Supervisory Control of Industrial Processes

5
Supervisory Control of Industrial Processes

Alexander A. Ambartsumyan
Institute of Control Sciences RAS,
Russia
1. Introduction
Modern production is complex, integrated and is constantly being adapted to the market
requirements by means of the reconfiguration of equipment structure and process alteration.
The development of such production is performed based on evolutionary strategy by
successively engaging (eliminating) stand-alone technological systems.
Evolutionary developed technical systems and facilities presently make up a considerable
share of technical systems. It is typical both for high-tech industries, namely: aviation, space
exploration, military equipment, machine-building (Sujeet, 2005), and for applications based
on large-scale interconnected production complexes (e.g. oil- and gas-producing industry,
oil and gas transportation, city economy engineering etc) (Gilard, 1999; Van Brussel et al.,
1999; Jo, 1999; Ambartsumyan, Prangishvili, Poletykin, 2003; Ambartsumyan, Kazansky,
2008; Ambartsumyan, Potehin, 2003; Ambartsumyan, Branishtov, 2006).
Evolutionary developed technical systems and facilities are featured by complex control
system availability. The latter integrates into a single whole different, as to the purposes,
automatic control loops (automatic control and regulation of physical process parameters,
automatic shielding and blocking, logical configuration control) as well as the functions of
supervisory control mainly aimed at coordination of different processes in a technical system.
Supervisory control (SC) is intrinsically logical and is to provide the required operational
sequence and exclude mutual blocking and deadlocks for stand-alone components
(operating according to their internal rules time scale). SC is discrete and asynchronous by
its nature and most commonly reveals itself as the change of event flow as required by
certain application (technical system functionality).
It is important to consider two "event" aspects: first, everything happens as the result of a
certain event; second, the change of states is regulated by events – there is no physical time
though the system is dynamic.
Though control systems are widely spread in the technical systems of such kind
(Sujeet, 2005; Gilard, 1999; Van Brussel et al., 1999; Jo, 1999; Ambartsumyan, Prangishvili,
Poletykin, 2003; Ambartsumyan, Kazansky, 2008; Ambartsumyan, Potehin, 2003;
Ambartsumyan, Branishtov, 2006), presently there is no appropriate theoretical base to solve
such supervisory control tasks as local control loops coordination, configuration of material
flows structure and interaction with operations staff.
Most spread concept of practical engineering of such systems is based on the model of
interacting ″black boxes″: a ″black box–control object″ and symmetrically connected with it
as to inputs and outputs a ″black box–control system (device)″. (Fig. 1).
Source: Process Management, Book edited by: Mária Pomffyová,
ISBN 978-953-307-085-8, pp. 338, April 2010, INTECH, Croatia, downloaded from SCIYO.COM
www.intechopen.com
3 AUDIOBOOK COLLECTIONS
6 BOOK COLLECTIONS
68 Process Management
Fig. 1. The scheme of transfer from the object data base and control requirements to the
mathematical description of the control
The first ″black box–control object″ is formed as a data base on the control object and
technique at the stage of the object examination and includes the requirements of this object
appropriate behaviour. The task of the required control search is tackled by the defining of a
″black box–control system″ able to monitor the behaviour – the event flow and, with the
control purpose taken into account, to affect the object inputs in such a way that an
appropriate behaviour of the object is achieved.
The question is how to search for a ″black box–control system″ with information on the first
black box available. Common engineering practice shows that information on control object
behaviour is only used indirectly.
What is the problem? We may speak about precise correspondence between a ″black box–
control object″ and a ″black box–control system″ only as far as inputs and outputs are
concerned, while behaviour is an approximate result of the designer’s informal, speculative
experiment with the initial data and limitations – the information the designer acquires
considering the process physics peculiarities and the object structure properties. At that,
there is not any confidence that a ″black box–control system″ can limit the behaviour of a
″black box–control object″ and provide its meeting the requirements since they, as a rule, are
specified as models of another (not "event") nature and the extent they are taken into
account depends on the designer’s skills. The above leads to serious problems: designer’s
uncertainty in the fact that the designed system complies with the control tasks set; the
necessity to make laborious verification of such compliance by computer simulation and the
refinement of the designed system at facilities.
For the last 10–15, a sophisticated interaction among computer-driven actuating devices
necessitates, when engineering, to analyze the design solutions safety and correctness, to
validate technical systems implementation techniques, to take other approaches actually
based on testing. It is a common knowledge that such approaches only can reveal a part of
errors but cannot guarantee the system as a whole is error-free.
Different engineering approach than that based on two black boxes concept is declared in
the theory of discrete event dynamic systems and supervisory control paradigm. The
abbreviation is often simplified to DES. The distinctive features of supervisory control
theory (all basic concepts and notions of this paper are borrowed from (Cassandras,
Lafortune, 2008)) are as follows:
• The controlled object is represented in DES model by three components: generator G of
L(G) language – proper control object, specification language К – limitations and G
functionality required, supervisor S – control component in DES;
• Setting and solving the task of formal synthesis of S on L(G) and K.
The above, in its turn, creates a theoretical basis for machine control engineering
fundamentally different from the deciphering of "black boxes" approximately fitting each
www.intechopen.com
Supervisory Control of Industrial Processes 69
other. What does it give as compared with the classic procedure of discrete process control
system synthesis according to two-black-boxes model?
First, the description of the object as L(G)-language generator G, limited by nothing, is more
simple than the object description with all the admissible behaviour limitations taken into
account. This work is performed as a separate stage – primary object examination and
constructing a model "as it is".
Second, to form the required functionality (К specifications) basing on a generator G model
already available is also easier than to consider all limitations and requirements in yet non-
existing control system.
Third, control task is solved formally: a supervisor (provided the initial data is correct) is
synthesized and does not require verification while the object and its behaviour are
specified by object and know-how specialist and he is responsible for the data correctness,
its verification and validation.
The present paper formulates the purpose of DES theory development, with the structural
properties of technical systems taken into account, thus creating effective methods to
synthesize a supervisor as an instrument to solve the task of consistency and co-ordination
control of stand-alone components in a technical system.
Here below is given a brief survey of basic concepts and major noted results, as to DES and
supervisory control, followed by the description of the present paper tasks and the results
obtained.
2. Basic concepts and definitions

DES behaviour is considered generally as behaviour of a certain generator (source) of strings
(sequences) of the events from a finite set of events E. The event e ∈ E is an abstraction for a
multitude of facts associated with DES "life". Events are instantaneous, occur spontaneously
in unpredictable moments, therefore the only thing that can be observed is their sequences
that are represented by strings. Event examples are: the facts of change in position and state
of separate object components; commands to which the object reacts by the change of its
state (position); characteristics of normal and abnormal states etc.
The main operation of strings forming is concatenation (we would like to remind that
concatenation is the appending of separate events or entire strings of events on the right to
the string, including – a space character). For the string, an integral function µ ( s ) = n is
defined, where n is the number of characters in string s. If n = 0, s = . A set of all string of
any finite length is designated by E* (it is endless but countable). Let a string s consist of
three parts: r, u, t ∈ E* connected by concatenation in such a way that s = rut, where r – a
prefix, t – a suffix, and u – a substring of string s. Any subset of strings L⊆E* is called a
language over E. If L includes and, jointly with any string s, contains all its prefixes, L is a
prefix–closed language. As usual, conventional language operations are defined, namely:
concatenation, prefix-closure and Kleene-closure.
In many constructions of DES theory, a couple of very important operations over languages
are used: a projection P and a back projection P-1. Let E1, E2⊂E be such that E1∪E2 = E
(possibly E1∩E2 ≠ Ø). Projection Pi of any string from Е* on Ei is defined in three steps:
1. Pi( ) = ; 2. Pi(e) = if e ∉ Ei, otherwise Pi(e) = e; 3. Pi(se) = Pi(s) Pi(e) for s ∈ E* and e ∈ E.
Conceptually, a projection of strings from larger alphabet E on smaller one Ei deletes from
the string all characters from E \ Ei (all characters outside Ei). Inverse function Pi-1(s) = {t ∈
www.intechopen.com
E*: Pi(t) = s}. Pi-1(s) correlates every string s ∈ Ei with some subset of strings E* the projects of
which on Ei equal s. Both operations are in natural manner extended to the languages L ⊆ E*
and Li ⊆ Ei*. Pi(L)={t ∈ Li: (∃s ∈ L) [Pi(s) = t]}; Pi−1(Li) := {s ∈ E*: (∃t ∈ Li) [Pi(s) = t]}.
In projection operation definition, instead of set indexes, for the sets, the events of which are
excluded from the result of this operation, we shall use the designation of the set itself:
PEi or PE−i 1 .
Languages are a good instrument to observe DES behaviour but in order to perform
analytical study and to set the task of providing the required dynamics (off-line behaviour),
it is necessary to present a countable string set as a mathematical operator. There are many
ways to present languages in the form of mathematical operators that generate or recognise
the language. In DES theory, for these purposes, as a rule, finite state machines are used. A
finite state machine is defined as G = (Q , E , δ , Γ , Qm , q0 ) , where Q – a set of states; E – a set of
events; – a transition function Q × E → Q ; Γ : Q → 2 E – a function of admissible events in
each state; Qm – a set of marked states; q0 – an initial state. We would like to note that in this
definition the function of outputs is missing. For every state qi the function of transitions is
specified for the events admissible in this state (e.g. for qi ∈ Q and e ∈ Γ i the function
δ (qi , e ) := q j ). This definition can be naturally extended also for the following event strings:
δ (qi , ε ) := qi , δ (qi , se ) := δ (δ (qi , s ), e ) for s ∈ E* and e ∈ E. Let’s denote by δ (qi , s )! the fact that
the function δ (qi , s ) is defined.
The function Γ : Q → 2 E is excessive in a model definition but it simplifies many
examination schemes and algorithms development when analysing the languages presented
by finite state machines, e.g. consistency definition. Qm ⊂ Q is a subset of marked states –
the states corresponding to a certain functionality of G, with one of them necessarily being
initiated in a specific variant of G use.
The language generated by G machine is designated as L(G ) := {s ∈ E∗ : δ (q0 , s )!} . This is a set
of all strings from E* admissible in the initial state q0. It is evident that L(G ) ⊆ E∗ . If the
machine is completely defined, L(G) = E*. It G is represented by a weighed graph of
transitions, L(G) is presented as a set of strings of the events weighing the edges of all the
paths originated from the initial state q0.
When a sophisticated DES is defined via components, two more operations on machines are
often applied: Cartesian product and parallel composition. Product definition
G1×G2 = (Q1×Q2, E1 ∩ E2, 1,2, Γ1×2, Qm1×Qm2, (q0 := q01, 02))

is conventional but there is one nuance: a function of transitions is defined on common
events for every pair of states. Isolated pairs and those unattainable from the initial state are
discarded together with their associated transitions. From the definition it follows that the
language L(G1×G2) of the Cartesian product of two machines is equal to L(G1) ∩ L(G2) – the
intersection of these machines languages.
Parallel composition (or just composition, let it be designated as ⊕) is defined on the union
of events of both machines G1⊕G2 = (Q1×Q2, E1 ∪ E2, 1,2, Γ1⊕2, Qm1⊕Qm2, (q01,q02)). At this, it
is possible that E1 ∩ E2 ≠ Ø, then on common events, transition synchronization takes place
in both components. If the event is individual, transition takes place in one component
(provided for this pair this event belongs to the value area of the corresponding function Г).
www.intechopen.com
Formally:
((q1, q2), e) = {( 1(q1, e), 2(q2, e)) if e ∈ Г1(q1) ∩ Г2(q2) │ ( 1(q1, e), q2) if e ∈ Г1(q1) \ E2 │ (q1, 2
(q2, e)) if e ∈ Г2(q2) \ E1 │ and indeterminate in other cases}.
It is obvious that both operations are associative and, provided parentheses are places
accordingly, may be easily generalized for n machines: a product –
n n
G = ×1Gi = G1 × ... × Gn ; a composition – G = ⊕ 1G i = G i ⊕ ... ⊕ G n .
The initial stage of object study (modelling) is dedicated to prognostication of possible
physical behaviour of the entire object or its subsystems, i.e. consideration of possible
actions and possible variants of behaviour in the absence of any control and restrictive
actions. At this stage, DES is represented by machine G as a language L(G) generator. Thus,
G generates event sequences of any kind reflecting control-free DES behaviour. In order to
specify and provide control in DES, a set of events E is subdivided into two disjoint subsets:
Ec – a subset of controllable events corresponding to the commands and Euc – a subset of
uncontrollable events for which the moments they occur are unpredictable.
The present-day view on DES was first worded in (Ramadge, Wonham, 1987) though then
the term "discrete event systems" was not used but a new technique of discrete process
modelling and control was stated. The term "discrete event systems (DES)" appears already
in (Ramadge, Wonham, 1989), where DES is represented by generator G of different
sequences of events from E. G is limited by nothing and therefore the sequences reflect the
behaviour L(G ) ⊆ E* unbounded by control. Any DES has some functionality to implement
which are required not all possible sequences but only those providing this functionality
and meeting the limitations specified. In order only to provide the required event sequences,
G is term "supplemented" by supervisor S, built-in a "feedback" manner (Fig. 2).
en, en-1,…,e1
G S
eu1, eu-1,…,euk
Fig. 2. The scheme of object – supervisor interaction

The scheme in Fig. 2 is no different from the conventional structure "control object – control
system" but the behaviour is absolutely different. First, a generator event sequence covers all
events in the system; second, a supervisor sequence includes only controlled events and
third, controlled event ek is incorporated into G output sequence conditioned to its presence
also in S sequence. This allowed to define S transparently enough as a function of strings
from the set L(G ) : S : L(G ) → 2 E .
Supervisor S is equipped with a mechanism of G sequences blocking provided they do not
meet limitations. For this purposes, S structure comprises one more component allowing for
G "free" behavior restriction – a specification K. For the real object, a certain functionality
(depending on G destination) must consider a multitude of all types of requirements and
limitations R = {ri | i=1,..,n}. As a rule, R is formed reasoning from physical, process and
www.intechopen.com
design limitations imposed on joint behaviour of separate G components. The allowance for
all restrictions R gives rise to K ⊆ L(G) – a language of specifications – a subset of sequences
dictated by G functionality. Actual control scheme stated in (Van Brussel et al., 1987) is
presented in Fig. 3. It took the name of "Supervisory control theory" or RW approach
(named after its authors J. Ramadge J. and W. Wonham W).
Fig. 3. Interrelationship of supervisory control components in DES

The functioning of G in the presence of S is denoted by S/G and a corresponding language –
L(S/G). The scheme symbolically shows that specification K is involved in S forming and in
providing blocking. Supervisor is designed, with K taken into account, in such a way that, in
accordance with L(G) observation results, S blocking mechanism provide the language
L(S/G) = K at DES output. We would like briefly to dwell upon the way L(S/G) generation is
realized. G is supposed to have its own controller that generates control events while a
supervisor blocks the events the occurrence of which runs counter to the specification
(Fig.4).
TCO
∩
E’c E=Ec Euc
actuators
S/Ec E’’ c
Process
∩ Ec
controller
E=Ec Euc
Ec supervizor E=Ec Euc
Fig. 4. Control scheme proposed in the paper (Ramadge, Wonham 1987)

Supervisor S monitors G output events and permits all Euc events, while as to Ec events, it is
"entitled" to permit or not permit them (to block by imposing limits on transition function
δ (qi , eс ) := q j ). For every string s ∈ L(G) generated by G under S control, a supervisor only
permits a set {S(s) ∩ Γ( (q0, s))} – a set of events admissible in G current state (q0, s) and not
conflicting with K. Hereinafter, (q0, s) will mean a state G transfers to from q0 as affected by
s. In other words, G cannot realize the event from its current active event subset Γ( (q0, s))
unless this event is contained also in S(s). However, making allowance for the fact that E is
subdivided into controllable and uncontrollable subsets and the appearance of the latter is
limited by nothing, supervisor S is called admissible if for all s ∈ L(G), always Euc ∩ Γ( (x0,
s)) ⊆ S(s), i.e. S is specified in such a way that in all states it is impossible to block an
www.intechopen.com
uncontrollable event and vice versa: S blocks the events not meeting limitations (irrelevant
to K). Further on, only admissible supervisors will be considered.
For the modelling of DES with passive actuators in paper (Chalmers, Golaszewski,
Ramadge, 1987) it is suggested that the model should be expanded with forced controllable
events and a new control scheme (Fig. 5), with controllable events generated by supervisor,
is developed. For such model, the terms of controllability for specification language are also
defined.
TCO
Ec ∩
actuators E’c E=Ec Euc
E’’c
supervizor E=Ec Euc
Fig. 5. Control scheme for DES with forced controllable events

For both models were developed the methods of supervisor synthesis as a finite state
machine (FSM) with output converters regulating blocking (or generation) of Ec events.
However, for the methods proposed the number of supervisor S states is less or equal to the
product of the number of states for G and K (Cassandras, Lafortune, 2008).
DES dynamics is interpreted in the sense that the system (a pair of G and S), once set to the
initial state, operates off-line, reacting to internal and external events, and provides a
resulting flow relevant to G structure and S control.
Since 1987, there have been a lot of publications on DES subject-matter. At three last world
IFAC Congresses, three sections on DES theory were working; IFAC Committee on DES
theory was established; symposiums on this subject-matter are held. The paper scope
limitation does not allow to survey the results on DES theory so we shall confine ourselves
to listing the basic research trends. They are as follows:
• Study of DES as a dynamic system with a certain range of states and a structure of event
transitions; the study of properties of the languages generating DES from the position of
general control theory and the definition, in terms of language properties, of
controllability, observability, attainability, safety (avoiding blocking situation) and
some others;
• Study of different models of G and K specification (finite state machines, Petry nets etc)
and the development of synthesis (engineering) methods for supervisor S on G and K;
• Assessment of supervisor complexity at synthesis with FSM models of G and K
involved;
• Study of different modular presentations of supervisor S in the form of parallel
generators of sub-languages with their subsequent combining via product operation
(conjunctive scheme), via parallel composition operation (disjunctive scheme) and others;
• Development of programming methods for logical controllers in industrial systems
with supervisor control theory applied;
• Creation of program verification methods for industrial systems with DES, as
simulation instrument, applied;
• Development of the methods of industrial system state diagnostics using DES as a
modelling instrument.
www.intechopen.com
A detailed survey of the results obtained on DES can be found in (Cassandras, Lafortune,
2008); herein the major results on controllability from (Ramadge, Wonham, 1987; Ramadge,
Wonham, 1989) are set forth:
• Is formulated the condition of controllability for the language: K ⊆ L(G ) is controllable
if KEuc ∩ L(G ) ⊆ K
• It is proved that if K is controllable, there exists a non-blocked S such that L(S/G) = K
• Are developed the methods to design supervisor S as a function of strings (Ramadge,
Wonham, 1987; Cassandras, Lafortune, 2008).
However, the direct practical application of the proposed models and methods is confined
to lab examples of dynamic DES engineering and supervisor synthesis. Such constraint is
explained by high dimensionality of the object states set. To analyze for controllability, a
complete DES specification of generator G is required. Even in the simple example given
here below (a machine with four mechanisms) the number of states equals 4356. (The
number can be considerably reduced with DES structural features taken into account).
Main direction of works focused on overcoming supervisor synthesis complexity is based on
different kind of modularity. Methods of modular supervisor synthesis for G, as a single
entity, are elaborated. At this, different control schemes are explored (disjunctive,
conjunctive, hierarchical, generalized). Pioneer work (Ramadge, Wonham, 1989) that
initiated the development of modularity, as applied to DES theory, was evolved and
generalized in (Yoo, Lafortune, 2002). Later, different authors (De Queiroz, Cury, 2000;
Gaudin, Marchand, 2003) developed the methods of modular supervisor synthesis on
modular description G=<G1, G2, …, Gn> and modular specification K=<K1, K2, …, Kn> of
modular S. However, the complexity of such synthesis and weak correspondence of the
initial specification structure to the resulting supervisor make the methods proposed
scantily attractive for practical implementation. Besides, controllability properties are
verified on language models K and L(G) defined for the object (Plant) as a whole, which
makes it difficult to apply these results to real industrial facilities.
The present paper sets the task to develop a prototype of structured dynamic DES by
structuring the object components according to their functionality. To operate the model, the
paper proposes the methods that will allow to raise the dimension of supervisor control
tasks and form a theoretical basis for a new supervisor control engineering technique.
Structured are all three DES components but mainly object model and specification.
3. Structured Discrete Event Systems (SDES)

3.1 Base concept – the structuring of events and specifications
The author considers it promising to develop a supervisory control theory in the direction of
structuring the events according to their role in production operations and in the required
object behaviour specification. This research is based on two specific machinery features
from DES-modelling point of view. The first feature relates to the fact that for discrete
machinery a set of events is usually subdivided into three sets. These are sets of controllable
and uncontrollable events Ec and Euc (typical for DES theory) and Ew is a set of expected
events. The events from Ew simulate states (positions) of actuator(s) or object components.
Supervisor cannot block Ew events as those controllable from Ec and thus Ew events are
traditionally referred to uncontrollable events as per Wonham's classification (Ramadge and
Wonham, 1987). However, Ew events are expected to occur as a response to Ec events – a
www.intechopen.com
confirmation of the fact that the commands sent to actuators were executed. So, the
foregoing gives the ground to mark out Ew events as a separate set. The second specific
feature is as follows: the behaviour of every actuator Gi is simulated by the language L(Gi) of
words over Ei = {Ewi ∪ Eci } and the specification of desired behaviour is formulated as a
language K over events Ed = Ec∪Euc, a totality of commands and conditions of their use.
Making the allowance for these specifics, makes it possible to get numerous advantages both
in defining DES and formulating controllability conditions and supervisor synthesis.
3.2 SDES definition

Definition 1: If the structure of DES is defined by: a collection of components
G=<G1,G2,…,Gn>; sets of Ei events, each being structured on Ei = {Ewi ∪ Eci } , and a set Euc of
general uncontrolled events; the behaviour of each DES component being defined by FSM
G i = Q i , Ei , δ i , Γ i , Qmi , q0i and L(G i ) language, then the DES with the above structure is
called well structured.
A set of common events for G=<G1,G2,…,Gn> is defined through the union of subsets
E = {Ew ∪ Ec ∪ Euc } , where Ew and Ec each are the unions of appropriate component subsets.
Note 1: Sets Ew and Ec for various mechanisms do not intersect, since various mechanisms
have their own actuators and their states are individual.
Note 2: Components of Gi define the behaviour of G that is not limited (controllable) by
anything, e.g. from the successive operation of <G1,G2,…,Gn> in any order up to their
independent work in parallel.
According to the theory of supervisory control, a parallel composition of all object
components is implemented, and, as the result, a model of uncontrollable object behaviour
is created (Ramadge & Wonham, 1987). The narrowing of free behaviour is carried out with
the constraints of purposeful joint behaviour considered. This, in essence, is the procedure
of adapting the initial unlimited behaviour i.e specifying the behaviour as required by
application. We would like to remind that the implementation of all restrictions generates a
language K ⊆ L(G ) called a language of specifications. Establishing the restrictions is a
creative process that requires an experimental approach to achieve a reliable result. Such
experiment is quite difficult to carry out as the number of states is increasing in the course of
composing. There is a collision.
On the one hand, a system analyst needs to get a general picture of all the transitions to
analyze their admissibility.
On the other hand, it is unreal to do it for complete composition, since the number of states
in it is too high (for practical applications this number is about n·103). Sequent revealing of
restrictions in the process of pair-wise composing, gives a ground to doubt of such
restrictions completeness or, on the contrary, of their extreme strictness. At the same time,
there is no possibility to consider the joint action of components with those absent in the
composition.
At the same time, it is known from the practice of discrete process engineering that the
efficient behaviour of discrete systems is achieved by solving two control tasks, namely:
operation control and control of operation sequence. Operation control is provided by the
execution of a certain command and monitoring the corresponding object response.
Commands and their reactions once defined, are iterated in various places of the sequence
of operations. In process modelling, it is important to set up the sequence of commands and
www.intechopen.com
to evaluate the completeness and correctness of conditions. With the above in view, herein is
proposed to create a specification of a well-structured DES with the events Ed = {Ec∪Euc}, i.e.
combination of commands and conditions for their execution in sequence.
Definition 2: The language K⊆Ed* defined by FSM Η = (Q h , Ed , δ h , Γ h , Qmh , q0 ) as a set of
strings defining the required specifications, is called a directive specification language (a
process specification tapes language).
It is assumed that FSM H has no deadlocks (Fig. 6) and livelocks (liveloops, within which H
fails to go out of a certain state subset and does not reach Qm and then q0), i.e. H is non-
blocking.
It is worthy to be noted that if a graph is strongly connected and q0 ∉ Qm , then q0
transitions only as shown in Fig. 6 are possible.
livelock
deadlock
Qm –
marked
out states
qo
Fig. 6. Types of fragments in the machine Н

The fact of non-blocking is easily verified. Contrary to the general DES theory (Cassandras,
Lafortune, 2008), where deadlocks and livelocks result from the excessive general
description via the product and composition, in SDES, there should be no hurry in cutting
down "bad" states and transitions but, vice versa, it is necessary to check if any transition is
missed to avoid deadlock or livelock situations.
Let’s define a supervisor for G and K. It is conceptually evident, that supervisor is an
operator that defins, for every string s, which of possible events, admissible for G, are
suitable as the next event not conflicting with K. At this, supervisor remains admissible in
terms of (Van Brussel et al., 1998) since it in no way limits Euc occurrence and affects only Ec.
Definition 3. Supervisor S is a converter of strings admissible for the system
G = G 1 , G 2 ,..., G n initial state to the events S(s ) = {ε ∪ Euc ∪ { ec }} such that: first, these are
any of uncontrollable events Euc (i.e. S is admissible for G); second, these are controllable
events ec admissible for the current G state; third, these events do not cause blocking of S
and G = G 1 , G 2 ,..., G n composition.
Let’s denote, as agreed, by L(S/G) the language generating G under S control. It is evident
that L(S/G) ⊆ L(G). Let’s also give a definition of L(S/G) language generating S/G, that is
consistent to the conventional definition of language generating G under S control.
Definition 4. The language L(G/S) generating G = G 1 , G 2 ,..., G n under S control contains
the following strings:
1. ∈ L(S/G);
2. ∀s , e(s ∈ L(S / G ) ∧ e ∈ S(s )) : se ∈ L(G ) ⇔ se ∈ L(S / G )
In other words, any string se belongs to L(S/G) provided it also belongs to L(G) being at the
same time the extension of string s which also enters L(S/G) by event e such that e ∈ S(s ) .
Possibly, s = ε .
www.intechopen.com
Definition 5. A well-structured DES, for which the uncontrollable part is set up by

definition 1, the desired behaviour is set by specification language K ⊆ Ed* (K ≠ Ø), and
which is supplied with a supervisor S such that K is fulfilled, is called a structured dynamic
discrete event system (SDES).
K fulfilment means that PEd (L(S / G )) = K , i.e. that K will be equivalent to the projection on
Ed of L(S/G) language that is generated by S/G.
3.3 Technical object modelling by structured DES

The events associated with real industrial objects, as a rule, are easily divided into groups
(types) as proposed herein. Such event grouping is typical for process systems of many
industrial spheres. Here below is the example which refers to the field of mechanical metal-
working. We consider this example most interesting since it is close to illustrative examples
frequently used in publications on DES (Ramadge, Wonham, 1987; Ramadge, Wonham,
1989; Chalmers, Golaszewski, Ramadge, 1987; Ambartsumyan, 2009).
The structuring of technical object (the first phase of study) includes as follows:
• enumerating actuators;
• defining for each of them the set of events necessary and sufficient for the outer
supervisor to identify actuators behaviour;
• defining the classification of marked out events;
• defining the components and object behaviour in the compact-form languages, e.g.
finite machine models.
In Fig. 7 a kinematical scheme of a small milling machine is presented. The machine consists
of 4 mechanisms: "workpiece clutch" - G1, "turntable" – G2, "spindle" – G3 and "cutter" – G4
To the left
To the right
Working (smooth)
Open clutch
Close clutch Turn on cutter
Parked (--+)
Ready to work (-+-)
Clutch is open Cutter is working End of the operation (+--)
Clutch is closed
Detail is on the table

Detail is ready to be send Rotate on 1/4
Table is fixed
Table is rotating
Table has closed full circle
Fig. 7. Kinematical model of the machine

Let’s enumerate the events and their semantics in the liveloop (behaviour) of each
mechanism.
www.intechopen.com
"Workpiece clutch" mechanism: e1-1 – to clamp, e1-2 – clutch closed, e1-3 – to unclamp, e1-4 –
clutch closed, e1-5 – clutch is moving.
"Turntable" mechanism: e2-1 – to lock the table, e2-2 – table locked, e2-3 – to unlock the table,
e2-4 – table unlocked, e2-5 – locker is moving, e2-6 – to make a ¼ turn, e2-7 – table is moving,
e2-8 – table is turned, e2-9 – to switch off turning gear, e2-10 – table stopped.
"Spindle" mechanism: e3-1 – to move spindle fast to the left, e3-2 – feed zone, e3-3 – working
position, e3-4 – to move spindle to the left, e3-5 – working zone, e3-6 – operation finished, e3-7 –
to move spindle to the right, e3-8 – to move spindle fast to the right, e3-9 – parked.
"Cutter" mechanism: e4-1 – to turn on cutter, e4-2 – cutter working, e4-3 – to turn off cutter, e4-4 –
cutter stopped, e4-6 – cutter unstable spinning.
Mechanisms behaviour, as agreed here above, will be considered as sequences (strings) of
possible events. These sequences will be defined as finite state machines (Fig. 8–11).
Hereinafter they are called component finite machines (CFM). It is easily seen that CFM
transition graphs and graph edges weighed by events, specify operation of each mechanism
quite transparently.
moving
e1-1 e1-5
clamp
2 e1-5 3 clutched
e1-4 e1-2
e1-1 e1-2
Clutch is
opened 1 4
e1-5 e1-3
e1-4 e1-3
6 e1-5 5 unclamp
moving
Fig. 8. G1 CFM – a model of "Workpiece clutch" mechanism
Start(Stopped and fixed)
e2-10 Locker moving fixed

e2-9 e2-1 e2-2 unfix
Stop turnable
e2-10 1
e2-1 e2-5
mechanism 11 2 e2-5 4 e2-3
3 e2-2 e2-3
e2-9
e2-6 5
e2-8 e2-7 e2-5
e2-4
e2-8 e2-7 8 e2-5
10 9 e2-6 6
7 e2-4
Rotated turn Locker
on 1/4 turning moving
Ready to
turn
Fig. 9. G2 CFM – a model of "Turntable" mechanism
www.intechopen.com
Start Feed
(Parked) <<
zone
e3-9 e3-1 e3-2
1 2 e3-2 3
e3-1 e3-3
e3-8
e3-2 Back >> e3-3
e3-9 e3-8
10
11 e3-2 4
e3-5 e3-3
e3-7
Back > 9 e3-4 e3-4
8 e3-5
e3-7
e3-6 5
e3-5
e3-5
7 e3-6 <
6
Working
End
zone
Fig. 10. G3 CFM – a model of "Spindle" mechanism
e4-1 e4-6
2 e4-6 3
e4-4 e4-2
e4-1 e4-2
1 4
e4-6 e4-3
e4-4 e4-3
6 e4-6 5
Fig. 11. G4 CFM – a model of "Cutter" mechanism

It is easy to make natural event grouping in all the CFM, namely:
G1 – Ec1 = { e1 − 1 ,e1 − 3 } , Ew1 = { e1 − 2 ,e1 − 4 , e1− 5 } ; G2 – Ec2 = { e2 − 1 ,e2 − 3 ,e2 − 6 ,e2 − 9 } ,
Ew2 = { e2 − 2 , e2 − 4 , e2 − 5 , e2 − 7 e2 − 8 , e2 − 10 } ; G3 – Ec3 = { e3 − 1 ,e3 − 4 ,e3 − 7 ,e3 − 8 } ,
Ew3 = { e3 − 2 ,e3 − 3 , e3 − 5 , e3 − 6 , e3 − 9 } ; G4 – Ec4 = { e4 − 1 ,e4-3 } , Ew4 = { e4 − 2 ,e4 − 4 , e4 − 6 } and to see the events
Euc = { eex − 1 , eex − 2 , eex − 3 , eex − 4 , eex-s , eex − w } common for all components (respectively: a
workpiece is on the table; a workpiece is removed from the table; processing is over, clutch
of s type , clutch of w type).
Note 3. Sets Ew and Ec for different mechanisms do not intersect.
It is evident, since different mechanisms have their own drivers and their positions for each
mechanism are individual.
The next stage of a technical system SDES-modelling is the defining of the system behaviour
specification based on the requirements to the system functionality and limitations. It is
www.intechopen.com
done by forming the behaviour of G as an uncontrollable system, as a whole, followed by

putting in limitations, thus "narrowing" G behaviour up to that required.
The traditional approach being applied, uncontrollable G behaviour is defined by
component machines combination. Let’s use two mechanisms of the above milling machine
(Turntable and workpiece Clutch) to illustrate this.
clutch clutch clutch
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
2 2 2
3 3 3
4 4 4
t 5 t 5 t 5
a a a
b 6 b 6 b 6
l l l
e e e
7 7 7
8 8 8
9 9 9
10 10 10
11 11 11
Fig. 12. CFM composition for G1 and G2: а) complete; b) with allowance for limitations r1
and r2; c) with allowance for limitations r1, r2, r3
Pursuant to SC theory, we should make a composition of all machines to achieve
"uncontrollable" G behaviour. DES, modelling "uncontrollable" behaviour of the first two
mechanisms, is represented by G1 ⊕ G2 composition, with relevant transition graph structure
illustrated in Fig. 12-а. Here a structure of initial components transitions is shown: across - G1
structure, down - G2 structure, and relevant pairs are represented by nodes at arrows
intersection. Edges weighing corresponds to weighing of transitions in the initial components.
Machine G 1⊗ 2 represents unlimited by anything, parallel operation of mechanisms G1 and
G2 originating L( G 1⊗ 2 ) language.
In our example, the following restrictions as to joint behaviour of the mechanisms take
place: r1: "turning of G2 "Turntable" mechanism is possible if a workpiece is clutched"; r2: "if
in the course of the table turning a workpiece unclasping begins , "Turntable" will only
terminate turning".
The implementation of these technological restrictions are formally realized by banning the
following state compositions: 1, 2, 3 of G1 CFM and 2-9 of G2 CFM. With these limitations
applied, all pairs of states under verticals 1, 2, 3 and a number of pairs under verticals 5, 6
are excluded (Fig. 12-b). The same refers to their incident transitions. As the result, we get
the machine K1 as shown in Fig. 12-b. More detailed analysis of admissible transitions
results in the necessity of one more limitation: r3 – "at table turning, а workpiece unclasping
is inadmissible", which makes specification more strict (K2) as shown in Fig. 12-c.
www.intechopen.com
Thus, we have DES of G 1⊗ 2 and it’s necessary to provide its operation within the
framework of language K. In what way is it possible to regulate a path choice in G 1⊗ 2
graph? In our example, for G 1⊗ 2 Ec1⊗ 2 = { e1 − 1 ,e 1-3 , e2 − 1 ,e 2-3 , e2 − 6 , e2 − 9 } ,
Ew1⊗ 2 = { e1 − 2 ,e 1-4 , e1 − 5 , e2 − 2 ,e 2-4 , e2 − 5 , e2 − 7 e2 − 8 , e2 − 10 } .
Graph transition trajectory can be regulated by a function of transitions G1⊗ 2 by blocking or
accepting the events from Ec set with the help of supervisor S (outer to G) which
dynamically interacts with G in a feedback manner. The way it can be realized is illustrated
by our example. In state, q1,4 in cycles 1, 2 and 3 of the table operation, a supervisor each
time enables e2-1 and disables e1-4, and, after the table returns to its initial position for the 4-th
time, it is e1-4 that is admitted and e2-1 that is banned.
So, CFM sequential merging and the detection of limitations for CFM joint operation are
quite a complicated procedure even in our case. We have already noted that the detection of
limitations in the course of pairwise component combination, gives the ground to doubt
about the completeness of such limitations or vice versa in their excessive strictness. Besides,
there is no possibility to predict the consequences of joint operation with the components
still absent in the composition. For example, should we start CFM merging with "Spindle"
and "Turntable" mechanisms, it will in no way possible to make allowance for the fact that
between their "activities" a locker actuation will take place.
At the same time, for technical objects, their required behaviour is always defined by their
functionality that is specified, for example, by text description. The required machine
behaviour is presented by informal specification in table 1.
1) on arrival, the piece is locked by clutch; 6) positioner makes a ¼ table rotation;

2) after clenching, the spindle moves from 7) after the table is fixed, the next operation
park position to work position (to the left); is carried;
3) the cutter is switched on; 8) after the table makes a turnover, the
4) smooth feed to the left utmost position spindle is parked, the clutch is unclamped,
(operation is over); the signal of the piece readiness is sent;
5) the spindle moves to the right back to 9) prior to parking, to switch off the cutter
work position; and wait for a stop.
Table 1. Text description of initial specification

At SDES-modelling, at this stage, a specification of joint behaviour in K language is applied.
A specification, compliant with the text specification, is presented by machine
H = (Q h , Ed , δ h , Γ h , Qmh , q0 ) shown in Fig. 13.
Since the verbal behaviour description, as a rule, is inaccurate, the resulting specifications
may vary. The example of another interpretation of verbal description is presented in Fig.
14. The specification is described in conformity with verbal description. Basing on the
information from table 1, it is possible to assume that at the beginning of operation, the table
is fixed, since otherwise is not specified and thus, the operation relevant to the transition
graph node 3 is omitted. However, should the order of operations as shown in Fig. 14 be
accepted, already the processing of the second workpiece will start with the table unfixed
since in the beginning of the large loop locker is not considered. The necessary operation is
missing.
www.intechopen.com
e3-1 e4-1 e3-4 e3-7

4 Work state 5 Turn on 6 7 8
Feed Smoothly
cutter
e2-1 Lock to back
table
e1-1 Lock 13 e4-3
piece Stop cutter
Stop table
3 e2-9 Unlock Work not
9
e2-6 table finished
12 11 e2-3 10 eex-3
Lock Rotate
e2-1 table table Work is
on 1/4 finished eex-4
eex-1 Piece is
put on Piece is Ulock Park spidle
table out of table piece
2 1 16 e1-3 15 e3-8 14
eex-1 eex-2
Fig. 13. The required machine behaviour in terms of directive specifications. The semantics
is as follows: eex-1 – a workpiece is on the table, eex-2 – a workpiece is removed from the table,
eex-3 – processing is not over, eex-4 – processing is over (other events semantics was given here
above in the mechanisms description).
e3-1 e4-1 e3-4 e3-7

4 Work state 5 Turn on 6 7 8
Feed Smoothly
cutter
e2-1 Lock to back
table
13 e4-3
Stop cutter
Stop table
Lock e2-9 Unlock Work not
e1-1 piece 9
e2-6 table finished
12 11 e2-3 10 eex-3
Rotate
table Work is
on 1/4 finished eex-4
eex-1
Piece is
put on Piece is Ulock Park spidle
table out of table piece
2 1 16 e1-3 15 e3-8 14
eex-1 eex-2
Fig. 14. H specification with erroneous missing of "Table locking" operation

Operation omission is far from being the only inconsistency in the required behaviour
specification. Here below (Fig. 15) another text description interpretation is given. The
specification is elaborated in accordance with the text but a "cutter halt" operation (node 8 of
Fig. 15) is performed prior to cutter parking in the "large" loop, which follows from item 9 of
the text description from Table 1. Cutter halt is performed in the "large" loop but on the
processing termination, therefore, while processing the second piece position, the attempt
will be made to switch on a working cutter.
Note3. The composition of modular hierarchic DES description of solely unblocked modules
may result in DES blocked operation.
This stage of SDES-modelling reveals a principle difference of discrete control engineering
with supervisor S on G and K given, as compared with a "black box" technique.
www.intechopen.com
3 e1-1 4 e3-1 5 e4-1

Working 6
Clamp the state Turn on e3-4
piece cutter
e2-1
Lock 7
Lock the Feed
the table e2-1
table
Rotate on 1/4 Unlock the

2 eex-1 table
13 e2-9 12 e2-6 11 e2-3 10 Smothly to

Piece is the backe3-7
Stop the table
placed on
Processing
eex-1 the table is NOT over
Unclamp
the piece e1-3 16 Unlock the
eex-3
11 table
eex-2 Stop the Processing
Piece e2-3 cutter
Park is over
removed
from the 17 the spindle e4-3
table 15 8 14 9
e3-8 eex-4
Fig. 15. Machine behaviour as described in the language of directive specifications, with a
"Cutter halt" operation moved to the large loop
Indeed, if we make quite a transparent substitution of CFM operations in the transition
graph of H specification and properly apply the functions of outputs (to be shown here
below), we shall get a controlling finite state machine. This machine, provided inputs are
independent (this being an indispensable condition for conventional logical control
according to the "black box" scheme), will precisely perform the operation sequences
specified. Note that substitutions can be made for each of three specifications and, thus,
three different controlling machines will be obtained. Later on, it will be possible to carry
out arbitrarily profound optimization applying all the methods used in the finite machine
theory and logical synthesis. However, at the attempt to unite a control object and
G = G 1 , G 2 ,..., G n machines, obtained as per specifications presented in Fig. 15, 16, the
errors, mentioned here before, will reveal themselves in blocking (non-fulfilment) of some
commands and a "hanging" – an unforeseen cyclic operation interruption will occur. At the
same time, with DES theory analytic methods applied, possible blocking situation will be
revealed analytically. It is evident that once DES theory methods are applied, a "dimension
damnation" will manifest itself: CFM parallel composition of the example in question
already gives a machine with the number of states equal to 4356 and its composition with H
machine results in the machine with dozens of thousands states.
So, we face the following problem: how to predict blocking situation without composition of
Gi in G followed by general composition with K. To tackle this problem, let’s continue
considering the theory of SDES-modelling.
4. Features of the models of G components and H specification

We would like to point out a number of important features of the models of
G = G 1 , G 2 ,..., G n components and specifications of industrial objects. Model components,
as a rule, simulate the behaviour of different actuators able to "perceive" events-commands,
www.intechopen.com
react to them by the change in the position (location, speed, pressure, level, temperature,
flow rate etc), with a set of space co-ordinates being split up into a number of intervals and
presented by events. Since space, though presented by a set of events remains physical, the
events in it may "happen" in a certain order.
Feature of expected events (F1). For the events e ∈ Ewi of one component, there exists
ordering based on consecution of ei1,ei2,…,ein such that in any chain of these events on graph,
the events are arranged in direct or reverse order (this also refers to ei1 and ein).
Furthermore, this relation is also valid for neighbouring graph chains.
Feature of operations (F2). The events e ∈ Ewi weigh on Gi graph the chains of transactions –
transitions (edges and states), with one edge and state, weighed by e ∈ Eci (event-command),
adjoining to this chain on the left side, and on the right side, either an edge and state, also
weighed by another command e ∈ Eci , or a fork with events e ∈ Euc i
. This feature allows to
unambiguously mark out process operations - the substrings relevant to the command and
the reaction expected, on Gi graph (i.e. to "colour" graph). Then, uncoloured will be left only
i
the edges corresponding to e ∈ Euc .
1
Example 1. G operations (Fig. 8) are as follows: To clench piece: states 1→4, chain –
e1 − 1 ,e1 − 5 , e1 − 2 ; to unclench piece: states 4→1, chain – e1 − 4 ,e1 − 5 , e1 − 6 .
Example 2. G3 operations (Fig. 9): Quick feed to the left: states: 1→4, chain - e3 − 1 ,e3-2 , e3 − 3 ;
operational feed to the left: states: 4→7, chain - e3 − 4 ,e3-5 , e3 − 6 ; slow retraction to the right:
states: 7→4, chain - e3 − 7 ,e3-5 , e3 − 3 ; spindle parking: states: 4→1, chain - e3 − 8 ,e3-2 , e3 − 9 .
Feature of forks separability in G and H (F3). Any fork in the transition graph (both for Gi
and H) is weighed by the events from {Eic ∪Eiuc} in a separate way, i.e., branching is always
either on e ∈ Euc i
or on e ∈ Eci : ∀j : [ Γ ij ≥ 2] ⇒ [∀e ∈ Γ ij : e ∈ Euc
i
|e ∈ Eci ] .
Forks in transition graphs are limited, as a rule, to provide one-to-one description and
implementation. For example, mixed branching (Fig. 16) is difficult to interpret. Since a
transition, particularly in object, has some delay, then, when analysing qi state (Fig. 16), it is
expedient to introduce a new fact – an event euc , negating the initial event euc , and to
transform the initial specification in corresponding transitions as shown in Fig. 16.
Fig. 16. Transformation of mixed branching

Feature (F4) of marked states q j ∈ Qmh for H. All edges leading to q j ∈ Qmh are weighed by
el ∈ Ec . This is a feature of terminated fragments: specific action is performed last.
Feature of uniqueness in use of operations in H (F5). Every edge of H graph can be
associated with one chain from Gi graph. Should in the description be any ambiguity, it can
be easily eliminated by duplicating the corresponding fragment of H graph. For actuators,
all operations of which are associated with different commands, this feature is always valid.
If there are still same commands executed at different "path" sections (a fragment of
sequence from Ew ), they can be always described in different fragments of Н graph by
duplicating the initial paths.
www.intechopen.com
The features presented are applied to choose the principle of role structuring, as a basis of
two-level SDES, and are used in ℜ(s ) algorithm of carrying out the experiment (refer to i.
5.1), actually, replacing the operation of component machine composition.
5. SDES study
It is natural to inquire, what properties the behaviour specification in language K should
possess to provide a supervisor which ensures behaviour G=<G1,G2,…,Gn> according to the
specification, and at the same time is admissible for G. The answer to the question is
associated with controllability study (in terms of Ramadge and Wonham, 1987)) of the
language K which is a specification of the required behaviour of G defined by a set of
components <G1,G2,…,Gn>. As the basic method to study joint G and H behaviour, it is
proposed to experiment with <G1,G2,…,Gn> by strings s∈K (such, that h(q0, s)!, i.e.
admissible for the initial state q0). The experiment point is to simulate operation of
component machines driven by events-commands from strings s∈K. The algorithm of such
experiment is given here below.
5.1 Algorithm ℜ(s ) of the experiment with SDES of G = G 1 , G 2 ,..., G n by string s

The assignments and functions used in ℜ(s ) algorithm are as follows: (s) – length of string
s; s(i) – current event in string s, with i being the symbol number in string s; θ(e)={c | uc | w}
– function of event type; N ( e ) ∈ {0,1,..., n} – number of component Gk in set
G=<G1, G2,…, Gn> such that e∈Ek. If N(e)=0, then e∈Euc refers to common uncontrollable
events of G. VIS = qr1,1 , qr2,2 ,..., qrn, n is called a n-dimension vector of initiated states of each
of G=<G1,G2,…,Gn> components.
As you see, the algorithm executes the experiment on <G1, G2,…, Gn> collection which is set
componentwise to the initial states by strings s∈K.
The algorithm has two kind of results:
1. Logical. ℜ(s ) = True|False. If, under the consecutive influence of symbols s, all
components fulfil their transitions successfully, ℜ(s ) = True. If in the course of the
experiment for a certain symbol s(i) a component k (k= v(s(i))) fails to fulfil its
transitions, i.e. s(i) ∉ Γ kj , then ℜ(s ) = False (refer to step 2, table 1). The latter means that
H is inconsistent with G=<G1,G2,…,Gn>.
2. Constructive. If ℜ(s ) =True, ℜO(s ) := v ; ℜr (s ) := r ; ℜg(s ) := k ; ℜb(s ) := j ; ℜe(s ) := l ,
where (if k≠0) k is a component number of Gk, j and l are numbers of its states at which
the experiment ends successfully by string s, with j being the beginning and l - the end
of a substring v corresponding to the last operation of component Gk as a reaction to
string s, and r is a resulting string. If k=0, then the experiment is successful, but v, j and l
point to the operation of the last component involved in the experiment. In any case,
r ∈ L(G ) is a string admissible in G and is one of s prototypes, i.e. r ∈ PE−d1 (s ) . (It should
be reminded that r ∈ PE−d1 (s ) is a string r, with its projection upon events Ew being equal
to s).
The examples of experiments on specifications:
1. For the experiment, let’s choose a string s1 covering the beginning and a small loop:
states 1, 2, …, 13, 5 (Fig. 10). Here is the string:
www.intechopen.com
Step Operation Comment

1 Set i=1; r:= ; VIS = q01 , q02 ,..., q0n Initial setting
2 k=N(s(i)); If k=0, then r:=rs(i); go to item 5,
else choose from VIS the state of steps 2-5: moving along string s
component k – q kj ;
Current event of string is not
If s(i)∉ Γ kj , then ℜ(s ) =False, go to item 6; admissible for Gk; the experiment
failed
Uncontrollable event injected into
Let δ k (q kj , s(i)) = qlk ; If θ (s( i )) = uc , then
output string. set command from s(i)
r:=rs(i), go to step 4, otherwise v := s(i ) ;
to operation
3 Choose er ∈ Γlk such that { δ k ( qlk ,e r ) = q pk Scrolling by ew – the expected
and l ≠ p}; If θ ( er ) = w , then v:=ver, l:=p events of operation performed as the
go to step 3 reaction to command in Gk
4 place qlk in VIS k-position; r:=rv; v:= ; Current operation is over
5 i=i+1; if i ≤ µ (s ) , then go to step 2; ℜ(s ) := Checking for the string end and
ℜO(s) := v ; ℜr (s) := r ; ℜg(s ) := k ; assigning of output experiment
True;
ℜb(s ) := j ; ℜe(s ) := l ; results
6 The end
Table 2. Algorithm ℜ(s )
s1 = e2 − 1 , e1− 1 , e3 − 1 , e4 − 1 , e3 − 4 , e3 − 7 , e4 − 3 , eex − 3 , e2 − 3 , e2 − 6 , e2 − 9 , e2 − 1 . It is easy to trace that all

transitions in component models will operate since the commands are given correctly. The
experiment result is ℜ (s1) = True. We shell not adduce the resulting string but the last
string operation is offered in full: ℜO(s ) := e2 − 1 , e2 − 5 , e2 − 2 ; ℜg(s1 ) = 2; ℜb(s1 ) = 1; ℜe( s ) = 4 .
String r will include similar extensions for all the events of sting s1.
For the experiment on erroneous graph (Fig. 13), let’s choose string
s2 = e1− 1 , e3 − 1 , e4 − 1 , e3 − 4 , e3 −7 , e4 − 3 , eex − 3 , e2 − 3 . It is easy to notice that G2 is addressed first at the
last event but this attempt fails since the transition from state 1 of G2 component is not
specified for the event e2 − 3 (Fig. 5). Therefore, ℜ (s2)= False.
5.2 Main SDES result

It is natural to ask, what properties a specification of behaviour in language K should
possess to provide a supervisor making G = G 1 , G 2 ,.., G n behave in conformity with
specification and not blocking G.
The answer as to the supervisor existence can be obtained using the following theorem.
Theorem of SDES controllability. Let a well-structured G=<G1,G2,…,Gn> be given, where
E = {Ew ∪ Ec ∪ Euc } , Ed= Ec∪Euc, and K ⊆ Ed* (K≠Ø). Non-blocking supervisor S, such that
PEd (L(S / G )) = K , exists if and only if for any s∈K (such that k(q0, s)!) ℜ (s)=True with
respect to G=<G1,G2,…,Gn>.
In other words, the theorem asserts, that for a well-structured G=<G1,G2,…,Gn> and a given
specification K, there exists a non-blocking supervisor S such, that the projection on Ed of the
www.intechopen.com
language, generated by G under S control, coincides with K provided that for any line s,
specified for the initial state of H, which defines the language of specification K, the
experiment on ℜ (s) algorithm is positive.
Proof The necessity is proved by contradiction: the theorem terms are satisfied, unblocking
supervisor S such that PEd (L(S / G )) = K exists but for a certain string s ∈ K (such that h (q0,
s)!) the experiment ℜ (s) = False with respect to G = G 1 , G 2 ,.., G n . There can be a lot of such
strings, but let s be the shortest of them. Let ω (s ) = e and s := ue. It is evident that for all
prefixes of string u and string u itself, the experiment is positive, i.e. ℜ (u) = True, but, in
case u is extended, ℜ (ue) = False and θ ( e ) = c (e – control event). It is revealed at 2.2 (Table
1). At this, Gk is in state q kj and e ∉ Γ kj (e is inadmissible for Gk in its current state q kj ).
On the other hand, as u, ue ∈ K and supervisor S, such that PEd (L(S / G )) = K , exists, let’s
choose from L(S / G ) strings u', u'e for which PEd ( u′) = u and PEd (u′e ) = ue . Since the sets of
CFM events do not intersect (refer to note 1), N ( e ) = k , from u' admissibility for Gk it follows
that at u' generation Gk will be transferred to state q kj , and then from u'e admissibility for Gk
it immediately follows that e∈ Γ kj (i.e. e is admissible for Gk in its current state q kj ). This
comes into conflict with the assumption that ℜ (s) = False. The necessity is proved.
Sufficiency. Let’s K ≠ Ø be a language such that for any s ∈ K ℜ (s) = True with respect to
G = G 1 , G 2 ,.., G n . We shall show that in such case there exists a supervisor unblocking for
G and providing PE ( L(S / G)) = K .
d
Let’s define language M on K in the following way:
⎧ ⎫
M = ⎨ε ∪ ∪ ℜr(s )⎬
⎩ ⎭
(1)
s∈K
For any u ∈ M let’s define:
⎧ ⎧(θ ( e ) = uc ∧ ℜ( PEd (u)e ) = True ),

⎪ ⎪ ⎫
⎪ ⎪(θ ( e ) = w ∧ e ∈ ℜO( PEd (u)e )), ⎪
S(u) = ⎨ e : ⎨ ⎬
⎪ ⎪(θ ( e ) = c ∧ ℜ( PEd (u)e ) = True ), ⎪
(2)
⎪ ⎩⎪other e not included in codomain S for u ⎭

⎩
The designed converter admits as follows:
- all Euc possible (as to transition function for G) after u (string 1 from (2));
- all Ew if they fall into the definition area of corresponding transition in a certain G k
component (string 2 from (2));
- all controlled events Ec for which the experiment on PEd (u)e is positive (string 3 from (2)).
Thus, the converter is a non-blocking supervisor such that L(S / G ) = M and
PEd (L(S / G )) = K (this follows from M definition (1) and option 3, step 2 of ℜ (s) algorithm
on which ℜr ( s ) is formed). Since finite state machine H (generating К) does not contain
deadlocks and liveloops, then S by construction also cannot contain deadlocks and
liveloops, thus, S is non-blocking. The theorem is proved.
Comments to the theorem. A natural question may arise: how this result is correlated with
the controllability condition by Wonham? First of all, it is quite correlated. If for any s ∈ K
ℜ (s) = True then the language M, built as per the algorithm (refer to (1)), will be
controllable, i.e. for it, a controllability condition by Wonham is satisfied.
www.intechopen.com
The controllability condition derived in the paper is formed with respect to specification
language K outside L(S\G) . Therefore, K is controllable with respect to <G1, G2,…,Gn > if it is
prefix-closed and ℜ experiment is positive on all s ∈ K. This requirement is more strict then
Wonham’s but it relates to the language К that is more expressive then L(S\G). The example
in section 4 illustrates SDES blocking by supervisor (in case the experiment is false). At the
same time, this result and, which is most important, the procedure of its verification
(algorithm ℜ (s)) are pragmatic, i.e. the number of checks cannot exceed the number of
simple paths to every edge of graph H) and the result is given in terms of conditions and
transitions of all the components involved in the experiment.
Our example is illustrated in Table 3, with supervisor S designed as a function of strings as
per the algorithm defined above in the theorem proof.
Table 3. Specifying S(s ) as a function of strings
6. Method of direct supervisor synthesis on the basis of SDES model for real-
time automation systems (RTAS)
The investigations set forth in sections 4, 5 were carried out for SDES with off-line
components that were controlled via blocking mechanism as pee the scheme shown in Fig.
www.intechopen.com
4. At the same time, RTAS has a number of features that are useful to apply for control
modelling and engineering.
• First, RTAS is featured by control subdivision into two sublevels of control: the level of
actuators that executes operations control and a process control level that provides
operation sequences.
• Second, actuators are passive but can receive operative commands, execute them
autonomously and provide feedback.
• Third, while RTAS engineering, a technologist defines specifications (the required
operation sequences), and it is advisable that in a synthesized supervisor, the structure
of sequences was preserved and the synthesis result, as to its complexity, was linearly
dependent on initial specification.
The papers on the synthesis of logical devices (Kuznetsov, 1975; Ambartsumyan, Potekhin
1977) contained similar requirements and synthesis methods were called standard
realization. According to the papers on standard realization methods, such approach has the
following advantages:
• The obtained result is always "recognizable" by the author of initial specifications;
• The result complexity is proportional to the scope of initial data;
• The number of operations in the synthesis procedures is also linearly dependent on
initial data.
Basic paradigm of standard realization is the synthesis of object control system (device) by
syntactic transformation of this object behaviour specification. Therefore, standard
realization is the engineering method that guarantees the engineering result of acceptable
complexity and for acceptable time, provided there is the initial description of the object
behaviour
With the above mentioned RTAS features and standard realization idea taken into account,
the present section pursues the objective to develop a supervisor synthesis method
providing dependability – acceptable complexity of the result (supervisor) achieved for
acceptable time (the number of operations).
This section is dedicated to the study of SDES with passive actuators. In such SDES, all
controlled events are forced from the point of view of operation (Chalmers, Golaszewski,
Ramadge, 1987) and the control is performed as per the scheme similar to that shown in Fig.
17.
Definition 6. A well-structured DES, for which the composition of uncontrollable part is
defined as per Definition 1, all are forced, the required behaviour is defined by the
specification language K ⊆ Ed* ,(K ≠ 0) , and which is provided by supervisor S generating
unambiguously controlled events Ec in such a way that K is fulfilled, will be called a
structured discrete event system with forced controlled events (SDESf).
Comments to the definition. SDESf should meet the condition of determinacy, i.e. for any
string s admissible for the initial state, if its extension by a controlled event is possible, such
extension for this string is unique. It is suggested that a structured DES with forced events
should be realized according to the scheme (Fig. 17) in which supervisor "perceives" all the
events generated by G but initiates only controlled events.
Based on introduced notions, let’s specify the tasks of this section.
For SDESf specified by G component set and K specification, they are as follows:
• Define a condition of K specification controllability.
• Examine the matter of a supervisor existence.
www.intechopen.com
Plant - Generator
1
G 1 Ec
Ew
Euc
2
G 2
Ew
Ec Ec E w Euc E
...
3
Ew
Gn
Supervisor E=Ec Ew Euc
Fig. 17. The scheme of supervisory control for SDES with forced controlled events
• Elaborate the method of specification realizability analysis that will indicate if G and K
are consistent.
• Develop the method of synthesis of supervisor S (if K specification is realizable)
providing control in G in such a way that К is fulfilled, with the method synthesising S
for acceptable time and S complexity having linear dependence on K complexity.
6.1 Study of SDES with forced events

In order to unambiguously define the behaviour of SDES represented by G = G 1 , G 2 ,...G n
collection, it is necessary to specify all the system states and their admissible transitions
(structure and weighing functions). A traditional tool used for such tasks in discrete systems
is the building of attainability tree. In section 5 of the present paper, as a basic instrument to
study SDES behaviour, it is proposed to use the algorithm ℜ that actually is a procedure of
H graph traversal. At this walk, for any reached state of H, is formed VIS = qr1,1 , qr2,2 ,..., qrn, n -
a vector of initiated states of each of G=<G1,G2,…,Gn> components, which appears sufficient
to build a tree of attainability. It would be logical to use intermediate results of algorithm ℜ
at supervisor synthesis. The way to this is set forth below. Important is the fact that the
synthesis task can be divided into two main subtasks: the analysis of H graph structure and
the analysis of complete states.
Structure analysis For further study, we shall need to examine states qi in which the
selection (fork) in the transition graph, defining machine H, takes place. Without the loss of
generality, we assume that there are no mixed forks in the transition graph of H. In other
words, if more than one edge originates from qi, these edges are always weighed either only
by uncontrollable events or, on the contrary, only by those controllable (feature F4 – forks
separability feature worded in section 4).
The last condition in the defining set definition ⎡⎣ω ( s ) = τ ( qi → q k ) ⎤⎦ states the fact of string s
termination on ( qi → q k ) transition.
Definition 7. Let τ ( qi → q k ) be a function with its value equal to the event weighing the
transition ( qi → qk ) , and ω(s) – the last event in string s. Then
Φ ik = {s [ δ h (q0 , s )!] ∧ [ω (s ) = τ ( qi → q k )]} will be called a defining set of strings of k direction
in fork qi .
www.intechopen.com
The last condition in the above definition ω (s ) = τ ( qi → q k ) states the fact of the string s end
on ( qi → q k ) transition.
Definition 8. Let Oi be a set of subscripts of states qj to which there is a direct transition from
qi ; let θ ( e ) = {c uc w} be a function of event type; let qi ∈ Q H ; ⎣⎢Oi ⎦⎥ ≥ 2 . Then qi is called a
correct selection (fork), if only one of the following conditions is fulfilled:
• For all the transitions incidental to qi, θ (τ ( qi → q k )) = uc is fulfilled (the selection on
uncontrollable events);
• For all the transitions incidental to qi θ (τ ( qi → q k )) = c is fulfilled (the selection on
∀m , n : [m , n ∈ Oi ⇒ Φ m
controllable events), with the defining sets on any pair of directions not intersecting, i.e.
i
i ∩ Φ n = ∅] .
The answer to the question, as to the existence of supervisor for SDES with forced controlled
events, is given by the following theorem.
Theorem of SDESf controllability. Let a well-structured G = G 1 , G 2 ,.., G n be given, for
which E = {Ew ∪ Ec ∪ Euc } , all Ec are forced Ed = {Ec ∪ Euc } and K ⊆ Ed* ,(K ≠ ∅ ) . Non-
blocking supervisor S such that PEuc (L( s / G )) = K } exists then and only then when for any
s ∈ K (such that δ h (q0 , s )! ) ℜ(s) = True as respects G = G 1 , G 2 ,.., G n and all selections in
the transition graph of H are correct.
Theorem proof Necessity. The first part of condition: ℜ(s) = True is valid as shown in the
proof of the theorem of controllability in section 5. Let’s prove the necessity and sufficiency
of the second condition: branching correctness. Proof is made by contradiction. Let there
exist an unblocking supervisor S but for H the condition of branching correctness is not met.
Then two options are possible, namely:
• Branching is mixed, i.e. for the transitions incident to qi θ [τ ( qi → q k )] = [c ∧ uc ] is
fulfilled. It is impossible as conflicting branching limitation.
• For qi branching, the condition of empty intersection of defining sets is not fulfilled. Let
si transfer H to qi state, than there exists at least one event e i + 1c that simultaneously
weighs two different edges originating from qi. On the other hand, since a supervisor
exists, it is determinate and is defined as a function of strings, thus, for this option, two
different values e i + 1c and [ e i + 1c ] , weighing the next pair of edges, must fit the same
argument si. We have arrived at a violation.
The necessity is proved.
Sufficiency is proved constructively. Let K ≠ ∅ be a language such that for any s ∈ K
ℜ(s) = True with respect to G = G 1 , G 2 ,.., G n . We shall show that in such case there exists a
supervisor unblocking for G and providing PEd (L(S / G )) = K .
Let’s define language M on K in the following way:
⎧ ⎫
M = ⎨ε ∪ ∪ ℜr (s )⎬
⎩ ⎭
(3)
s∈K
Any string u ∈ M is admissible for L(G ) as per construction in ℜ . For this reason,
M ⊆ L(G ) .
For any u ∈ M let’s define:
www.intechopen.com
⎧
⎪
e : θ ( e ) = c ∧ ℜ( PEd (u)e ) = True
S(u) = ⎨
⎪other e not included in codomain S for u
e : θ ( e ) ≠ c ∧ ℜ( PEd (u)e ) = True (4)
⎩
The designed converter admits the following:
• all Ec possible (as to transition function for G) after u (string 1 from (4));
• instead of any event of e ∈ {Ew ∪ Euc } , if this event enters the definition area of
corresponding transition in H or a certain component of G k (string 2 from (4)).
Since machine H (generating K) does not contain deadlocks and liveloops, L(S / G ) = PE−d1 (K )
by designing, and all s ∈ PE−l 1 (K ) are admissible for G (as ℜ=True), than S by construction also
cannot contain deadlocks and liveloops, thus, S is non-blocking. The theorem is proved.
Comments to the theorem.
1. The proposed condition of controllability is formulated with respect to specification
language К that is more expressive and compact then L(S\G), at least, because it is a
projection of L(S/G) on Ed.
2. For controllability, besides the requirement of positive experiment, it is necessary that
all forks in the transition graph of H should be correct, which is effectively verified by
the graph nodes review.
3. The condition of controllability for SDESf is worded as a limitation imposed only on
specification language K but does not restrict language L(S/G). Nevertheless, K is
controllable as relates G = G 1 , G 2 ,.., G n , provided it is prefix-closed, algorithm ℜ=True
for all s ∈ K , and all forks are correct. This condition is more strict than that in paper
(Chalmers, Golaszewski, Ramadge, 1987) as it admits branching on controllable events,
in case the selection is correct.
Let’s consider possible branching variants on e ∈ Ec - controlled events. Practically, the
following situations are possible:
• Logical substantiation for choosing the continuation is in the pre-history.
• There is no logical substantiation in the past (the defining sets for both directions
intersect but, at this, sequences are admissible for both branches.
This situation will be illustrated by the structure of transition graph shown in Fig. 18.
Semantics of events, states and sequences will be described later in section 6.4. Herein we
shall discuss a few peculiarities of forks in the transition graph. The edges of forks
originating in states q1,q4 and q22, in Fig. 18, are outlined by firm ellipses. The events: eex-1 – a
round piece or eex-2 – a hexahedral piece, took place in the first outlined fragment, but in the
situation of the following firm ellipses, there are no longer such events and a clamp choice
should be made from memory of those events.
Another branching variant is referred to in the description of cutter-type choice – in Fig. 18,
corresponding forks are marked by dashed ellipses. From the point of view of event
sequence, both variants are admissible for G = G 1 , G 2 ,.., G n and there is no data to choose
the variant of the process continuation. In principle, the second situation, in conformity with
the theorem condition, testifies that K is not coordinated with G = G 1 , G 2 ,.., G n and SDES
is not controllable with К. However, for practical tasks, such situation is settled by the
addressing of algorithm to the external, as relates to given SDES, system (e.g. to operator).
Thus, when analysing forks (selection) of H graph on controllable events, two aspects,
important for supervisor S engineering, were revealed. First, for every branching on
www.intechopen.com
⎧→ q i '
⎪
controllable events qi ⎨→ ... , it is a unique extension (sq' i ≠ sq' n ) that corresponds to string
⎪→ q '
⎩ i
sq' , and this provides determinacy of S. Second, the events defining the condition of
i
selection (direction) either happened in the past or lie outside SDES structure.
eex-5 e5-1
eex-2 3 e2-1 6 e3-1 e6-1

eex-5 e5-1 e3-1
e2-1 e6-1
1
16 4 7 8 e3-4
eex-1 e1-1 e4-1 e3-4
eex-1 e2-1 e1-1 e3-1 e2-1 e4-1
e6-1 10
2 5
e4-1 e3-4
17 9
eex-2
e5-3 eex-2 e1-3 e2-9 e2-1 e2-11 eex-3 e6-3 e3-7
e2-1 eex-3
e3-7
24 23 19 16 14 12
eex-3 e6-3
e5-3 e 11
e3-8 1-3
e3-8 e2-9 e2-6 e2-11 e2-3 e2-3 eex-4 e4-3
eex-4 e4-3
22 e2-3 21 18 e2-6 15 20 13
eex-4
e3-8
Fig. 18. Graph of H specifications

In any case, as of the moment of the analysis of situation with branching, the events,
conditioning it, either happened, and the selection should be requested as a new event, or
the selection is impossible to define by prehistory and then it should be requested from
external sources. For providing such request, let’s design machines of special kind: selection
Aqi=〈Qa,Ec,δa,Γia,λja,q0〉 kind is built on controllable variables. A transition graph Aqi is shown

agents. For all states of branching origin qi, an individual agent – a machine of
in Fig.19.
qk Ф ik Фij qI qj
λa(qk)=k eCSi qI qj eCSi λa(qj)=j
Fig. 19. Machine-agent of selection

The discipline of interconnection of supervisor S with branching machine-agent is as
follows: on S reaching state qi, after which it is necessary to make choice (arrive at a
decision), S, through its special output, sends a request to Aqi . In accordance with the
discipline accepted, Aqi replies issuing a direction index aqi = { k j} , (not necessarily 1 of 2,
possibly 1 of many). At this, ai will be used as index. Fig. 20 illustrates a control scheme.
Analysis of complete states To answer the question about the consistency of supervisor S
and object G, SDES state analysis is required. The state of the set of primary components
VIS = q 1j 1 , q 2j 1 ,..., q rjn reflects the state of control object G = G 1 , G 2 ,.., G n . The current state
www.intechopen.com
External External
query eex feedback eex
Aq
H
S
G1 G2 G... Gn
Fig. 20. Control scheme with selection agent

of H machine - qrh reflects the intentions to control (to limit free behaviour of G) for the
purpose of solving some technological tasks. Thus, a complete SDES state is
VFS = qrh , q 1j 1 ,..., q rjm . A conventional technique of listing the complete attainable states in
similar situations is evident: it is necessary to build a tree of attainability and, guided by the
tree, to carry out the experiment on H machine and a set of G 1 , G 2 ,.., G n component
machines. In the paper, it is proposed to analyze complete states by their "projection" on Н
transition graph, in parallel with the experiment on controllability.
For this purpose, let’s set H and all the component machines to their initial states. We shall
get a vector of initial states. Let’s weigh H initial state by this vector. Then we shall make an
experiment with a string composed of events Ed (let it be eci ), weighing H graph edges, and
of the events corresponding to the reaction from one of the components G 1 , G 2 ,.., G n of a
definite component machine Gi. At this, H will transit to another state qrh , with the
following states adjoined: first, all the states of component machines G 1 , G 2 ,.., G n that are
unchanged at this transition; second, by turns, all Gi component machine states
corresponding to its reaction to eci . These vectors form a block of states that weighs qrh
states. The experiment fragment is shown in Table 4. From the fragment of the table of
complete states, it is clear that one H node is associated with blocks of complete states and
this results not only from including states and components "put into action" but also from
the availability of multiversion attainment of the given state. For example, in state 10, there
are 4 blocks corresponding to different state combinations for components G1 and G2. It is
important that all the states inside blocks should successfully operate on subsets of events
admissible in this state.
Reasoning from the above, the requirements to the method of complete states design and
analysis are set forth as follows:
1. Complete states are formed as they are required in H and operations are processed in G;
2. The method should provide for the structure of data on H complete states and
transitions. This structure should provide easy access to complete states at H transition
graph traversal; the advancement and distribution of complete states along H must be
accompanied by H and G consistency analysis.
The major novelty of this idea is that a set of complete states VFS moves and spreads along
H graph structure in compliance with the flow of possible events. It is important to note the
following:
www.intechopen.com
qh G1 G2 G3 G4 G5 G6
1 1 1 1 1 1 1
1 1-4 1 1 1 1
4
1 1-4 1 1 1 1
4 4 4-7 1 1 4
4 4 4-7 4 1 1
10
1 4 4-7 1 4 4
1 4 4-7 4 4 1
Table 4. Full states (fragment)
1. If, first, the experiment was made for every complete state in every node and it, at least
once, was positive, and, second, all H edges were walked through, than the set of all
attainable complete states was obtained. This is equivalent to the building of the tree of
attainability.
2. If all transitions came into action, than the specification defined by H is controllable and
the data acquired is sufficient to form S basis.
It is suggested that a supervisor should be synthesised via weighing H graph edges by new
operations and output functions λs (qi , ti , j ) = { ec , ε } defined on pairs of states and transitions.
Output functions will "return" either controllable event ec or empty symbol ε and they
must keep their value over the whole transition qi → q j .
6.2 Synthesis of supervisor for SDESf

Supervisor engineering will be made based on FSM of special type.
Definition 9. A machine active at transitions (TAM) is a finite machine
S = (Qs , E, δ s , ρ s , λs , Γ s , Γτ , Q,s , q0 ) in which the set of states Q s , set of events E, transition
functions δ s , functions of admissible events Γ s a set of necessarily attainable states Qms and
the initial state q0 are defined conventionally. Control functions Γτ , ρ s , λs of TAM are
defined in the following manner:
• Γti : Q s → {ti , j { j : δ s (qi , Γ i )} = qi } is a function of possible transactions (transitions). This
function associates every edge originating from qi with string ti , j ∈ E* which, when
performed, initiates in S a transition qi → q j ; thus, at S operation, both states and
transitions are active.
• ρs (qi ) = eq i Moor-type function of outputs defined only at forks on controllable events
( Oi ≥ 2; θ (qi ) = c ); for the rest of qi , ρ s (qi ) = ε .
• λs (qi , ti , j ) = { ec ε } - a machine output function, defined on a pair [state– string of
transition], equal either to controlling event or to empty symbol and keeps its acquired
value over the whole transition qi → q j
Apparently, a machine, active at transitions (TAM), has a number of destinations:
• "Language generator" L(S). Let’s present any path incident to the initial state q0 in S, as
iteration of concatenations of pairs [state–outgoing edge] expressed as r0, l = •li = 0 [qi ri , j ] ,
than L(S ) = {s : s = •li = 0 [ qi := λ s (qi , ri , j • ti , j )]} is a set of strings obtained from the set of
paths by substitutions of corresponding events and strings.
www.intechopen.com
• "Direction pointer" – Milly-type output function successfully relevant to the definition

of operator able to admit or turn down controllable events. However, it is very
⎪⎧ ε for q0
important that the function is defined on states and substrings S(s0, l ) = ⎨ s
⎪⎩λ (qi , ti , l )
.
• "Internal interconnections". Provides interconnection with selection machines-agents

that interface SDESf with external media - a supplier of incontrollable events of euc type.
From the point of view of the theory of finite machines, S is a machine of mixed– type: it has
output ρs (qi ) = eq i depending only on states, which is typical for Moor machine, as well as
output λs (qi , ti , j ) = { ec ε } typical for Milly machines; furthermore, the latter is defined at
transition.
Supervisor S is designed, basing on specification H, by the special algorithm of syntactical
transformation ℑ(G , H ) → S . Machines-agents are supposed already defined, so, the
transformation is made according to the following scheme:
H = ( Q h , Ed , δ h , Γh , Qmh , q0 )
↓ ↓ ↓ ↓ ↓ ↓
S = ( Qs , E, δ s , ρs , λs Γ s , Γt , Qms , q0 )
Thus, supervisor synthesis method must solve the following tasks:
• form the vectors of complete states for every state Qh;
• put in action all the transitions (edges) of H;
• should the experiment be positive, to build for S main missing constructions ρs , λs , Γτ .
en . . e3 e 2 e 1
2 4
2
2
VIS c
q0 c
q4 c
q0 q1 c c c
q43 q4
q0 3 5 c
3 q3
q4q1 5
c c K
q1 q3
q1 q3
1
qq2
3 G
3
Command from specification is sent to CFM
q2 1 1
2 Putting in action its internal states, the CFM issues the
q2 G 1 2 events of uc type
G1 3
After performing the command, CFM current states
form a vector of states
4 Words weigh specification edges
Every new vector of states is "pushed" through the rest

5 part of specification
Fig. 21. Data scheme of algorithm ℑ
www.intechopen.com
The proposed algorithm ℑ which fulfils the method of direct supervisor engineering, for the
purpose of clearness, will be depicted graphically as data structure and the algorithm
diagram (Fig. 21). Algorithm ℑ processes specification K, defined by a transition graph,
carries out the simulation of control command operation (arrows 1 in Fig. 21) in component
machines Gi. The component machines "put into action" the transitions, in conformity with
the command. As the result, algorithm ℑ forms:
• word ec ew 1 ...ew n representing operation fulfilled (arrows 2 in Fig. 21);
• next complete state VIS := q 1j 1 , q 2j 1 ,..., q rjm (arrows 3 in Fig. 21).

Specification graph is processed as shown in Fig. 22. Block inscriptions correspond to
algorithm steps. We would like to draw your attention to the fact that the algorithm is
constructed as the traversal (block 4) of graph with unprocessed complete states (the first
unprocessed complete state ‹qh0,q10,q20…qn0› is created at initialization – block 1). For every
Х Х
To process the next edge
7
To take e from the edge

8
To send a command to 1
To initiate components, corresponding CFM
2
vector of states,
no specification 9 3
1
To get a feedback (word)
Are there nodes with no 10

unprocessed complete 2 no
states?
To renew vector of states
yes
11
yes
Was there at least one
cursal edge?
3
Is there the same vector in the
set of vectors of parent-node? 12
To process next nodeAre all nodes

passed through? no
4
To weigh a descendant-node by the
vector of complete states
yes yes 13
Have all vertexes
been processed ?
5
To weigh a parent-node by ρ
14
To mark the edge as passed
To weigh a node by the last word
6 and λ
15
Fig. 22. Block-diagram of algorithm ℑ
www.intechopen.com
state, a loop on outgoing edges (block 5) is formed. For every edge, the weighing by
functions ρ, and operation ec ew 1 ...ew n takes place (blocks 14, 15) once. On termination of
the traversal of all H nodes and edges, we get the set of all admissible states. If all the
transitions took place, than H is controllable (ticked-off output) and the acquired data define
the weighing of machine - supervisor S. If graph traversal is terminated ahead of schedule
(the output marked with "!" symbol) than specification K and object G=<G1, G2, …, Gn> are
incompatible.
6.3 Study of algorithm ℑ

The purpose of the method study is to assess its complexity and time characteristics. The
main question is whether the main features of standard realisation methods, are preserved,
namely: linear dependence of the result complexity and spent time on the initial data scope.
Partially, the answer to this question may be given by the following theorem:
Theorem on standard realization. Given: G = G 1 , G 2 ,.., G n , language K specified by
machine H = (Qh , Ed , δ h , Γ h , Qmh , q0 ) for which the set of transition graph edges is designated
by R. Than algorithm ℑ, based on G = G 1 , G 2 ,.., G n and H = (Qh , Ed , δ h , Γ h , Qmh , q0 )
specification, constructs (as per the scheme of transition-active machine) supervisor S such
that project PEd (L(S / G )) = K and, at this, S complexity on the data scope
≤ O( MAX( Q h , R )) , and the number of operations to design S ≤ O( R )
Proof. The fact that the first theorem part is true follows from the method, though the proof
by induction on string length, can be easily developed. The validity of the second part – the
complexity of S presentation (specification), namely: ≤ O( MAX( Q h , R )) 1 follows from the
fact that all S constructions are obtained by redefining H constructions and by including
new constructions (block 6 of the algorithm) associated with states and edges of H graph
transitions and limited by the data sets from G and H. From the above, it follows that the
complexity of S ≤ O( MAX( Q h , R )) . The validity of the last theorem statement on the
blocks from 2 to 6 on all states ∑ 1 Oi equals the number of H edges - |R|. Which required.
number of operations follows from the fact that the total number of the executed algorithm
n
Thus, for the case, when the complete states one by one take their places in H states, the
linearity of dependence of the number of operations on the number of edges, is maintained
but in more complicated cases, the occurrence of "additional" complete states in H weighing
structure results in the iteration of transitions analysis.
6.4 Example 2
The proposed method will be illustrated by the example of supervisor synthesis for a
milling machine with 6 mechanisms: a clamp for round pieces (1), turntable (2), spindle (3),
rectangular cutter (4), angle clamp (5), round cutter (6). Kinematics of this machine is similar
to that of the machine from section 3.3 (Fig. 4) but in the considered machine, the processing
of 2 types of pieces with different fastening and by different tools (mechanisms G5 and G6)
are foreseen. Each mechanism is simulated by corresponding CFM {Gn}. The nodes
1 The formula runs as follows: S has the order of magnitude maximal of two values: the
cardinal number of the set of states or the cardinal number of the set of edges.
www.intechopen.com
correspond to the space position and edges – to events. Event semantics is presented in
Appendix 1.
Respective CFM is shown in Fig. 23, by colour, in transitions, are marked the events of ec
type, ticked-off is the initial state q0.
e1-4
1 e 1 -1 2 e1-5 3 e1-2 4 e1-3 5 e 1 -5 6
G 1- round clutch
2 e2-5 3 e2-2 4 e 2 -3 5 e2-5 6
e 2 -1 e2-4
1 e 2 -6 7 e2-7 8 e2-8 9 e 2 -9 10
e2-10
e 2 -1 1 11 e2-12 12
e2-13
G2- turntable
1 e 3 -1 2 e3-2 3 e3-3 4 e3-3 9 e3-5 8
e3-9 e 3 -8 e 3 -4 e 3 -7
11 e3-2 10 5 e3-5 6 e3-6 7
G3- spindle
e4-4
1 e 4 -1 2 e4-5 3 e4-2 4 e 4 -3 5 e4-5 6
G 4- cutter
e5-4
1 e 5 -1 2 e5-5 3 e5-2 4 e 5 -3 5 e5-5 6
G 5 – corner clutch
e6-4
1 e 6 -1 2 e6-6 3 e6-2 4 e 6 -3 5 e6-6 6
G 6 – round cutter
Fig. 23. Component finite machines G1, G2, …, G6
www.intechopen.com
The required machine behaviour is informally presented by text specification in Appendix 2.

This behaviour is formalized by finite machine H = (Qh , Ed , δ h , Γ h , Qmh , q0 ) , with a graph of
transitions shown in Fig. 18, section 6.1. The states of machine Q h = { q1 ,..., q23 } correspond to
the steps of processing and the edges – to the operations of component machines.
The proposed method was applied to analyse G=<G1, G2, …, Gn> object and H specification.
The experiment ℜ showed the consistency of H and G. Then, the graph of supervisor (Fig.
24) was obtained. Entering a node, the edges relevant to the same operation have common
marking (e.g. the edges entering node 7). In the supervisor graph, every edge is weighed by
Gi CFM component operation and supplemented by the events of relevant reaction. Events
eex are incontrollable, the edges with these nodes do not change. The comparison of two
graphs reveals their structure identity. This illustrates that the complexity of supervisor
designed by the proposed method, linearly depends on the complexity of initial
specification.
e6-1
3 e5-1 6 8
e6-6
eex-5 e2-1 e5-5 e3-1 e67-8- 2 e3-4
e2-5 e5- 2 e3-2 e3-5
1
16 4 7 10
e3-4
2- 2 e1-1 e6-7
3 -3 e4-1 e8-10
3- 6
eex-1 e1-5 e7-94-5 9
2 e1-2 5 e4-2 e3-7
e2-1 e3-5
eex-2
24-1 e2-9 17
e2-5 e3-3
18 e2-10 19 e19-17
2 -2 e6-3
e2-11 12 e6-6
24 23 e2-6 eex-3
e2-12 12-14
e6- 4
e2-7 16 e2-3 14 11
e1-3 e2-13 e4-3
e5-3 e2- 8 e2-5
e5-5 e1-5 15 e2 - 4 13 e4-5
e1-4 eex-4 e4-4
e5 - 4 12-20
e3-8 20
e2-3 e3-2
22 e2-5 21
e3-9
e2- 4
Fig. 24. Supervisor graph
7. Conclusions
SDES model proposed herein, does not use component composition in the explicit form but
operates G 1 , G 2 ,..., G n set and K⊆Ed* specification language specified by
H = (Q , Ed , δ h , Γ , Qm , q0 ) machine (recall that Ed = Ec ∪ Euc is a language over a set of
h h h
commands and conditions). Such approach to the description model is more economical,
than that in L(G) language and is much more expressive than the one based on parallel
n
composition ⊕ 1G i = G i ⊕ ... ⊕ G n and K⊆L(G).
Thus, the proposed SDES model and the procedure of its operation take maximum account
of the SDES (real-time automation system) peculiarities mentioned in the introductory part.
There is a ground to believe that thereby it will be possible to avoid the «explosion of states»
at supervisor synthesis. The proved theorem of controllability for SDES builds a theoretical
www.intechopen.com
basis for further studies and a base for programming and experiments on the stream of real
tasks.
Thus, (turning back to the problem stated in Introduction) it can be declared that herein is
developed a theoretical basis for a new technique of machine control engineering that
excludes ambiguity and mistakes in the initial specification of a control object as a "black
box".
As the result of research pursued, the conditions of SDES and SDESf controllability were
formulated, the matter of supervisor existence was studied, the method of specification
realizability verification was shown.
The condition of controllability was worded with respect to specification language K that is
more expressive and compact (being a project of L(S/G)) than language L(S/G) traditionally
used in the models with parallel composition.
The paper suggests the structure of supervisory control, contains the study of the method of
supervisor S synthesis based on the object model and specification (G and K). It also
illustrates a linear dependence of supervisor S complexity on the number of edges of H
machine.
At the same time, the number of synthesis operations (time complexity) remains linear only
for the specification in which complete states, one at a time, are disposed on H, i.e. the
number of operations for the designing of S ≤ O( R ) . Generally, the appearance of "second"
complete states in the structure of H weighing, results in the repeated analysis of transitions
and the linearity is violated. However, practically, for real tasks, this phenomena, reflecting
a designer’s aspiration to specify commands sent to aggregates (actuators) more
economically, does not lead to a considerable growth of the number of operations.
8. References
Ambartsumyan A.A. (2009) Supervisory Control of Dynamic Discrete-Event Systems.
Automation and Remote Control, No.8 (August 2009), pp. 156-166
Ambartsumyan, A.A., Kazansky, D.L. (2008). Complex automation of technological
processes with the event model involved. Procedings of 17th IFAC World Congress
Ambartsumyan A.A., Bronishtov S.A. (2006) Event models of process control directed at the
protection against personnel’s erroneous actions. «Greenwich LTD », Moscow
Ambartsumyan A.A., Potehin A.I. (2003) Development of control mechanisms of objects
with continuous technology on base of channel event models. Automation and
Remote Control, No 4 (April 2003)
Ambartsumyan, A.A., Prangishvili, I.V., Poletykin, A.G. (2003). Power plants: the analysis of
state and the proposal of enhance automation. Problems of control, No.2 (March
2003)
Ambartsumyan A.A., Potekhin A.I. (1977) Standard realization of asynchronous machine
Automation and Remote Control; No.5, (May 1977) pp.67-83
Cassandras C.G., Lafortune S. (2008) Introduction to discrete event systems « Springer
Science+Business Media, LLC ». USA.
Chalmers Golaszewski C. H., Ramadge P. J. (1987) Control of discrete-event processes with
forced events. Proc. 28th Conf. Decision Control. p. 247-251, Los Angeles
www.intechopen.com
De Queiroz, M.H., Cury, J.E.R. (2000) Modular supervisory control of large scale discrete-
event systems. Discrete Event Systems: Analysis and Control. Proc. WODES'00, pp.
103-110.
Gaudin, B., Marchand, H. (2003) Modular supervisory control of asynchronous and
hierarchical finite state machines. Proc. ECC, pp. 13-25
Gilard Langer. HoMuCS (1999) A methodology and architecture for Holonic Multicell Control
Systems Preface. KPB -2-99; ISBN 87-90855-00.
Jo Wyns. (1999) Reference architecture for Holonic Manufacturing Systems - the key to support
evolution and reconfiguration. ISBN 90-5682-164-4, K.U.Leuven
Kuznetsov O.P. (1975) Logical machine graphs and their transformations Automation and
Remote Control, No. 9, (September 1975) pp.149-158
Ramadge J. G., Wonham W. M. (1989) The control of discrete-event systems IEEE Trans.
Automat. Control.. 77(1). p. 81-98.
Ramadge J. G., Wonham W. M. (1987) Supervisory control of a class of discrete event
processes SIAM J. Control Optimization. 25(1). p. 206-230.
Sujeet Chand. (2005) From electric motors to flexible manufacturing: control technology
drives industrial automation. IFAC
Van Brussel H., Valckenaers P., Bongaerts L.,Peeters P. (1998) Reference Architecture for
Holonic Manufacturing Systems: PROSA. Computers in Industry. p.37- 47.
Yoo, T.-S., Lafortune, S A (2002) General architecture for decentralized supervisory control
of discrete-event systems. Discrete Event Dynamic Systems: Theory & Applications –
No. 12(3), pp. 335-377
www.intechopen.com
Appendices
Appendix 1
event operation event operation

e1-1 To clench e3-8 Feed ->>
e1-2 Clamp is closed e3-9 Parked
e1-3 To unclench e4-1 To switch on
e1-4 Clamp is open e4-2 Working
e1-5 clamp is moving e4-3 To switch off
e2-1 To fix e4-4 Stopped
e2-2 Table is fixed e4-5 Unstable rotation
e2-3 To unfix e5-1 To clench
e2-4 Table is unfixed e5-2 clamp is closed
e2-5 Locker is moving e5-3 To unclench
e2-6 To make a ¼ turn e5-4 Clamp is open
e2-7 Table is moving e5-5 Clamp is moving
e2-8 Table is turned through e6-1 To switch on
¼
e2-9 To switch off turning e6-2 Working
mechanism
e2-10 Table is stopped e6-3 To switch off
e2-11 To tilt plane e6-4 Stopped
e2-12 Is tilting e6-5 Unstable rotation
e2-13 Angle is achieved eex-1 Piece is on the table (round)
e3-1 <<- to feed eex-2 Piece is removed from the table
e3-2 Feed zone eex-3 Processing is not finished
e3-3 Operating position eex-4 Processing is over
e3-4 <- to feed eex-5 Piece is on the table (hexahedral)
e3-5 Operational zone eex-6 To choose usual cutter
e3-6 Operation is over eex-7 To choose round cutter
e3-7 Feed ->
Table of system events
www.intechopen.com
Appendix 2
1 On arrival, a piece is clenched by a clamp

After clamp operated, spindle is transferred from parking to working position
2
(to the left)
3 After spindle shifted to working position, cutter is turned on
After cutter is turned on, a smooth feed to the left utmost position takes place
4
(end of operation)
5 After operation is over, spindle is fed back to the right up to working position
6 After spindle occupied working position, a positioner turns the table through ¼.
After the table is fixed, the next operation is executed until the processing is
7
over
After the round is made, a spindle parks, a clamp is unclenched, a signal of
8
piece readiness is sent
9 Before parking, to switch off cutter, to wait until it stops
10 Operator chooses a cutter type and a cutting angle
Table of text specifications
www.intechopen.com
Process Management
Edited by Maria Pomffyova
ISBN 978-953-307-085-8
Hard cover, 338 pages
Publisher InTech
Published online 01, April, 2010
Published in print edition April, 2010
The content of the book has been structured into four technical research sections with total of 18 chapters
written by well recognized researchers worldwide. These sections are: 1. process and performance
management and their measurement methods, 2. management of manufacturing processes with the aim to be
quickly adaptable after real situation demands and their control, 3. quality management information and
communication systems, their integration and risk management, 4. management processes of healthcare and
water, construction and demolition waste problems and integration of environmental processes into
management decisions.
How to reference
In order to correctly reference this scholarly work, feel free to copy and paste the following:
Alexander A. Ambartsumyan (2010). Supervisory Control of Industrial Processes, Process Management, Maria
Pomffyova (Ed.), ISBN: 978-953-307-085-8, InTech, Available from:
http://www.intechopen.com/books/process-management/supervisory-control-of-industrial-processes
InTech Europe InTech China

University Campus STeP Ri Unit 405, Office Block, Hotel Equatorial Shanghai
Slavka Krautzeka 83/A No.65, Yan An Road (West), Shanghai, 200040, China
51000 Rijeka, Croatia
Phone: +385 (51) 770 447 Phone: +86-21-62489820
Fax: +385 (51) 686 166 Fax: +86-21-62489821
www.intechopen.com
3 AUDIOBOOK COLLECTIONS
6 BOOK COLLECTIONS

Supervisory Control of Industrial Processes

Uploaded by

Copyright:

Available Formats

Supervisory Control of Industrial Processes

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

Supervisory Control of Industrial Processes

Uploaded by

Copyright:

Available Formats

5

Supervisory Control of Industrial Processes

2. Basic concepts and definitions

G1×G2 = (Q1×Q2, E1 ∩ E2, 1,2, Γ1×2, Qm1×Qm2, (q0 := q01, 02))

Fig. 2. The scheme of object – supervisor interaction

Fig. 3. Interrelationship of supervisory control components in DES

Ec supervizor E=Ec Euc

Fig. 4. Control scheme proposed in the paper (Ramadge, Wonham 1987)

supervizor E=Ec Euc

Fig. 5. Control scheme for DES with forced controllable events

3. Structured Discrete Event Systems (SDES)

3.2 SDES definition

Fig. 6. Types of fragments in the machine Н

Definition 5. A well-structured DES, for which the uncontrollable part is set up by

3.3 Technical object modelling by structured DES

Detail is on the table

Fig. 7. Kinematical model of the machine

Fig. 8. G1 CFM – a model of "Workpiece clutch" mechanism

Start(Stopped and fixed)

e2-10 Locker moving fixed

Fig. 9. G2 CFM – a model of "Turntable" mechanism

Fig. 10. G3 CFM – a model of "Spindle" mechanism

Fig. 11. G4 CFM – a model of "Cutter" mechanism

done by forming the behaviour of G as an uncontrollable system, as a whole, followed by

1) on arrival, the piece is locked by clutch; 6) positioner makes a ¼ table rotation;

Table 1. Text description of initial specification

e3-1 e4-1 e3-4 e3-7

e3-1 e4-1 e3-4 e3-7

Fig. 14. H specification with erroneous missing of "Table locking" operation

3 e1-1 4 e3-1 5 e4-1

Rotate on 1/4 Unlock the

13 e2-9 12 e2-6 11 e2-3 10 Smothly to

4. Features of the models of G components and H specification

Fig. 16. Transformation of mixed branching

5.1 Algorithm ℜ(s ) of the experiment with SDES of G = G 1 , G 2 ,..., G n by string s

Step Operation Comment

s1 = e2 − 1 , e1− 1 , e3 − 1 , e4 − 1 , e3 − 4 , e3 − 7 , e4 − 3 , eex − 3 , e2 − 3 , e2 − 6 , e2 − 9 , e2 − 1 . It is easy to trace that all

5.2 Main SDES result

For any u ∈ M let’s define:

⎧ ⎧(θ ( e ) = uc ∧ ℜ( PEd (u)e ) = True ),

⎪ ⎩⎪other e not included in codomain S for u ⎭

Table 3. Specifying S(s ) as a function of strings

Supervisor E=Ec Ew Euc

6.1 Study of SDES with forced events

eex-2 3 e2-1 6 e3-1 e6-1

Fig. 18. Graph of H specifications

Aqi=〈Qa,Ec,δa,Γia,λja,q0〉 kind is built on controllable variables. A transition graph Aqi is shown

Fig. 19. Machine-agent of selection

Fig. 20. Control scheme with selection agent

6.2 Synthesis of supervisor for SDESf

• "Direction pointer" – Milly-type output function successfully relevant to the definition

• "Internal interconnections". Provides interconnection with selection machines-agents

4 Words weigh specification edges

Every new vector of states is "pushed" through the rest

Fig. 21. Data scheme of algorithm ℑ

• next complete state VIS := q 1j 1 , q 2j 1 ,..., q rjm (arrows 3 in Fig. 21).

To take e from the edge

Are there nodes with no 10

To process next nodeAre all nodes