Sunteți pe pagina 1din 6

A Comparative Study of Two Deadlock Avoidance

Controller Synthesis Methods for Assembly Processes *


Fu-Shiung Hsieh
Overseas Chinese Institute of Technology
Taiwan, R.O.C.

shiel2lO@ocit.edu.tw
analyze the control policies generated by applying the two
Abstract
Although there are a few results regarding
sufficient conditions to the same class of assembly processes.
deadlock avoidance problem for non-sequential producrion
Our analysis shows that the synthesis method proposed in [4] is
processer in literature, there is a lack of camparicon for the
less conservative than that proposed in [8]-[9] for assembly
existing results. The goal of this paper is to compare hw
processes.
existing results concerning deadlock awdance in assembly
Organization of the remainder of this paper is as follows.
nianufaccnrring system, This paper considers a subclass of
Section I1 reviews the CAPN model. Section 111 presents
the existing Controlled Assembly Petri Nets (CAPN) with
the liveness condition for CAPN. Section IV compares with
each operation requiring only one unit of arbitrary number of
an existing result. Section V concludes this paper.
Wpes
.. of- resources called WW. Analysis shows that the
svnthesis method .
Drouosed
in our previous work is less
.
conservative than that proposed in an existing result for the
class of CAPNU.
2 CAPNModel

Keywords: Flexible manufacming system, deadlock,


control policy, assembly process.

Let J denote the set of assembly processes in the system.


Let be the set of
types in the system. To capcure
the interactions among resources and jobs in assembly
processes, an operation denoted as
that merges two

11

Introduction

Deadlocks cripple the progress of production activities.


Guarantee of deadlock-free operations is essential for
achieving high resource utilization in flexible
manufacturing systems. Although there are a few results for
non-sequential production processes in literature ([I)-[4], [SI[9]), there is a lack of comparison for the existing results. The
goal of this paper is to compare the results of 181-[9] with those
appeared in [4]. In [8] and [9], the authors defined a Petri net
model for assembly/disassembly processes to be realizable if and
only if there exists a feasible execution sequence. The authors
proved that the computational complexity to determine the
realizability of an assembly process (RAP) is NP-complete. The
authors also proved that to find the a l l y permissive
deadlock avoidance control policy for assembly/disassembly
systems is NP-hard. A sufficient condition (Theorem 1 of [SI)
was proposed to maintain the deadlock free p
r
o
m of this class
of assemhlyidisassembly systems. The sUacient condition is
based on the zoned structure of the production processes and
states that if the resource capacity is no less than the number of
zones involved, there exists a tansformation to maintain the
deadlock k e propem of the given system. In [4], the author
proposed a conlrolled Petri net model called Controlled
Assembly Petri Net (CAPW for a class of assembly processes
and a suboptimal polynomial complexity deadlock avoidance
algorithm based on a sufficient liveness condition for CAPNs.
However, no comparison has been made for [8]-[9] and [4]. In
this paper, we will fist re-examine the sufficient conditions
proposed in [SI-[9] and [4] for the class of CAPNs. Then we

PNs through common places, transitions, or arcs is defined


two
Petri
nets GI =
as
follows.
Given

4,
m20 ), where
we assume that m l o ( p )= mzo( p ) V p E 4 nP 2 .

( P , ,T, ,1 1 , 0 1mlo
, )and Gz = ( P 2 T 2 ,I,,
I

We define GlIIG2 =( P ,T ,I , 0 ,mo ), whereP= P, u p 2 ,

The operationI1

also appeared in [3]-[7]. By extending

the CPPN model proposed in [SI, we propose a CAPN


model
based
on
the
Petri
net G = IIrEnGR, IIjeJGJ, constructed by merging the
resource subnets GR, , r

R , with job subnets GJ, ,

j E J , described as follows. The procedure to construct


CAPN also resembles that of RCN-merged nets ([7]).
Definition 2.1: A job subnet G J j = ( P , , T j , I j , O j , m j o )
is an acyclic marked graph ( M G ) and is of tree structure as
each place has exactly one input and one output transition,

0-7803-7952-7/03/$17.00 0 2003 IEEE.


3458

Authorized licensed use limited to: WASEDA UNIVERSITY. Downloaded on May 8, 2009 at 22:24 from IEEE Xplore. Restrictions apply.

;if;+1

where m j o ( p )= 0 Vp E Pj

That is, there is no job in

process under m j o .
1;

Definition 2.3: A control place p, is a control point to


enable or disable a controlled transition. We use a small
square box to represent a control place. There is a transition
input arc between p e and the corresponding controlled
hansition. A controlled transition is disabled if no token is
placed in the control place preceding it and may he fxed as
many times as the number of tokens in the control place.

P>

1;

T f*

p,

t5

Figure 1 A job subnet


Assume that a unit of resource can only he involved in one
operation at a time. A type- r resource, r E R , may he
involved in NR,. activities, where each activity consists of a
sequence of operations using type- r resources sequentially.
Suppose there are n," distinct operations in the k -th activity
of type- r resources. The set T,"

Figure 2

{ tf(1) , ((2) ,...,

r,"(n:) }of transitions are used to represent the operations


in the k -th activity of type- r resources. The states of the
resource before and after transition t,"(n) are denoted by
places p : ( n - I ) and p : ( n ) , respectively. Letp,(O) be the
idle state place of type- r resources. The Petri net of the k th activity can he represented by a type- r resource activity
circuit

c,k =

P,(O)t;(I)P:(1)t;(2)P:(2)...
p f ( n l -l)t;(n,")p,(O) . To model synchronization of
operations, we assume two distinct resource activity
circuits C> and C? may have multiple common transitions
hut have one and only one common place p r ( 0 ) , the idle
state place. The type- r resource suhnet GR, is GR, =

C: 11 C: /I ... CrRr. Remark that GR, allows modelling of


production activities that cannot be modeled with state
machines.
Definition 2.2: Let GR, =
T,,I,, O r ,m,,) denote the
type- r resource suhnet, where m,,(p,(O)) > 0
and m,o(p) = 0 Vp E P, \ {p,(O)} . We will use

(e,

Po = {p,(O)lr E R} to denote the set of resource idle state


places.
Given a set of job subnets G J j , j

J , and a set of

resource subnets GR, , r E R , we construct a Petri net


model G =

GR,

jeJ

G J j = (P,T,I , 0,mo ) .

Figure 3
Defmition 2 . 4 A CAPN is defined as an eight tuple G,=
(P,Pc,T,,T,,I,O,mo,u)
, abbreviated as Gc(mo,u),
where mo is the initial marking of G , and mo E !Mo(G,),
U,(C,) denotes the set of initial markings of G,, andu is
a control policy defined based on control action ofG, as
follows.
2.5: Mo(G,)
~ { mI m ( p ) = O
Defmition
V p E P-Po and m(p)> 0 V p E P o } , where Po denotes the
set of all idle state places of all types of resources of G, .
We will abhreviateM,(G,) asMOwhenever it is clear from
the context. U,(G,) denotes the set of initial markings
ofGCunderwhich all resources are in idle state.
Delinition 2.6 A conhul action a is a vector in Z'" that
determines how many times each transition in T, may be fired
simdtaneously under a reachable marking m of a CAPN G, .

3459

Authorized licensed use limited to: WASEDA UNIVERSITY. Downloaded on May 8, 2009 at 22:24 from IEEE Xplore. Restrictions apply.

The number of tokens in the control place p , of


transitionfunderaisdenoted asa(p,)ora(t). Atransitionthat
can be fued undera is called an admissible transition. A control
policy U is a mapping that generates a control action a
for G, based on its current marking. That is, U : R(mo)--f 2"' .
G,(mo,u) evolves by firing the admissible transitions under
no = u(mo) . A marking m, is reached and a, = u ( m l ) is
applied nexf etc. Hence a sequence { ai} of control actions will
be generated for G, based on current marking under control

policyu . Let R ( m o , u ) denote the set of all reachable markings


of G,(mo,u) fromm,

Mo(Gc1. Let R(mo) = U R ( m o , u ),

somem E M i ( G , ) ,there does not exist any control policy U '


under which G,(m',u') is live.
Property 3.1: Given a CAPN G, with marking m E N ( m o ) ,
there exists a control policy U such that G,(m,u)is live if
and only if there exists a m* EM,' and a sequence of
control actions that bring m to a marking m' E M O
with m' 2 m* .
As a CAF" G, consists of a set J of processes, an upper
bound of MRRs of G, can be calculated based on a h4RR of
each process in J . To find a MRR of type- j process, every
transition in T j need to be fired. Figure 9 illustrates an

UEU

where U is the set of all control policies of G, .


A necessary condition for the existence of a control policy
to keep a given CAPN G, live is to have sufficient resources
available to fire each transition inGC.We defineMo(G,) as
the set of initial markings of G, with sufficient resources to
keep G,live as follows, where M , ( G , ) c M O .
Definition 2.9: Mo(G,)=( m 1 m E Mo(G,)and there exists
a control policy U under which G,(m,u) is live.}. We will
abbreviate M o ( G , ) as M Owhenever it is clear from the
context.
Given a CAF" G, with marking m E R ( m o ) ,
wberem0 E M o ( G , ) , determine a least restrictive allowed
control action a such that G, can be kept live under the
marking reached after executing a under m . The problem
was proven to be NP-hard in [SI and [9].Therefore, we will
develop a suboptimal algorithm with polynomial
complexity to maintain the liveness of G, .

assembly transitiont of type- jprocess. To fire an assembly


transition requires all its immediate subassembly transitions
to be fired. We present a heuristic algorithm to compute an
upper bound of MRR for an assembly transition based on
the resource requirement to fire each immediate
subassembly transition o f t . We use +t to denote the set of
immediate subassembly transitions precedent t o t . A formal
definition of't is as follows.
Definition 3.3: The set of input places of f , not including
its control place, is denoted as't. The set of output places
o f t is denoted as'f . The set of immediate subassembly

.
transitions precedent tot is denoted as +t = { t ' / t ' * s ' f }For
the type-1 job subnet shown in Figure I,

f 4 = ( f 3 ) and +tS = { t 2 , f 4 } .
Defmition 3.4: Transitions in the subassembly processes of
a tmnsitiont
The set (t) of all transitions in the subassembly processes

rj

oft is the subset of transitions of

3 A Liveness Condition for CAPN

+f2 = ( f l )

--t

T j with each transition

in T j ( t ) connected to t with at least one directed path

Definition 3.1:Let VI E Z N and V2 E Z N . We denote


VI 2 V, ifV,(i) 2 V 2 ( i )for eachi E {1,2,3,.._..
N) . We denote
V, > V2 if VI 2 V2 and there exists at least
N}suchthatV,(i)> V 2 ( i ) .
oneiE{1,2,3 ,_____

r T j ( t ) . Let s(Tj(t))denote the


sequence s on the set T j ( f )of

in GJ, . Remark that +f


projection of firing
transitions.

1:

For

the

CAF"

in

Definition 3.2:Given a CA" G,, M,'(G,) denotes the subset


of initial markings of G, with minimal resources for the
existence of a control policy to keep G, Live. Obviously,

Example

M i ( G , ) c M o ( G , ) . M,'(G,)is abbreviated as M i w h e n it
is clear from the context. The set of resources in idle state

a n d t = t S . Thens(q(fS))= f ; t ; t l f 2 f j t 4 f s .

under m E M,'(G,) can be represented by a vector in ZIRl


called a MRR of G, .
For each m E M i ( G , ) , there exists a control policy U under
which G,(m,u) is live. For any marldng m' , if m' < m for

. Suppose

3,

TI (ts 1=

TI = {f[,tg ,f1,t2,f3~t4.fs.f(l
{t;,t;,r,,t2,t3,t,,t,f

Fig.

s = t;t;tlt2t3t4tst[

Let N , be a vector in ZIRl that denotes the resource


requirement to fire a sequence s of transitions, with N,(r),
r E R , as the number of type- r resources required. We
will let R, , a vector in 21Rl, denote the resources involved in
firingjwtt,withR,(r)asthenumber oftype-r resources
required, where r E R .

3460

Authorized licensed use limited to: WASEDA UNIVERSITY. Downloaded on May 8, 2009 at 22:24 from IEEE Xplore. Restrictions apply.

Definition 3.5: A firing seqwnces is said to be a m i n i i

End For
End For

firing sequence to lire t E T, if (I) S is an enumeration of the

s, t S J

set T, (t) of transitions, (2) t is the last transition ins , and (3)
there does not exist any enumeration

S'

A t A U {t)
End For

of the set T, ( t ) of

Update5 = {tit E T, \Aand't

< N , . A minimal firing sequence to f r e t is


transitions with Ns,

denoted ass; .Remark that s: contains t

The following heuristic Pair-wise Interchange algorithm


finds an upper bound of MRR for GJ with polynomial
complexity. Let s(- t) denote the prefvt of s preceeding
t E sand s(r -) denote the suffut o f s afiert . Lets, denote
the firing sequence constructed by the algorithm to fire t .
Let A be the set of transitions whose firing sequence have
been constructed by the algorithm in some iteration.
Let B be the set of transitions whose firing sequence is to be
constructed in some iteration. The idea of the Pair-wise
Interchange Algorithm is very intuitive. Consider a
tET,
. Let ' t = { f i 2 f 2 , h , . . . J ~ }
.
transition
~~

Some

t'ETj(tk)-{tkl

fuins f' after ti is

fired may require more resources as the


same h e of resources are required byf, andr' . In this case,

it is reasonable to fue t' before ti is fired. Although the


algorithm does not guarantee generation of minimal f i n g
sequence, an upper bound of MRR can be obtained wing
the firing sequence generated by it.
Pair-wise Interchange Algorithm
Step 0: A = LT,
Update5 =

{tit E Tj \ A and't c A

st2 st,

...St,.)

SI, SI,., '.' S I ,

sr,

st,*,"' SIN

CJ andt'@ sl(- t i )

(T,(f')) I Where a b is the


of
removing firing
sequence b from it.
s 2 = s i ( - t i ) s,(T,(t')) ti $,(ti -)

Let

Si

st

t St

+52

St

the Pair-wise Interchange Algorithm to fire the final

we-j Process. An

transition f:Of
for
G,
can

be

N = N , @N, @N ,

calculated

bound Of Mm
as
follows.

e...@ N'IJI .

Example l(Continued): Consider the CAPN GI shown in


Figure 3. The algorithm initializes A with { ! : , t i } .
Fort,,s,, =t;ti,andfortj,s,,=f;t). Fort,,s,~=t;r,t,, and

for

t4

, sq

= t;t3t4

For fs , the above algorithm

s,, = t;tlt2t;t3t4 .
As R((t2 1) n R(Ti(t4)) t.CJ , the firing sequence found to
initializes s , ~ with s , ~ = sl,

fire ts is s , ~= t;t3r4t;tlt2t5. The resource requirement to


firet;r3t4r~tlt2tsis[21 I].

[SI and [9] deal with deadlock avoidance problem in


assembly systems. The papss focus on the problems of process
realizability and of the least restrictive deadlock avoidance
policy. In 181 and 191, the anthors defined a Petri net model for
assembly processes to be realizable if and only ifthere exists an
execution sequence U =mOtlm,t2mzt3..fnmn such that

Fori E {l,2,..,,N)
Fork E (i+ l,i+ 2,,.., A'}
For PE T,(tk) \ {fk}
IfR({fi}) nR({t'))

Go to Step 1
Else
Exit
End If
To calculate an upper bound of MRR for G, with a set J of
processes, we define the following operator.
Definition 3.7: Operator@ takes the larger of two vectors
element by element.
Let s j denote the firing sequence constructed by applying

4. Comparison with An Existing Result

Step 1: For each t E E


Let't = { t i , t z , t 3,...,t N )
s: =st,

If E # CJ Then

Defmition 3.6 Let R(S) denote the set of resources


involved with the set S of transitions. That is,
R(S)={rl R,(r) # 0 , wheret E S andr E R } ,

If R((til)nR({r'))fCJ for

EA

,...,n}=T,whereTdenotesthesetoftransitionsin
{fili~l,Z,,3
the assembly process. The authors also deiined a system is
reversible if and only if for eachm E R(mo) , mois reachable
fromm . In 181 and [9], the authors stated that to distinguish the
least restrictive model to which the deadlock avoidance problem
can be addressed and to develop the least restrictive deadlock
avoidance policy, it's necessary to find algorithms to solve the
following two problems:
(1)RedimbiIity of Assembly Rocess (RAP): Given a
system PT ,is it &ble?

End If
End For

3461

Authorized licensed use limited to: WASEDA UNIVERSITY. Downloaded on May 8, 2009 at 22:24 from IEEE Xplore. Restrictions apply.

(2)Reachability of ma (RIM): Given a realizable system PT ,


suppose m E R(m,) and f is enabled under marking m .
Let m' be the marking reached after firing f under marking m .
Is mo reachable h m m ' 7
The authors proved that Problem RAP is NP-complete
(Theorem 4 of [9]) whereas Problem RIM is NP-hard (Theorem
2 of [9]). Based on the above results, the " a l l y permissive
deadlock avoidance control policy is NP-hard To reduce the
computational complexity, [8] and 191 proposed a sufficient
condition (Theorem 1 of [8]and Theorem 3 of [9]) for a Petri
net model of assembly processes to be realihle. The sufficient
condition proposed in [8] and [9] is as follows.
Theorem 3 of [9]: If c(b) L v(b) for each b E B then PT is
realizable, where B denotes the set of buffers, c(b) denotes the
capacity of buffer b E B and v(b) is a parameter that denotes the
number of zones in the job model involving buffer b E B .
The authors also proposed a sufficient condition (Theorem 2 of
[SI and Theorem 4 of [9]) to guarantee deadlock free property
of the system by making the assembly processes realible and
reversible. Application of the sufficient condition
requires c(b) 2 v(b) for eachb E B .
Theorem 4 of [9]: For each system AP such
that c(b) L v(b) V b E B , the system under the control of
Definition 5.4 in [9] is realizable and reversible.
We compare the result of this paper with the existing result of [8]
and [9]. We show that our result is less conservative than the
sufficient condition proposed in [8] and [9] for the class of
CAPNs as the resources required by DAC algorithm is no more
than those required hy the sufficient condition of [8] and [9] .
We first illwkate this result by an example.
Example 1 (Continued): Consider the example in Figure 2,
which is similar to one of the assembly process in the example
that appeared in [SI (Figure 1) and [9]. For this example, there
are three types of resources and the set Po = ( ps,p,.ps) .
According to the delinition of a zone in [8] and [9], there are
three zones in this example: zi = p I p 2, z =p,p4.2 3 = p s .

As there are two zones ( z i = p 1 p 2andz2 = p 3 p 4 ) requires


bufferbi , v(bl) = 1 . Similarly, v(b2)= 2, and v(b3) = 1 (See
Example 6.2 on page 419 of [SI). The sufficient condition of [9]
requires the capacities of buffers b, ,b, and b3 to be c(b,)2 2 ,
c(b2) 2 2 , andc(b,) 2 1 , respectively. That is, the resource
requirement is [2 2 11. By applying the Pair-wise Interchange
Algorithm, an upper bound of MRR for this example is [2 1 I].
As [2 1 I] 5 [2 2 11. The Pair-wise Interchange Algorithm yields
a less conservative result than Theorem 3 of [9]. To formally
compare our results with Theorem 3 of [9], we consider the
following subclass of CAP".
D e f ~ t i o n4.1: The subclass of CAF'Ns with each operation
requiring only one unit of arbitrary number of types of resources
is denoted as CAPNU.
For the class of CAPNU, although the Pair-wise Interchange
Algorithm does not.guarantee MRR can be found, it always

yields less conservative results than Theorem 3 of 191. That is,


we have the following result.
Theorem 4.1: For CAPNU, the Pair-wise Interchange Algorithm
always h d s a firing sequences whose MRR is no greater than
the resource required by Theorem 1 of [SI. That is,

N,( r )< v ( r )V r E R .
Pro03 We prove by induction on the levell, of an assembly
&ansition t , l't152 , in the directed paths of GJ, ,
where 1, = max len(rr) , Ien(rr) =

TI and let\ t 2}1 .

(1')

We first prove it holds for the case 1, = I . Lett denote the only
transition

in G J j with l'tIL2

, where

tET,

Let

+r = ( t , , f 2 , t3 ,___,
t N} , where N L 2 . Step 1 of the Pair-wise

Interchange
s,

, ...s,,.

s,, sIit,

Algorithm
sets
s,
=
s,,
&*
... s,*-,sIk s,*+<... s,,, . Lets denote the firing

sequence constructed by applying the Pair-wise Interchange


Algorithm. AU we need to prove is that N,( r )< v(r) V r E R .
According to Definition 6.1 of [SI, the job modelGJjcan be
partitioned into a number of zones. For I, =I, there
are N + 1 zones. Let us number the zones as follows. Then -th
zone consists of the set of input places of the transitions ins," ,
where n E (l,2,3,._.,N } . The N + 1 -th zone consists of the input
places of the set T, \ T, (r) of transitions, where T, \ T j ( t ) is
actually the set
preceding t,/ in GJ,

of

transitions succeeding f and


be the fning sequence that

. Let s,-

sequentially fires the transitions between t and t i in GJ, . As


type- r resource involved in q1zones,

V(T)

= q, and Theorem 1

of [8] requires at least q1units of type-r resources. We have to


prove the type- r resource requirement N,, ( r )to fires, is no
greater thanq, , No that for CAPNU, each operation requiring
only one unit of arbitrary number of types. Firings,,'requires at
most one unit of type- r resource. If type-r resource is not held
afters,l requks at most one unit of
after firings,) ,
type- r resource. Otherwise, suppose the type r resource is held
after firings,, , firingsa afters,, requires at most two units of

type- r resources.
Based on similar reasoning, fning
s , ~s, ... st,-, s , requires
~
no greater than q1units
s,,,
of type-r resources. Firings,- after s, quires at most q1units
of type-r resources as all with the exception of at most one unit
wig held) of type-r resources will be released alter s,t is fied.
Therefore it holds forl =I.
To prove by induction, assume that it holds forl, =k. We have to
prove it holds for I, =k+l. Let t be a transition

3462

Authorized licensed use limited to: WASEDA UNIVERSITY. Downloaded on May 8, 2009 at 22:24 from IEEE Xplore. Restrictions apply.

w i t h l t l 2 2 , t ~ T ~ a n 1,d =k+l. Lett=(t,,f,,t, ,...,f N } ,


where N 2 2 . Lets, denote the-firing sequence conshucted by
the Pair-wise Interchange Algorithm. Lets,- be the sequence that
sequentially fires the transitions in GJj , starting with f and
ending
st,

... s,,-,

with t{

,,+,... s,,.]

s,, s

Let

s=

s,

,,+,... sly

s , ~s

s,.
s,-

. AU

st,

sr,

we need to

prove is that N,(r)iv(r) V r c R . As 1, =k for each


f E { t , , t 2 , f 3,..., t N }, according to the inductive assumption,
NS,( r ) 5 v5 ( r ) V r E R . Firing s , ~ requires at most

N,,, ( r ) units of type- r resources. If no type-r resource is held


afier firings,, ,firings,, afters,, requires at mostN,,> (r)unitsof

type- r resources. Otherwise, firing s,, after

s,

requires at

most I Y ~( r, )~+ N $1 ( r ) units of type- I resources. Based on

similar reasoning, firings,(+,


afters,, s, s,, ... s,-,s , ~requires
no greater than Ns,,( r ) + N , ( r ) +...+
( r )units of

*-

r resources. Firing s,- &er


s,
requires at most
Nso ( r ) + N 542 ( r )+...+ N,,, ( r )units of type-r resources as all
with the exception of at most one unit (being held) of typer resources will be released after S J is fired. ?herefore, the
number of type- r resoms requmd to fire s is no greater
than I f s , , @ ) + N,,>(r) +...+ N s , N ( r ) . As N , , ( r ) +
N

N s r 2 ( r ) + , . . + N s , J r )Cv,(r),itholdsforl,=k+l.
5
4

Processes,Proceedings of 2000 IEEE Int. Conj on Sysf.,


Man, Cybern, pp. 3045-3050, Oct. 8-1 1,2000.

[4] F.S. Hsieh, Deadlock Avoidance Algorithms For


Assembly Processes With Unreliable Resources,
Proceedings of the 2001 IEEE International Conference on
System, Man and Cybernetics, Tucson, Arizona, pp. 19601965, Oct. 7-10, 2001.
[5] F.S. Hsieh and S.C. Chang, Dispatching driven
deadlock avoidance controller synthesis for flexible
manufacturing systems, IEEE Trans. Robof. Automat.,
vol.lO, pp. 196-209, 1994.
[6] F.S. Hsieh, Robustness of Deadlock Avoidance
Algorithm for Sequential Processes, To appear on
Proceedings of the 15th IFAC World Congress.
[7] MuDer Jeng and Frank DiCesare,Synthesis Using
Resource Controlled Nets for Modeling Shared Resource
Systems, IEEE Trans. Robot. Automaf., vol.ll, no. 3, pp.
317-327, June. 1995.
[SI E. Roszkowska, CAST-Tools for automatic FMS
dynamics models synthesis, In Compufer Aided .Svstents
Theoiy - EUROCAST91, volume 585 of Lecture Notes on
Computer Science, pp. 412421, Springer-Verlag, 1992.
[9] E. Roszkowska and R. Wojcik, Problems of process
flow feasibility in FAS, In ClA4 in Process and
Manufacmring Industries, pp. 115-120. Oxford: Pergamon
Press. 1993.

QE.D.

5. Conclusion
In this paper, we compare two existing liveness
conditions for the subclass of CAPNs with each operation
requiring only one unit of arbitrary numher of types of resources
called CAPNU. Analysis shows that our condition is less
conservative than the existing result for CAFNU.

References
[ 11 J. Ezpeleta and L. Recalde, A deadlock avoidance
approach for non-sequential resource allocation systems,
Proceedings of the IEEE Int. ConJ on Syst., Man, Cybern.,
Tunisia, Oct. 2002.

[2] M.P. Fanti, B. Maione, and B. Turchiano, Event


Control for Deadlock Avoidance in Assembly Systems,
Proceedings of 1997 IEEE Int. ConJ on Syst., Man,
Cybem., vol. 4, pp. 3756-3761, Oct. 12-15, 1997.

[3] F.S. Hsieh,Reconfigurable Fault Tolerant Deadlock


Avoidance
Controller
Synthesis
for
Assembly

3463

Authorized licensed use limited to: WASEDA UNIVERSITY. Downloaded on May 8, 2009 at 22:24 from IEEE Xplore. Restrictions apply.

S-ar putea să vă placă și