IEC 61131 standard differently making it harder
to generate code that will have the same behavior on different
platforms. The main problem is whether they rst choose to
evaluate all transition conditions in all SFCs before they do
the transition, or if they evaluate the transition condition for
a single SFC and then change the active steps before they go
on to evaluate the conditions in the following SFCs. The rst
author of [18] has contributed to Supremica an implementation
of a robust algorithm that generates code that can be executed
in ABB Control Builder. Currently, this implementation only
supports full synchronization and does not allow self-loops
in the input automata. A more straightforward approach is
to generate Structured text or Instruction List code. This
approach avoids the problem with interacting SFCs and instead
generates a single code block that is responsible for computing
the enabled events.
The basic approach to code generation is to rst check
which events that are enabled in the automata. This is done
by doing an online synchronization in the PLC. After this is
checked if there are external conditions, i.e. boolean conditions
on input signals, that must be true in order for the event to
be enabled. Events can also be associated with internal timers
that must have done a timeout before an event is enabled, so
this is also checked. At this point zero, one, or more events
may be enabled. If zero events are enabled then we do not
have to do anything, if one event is enabled then we change
state in the automata and execute actions associated with that
event. If more than one event is enabled it is in general not
safe to execute all events. Instead a selection must be done.
One approach is to execute one of enabled events randomly,
another approach is to have the events ordered and execute the
rst event in this order. However, this is an area that need to
be explored more but the paper [19] presents some properties
that the automata should have in order to be safe to execute
with this approach.
At the time of writing it is possible to generate IEC
61131 code as Instruction List and Structured Text. Currently,
it is only possible to compile and execute Instruction List
code within Supremica. The execution of the generated code
is discussed in more detail in the following section. The
generated PLC code may have a modular structure in the
sense that the synchronization of the automata is done by
the PLC during run-time. It is important to be able to do
the synchronization between multiple automata in the PLC
in order to avoid having to generate a single automaton that
explicitly contains all states since the number of states may
be huge even for small problems.
E. Code Execution
Usually a special hardware device, a PLC, does execute the
control code. It is increasingly popular to use standard PCs as
PLCs. The PLC functions are then implemented in software
that is then executed by the PC.
Supremica contains a soft-PLC implemented in Java [20].
The compiler transforms/compiles IEC 61131 Instruction List
code to Java bytecode. The runtime system uses a standard
Java Virtual Machine (JVM) to execute the code. One part of
the runtime system is to enforce the PLC execution behavior
onto the JVM; the other part is to interact with the I/O system.
Currently, the soft-PLC is not able to handle hard real-time
constraints due to limitations in the JVM. A standard for real-
time Java has recently been developed, see [21] but it is not
currently used in Supremica.
Related to the execution of a supervisor is the problem of
how do the controller select among multiple enabled events. A
naive approach might result in a blocking system even though
verication indicates that the system is non-blocking. This
problem is analyzed in [19].
In many applications and especially in soft-PLC appli-
cations, it is desirable to be able to call general purpose
programming languages for example to implement a graphical
user interface. In the Supremica soft-PLC the PLC code is
compiled into Java bytecode and thus it is possible to call Java
methods. The main difference with a standard Java program
and a PLC program is the way a PLC program is executed.
To allow the PLC code compiled into Java bytecode to
behave like a proper PLC, a class was written to execute the
PLC-program at regular time-instants and also to do input and
output copying of variables. This approach is implemented in
Supremica and allows the user to switch between running the
soft-PLC, i.e. the Java Virtual Machine, against a physical I/O-
card or against a software module that exhibits the same be-
havior as a physical I/O-card. Since events are associated with
transition between states in the internal model but signals are
used to communicate to the environment, it is necessary deal
with the differences between events and signals. In Supremica
this is handled by making it possible to associate actions and
conditions to events. An action is a set of signals that should
be set to high or low when the associated event is executed.
A condition is a Boolean expression of variables, associated
with input signals, which must be evaluated to true before the
associated event is allowed to execute. It is important to note
that it must be known a priori that if an event is enabled in
the nite automata model then the corresponding condition
must sooner or later evaluate to true without executing any
other events. If this is not fullled then the supervisor may
not be non-blocking even though the verication says it is.
Experience shows that it is generally not a problem to write
code conditions that fullls the condition above. It is also
possible to associate timers to the conditions.
III. APPLICATIONS
Supremica has been evaluated on several different problem
classes. Supremica was originally used to supervise the re-
source allocations in a commercial batch control system. This
implementation is presented in more detail in [11]. Later the
possibility for code generation and code execution was added
and this made it possible to use Supremica to control several
different lab processes. The rst application is a ball process
in our lab. We present this since it is an application originally
controlled by a standard PLC and also used to teach students
the basics of PLC programming using IEC 61131 languages.
In this application we have replaced the PLC with a digital I/O
in a standard PC and the uses Supremica to verify and execute
state automata against both a graphical simulation of the ball
process and also against the physical ball process. Secondly
we present an application with automated guided vehicles.
This is a standard example that has been used to evaluate
many different control strategies. We here use the modular
synthesis algorithms to synthesize a set of supervisors that
together guarantees collision-free coordination of AGVs. The
number of reachable states in this application is approximately
26 millions that make the use of non-modular verication
and synthesis algorithms unfeasible. The last application is
an implementation in a commercial chemical batch control
system.
q0
q1 !w21 q2
!w22
q3
!w23
!w24
Fig. 1. The automaton modeling Workstation 2.
A. Ball Process
The Ball Process is a lab process used to teach students at
Chalmers the basics of PLC programming. This implementa-
tion uses the code generation and code execution possibilities
to generate IEC 61131 compatible IL code that can be com-
piled into Java bytecode. The soft PLC is used to execute the
Java bytecode and all I/O is done with a digital I/O-card, and
thus used to control the ball process.
B. Automated Guided Vehicles
A model of a exible manufacturing cell was introduced
in [22]. The cell consists of three workstations, two input
stations and one output station. Five Automated Guided Vehi-
cles (AGVs) are responsible for routing the parts through the
cell. The oor space is shared between the AGVs and thus
the AGVs may collide. To complicate things the controller
only has the possibility to prevent an AGV from proceeding
at certain positions. The positions where the supervisor can
prevent an AGV from proceeding are associated with c
i
events.
All other events are uncontrollable. It is assumed that all
events are observable to the supervisor. The input-, output-, the
workstations, and the AGV routes are easily modeled as state
automata. The automaton modeling Workstation 2 is shown
in Figure 1 and the automaton modeling Zone 2 is shown in
Figure 2. The restriction that at most one AGV can be inside a
shared zone at a time can be viewed as a specication, while
the AGV routes can be viewed as the plant. Supremica can now
generate a set of interacting supervisors that together supervise
the plant in a maximally permissive manner. In our example
we will get four supervisors, one for each specication. The
sizes of the supervisors are 72, 72, 116, and 68 states. The
supervisor computations are completed in less than half a
second on a standard PC. Note that solving this problem with
a brute-force approach will take a lot of computing resources
since the system has over 26 million reachable states. For this
example no physical process were built, but a screenshot of
Supremica running a graphical simulation of the process is
shown in Figure 3.
C. Resource Allocation Systems
In exible manufacturing systems both the products to be
produced and the resources used to produce the products
changes over time. Typically the resources are versatile in the
sense that they can be used to carry out several different pro-
duction steps. A goal when designing exible manufacturing
systems is to allow the system to continue producing products
when one or more of resources fail, if possible. Changing
products, exible resources, and robustness to failure sets new
requirements on the control system. A control system must be
adaptable to all these changes and thus it is no longer a viable
alternative to have control functions that does not have full
q4
q0
!u30 c5
q1
c6
q2
!u21
q3
!u26 !u31 !u22 !u27
Fig. 2. The automaton modeling Zone 2.
knowledge of the set of products to be produced, their resource
requirements, and the resources use to produce the products.
A danger with transforming a manufacturing system from a
manual system to an automatic system is that the exibility
of the production process suffers. The reason for this is the
high complexity involved with building a fully automatic and
exible manufacturing control system. In the work presented
in [11], we show how to build efcient models of products
and resources, and then how to synthesize supervisors such
that all products can always be successfully produced. When
multiple products are produced within the same plant conicts
of resources might emerge. The smallest situation is when
two products both holds a resource and wants access to the
resource hold by the other product. All models needed by
the synthesis algorithms can be automatically computed based
on data available in a tool suitable for end-users. In our
evaluation application a commercial chemical batch control
system, called SattBatch, from ABB was used. SattBatch is
used to edit recipes, which are descriptions of how to produce
each product, SattBatch also holds a model of the plant, and
SattBatch is also used to start and execute recipes or products.
In SattBatch, the typical scenario is that the entire recipe is
entered before the production of the recipe starts. However, in
SattBatch it is possible to manually control the production of
batches while the control system is responsible for producing
the other batches. This implies that it is in general possible
for the manual user to run the system into a deadlock. It is
also possible to use the optimization functions in Supremica
to compute the time-optimal coordination of products in the
resource allocation systems. The optimization algorithms used
is partially presented in [23].
IV. CONCLUSIONS
A new tool, Supremica, for verication and synthesis of
discrete event supervisors according to the Supervisory control
theory was presented. Supremica has been used to teach
students the basics of the Supervisory control theory as well
as communicate the main ideas to the students. It is possible
to evaluate the synthesized control code against a graphical
simulation of the environment as well as generate standard
compliant code that can be executed in this tool against a phys-
ical environment or by a standard compliant Programmable
Fig. 3. Supremica doing a graphical simulation of the AGV system.
logic controller. Efcient algorithms and data-structures make
it possible to verify and synthesize control functions for
problems of industrial size. The tool has been used in a number
of applications including handling the resource allocations in
a commercial chemical batch control system. Ongoing work
will let the tool be responsible for the optimal collision free
coordination of a multi-robot system.
REFERENCES
[1] P. Ramadge and W. Wonham, The control of discrete event systems,
Proc. of IEEE, vol. 77, no. 1, pp. 8198, 1989.
[2] K.

Akesson, H. Flordal, and M. Fabian, Exploiting modularity for
synthesis and verication of supervisors, in Proc. of the 15th Triennial
World Congress of the International Federation of Automatic Control,
Barcelona, Spain, 2002.
[3] R. Bryant, Symbolic manipulation with ordered binary decision dia-
grams, ACM Computing Surveys 24, vol. 24, pp. 293318, 1992.
[4] Z. Zhang and W. Wonham, STCT: An efcient algorithm for supervi-
sory control design, in In B. Caillaud, P. Darondeau, L. Lavagno, X.
Xie, editors, Synthesis and Control of Discrete Event Systems, 2001, pp.
7798.
[5] J. Gunnarsson, Symbolic methods and tools for discrete event dynamic
systems, Ph.D. dissertation, Dept. of Electrical Engineering, Link oping
University, Link oping, Sweden, 1997.
[6] V. Chandra, B. Oruganti, and R. Kumar, UKDES: A graphical software
tool for the design, analysis & control of discrete event systems, 2002,
submitted for publication.
[7] E. Tronci, Automatic synthesis of controllers from formal specica-
tions, in Proc. of 2nd IEEE Int. Conf. on Formal Engineering Methods,
1998.
[8] S. Balemi, G. Hoffmann, P. Gyugyi, H. Wong-Toi, and G. Franklin, Su-
pervisory control of a rapid thermal multiprocessor, IEEE Transactions
on Automatic Control, vol. 38, no. 7, pp. 10401059, 1993.
[9] S. Balemi, Control of discrete event systems: Theory and application,
Ph.D. dissertation, Automatic Control Laboratory, Swiss Federal Insitute
of Technology, Z urich, Switzerland, 1992.
[10] G. Hoffmann and H. Wong-Toi, Symbolic synthesis of supervisory con-
trollers, in Proc. of the 1992 American Control Conference, Chicago,
IL, 1992, pp. 27892793.
[11] K.

Akesson, Methods and tools in supervisory control theory: Operator
aspects, computation efciency and applications, PhD thesis, Chalmers
University of Technology, England, 2002.
[12] M. Fabian and A. Hellgren, Desco - a Tool for Education and Control
of Discrete Event Systems. Kluwer Academic Publishers, 2000.
[13] C. Hoare, Communicating Sequential Processes. Prentice-Hall Inter-
national Series in Computer Science, 1985.
[14] A. Vahidi, A study of symbolic tools in automatic control, Department
of Signals and Systems, Chalmers University of Technology, Tech. Rep.,
2002, licentiate thesis.
[15] G. Frey and L. Litz, Formal methods in PLC programming, in
Proceedings of the IEEE SMC 2000, Nashville, USA, 2000, pp. 2431
2436.
[16] N. Pryce and J. Magee, Scenebeans: A component-based animation
framework for java, 2000, draft paper, Department of Computing,
Imperial College.
[17] IEC, IEC 61131 programmable conrollers part 3: Programming lan-
guages, International Electrotechnical Commission, Tech. Rep., 1993.
[18] A. Hellgren, M. Fabian, and B. Lennartson, Synchronized execution of
discrete event models using sequential function charts, in Proc. of the
1999 IEEE Conference on Decision and Control, Phoenix AZ, USA,
1999.
[19] P. Dietrich, , R. Malik, W. Wonham, and B. Brandin, Implementation
considerations in supervisory control, in In B. Caillaud, P. Darondeau,
L. Lavagno, X. Xie, editors, Synthesis and Control of Discrete Event
Systems. Kluwer Academic Publishers, 2001, pp. 185201.
[20] B. Joy, G. Steele, J. Gosling, and G. Bracha, Java(TM) Language
Specication, 2nd ed. Addison-Wesley, 2000.
[21] G. Bollella, J. Gosling, B. Brosgol, P. Dibble, S. Furr, and M. Turnbull,
The Real-Time Specication for Java. Addison-Wesley, 2000.
[22] L. Holloway and B. Krogh, Synthesis of feedback control logic for a
class of controlled Petri nets, IEEE Transactions on Automatic Control,
vol. 35, no. 5, pp. 514523, 1990.
[23] T. Liljenvall, Scheduling for production systems, Department of Sig-
nals and Systems, Chalmers University of Technology, Tech. Rep., 1998,
licentiate thesis.