1. Petri nets Classical Petri nets: The basic model

2. Process modeling
Emphasis on dynamic behavior rather than structuring the state space
Transition system is too low level
We start with the classical Petri net
Then we extend it with:
Color
Time
Hierarchy

3. Classical Petri net Simple process model History:
Just three elements: places, transitions and arcs.
Graphical and mathematical description.
Formal semantics and allows for analysis.
History:
Carl Adam Petri (1962, PhD thesis)‏
In sixties and seventies focus mainly on theory.
Since eighties also focus on tools and applications (cf. CPN work by Kurt Jensen).
“Hidden” in many diagramming techniques and systems.

4. Elements

5. Rules Connections are directed.
No connections between two places or two transitions.
Places may hold zero or more tokens.
First, we consider the case of at most one arc between two nodes.

6. Enabled
A transition is enabled if each of its input places contains at least one token.
enabled
Not enabled
Not enabled

7. Firing An enabled transition can fire (i.e., it occurs).
When it fires it consumes a token from each input place and produces a token for each output place.
fired

8. Play “Token Game”
In the new state, make_picture is enabled. It will fire, etc.

9. Remarks Firing is atomic.
Multiple transitions may be enabled, but only one fires at a time, i.e., we assume interleaving semantics (cf. diamond rule).
The number of tokens may vary if there are transitions for which the number of input places is not equal to the number of output places.
The network is static.
The state is represented by the distribution of tokens over places (also referred to as marking).

10. Non-determinism
Two transitions are enabled but only one can fire

11. Example: Single traffic light

12. Two traffic lights OR

13. Problem

14. Solution
How to make them alternate?

15. Playing the “Token Game” on the Internet
Applet to build your own Petri nets and execute them: pn_applet/pn_applet.html
FLASH animations: /flash/

16. Exercise: Train system (1)‏
Consider a circular railroad system with 4 (one- way) tracks (1,2,3,4) and 2 trains (A,B). No two trains should be at the same track at the same time and we do not care about the identities of the two trains.

17. Exercise: Train system (2)‏
Consider a railroad system with 4 tracks (1,2,3,4) and 2 trains (A,B). No two trains should be at the same track at the same time and we want to distinguish the two trains.

18. Exercise: Train system (3)‏
Consider a railroad system with 4 tracks (1,2,3,4) and 2 trains (A,B). No two trains should be at the same track at the same time. Moreover the next track should also be free to allow for a safe distance. (We do not care about train identities.)‏

19. Exercise: Train system (4)‏
Consider a railroad system with 4 tracks (1,2,3,4) and 2 trains. Tracks are free, busy or claimed. Trains need to claim the next track before entering.

20. WARNING It is not sufficient to understand the (process) models
WARNING It is not sufficient to understand the (process) models. You have to be able to design them yourself !

21. Multiple arcs connecting two nodes
The number of arcs between an input place and a transition determines the number of tokens required to be enabled.
The number of arcs determines the number of tokens to be consumed/produced.

22. Example: Ball game

23. Exercise: Manufacturing a chair
Model the manufacturing of a chair from its components: 2 front legs, 2 back legs, 3 cross bars, 1 seat frame, and 1 seat cushion as a Petri net.
Select some sensible assembly order.
Reverse logistics?

24. Exercise: Burning alcohol.
Model C2H5OH + 3 * O2 => 2 * CO2 + 3 * H2O
Assume that there are two steps: first each molecule is disassembled into its atoms and then these atoms are assembled into other molecules.

25. Exercise: Manufacturing a car
Model the production process shown in the Bill- Of-Materials.
car
subassembly2
engine 2
chair
subassembly1 4
chassis
wheel

26. Formal definition
A classical Petri net is a four-tuple (P,T,I,O) where:
P is a finite set of places,
T is a finite set of transitions,
I : P x T -> N is the input function, and
O : T x P -> N is the output function.
Any diagram can be mapped onto such a four tuple and vice versa.

27. Formal definition (2)‏
The state (marking) of a Petri net (P,T,I,O) is defined as follows:
s: P-> N, i.e., a function mapping the set of places onto {0,1,2, … }.

28. Exercise: Map onto (P,T,I,O) and S

29. Exercise: Draw diagram
Petri net (P,T,I,O):
P = {a,b,c,d}
T = {e,f}
I(a,e)=1, I(b,e)=2, I(c,e)=0, I(d,e)=0, I(a,f)=0, I(b,f)=0, I(c,f)=1, I(d,f)=0.
O(e,a)=0, O(e,b)=0, O(e,c)=1, O(e,d)=0, O(f,a)=0, O(f,b)=2, O(f,c)=0, O(f,d)=3.
State s:
s(a)=1, s(b)=2, s(c)=0, s(d) = 0.

30. Enabling formalized
Transition t is enabled in state s1 if and only if:

31. Firing formalized
If transition t is enabled in state s1, it can fire and the resulting state is s2 :

32. Mapping Petri nets onto transition systems
A Petri net (P,T,I,O) defines the following transition system (S,TR):

33. Reachability graph
The reachability graph of a Petri net is the part of the transition system reachable from the initial state in graph-like notation.
The reachability graph can be calculated as follows:
Let X be the set containing just the initial state and let Y be the empty set.
Take an element x of X and add this to Y. Calculate all states reachable for x by firing some enabled transition. Each successor state that is not in Y is added to X.
If X is empty stop, otherwise goto 2.

34. Example (3,2)‏ (3,1)‏ (3,0)‏ (1,3)‏ (1,2)‏ (1,1)‏ (1,0)‏
Nodes in the reachability graph can be represented by a vector “(3,2)” or as “3 red + 2 black”. The latter is useful for “sparse states” (i.e., few places are marked).

35. Exercise: Give the reachability graph using both notations

36. Different types of states
Initial state: Initial distribution of tokens.
Reachable state: Reachable from initial state.
Final state (also referred to as “dead states”): No transition is enabled.
Home state (also referred to as home marking): It is always possible to return (i.e., it is reachable from any reachable state).
How to recognize these states in the reachability graph?

37. Exercise: Producers and consumers
Model a process with one producer and one consumer, both are either busy or free and alternate between these two states. After every production cycle the producer puts a product in a buffer. The consumer consumes one product from this buffer per cycle.
Give the reachability graph and indicate the final states.
How to model 4 producers and 3 consumers connected through a single buffer?
How to limit the size of the buffer to 4?

38. Exercise: Two switches
Consider a room with two switches and one light. The light is on or off. The switches are in state up or down. At any time any of the switches can be used to turn the light on or off.
Model this as a Petri net.
Give the reachability graph.

39. Modeling Place: passive element Transition: active element
Arc: causal relation
Token: elements subject to change
The state (space) of a process/system is modeled by places and tokens and state transitions are modeled by transitions (cf. transition systems).

40. Role of a token Tokens can play the following roles:
a physical object, for example a product, a part, a drug, a person;
an information object, for example a message, a signal, a report;
a collection of objects, for example a truck with products, a warehouse with parts, or an address file;
an indicator of a state, for example the indicator of the state in which a process is, or the state of an object;
an indicator of a condition: the presence of a token indicates whether a certain condition is fulfilled.

41. Role of a place
a type of communication medium, like a telephone line, a middleman, or a communication network;
a buffer: for example, a depot, a queue or a post bin;
a geographical location, like a place in a warehouse, office or hospital;
a possible state or state condition: for example, the floor where an elevator is, or the condition that a specialist is available.

42. Role of a transition
an event: for example, starting an operation, the death of a patient, a change seasons or the switching of a traffic light from red to green;
a transformation of an object, like adapting a product, updating a database, or updating a document;
a transport of an object: for example, transporting goods, or sending a file.

43. Typical network structures
Causality
Parallelism (AND-split - AND-join)‏
Choice (XOR-split – XOR-join)‏
Iteration (XOR-join - XOR-split)‏
Capacity constraints
Feedback loop
Mutual exclusion
Alternating

44. Causality

45. Parallelism

46. Parallelism: AND-split

47. Parallelism: AND-join

48. Choice: XOR-split

49. Choice: XOR-join

50. Iteration: 1 or more times
XOR-join before XOR-split

51. Iteration: 0 or more times
XOR-join before XOR-split

52. Capacity constraints: feedback loop
AND-join before AND-split

53. Capacity constraints: mutual exclusion
AND-join before AND-split

54. Capacity constraints: alternating
AND-join before AND-split

55. We have seen most patterns, e.g.:
Example of mutual exclusion
How to make them alternate?

56. Exercise: Manufacturing a car (2)
Model the production process shown in the Bill-Of-Materials with resources.
Each assembly step requires a dedicated machine and an operator.
There are two operators and one machine of each type.
Hint: model both the start and completion of an assembly step.
car
subassembly2
engine 2
chair
subassembly1 4
chassis
wheel

57. Modeling problem (1): Zero testing
Transition t should fire if place p is empty. ? t p

58. Initially there are N tokens
Solution
Only works if place is N-bounded t
Initially there are N tokens
N input and output arcs p’ p

59. Modeling problem (2): Priority
Transition t1 has priority over t2 t1 ? t2
Hint: similar to Zero testing!

60. A bit of theory
Extensions have been proposed to tackle these problems, e.g., inhibitor arcs.
These extensions extend the modeling power (Turing completeness*).
Without such an extension not Turing complete.
Still certain questions are difficult/expensive to answer or even undecidable (e.g., equivalence of two nets).
* Turing completeness corresponds to the ability to execute any computation.

61. Exercise: Witness statements
As part of the process of handling insurance claims there is the handling of witness statements.
There may be 0-10 witnesses per claim. After an initialization step (one per claim), each of the witnesses is registered, contacted, and informed (i.e., 0-10 per claim in parallel). Only after all witness statements have been processed a report is made (one per claim).
Model this in terms of a Petri net.

62. Exercise: Dining philosophers
5 philosophers sharing 5 chopsticks: chopsticks are located in-between philosophers
A philosopher is either in state eating or thinking and needs two chopsticks to eat.
Model as a Petri net.

63. High level Petri nets Extending classical Petri nets with color, time and hierarchy (informal introduction)‏
Prof.dr.ir. Wil van der Aalst
Eindhoven University of Technology, Faculty of Technology Management,
Department of Information and Technology, P.O.Box 513, NL-5600 MB,
Eindhoven, The Netherlands.

64. Limitations of classical Petri nets
Inability to test for zero tokens in a place.
Models tend to become large.
Models cannot reflect temporal aspects
No support for structuring large models, cf. top- down and bottom-up design

65. Inability to test for zero tokens in a place? t p
“Tricks” only work if p is bounded

66. Models tend to become (too) large
Size linear in the number of products.

67. Models tend to become (too) large (2)‏
Size linear in the number of tracks.

68. Models cannot reflect temporal aspects
Duration of each phase is highly relevant.

69. No support for structuring large models

70. High-level Petri nets To tackle the problems identified.
Petri nets extended with:
Color (i.e., data)‏
Time
Hierarchy
For the time being be do not choose a concrete language but focus on the main concepts.
Later we focus on a concrete language: CPN.
These concepts are supported by many variants of CPN including ExSpect, CPN AMI, etc.

71. Running example: Making punch cards
free desk employees
waiting patients
served patients
patient/ employees

72. Extension with color (1)‏
Tokens have a color (i.e., a data value)‏

73. Extension with color (2)‏
Places are typed (also referred to as color set).
record Brand:string * RegistrationNo:string * Year:int * Color:string * Owner:string

74. Extension with color (3)‏
The relation between production and consumption needs to be specified, i.e., the value of a produced token needs to be related to the values of consumed tokens.
The value of the token produced for place sum is the sum of the values of the consumed tokens.

75. Running example: Tokens are colored

76. Running example: Places are typed

77. Running example: Initial state
start is enabled

78. Running example: Transition start fired
New value is created by simply merging the two records.
stop is enabled

79. Running example: Transition stop fired
New values are created by simply spliting the record into two parts.

80. The number of tokens produced is no longer fixed (1)‏
Note that the network structure is no longer a complete specification!

81. The number of tokens produced is no longer fixed (2)‏
The number of tokens produced for each output place is between 0 and 3 and the sum should be 3.

82. Example
Model as a colored Petri net.

83. Product and quantity are in the value of the token
The entire stock is represented by the value of a single token, i.e., a list of records.

84. Types StockItem Stock color Product = string; color Number = int;
color StockItem = record prod:Product * num:Number;
color Stock = list StockItem;
StockItem

85. Extension with time (1)‏
Each token has a timestamp.
The timestamp specifies the earliest time when it can be consumed.

86. Extension with time (2)‏
The enabling time of a transition is the maximum of the tokens to be consumed.
If there are multiple tokens in a place, the earliest ones are consumed first.
A transition with the smallest firing time will fire first.
Transitions are eager, i.e., they fire as soon as they can.
Produced token may have a delay.
The timestamp of a produced token is the firing time plus its delay.

87. Running example: Enabling time
Transition start is enabled at time 2 = max{0,min{2,4,4}}.

88. Running example: Delays
Tokens for place busy get a delay of 3
@+3 = firing time plus 3 time units

89. Running example: Transition start fired
Transition start fired a time 2.
Continue to play (timed) token game…

90. Exercise: Final state?

91. Exercise: Final state?

92. Extension with hierarchy
Timed and colored Petri nets result in more compact models.
However, for complex systems/processes the model does not fit on a single page.
Moreover, putting things at the same level does not reflect the structure of the process/system.
Many hierarchy concepts are possible. In this course we restrict ourselves to transition refinement.

93. Instead of

94. We can use hierarchy

95. Reuse Reuse saves design efforts.
Hierarchy can have any number of levels
Transition refinement can be used for top-down and bottom-up design

96. Exercise: model three (parallel) punch card desks in a hierarchical manner

97. Analysis of Process Models: Reachability graphs, invariants, and simulation
Prof.dr.ir. Wil van der Aalst
Eindhoven University of Technology, Faculty of Technology Management,
Department of Information and Technology, P.O.Box 513, NL-5600 MB,
Eindhoven, The Netherlands.

98. Questions raised when considering the handling of customer orders
How many orders arrive on average?
How many orders can be handled?
Do orders get lost?
Do back orders always have priority?
What is the utilization of office workers?
If the desired product is no longer available, does the order get stuck?
Etc.

99. Questions raised when considering the handling of customers in the canteen
What is the average waiting time from ?
What is the variance of waiting times?
What is the effect of an additional cashier on the queue length?
Etc.

100. Questions raised when considering the an intersection with multiple traffic lights
How much traffic can be handled per hour?
Give some volume of traffic, what is the probability to get a red light?
Is the intersection safe, i.e., crossing flows have never a green light at the same time?
Can a light go from yellow to green?
Is the intersection fair (i.e., a traffic light cannot turn green twice while cars are waiting on the other side)?

101. Questions raised when considering a printer shared by multiple users
Can two print jobs get mixed?
Do small jobs always get priority?
Can the settings of one job influence the next job?
Do out-of-paper events cause jobs to get lost?
How many jobs can be handled per day?
What is the probability of a paper jam?

102. Questions raised when considering a teller machine
What is the average response time?
Is there a balance, i.e., the amount of money leaving the machine matches the amount taken from bank accounts?
How often should the machine be filled to guarantee 90% availability?
Is fraud possible?
Etc.

103. Analysis validation verification performance analysis
Analysis is typically model- driven to allow e.g. what-if questions.
Models of both operational processes and/or the information systems can be analyzed.
Types of analysis:
validation
verification
performance analysis

104. Three analysis techniques (Chapter 8)‏
Reachability graph
Place & transition invariants
Simulation
Each can be applied to both classical and high-level Petri nets. Nevertheless, for the first two we restrict ourselves to the classical Petri nets.
Use:
reachability graph (validation, verification)‏
invariants (validation, verification)‏
simulation (validation, performance analysis)‏

105. Traffic lights are safe!
Reachability graph
(0,0,1,1,0,0,0)‏
(1,0,0,0,0,1,0)‏
(1,0,0,1,0,0,1)‏
(0,1,0,1,0,0,0)‏
(1,0,0,0,1,0,0)‏
Five reachable states.
Traffic lights are safe!

106. Alternative notation
o1+r2
r1+o2
r1+r2+x
g1+r2
r1+g2

107. Reachability graph (2)‏
Graph containing a node for each reachable state.
Constructed by starting in the initial state, calculate all directly reachable states, etc.
Expensive technique.
Only feasible if finitely many states (otherwise use coverability graph).
Difficult to generate diagnostic information.

108. Infinite reachability graph

109. Exercise: Construct reachability graph

110. Exercise: Dining philosophers (1)‏
5 philosophers sharing 5 chopsticks: chopsticks are located in-between philosophers
A philosopher is either in state eating or thinking and needs two chopsticks to eat.
Model as a Petri net.

111. Exercise: Dining philosophers (2)‏
Assume that philosophers take the chopsticks one by one such that first the right-hand one is taken and then the left-hand one.
Model as a Petri net.
Is there a deadlock?

112. Exercise: Dining philosophers (3)‏
Assume that philosopher take the chopsticks one by one in any order and with the ability to return a chopstick.
Model as a Petri net.
Is there a deadlock?

113. Structural analysis techniques
To avoid state-explosion problem and bad diagnostics.
Properties independent of initial state.
We only consider place and transition invariants.
Invariants can be computed using linear algebraic techniques.

114. Place invariant 1 man + 1 woman + 2 couple
Assigns a weight to each place.
The weight of a token depends on the weight of the place.
The weighted token sum is invariant, i.e., no transition can change it
1 man + 1 woman + 2 couple

115. Other invariants 1 man + 0 woman + 1 couple
(Also denoted as: man + couple)‏
2 man + 3 woman + 5 couple
-2 man + 3 woman + couple
man – woman
woman – man
(Any linear combination of invariants is an invariant.)‏

116. Example: traffic light
r1 + g1 + o1
r2 + g2 + o2
r1 + r2 + g1 + g2 + o1 + o2
x + g1 + o1 + g2 + o2
r1 + r2 - x

117. Exercise: Give place invariants

118. Transition invariant 2 marriage + 2 divorce
Assigns a weight to each transition.
If each transition fires the number of times indicated, the system is back in the initial state.
I.e. transition invariants indicate potential firing sets without any net effect.
2 marriage + 2 divorce

119. Other invariants 1 marriage + 1 divorce
(Also denoted as: marriage + divorce)‏
20 marriage + 20 divorce
Any linear combination of invariants is an invariant, but transition invariants with negative weights have no obvious meaning.
Invariants may be not be realizable.

120. Example: traffic light
rg1 + go1 + or1
rg2 + go2 + or2
rg1 + rg2 + go1 + go2 + or1 + or2
4 rg rg2 + 4 go go2 + 4 or1 + 3 or2

121. Exercise: Give transition invariants

122. Exercise: four philosophers
Give place invariants.
Give transition invariants

123. Two ways of calculating invariants
"Intuitive way": Formulate the property that you think holds and verify it.
"Linear-algebraic way": Solve a system of linear equations.
Humans tend to do it the intuitive way and computers do it the linear-algebraic way.

124. Incidence matrix of a Petri net
Each row corresponds to a place.
Each column corresponds to a transition.
Recall that a Petri net is described by (P,T,I,O).
N(p,t)=O(t,p)-I(p,t) where p is a place and t a transition.

125. man
Example
woman
marriage
couple
divorce

126. Place invariant
Let N be the incidence matrix of a net with n places and m transitions
Any solution of the equation X.N = 0 is a transition invariant
X is a row vector (i.e., 1 x n matrix)‏
O is a row vector (i.e., 1 x m matrix)‏
Note that (0,0,... 0) is always a place invariant.
Basis can be calculated in polynomial time.

127. Example
Solutions:
(0,0,0)‏
(1,0,1)‏
(0,1,1)‏
(1,1,2)‏
(1,-1,0)‏

128. Transition invariant
Let N be the incidence matrix of a net with n places and m transitions
Any solution of the equation N.X = 0 is a place invariant
X is a column vector (i.e., m x 1 matrix)‏
0 is a column vector (i.e., n x 1 matrix)‏
Note that (0,0,... 0)T is always a place invariant.
Basis can be calculated in polynomial time.

129. Example
Solutions:
(0,0)T
(1,1)T
(32,32)T

130. Exercise Give incidence matrix. Calculate/check place invariants.
Calculate/check transition invariants.

131. Simulation Most widely used analysis technique.
From a technical point of view just a "walk" in the reachability graph.
By making many "walks" (in case of transient behavior) or a very "long walk" (in case of steady-state) behavior, it is possible to make reliable statements about properties/ performance indicators.
Used for validation and performance analysis.
Cannot be used to prove correctness!

132. Stochastic process
Simulation of a deterministic system is not very interesting.
Simulation of an untimed system is not interesting.
In a timed and non-deterministic system, durations and probabilities are described by some probability distribution.
In other words, we simulate a stochastic process!
CPN allows for the use of distributions using some internal random generator.

133. Uniform distribution
pdf
cumulative

134. Negative exponential distribution

135. Normal distribution

136. Distributions in CPN Tools
Assume some library with functions:
uniform(x,y)‏
nexp(x)‏
erlang(n,x)‏
Etc.
A nice function is also C.ran() which returns a randomly selected element of finite color set C, e.g., color C = int with 1..5; fun select1to5() = C.ran() returns a number between 1 and 5

137. Example

138. Example(2)‏

139. Subruns and confidence intervals
A single run does not provide information about reliability of results.
Therefore, multiple runs or one run cut into parts: subruns.
If the subruns are assumed to be mutually independent, one can calculate a confidence interval, e.g., the flow time is with 95% confidence within the interval 5.5+/-0.5 (i.e. [5,6]).

140. Example of a simulation model
Gas station with one pump and space for 4 cars (3 waiting and 1 being served).
Service time: uniform distribution between 2 and 5 minutes.
Poisson arrival process with mean time between arrivals of 4 minutes.
If there are more than 3 cars waiting, the "sale" is lost.
Questions: flow time, waiting time, utilization, lost sales, etc.

141. Top-level page: main

142. Subpage gas_station

143. Assuming pages for the environment and measurements the last two pages allow for ...
Calculation of flow time (average, variance, maximum, minimum, service level, etc.).
Calculation of waiting times (average, variance, maximum, minimum, service level, etc.).
Calculation of lost sales (average).
Probability of no space left.
Probability of no cars waiting.
For each of these metrics, it is possible to formulate a confidence interval given sufficient observations.

144. Alternatives Model the following alternatives: 5 waiting spaces
2 pumps
1 faster pump

145. Simulation of a Production systemX Y Z C A B C

146. Use distributions
Data

147. Top level page: main

148. Sub page: supplier

149. Sub page: customer

150. Sub page: work_center

151. Overview Results: response time utilization % backorders average stock
etc.

152. Classical versus high-level Petri nets
Simulation clearly works for all types of nets.
Hierarchy is never a problem.
Time allows for new types of analysis.
Reachability graphs and invariants can also be extended to high-level nets.
More complex (both technique and computation)‏
Sometimes abstraction from color is possible to derive invariants (consider previous example).

153. Exercise: Five Chinese philosophers
Recall hierarchical CPN model of five Chinese philosophers alternating between states thinking and eating.
Give place invariants
Give transition invariants
Change the model such that philosophers can take one chopstick at a time but avoid deadlocks and a fixed ordering of philosophers.

154. Top-level page

155. Page philosopher

156. Flat model is obtained by replacing substitution transitions by subpages
Naming:
PH3.think
PH3.eat
PH3.take_chopsticks
PH3.put_down_chopsticks
Repeat 5 times...

157. Alternative page

158. You should be able to ...
Construct a reachability graph for a classical Petri net.
Give meaningful place and transition invariants for a classical Petri net.
Construct a reachability graph and give meaningful place and transition invariants for a hierarchical CPN after abstracting from data and time and removing hierarchy.
Build a simple simulation model using CPN.
Motivate the use of each of the analysis techniques.