You are on page 1of 6

A Deadlock Prevention Policy for FMS Based on Time Constraints

Jinwei Guo Jingyang Fan


School of Electro-Mechanical Engineering School of Electro-Mechanical Engineering
Xidian University Xidian University
No.2 South Taibai Road, Xi’an 710071, China No.2 South Taibai Road, Xi’an 710071, China
Email: jw guo525@126.com Email: fan edu@126.com

Abstract— To solve the problem of deadlock for flexible the two approaches aforementioned, the deadlock prevention
manufacturing systems (FMS), an effective deadlock prevention strategies require no run-time cost, since problems are solved
policy based on time constraints is presented in this paper. in the system design and planning stages either by effective
The proposed approach is for a special class of Petri nets,
named S3 PR. Reachability analysis is a fundamental method system design or by using an off-line mechanism to control
applied widely into analyzing Petri nets model. The deadlock- the requests for resources to ensure that deadlocks never
freedom is achieved by preventing the system from reaching the occur. Because of the advantage of the deadlock prevention
bad, deadlock markings and local isolated loops (LIL). By time over the deadlock detection and recovery and deadlock
constraints, the firing sequences of transitions are restricted avoidance, many scholars have made a great deal of research
to avoid the system entering the forbidden states. Compared
with the traditional deadlock control policies, the controlled on the deadlock prevention and obtained many achievements
net by the presented method has simpler structure. Finally, an over the last two decades. A comprehensive survey and
example is used to demonstrate the applications of the proposed comparison of Petri net-based deadlock prevention policies
approach. for FMS can be found in [10]. From the recent literatures, in
KeyWords-Petri nets; deadlock prevention; flexible man- summary, the main three approaches to deadlock prevention
ufacturing system; time constraints are siphon-based [5] [11], reachability graph-based [12] [13]
and hybrid approaches which combine the theory of regions
I. I NTRODUCTION and siphon-based control methods [9].
The flexible manufacturing systems (FMS), consisting of
the uniform information system, material storage and trans-
portation system and a set of numerical control processing In this paper, we propose a novel deadlock prevention
equipments, is a kind of automated manufacturing systems. strategy based on time constraints for FMS. First of all, for
The introduction of FMS is intended to tackle the balance the given Petri net model of an FMS which exists deadlocks,
problems between the efficiency and the flexibility of highly its reachability graph is computed. All the reachable states
automated manufacturing systems, especially, in the small are classified into dead, bad, dangerous and good markings.
batch manufacturing. However, in the practical producing The bad markings are defined as the states that inevitably
activities, the existences of insufficient system resources, enter into the deadlock or local isolated loops (LIL). By
inappropriate processing sequences and improper resource contrast, from a dangerous marking, the system can possibly
allocations can lead to the competition for resources and enter into dead markings. This study aims to control the firing
result in deadlock which is an undesirable situation. sequence by adding proper time constraints to the transitions.
Petri nets, a graphical and mathematical modeling tool, With the time constraints, the firing priorities of enabled tran-
have been widely used to model flexible manufacturing sitions can be controlled to ensure the deadlocks never occur.
systems. In general, within the Petri net formalism, there The introducing of time Petri nets [14] makes the control
are mainly three methods to handle deadlocks in FMS: strategy here different from traditional reachability graph-
deadlock detection and recovery [1] [2], deadlock avoid- based methods. The controlled net has simpler structure.
ance [3] [4] and deadlock prevention [5] [6] [7] [8] [9].
The deadlock detection and recovery approach is to cope
with the system when a deadlock occurs. It permits the The rest of the study is organized as follows. Section II
occurrence of deadlocks. An effective recovery mechanism reviews the basics of Petri nets (PN) and Time Petri nets
can be initialized to tackle the undesirable situation once a (TPN), including the definitions and properties. A subclass
deadlock is detected. The deadlock avoidance approach is of Petri nets, named S3 PR, is also introduced in this section.
an on-line control strategy aiming to avoid the occurrence Section III provides the control strategy in detail. A flexible
of undesirable states by imposing some constraints on the manufacturing example is given in Section IV. Section V
system before the forthcoming deadlock. Compared with presents the conclusions and discussions.
____________________________________
978-1-4799-- /14/$31.00 ©2014 IEEE



Authorized licensed use limited to: Santa Clara University. Downloaded on October 19,2022 at 08:23:53 UTC from IEEE Xplore. Restrictions apply.
II. P RELIMINARIES C. S 3 P R
A. Basics of Petri nets The proposed supervisor design approach targets at sys-
A Petri net [8] is a four-tuple N = (P, T, F, W ) where P tems that can be modeled by some classes of ordinary Petri
and T are finite and nonempty sets. P is a set of places and nets. This subsection introduces Systems of Simple Sequen-
T is a set of transitions with P ∪ T = ∅ and P ∩ T = ∅. tial Processes with Resources (S3 PR). It is an important
F ⊆ (P × T ) ∪ (T × P ) is called a flow relation of the net, subclass of Petri nets proposed by Ezpeleta et al.[16] that
represented by arcs with arrows from places to transitions or can model a typical class of resource allocation systems
from transitions to places. W : (P × T ) ∪ (T × P ) → N is a and draws general interests from academic researchers and
mapping that assigns a weight to an arc: W (f ) > 0 if f ∈ F industrial practitioners. It is defined as follows [8].
and W (f ) = 0 otherwise, where N = {0, 1, 2, . . .} is the Definition 1: A simple sequential process (S2 P) is a Petri
set of nonnegative integers. N = (P, T, F, W ) is called an net N = (PA ∪ {p0 }, T, F ) where (1) PA = ∅ is called the
ordinary net, denoted as N = (P, T, F ), if ∀f ∈ F, W (f ) = set of operation places; (2) p0 ∈ PA is called the idle process
1. If ∀t ∈ T, |• t| = |t• | = 1, the ordinary Petri net N = place; (3) N is a strongly connected state machine; (4) every
(P, T, F ) is called a state machine. circuit of N contains place p0 .
A marking M of N is a mapping from P to N. (N, M0 ) Definition 2: A simple sequential process with resources
is called a net system or marked net. For economy of space, (S2 PR) is a Petri net N = ({p0 } ∪ PA ∪ PR , T, F ) such that:

M (p)p is used to denote vector M . The preset of a 1. The subnet generated by X = PA ∪ {p0 } ∪ T is an S2 P.
p∈P
node x ∈ P ∪ T is defined as • x = {y ∈ P ∪ T |(y, x) ∈ F } 2. PR = ∅ and (PA ∪ {p0 }) ∩ PR = ∅.
and its postset is defined as x• = {y ∈ P ∪ T |(x, y) ∈ F }. 3. ∀p ∈ PA , ∀t ∈ • p, ∀t ∈ p• , ∃rp ∈ PR ,• t ∩ PR =
•
A transition t ∈ T is enabled at a marking M if ∀p ∈ • t, t ∩ PR = {rp }.
M (p) ≥W (p, t). This fact is denoted as M [t . Firing t yields 4. The following statements are verified: (a) ∀r ∈ PR ,
••
a new marking M  such that ∀p ∈ P , M  (p) = M (p) − r ∩ PA = r•• ∩ PA = ∅ and (b) ∀r ∈ PR ,• r ∩ r• = ∅.
W (p, t) + W (t, p), as denoted by M [t M  . M  is called 5. •• (p0 ) ∩ PR = (p0 )•• ∩ PR = ∅.
a reachable marking from M . Marking M  is said to be Definition 3: A system of S2 PR, called S3 PR for short, is
reachable from M if there exists a sequence of transitions defined recursively as follows:
σ = t0 t1 . . . tn and markings M1 , M2 , . . ., and Mn such that 1. An S2 PR is an S3 PR.
M [t0 M1 [t1 M2 . . . Mn [tn M  holds. The set of markings 2. Let Ni = (PAi ∪ {Pi0 } ∪ PRi , Ti , Fi ), i ∈ {1, 2}, be
reachable from M in N is called the reachability set of Petri two S3 PR such that (PA1 ∪ {P10 }) ∩ (PA2 ∪ {P20 }) = ∅,
net (N, M ) and denoted by R(N, M ). A transition t ∈ T is PR1 ∩ PR2 = PC = ∅, and T1 ∩ T2 = ∅. Then, the net
live at M0 if ∀M ∈ R(N, M0 ), ∃M  ∈ R(N, M ), M  [t . N = (PA ∪P 0 ∪PR , T, F ) resulting from the composition of
N is dead at M if  ∃t ∈ T , M [t . (N, M0 ) is deadlock- N1 and N2 via PC defined as follows: (1) PA = PA1 ∪ PA2 ,
free if ∀M ∈ R(N, M0 ), ∃t ∈ T , M [t . (N, M0 ) is live (2) P 0 = {P10 }∪{P20 }, (3) PR = PR1 ∪PR2 , (4) T = T1 ∪T2 ,
if ∀t ∈ T , t is live at M0 . (N, M0 ) is bounded if ∃k ∈ N, and (5) F = F1 ∪ F2 is also an S3 PR.
∀M ∈ R(N, M0 ), ∀p ∈ P , M (p) ≤ k. ∀p ∈ P 0 , t ∈ p• is called a source transition. Unless
otherwise stated, only the S3 PR net which has an unique
B. TPN source transition for each process is considered in this work.
Time Petri nets [15] are classic Petri nets where each tran- Some researchers also defined this kind of Petri nets as
sition t is associated with a time interval [Ef t(t), Lf t(t)]. LS3 PR (linear S3 PR).
When t becomes enabled, it cannot fire before Ef t(t) time
units have elapsed, and it has to fire no later than Lf t(t) time III. S UPERVISOR D ESIGN
units after being enabled, unless it has meanwhile become The deadlock prevention approach presented here is based
disabled by the firing of another transition. Here Ef t(t) on the theory of regions [17]. It is an important method
and Lf t(t) are relative to the moment when t just became for supervisory control of discrete-event systems. First, the
enabled. The time Ef t(t) is the earliest possible firing time reachability graph of a Petri net system is computed. In the
for t and is called earliest f iring time of t, and Lf t(t) is reachability graph, markings are generally categorized into
the latest possible firing time for t and is called latest f iring four classes in the sense of deadlock control: deadlock, bad,
time of t. The firing of a transition itself does not take up any dangerous, and good markings. A deadlock one is a dead
time. The interval bounds are non-negative rational numbers system state by which no successive marking is followed. A
or ∞ in the case of Lf t(t) = ∞ while the time interval bad marking is defined as the one that inevitably enters into
itself is given in real numbers. Without loss of generality the deadlock or local isolated loops. A dangerous marking
we can require the interval bounds to be natural numbers. is the one that one or some (but not all) of its son nodes
Therefore, we consider interval bounds Ef t(t) and Lf t(t) are either bad or deadlock ones. Good markings are the
of a transition t as non-negative integers including zero such markings except deadlock, bad, and dangerous ones. Good
that Ef t(t)  Lf t(t) or Lf t(t) = ∞. and dangerous markings are included in the maximal strongly



Authorized licensed use limited to: Santa Clara University. Downloaded on October 19,2022 at 08:23:53 UTC from IEEE Xplore. Restrictions apply.
 
transition sequences rather than markings or states. As shown

  in [18] [19], an enabled transition t is said to be schedulable
under marking M if t can fire before any other transitions
 

enabled simultaneously. Generally, the schedulability of an
     
individual transition depends on the time constraints of all
 
enabled transitions under the current marking. Transition
 t is schedulable under the initial marking M0 if Ef t(t)
 
≤ min{Lf t(t ): t is enabled under M0 }. According to
  the statements on schedulable transitions, two definitions are
3
given as follows.
Figure 1. An S PR model. Definition 4: ∀ti , tj ∈ T , ti , tj are enabled simultaneous-
ly, i, j ∈ N, ti is weakly schedulable if Ef t(ti ) ≤ Lf t(tj ),
denoted by S(ti )  S(tj ).
M0

t4
t4 t1 t5 t8
t8 Definition 5: ∀ti , tj ∈ T , ti , tj are enabled simultaneous-
M 3 t2 M 1
t5
M4
t1
M2
t6
M5
ly, i, j ∈ N, ti is strongly schedulable if Lf t(ti ) < Ef t(tj ),
t4 t1 t3
t5
M8
t2 t6
M9
t1
t5 t8
denoted by S(ti ) > S(tj ).
t7
t5 t1 In Figure 3(a), both t1 and t2 in the conflict structure are
M 6 t3 M7 t1 t5 M 10 M 11
t1 t5 t7 weakly schedulable under the current marking, though the
t4 M 12 t2 M 13 M 14 t6 M 15 t8 firing of one transition makes the other disabled under the
M 16 M 17 new marking. In Figure 3(b), t1 is strongly schedulable, t2
t1 t5
M 18 M 19 is not schedulable because t1 fires doubtlessly ahead of t2
good marking
and t2 can never fire.
bad marking
Definition 6: A local isolated loop (LIL) is defined as
dangerous marking ti tj tk tl
follows. Let −→ Mi −→ Mj · · · −→ Mk −→
deadlock marking
Mi be a finite
Figure 2. The reachability graph of Figure 1. directed loop in the reachability graph of a Petri net where
arcs are labeled by transitions in T . The set of markings
among the loop is denoted as M. The loop is called an LIL,
connected components containing the initial marking M0 . if ∀M ∈ M, there does not exist a path from which the
They constitute the legal behavior of a system. initial marking M0 is reachable.
Figure 1 is an S3 PR Petri net model. Its reachability For example, Figure 4 is a part of the reachability graph
graph is shown in Figure 2. A reachability graph presents of a Petri net. In Figure 4, the loop constructed by M4 − M7
the complete dynamic behavior of a Petri net model. The is an LIL. Obviously, the existence of LILs makes a net’s
nodes denote the set of reachable states and the edges are liveness impossible.
labeled with firing transitions. Given a net system (N, M0 ),
all the reachable states can be divided into four types as M3 M0 M4 M5
t4 t6 t7
aforementioned: deadlock, bad, dangerous and good ones.
According to the theory of regions, dangerous and good t3 t1 t10 t8
t2 t9
markings constitute the legal behavior of a system. Hence,
M2 M1 M7 M6
in order to realize the liveness of a system, what one needs
to do is separating the deadlock and bad markings from
the reachability graph by adding time constraints to the Figure 4. A part of the reachability graph of a Petri net
transitions.
The concept of time schedulable is mainly concerned with By the analysis of all the possible conditions, the classes of
structures related to deadlocks and LILs in the reachability
graph are presented in Figure 5 and Figure 6. As known,
  according to the theory of regions, dangerous and good
markings constitute the legal behavior of a system. To obtain
          
 the supervisor of a Petri net, deadlock, bad markings and
    LILs must be forbidden. Based on time constraints, such
purposes can be achieved by setting proper firing intervals
for transitions. According to different conditions, a control
   
policy is developed as follows.
 
Algorithm - Deadlock prevention policy based on time
Figure 3. The time schedulability of transitions. constraints of transitions



Authorized licensed use limited to: Santa Clara University. Downloaded on October 19,2022 at 08:23:53 UTC from IEEE Xplore. Restrictions apply.
        




 
    



 


        
   



   


     

 



     



 
     

  Figure 6. The classes of the structures related to LILs in a


Figure 5. The classes of structures related to deadlocks in a reachability graph.
reachability graph.

end if
if it is covered by subclass II(d) as shown in Figure 6 then
Input: A marked Petri net (N, M0 ) with deadlock states
S(ti ) > S(tj ) ∩ S(ti ) > S(tk ).
Output: The controlled net with time constraints.
end if
Step 1: Compute the reachability graph of the given Petri net, Remove Lq from L.
denoted as RG(N, M0 ). End while
Step 2: Distinguish the sets of good, dangerous, bad, and deadlock Step 6:
markings, denoted by P1 , P2 , P3 , and P4 , respectively. According Add a control place pc with one token. Add arcs from pc to any
to the structural properties of the considered class of Petri nets, source transition t and from t•• to pc with unit weight.
there are three classes of structures related to deadlocks and two The supervisor is obtained.
classes of structures related to LILs. They are also divided into nine
subclasses (a)-(i) totally as shown in Figure 5 and four subclasses By the algorithm above, in Step 4, when the states meet
(a)-(d) that are shown in Figure 6. a dangerous marking whose son nodes may reach a good,
Step 3: In order to make the net live, dead, bad markings and bad, or deadlock one, we can separate the deadlock and
LILs should be removed. Aiming at different conditions, time bad markings by making the transitions, which lead to
constraints are imposed on transitions to make some transitions good markings, to be strongly schedulable. According to
strongly schedulable.
Step 4: the time constraints and scheduling, the undesired states can
While P3 ∪ P4 = ∅ do be forbidden to satisfy the deadlockfree request. For LILs,
Select an element Mq ∈ P3 ∪ P4 , q ∈ {1, 2, · · · , |P3 | + |P4 |} Step 5 can eliminate them by time scheduling. For some
if it is covered by subclass I(a) or I(b) or II(e) or II(f) as shown net models, Step 6 is not necessary. If the controlled net
in Figure 5 then is already live after Step 5, Step 6 should be neglected.
Let S(ti ) > S(tj ).
end if Actually, Step 6 comes at the expense of concurrency.
if it is covered by subclass I(c) or I(d) or II(g) as shown in Theorem 1: For an S3 PR net with an unique source tran-
Figure 5 then sition for each process, the controlled net by proposed Algo-
Keep ti and tj unchanged. rithm is live if its source transitions are weakly schedulable.
end if
if it is covered by subclass III(h) as shown in Figure 5 then IV. A N FMS EXAMPLE
Let S(ti ) > S(tj ) ∩ S(ti ) > S(tk ).
The flexible manufacturing cell shown in Figure 7 has
end if
if it is covered by subclass III(i) as shown in Figure 5 then two machines, two robots, three loading buffers and three
Let S(ti ) > S(tj ) ∪ S(tj ) > S(tk ). unloading buffers. Three part types are produced. The net
end if model is an S3 PR and contains deadlocks. It has 30 reach-
Remove Mq from P3 ∪ P4 able markings including two deadlock markings, three bad
End while markings, ten dangerous markings and an LIL. Their detailed
Step 5:
The set of LILs is denoted as L. classifications can be seen from Table I.
While L = ∅ do Furthermore, it is needed to determine all the possible
Select an element Lq ∈ L, q ∈ {1, 2, · · · , |L|}. conditions to separate the the deadlock, bad markings and
if it is covered by subclass I(a) or I(b) as shown in Figure 6 LIL. As shown in Figure 8, Class I represents the conditions
then which may lead to deadlocks. From Class II, all the bad
Let S(ti ) > S(tj ).
end if markings can be found. Class III describes an LIL generated
if it is covered by subclass II(c) as shown in Figure 6 then from the net model. And it shows how to prevent the system
S(ti ) > S(tk ) ∪ S(tj ) > S(tk ). from entering into the LIL.



Authorized licensed use limited to: Santa Clara University. Downloaded on October 19,2022 at 08:23:53 UTC from IEEE Xplore. Restrictions apply.
t12 No. I M 10 M 12 M 15 M 13
Class

p12 t7 t3 t7 t1 t11 t10 t8


p13 t2

p10 t11
M 17 M 20 M 21 M 17 M 23 M 17 M 21 M 22
t1
p11 M4 M6 M8
No. II
p14 Class
t10 p2
t6 t10 t2 t7 t1 t11 t7

t9 t2
M 10 M 10 M 12 M 13 M 10 M 14 M 15
p9 p3
M1 M5 M2

t8 t3
p15 p1 t10 t2 t6 t3 t6 t1 t11 t6
p4
p8
p6
M4 M5 M6 M 11 M 12 M4 M7 M8
t7 t4
p16 No. III M 11 M 18
Class
p7 P5
t10 t4 t6 t11 t4 t6
t6 t5
M 29 M 26
t11
M 18 M 19 M 20 M 24 M 25 M 26
3
Figure 7. An S PR example.
t12 t10 M 12 M 24

M 20
LIL t3 t7 t12 t4 t6

With the decomposed reachability graph in Figure 8,


the proposed method can be understood intuitively. Some M 20 M 21 M 11 M 28 M 29
transitions should be controlled to avoid the undesired states
by restricting their firing sequences. By the algorithm afore-
mentioned, the firing time constraints of transitions can be good dangerous bad deadlock LIL
determined as follows:
⎧ Figure 8. The reachability analysis of the net system.

⎪ S(t11 ) > S(t1 ).



⎪ S(t8 ) > S(t10 ) ∩ S(t8 ) > S(t2 ).



⎪ S(t 7 ) > S(t10 ) ∩ S(t7 ) > S(t2 ).

⎪ scheduling is realized. Moreover, considering the concurren-

⎪ S(t 11 ) > S(t1 ) ∪ S(t7 ) > S(t1 ).
⎨ cy and synchronization of S3 PR, it should be guaranteed that
S(t3 ) > S(t6 ).
(1) the source transitions of different manufacturing processes

⎪ S(t2 ) > S(t10 ) ∪ S(t6 ) > S(t10 ).

⎪ are weakly schedulable each other. Taking all the factors into

⎪ S(t11 ) > S(t1 ) ∪ S(t6 ) > S(t1 ).

⎪ account, we impose proper time constraints on transitions.

⎪ S(t4 ) > S(t6 ) ∪ S(t10 ) > S(t6 ).

⎪ And a group of the time constraints is determined as follows.

⎪ S(t4 ) > S(t6 ) ∪ S(t11 ) > S(t6 ).
⎩ The final supervisor is shown in Figure 9 and the controlled
S(t4 ) > S(t6 ) ∪ S(t12 ) > S(t6 ).
net is live.
According to the possible solutions to Eq.1, proper time t1 [3, 5] t2 [2, 3] t3 [0, 1] t4 [0, 1]
constraints can be added to the transitions such that time
t5 [0, 0] t6 [3, 5] t7 [0, 1] t8 [0, 1]
t9 [0, 0] t10 [4, 5] t11 [1, 2] t12 [0, 0]
TABLE I: Classifications of the reachable markings

classifications No. of reachability markings


dead M17 , M21 V. C ONCLUSIONS
bad M4 , M10 , M12 This paper focuses on the deadlock prevention for a class
dangerous M1 , M2 , M5 , M6 , M8 , M11 , M13 , of Petri nets. Traditional deadlock prevention methods need
M15 , M18 , M24 adding monitors to the original net system which makes the
good M0 , M3 , M7 , M9 , M14 , M16 , M19 , supervisor more complex. In real-world systems, traditional
M20 , M22 , M23 , M25 − M29 methods may bring exorbitant expenditure. The aim of this
LIL M20 , M26 , M29 research is to design the supervisor with simple structure.
First, the existence of deadlock and bad markings in a Petri



Authorized licensed use limited to: Santa Clara University. Downloaded on October 19,2022 at 08:23:53 UTC from IEEE Xplore. Restrictions apply.
[0, 0] R EFERENCES
t12
p13 [1] R. A. Wysk, N. S. Yang, and S. Joshi, ”Detection of deadlocks in
p12 flexible manufacturing systems”, IEEE Transactions on Robotics and
Automation, vol.7, no.6, pp.853-859, 1991.
p10 [1, 2] t11 p17 [2] M. P. Fanti and M. C. Zhou, ”Deadlock control methods in automated
[3, 5]
manufacturing systems”, IEEE Transations on Systems, Man, and
t1
p11 Cybernetics-Part A, vol.34, no.1, pp.5-22, 2004.
[4, 5] [3] K. Y. Xing, B. S. Hu, and H. X. Chen, ”Deadlock avoidance policy
t10 p2
for Petri-net modelling of flexible manufacturing systems with shared
[0, 0] p14 [2, 3] resources”, IEEE Transations on Automatic Control, vol.42, no.2,
t9 t2
pp.289-295, 1996.
p9 p3 [4] J. Park and S. A. Reveliotis, ”Deadlock avoidance in sequential
[0, 1]
resource allocation systems with multiple resource acquisitions and
[0, 1] t8 t3 flexible routings,” IEEE Transactions on Robotics and Automation
p1
p15 vol.46, no.10, pp.1572-1583, 2001.
p4
p8 [5] Z. W. Li and M. C. Zhou, ”Elementary siphons of Petri nets and their
p6 application to deadlock prevention in flexible manufacturing systems”,
[0, 1]
[0, 1] t7 t4 IEEE Transactions on Systems, Man, and Cybernetics-Part A vol.34,
p16 no.1, pp.38-51, 2004.
p7 P5 [6] M. V. Iordache, J. O. Moody, and P. J. Antsaklis, ”Synthesis of
[3, 5] [0, 0] deadlock prevention supervisors using Petri nets”, IEEE Transactions
t6 t5 on Robotics and Automation, vol.18, no.1, pp.59-68, 2002.
[7] Z. W. Li, H. S. Hu, and A. R. Wang, ”Design of liveness-enforcing
Figure 9. The controlled net for Figure 7. supervisors for flexible manufacturing systems using Petri nets”, IEEE
Transactions on Systems, Man, and Cybernetics-Part C vol.37, no.4,
pp.517-526, 2007.
[8] Z. W. Li and M. C. Zhou, ”Deadlock resolution in automated manu-
net is a crucial reason causing the net to be in a deadlock facturing systems: a novel Petri net approach”, Berlin:Springer, 2009.
[9] Z. W. Li, M. C. Zhou and M. D. Jeng, ”A maximally permissive
state. However, the removing of the deadlock and bad deadlock prevention policy for FMS based on Petri net siphon control
markings can only ensure that the controlled net is deadlock- and the theory of regions”, IEEE Transactions on Automation Science
free and not necessarily live. The existence of LILs makes and Engineering, vol.5, no.1, pp.182-188, 2008.
[10] Z. W. Li and M. C. Zhou, ”A survey and comparison of Petri net-based
the net not live. An algorithm based on time constraints is deadlock prevention policy for flexible manufacturing systems”, IEEE
developed to eliminate the bad, deadlock markings and LILs. Transactions on Systems, Man, and Cybernetics-Part C vol.38, no.2,
Considering the time property of TPN, an additional control pp.172-188, 2008.
[11] Y. S. Huang, M. D. Jeng, X. L. Xie and S. L. Chung, ”Siphon-based
place related to the source transitions is added. After that, deadlock prevention policy for flexible manufacturing systems”, IEEE
the supervisor with simple structure is obtained. Transations on Systems, Man, and Cybernetics-Part A, vol.36, no.6,
However, inevitably, there are also some defects in our pp.2152-2160, 2006.
[12] M. Uzam, ”The use of Petri net reduction approach for an optimal
research. And some problems can be investigated in the deadlock prevention policy for flexible manufacturing systems” In-
future. First, the supervisor by the proposed method may be ternational Journal of Advanced Manufacturing Technology vol.23,
so conservative that some legal behavior is removed. Hence, no.3-4, pp.204-219, 2004.
[13] M. Uzam and M. C. Zhou, ”An improved iterative synthesis method
in the future study, the presented method should be optimized for liveness enforcing supervisors of flexible manufacturing systems”,
and synthesized with other methods, e.g., siphon-based ap- International Journal of Advanced Manufacturing Technology vol.44,
proaches. Second, it is probable that the time constraints by no.10, pp.1987-2030, 2006.
[14] J. W. Guo and Z. W. Li, ”A deadlock prevention approach for a class
the proposed method cannot meet the requirements of the of time Petri nets based on SCG”, Proceeding of the 2010 IEEE
real-world systems. In fact, the feasible solutions of the time Inernational Conference on Mechatronics and Automation, Xi’an,
factors in this paper are not unique, which makes it possible China, pp.1641-1646, 2010.
[15] L. P. Zeugmann, ”Time and Petri nets”, Berlin:Springer, 2013.
to choose proper time factors for practical systems. Third, the [16] J. Esparza and M. Silva, ”A polynomial-time algorithm to decide
developed method may be applied to other subclasses of Petri liveness of bounded free choice nets”, Theoretical Computer Sciences,
nets, such as PPN and RCN-merged nets. In these subclasses, vol.102, no.1, pp.186-205, 1992.
[17] M. Uzam, ”An optimal deadlock prevention policy for flexible man-
the conditions under which the proposed method can be ufacturing systems using Petri net models with resources and the
applied should be further considered. Finally, applications of theory of regions”, International Journal of Advanced Manufacturing
the proposed method to industrial FMS need to be pursued. Technology, vol.19, no.3, pp.192-208, 2002.
[18] D. X. Xu, X. D. He and Y. Deng, ”Compositional schedulability
analysis of real-time systems using time Petri nets”, IEEE Transactions
VI. ACKNOWLEDGMENT on Software Engineering, vol.28, no.10, pp.984-996, 2002
[19] J. C. Wang, Y. Deng and G. Xu, ”Reachability analysis of real-time
This work is supported in part by the Natural Science systems using Time Petri nets”, IEEE Transactions on Systems, Man,
Foundation of China under Grant No. 61403295, the Fun- and Cybernetics-Part B, vol.30, no.5, pp.725-736, 2000.
damental Research Funds for the Central Universities under
Grant No. JB140416 and 7214468806, and Provincial Nat-
ural Science and Fundamental Research Funds of Shannxi
Province, China under Grant No. 2012JM8038.



Authorized licensed use limited to: Santa Clara University. Downloaded on October 19,2022 at 08:23:53 UTC from IEEE Xplore. Restrictions apply.

You might also like