J. Software Engineering & Applications, 2008, 1: 68-75
Published Online December 2008 in SciRes (www.SciRP.org/journal/jsea)
Towards Automatic Transformation from UML Model
to FSM Model for Web Applications
Xi Wang, Huaikou Miao, Liang Guo
School of Computer Engineering and Science, Shanghai University, Shanghai, 200072, China
Received November 17
th
, 2008; revised November 26
th
, 2008; accepted November 30
th
, 2008.
ABSTRACT
The need for automatic testing of large-scale web applications suggests the use of model-based testing technology.
Among various modeling languages, UML is widely spread and used for its simplicity, understandability and ease of
use. But rigorous analysis for UML model is difficult due to its lack of precise semantics. On the other hand, as a
formal notation, FSM provides an avenue for automatic generation of test cases, but the requirement for mathematical
basis makes itself academic inventions divorced from real applications. This paper proposes an approach to
transforming UML model to FSM model, taking advantage of both languages. As our work focuses on the
transformation of UML state diagrams to FSM models, a specific transformation mechanism is presented, which deals
with different elements with different mapping rules. To illustrate the mechanism we proposed, an example of a web
application for software download is presented. Finally, we give a method for implementation of the mechanism and a
tool prototype to support the method.
Keywords: UML Model, FSM Model, Model transformation
1. Introduction
Providing greater assurance that the software is of high
quality and reliability, testing has been considered more
and more important as people gradually realize the great
effect on their daily life made by software products.
Hand-crafted methods are acceptable until the coming of
age when there are full of large-scale manufactures with
high complexity, especially the appearance of web
applications which labeled for their additional
heterogeneity, concurrency and distribution.
Web applications are usually composed of front-end
user interfaces, back-end servers including web servers,
application servers and database servers, which build up a
new way for deploying software applications. Components
called for supporting task completion of web applications
by each server may be programmed in different
languages and executed on different platforms. In
addition, web applications are frequently modified due to
continuous updates of its components, high-speed
developing technologies and changes of the needs of its
users. All of these characteristics are challenging the
traditional testing method which largely depends on the
testers. On the other hand, most companies keep the
minimum amount of time as their primary priority to
meet market demand while customers pay their much
attention to the reduction of the cost during maintenance,
leading directly to the calls for effective testing within a
relative short period of time.
Generation of test cases is the main task of testing;
since detections of faults are operated by comparing
expect outputs with actual ones obtained from running of
these test cases. Model-based testing, which involves
developing and using a model describing the structural
and behavioral aspects of the system to generate test
cases automatically, is an effective method for testing
various software artifacts including web applications. As
the models are developed early in the cycle from
requirements information [1], the generation of test cases
can be conducted in parallel with the implementation of
the System Under Test (SUT), rather than sequentially,
saving the time supposed to be spent for waiting. Also, it
supports re-use in future testing as these models capture
the behavior of a software system and in contrast to a test
suite, they are much easier to update if the specification
changes [2].
The critical part of model-based testing is the
construction of models. Among various modeling
languages, UML has been widely spread and used in
industry for its simplicity and ease of use. It enables
modelers to address all the views needed to analyze and
develop the corresponding system. Further more, as a
visual language, it can clearly show the structure and
functions of the system, facilitating understanding and
communication between designers, modelers, developers

Towards Automatic Transformation from UML Model to FSM Model for Web Applications 69
and users. Besides, many powerful tools have been
developed and used to support UML modeling such as
argoUML. But unfortunately, it is widely acknowledged
that UML can hardly provide formal semantics, as it
comprises several different notations with no formal
semantics attached to the individual diagrams. Therefore,
it is not possible to apply rigorous automated analysis or
to execute a UML model in order to test its behavior,
short of writing code and performing exhaustive testing
[3].
As one of the formal notations, FSM (Finite State
Machine) provides a significant opportunity for testing
because it precisely describes what functions the software
is supposed to provide in a form that can easily be
manipulated by automated means [4]. Being applied to
the testing process, its relative theory could be helpful
and supported for enhancing efficiency. Furthermore, in
addition to traditional software, a web application’s
behavior could also be modeled using FSMs theoretically
and then test cases could be automatically generated by
traversing the path through the FSM model of the
application, with each distinct path comprising a single
test case [5]. Besides, FSM model can be visualized to
tell intuitively the direction to which a test case is going,
since state-based specification languages are fairly easy
to translate into a specification graph as they have natural
graph representations [4]. Last, the transformation to
FSM facilitates model checking which verifies certain
property of the model. However, its requirement for
mathematical basis limits the range of utilization.
This paper proposes a method for transformation from
UML model to FSM model, taking advantage of both: the
simplicity and intelligibility of UML and the accuracy
and derivability of FSM. It also enables the reuse of the
existing and well-established tools for UML and theories
for FSM. There’re several kinds of diagrams within UML
corresponding to different views of the system, our job
focuses on the transformation of state diagram, as it is
most often used to model the behavior of an individual
object.
The remainder of this article is organized as follows:
Section 2 reviews existing works in transformation of
UML models. Section 3 presents a transformation
mechanism from UML state diagrams to FSM models.
To illustrate the transformation mechanism, an example
of transforming from a state diagram representing a web
application for software download is given in Section 4.
In section 5, a method for implementing the
transformation mechanism we proposed is given, together
with a brief introduction to a tool prototype based on this
method. Finally, concluding remarks and discussions
about future works are presented in section 6.
2. Related Works
Automatic testing has become a hot spot in the software
engineering field for facilitating development process of
software products. But most of the current technologies
are based on “capture/replay” mechanism, which costs
too much time and manual works while recording testing
scenarios and handling with small changes on the
functional design or user interfaces. Tools running on this
mechanism will not design or generate test cases
themselves and will not provide any instruction on the
coverage situation of the generated test cases. Further
more, there are even fewer automatic testing tools for
web applications which requires for even more
automatism. Most of the present tools [13] do not support
the function test of web applications including Link
Checks for checking links of the web application, HTML
Validators for providing standard HTML syntax
validation, Web Functional/Regression Test Tools, Web
Site Security Test Tools, Load and Performance Test
Tools and etc. Since most of them rely on information
obtained from codes of the web applications and only
concentrate on verification of static aspects, we need a
tool to help verifying the behavior of them while paying
least price.
With the appearance and popularity of the concept of
object-oriented and model-driven, model-based testing
for software products has aroused much attention in
industry. Though many researches are done in this field,
tools developed under their theories still have certain
gaps with applying to real uses due to their lack of
systematism and low automatic level [14,15,16,17,18,19].
Construction of models is the beginning of
model-based testing for web applications. The most
common one is to use Entity Relation Diagrams or UML
Class Diagrams to model web pages of a web application
and relationships between them. Isakowitz et al describe
web applications with a method called Relationship
Management Methodology [20]. Coda et al proposes a
model WOOM for modeling web applications in a higher
level of abstraction [21]. Gellersen et al introduce the
WebComposition Markup Language for implementing a
model for Web application development called Web
Composition [22]. Conallen et al extend UML modeling
language to model the structure of web applications [23].
However, these methods rarely construct models on the
behavioral and functional aspects of the web applications
and few testing approaches are figured out for these
models.
The model language we use when designing the web
applications is UML which strongly supports users to
describe complicated software including web applications.
But till now, no such complete testing tool has ever been
implemented as its semi-formal semantics prevents it
from automatic testing. On the other hand, many methods
for generation of test cases from formal models are
presented. [24] generates test cases from an Object-
Oriented Web Test Model which is a combination of
Object Relation Diagram, Page Navigation Diagram,
Object State Diagram, Block Branch Diagram and
Function Cluster Diagrams, but it will be trapped if there
are too many objects in the software. Ricca et al models

70 Towards Automatic Transformation from UML Model to FSM Model for Web Applications
web applications by modeling for each web page and
obtain test cases according to proposed rules. Still, it
would only be useful dealing with simple applications
[5,25], models web applications with FSM which will
then be used for test cases generation by search for
different path of the model under different criteria.
Considering that FSM model is also the most common
used object for model checking, we choose it for
destination of our model transformation process and
origin for test cases generation.
Formalization of UML models has aroused much
attention in industry. One of the most active group is the
precise UML group [6], which is made up of
international researchers who are interested in providing
a precise and well-defined semantics for UML, by using
model-oriented notations, such as Z or VDM. There are
also works done by other researchers, Borges et al. [7]
integrate UML class diagrams and a formal specification
language OhCircus by written UML elements in terms of
OhCirus. Latella et al. [8] converts UML state diagrams
into the formal language Promela. Traore et al. [9]
proposes a transformation mechanism from UML state
diagrams to PVS which facilitates automatic model
checking.
However, few researches on the transformation to the
FSM model can be found. Erich et al. [11] gives a
hierarchical finite state machine model for state diagrams,
which is capable of acquiring the hierarchical information,
but it does not mention the method for transformation to
FSM models with the removal of hierarchy. [10]
transforms time-extended UML state diagram into timed
automata, but special elements of the state diagram are
not under its consideration.
The method we proposed enables the transformation of
state diagrams with special elements, such as completion
transition, fork, join and history state. Besides, the
flatness of the resulting FSM model can greatly support
the automation of the generation of test cases.
3. Transformation Mechanism from UML
State Diagram to FSM Model
As UML and FSM are source and target models of the
transformation mechanism respectively, a brief
introduction of both is given below.
3.1 UML
The Unified Modeling Language (UML) is becoming a
standard language for specifying, constructing and
documenting the artifacts of a software-intensive system.
It can model from different perspectives with several
kinds of diagrams that express static and dynamic aspects
of a system. As a visualized model, UML conveys
information intuitively to our human beings who can get
better understanding through graphics. Besides, it is easy
to learn and use, making it more attractive to those who
model. Because of the characteristics mentioned above
UML severs as the ideal model for describing the real.
Class diagram, object diagram, use case diagram,
sequence diagram, communication diagram, activity
diagram and state diagram are the most commonly used
diagrams in UML. Class and object diagrams model the
static design view of a system, mostly about relationships
between objects, while rest of them focus on dynamic
aspects. For the purpose of capturing unexpected outputs,
we obtain most of the information needed for testing
from behavioral models.
As one of the behavioral models, state diagram is often
used to model the life cycle of certain object, from its
motivation to termination. Since most systems involve
more than one object, state diagrams are considered to be
the minimal unit for representing behaviors. We therefore
begin our research with UML state diagrams.
3.2 UML State Diagram
State diagram, which has been mainly discussed in this
paper, specifies the sequences of situations an object goes
through during its lifetime in response to events, together
with its responses to those events. Many elements are
involved for expressing semantics of the diagram.
States represent certain situations the object stays, each
with a name for distinguishing itself from others. There
are several types of states within state diagrams.
States that have no substructures are called simple
states, others are called composite states. A composite
state may contain nested states either concurrent or
sequential which are called orthogonal substates and
nonorthogonal substates respectively. Given a set of
nonorthogonal substates in the context of an enclosing
composite state called OR-state, the object is said to be in
the composite state and in only one of those substates at a
time [12]. In the case of orthogonal substates, the concept
of region is introduced which specifies each state
machine that execute in parallel in the context of the
enclosing composite state called AND-state. Only one
substate from each of the orthogonal regions is active as
long as the object remains in the corresponding AND-
state.
Initial state indicates the default starting place for the
state diagram or substate while final state indicates that
the execution of the state machine or the enclosing state
has been completed. Another special state is the history
state which allows an OR-state to remember the last
substate that was active prior to the leaving from the
OR-state.
Transitions are relationships between a pair of states
indicating that an object in the first state will enter the
second state when a specified event occurs under certain
condition. Therefore, a transition t comprises three parts:
source state denoted by src(t) which is the state affected
by the transition; target state denoted by dst(t) which the
object enters after the completion of the transition; label
denoted by EGA(t) which contains events, guards, and
actions.

Towards Automatic Transformation from UML Model to FSM Model for Web Applications 71
Semantics of transitions varies according to its source
and target state. When leading out of a composite state, a
fired transition leaves the active nested states before
leaving the composite one. When targeting a composite
sate, a fired transition would lead the object to the initial
state of each nested machine running in parallel after
entering the composite state.
In addition to these regular transitions, there exist some
special ones. Completion transition is a transition with
no event trigger, the fire of which depends on the
completion of the behavior within its source state.
Transition join which sources multiple states allows the
object to leave all the orthogonal regions of an AND-state
at one time. Similarly, transition fork which targets
multiple states enables passing directly to all the
orthogonal regions of an AND-state. The initial state of
the regions which have no target states of the fork will be
activated.
With clear semantics of each element, the
transformation mechanism which deals with different
elements with different mapping rules can be determined.
3.3 FSM Model
Finite State Machines (FSM) are models each built with a
set of states, as well as transitions going from one state to
another, which are triggered either by inputs from outside
or changes within the system itself. The execution would
start from a state called start state and keep running until
reaching a state called accept state. As its mathematic
nature, we can establish a formal representation for FSM
which is the target model during the transformation
process for facilitating automation.
Definition1. A FSM (Finite State Machine) A is a
quintuple (Q, L, δ
δδ
δ , q
0
, q), where Q is a finite set of states
of A, L is a finite set of transition labels of A, δ
δδ
δ : Q×L→
→→
→
Q is the transition function relating two states by the
transition going between them, q
0
∈ ∈∈ ∈Q is the start state,
q∈ ∈∈ ∈Q is the accept state.
If transition t∈
∈∈
∈ δ
δδ
δ represented as (s, l, s ′
′′
′ ), then source (t)
= s, target (t) = s ′ ′′ ′ , label (t) = l.
3.4 Transformation from State Diagram to FSM
Model
As can be seen from the definition of FSM model, states
involved are all basic ones, indicating that the removal of
hierarchy is needed during the transformation process.
For the sake of being conformed to the semantics of
original models, the hierarchical relations between states
of the state diagram should be obtained as critical
information for generating corresponding FSM model
without hierarchy. We therefore take the translation of
topological structures of state diagrams to mathematic
models of Hierarchical Finite State Machines (HFSM) as
a preliminary step towards model transformation due to
the fact that HFSM provides a simple and precise manner
to illustrate the topological structure of a state diagram.
Different from FSM, HFSM contains states with inner
structures. We could take HFSM as parallel and/or
hierarchical composition of FSMs with states of higher
hierarchy representing FSMs of lower hierarchy. A
definition of HFSM is given bellow according to this
point of view.
Definition2. Given a finite set of FSMs F = {A
1
,.., A
n
}
with mutually distinct state spaces Q(A
i
),
φ φφ φ : U UU U
A ∈ ∈∈ ∈ F
Q(A) → →→ → P(F) is a composition function on F iff
− ∃
1
A∈F ∧ A∉U ran(φ), which indicates a unique
root FSM denoted by φ
root
− ∀A∈U ran(φ) • ∃
1
s∈U
A′∈F\ {A}
Q(A′) • A∈φ(s)
− ∀S ⊆ U
A∈F
Q(A) • ∃s∈S • S ∩ U
A∈φ (s)
Q(A) = ∅.
Definition3. Hierarchical finite state machine (HFSM)
is a pair (F, φ φφ φ ) where F is a set of FSMs with mutually
distinct state spaces, φ φφ φ is a composition function on F.
With the definition of HFSM, the topological structure
of the original state diagram could be obtained in a
formal representation, which is specified by the
composition function φ φφ φ . Construction of such structure
starts from the top hierarchy, and then gradually comes to
completion by detailing each composite state that belongs
to the state diagram level by level. Establish φ φφ φ (s) = A
i
and F = F ∪
∪∪
∪ {A
i
} if the composite state s is an OR-state
with a sub-machine A
i
enclosed, while φ
φφ
φ (s) = {A
1
, A
2
,…,
A
n
} and F = F∪ ∪∪ ∪{A
1
}∪ ∪∪ ∪{A
2
}∪ ∪∪ ∪…∪ ∪∪ ∪{A
n
} if the composite
state s is an AND-state with sub-machines A
1
, A
2
,…, A
n
each located in the corresponding orthogonal region of s.
The state pointed by initial state turns to be the start state
of the corresponding FSM, while the state which points at
final state becomes the accept state.
Once the representation for topological structure is
present, we can get to know the hierarchical relation
between states which can be specified by the following
function. When given a HFSM (F, φ
φφ
φ ):
χ : U
A∈F
Q(A) → P ( U
A∈F
Q(A))
χ (s) = {s′ | ∃A∈F • A∈φ (s) ∧ s′∈Q(A)}
With hierarchical information represented in
mathematic form, the transformation to the resulting
FSM model starts from that of transitions of the original
state diagram. But some preliminary conceptions have to
be introduced first.
Definition4. A set C ⊆ U
A∈F
Q(A) is a configuration
of a given HFSM (F, φ φφ φ ) iff
−∃
1
s∈Q(φ
root
) • s∈C
−s∈C ∧ A∈φ(s) ⇒ ∃
1
s′∈Q(A) • s′∈C
−s∈C ∧ ∃ s′ • s ∈χ(s′) ⇒ s′∈C
Definition5. Given a HFSM (F, φ φφ φ ) with C as the set of
all its configurations and s as one of its states, function
config: U
A∈F
Q(A) → P (U
A∈F
Q(A))
config (s) = { ci | ci ⊆ C ∧ s∈ci }
Definition6. Given a HFSM (F, φ φφ φ ), the default
configuration of certain state sd is denoted as a function
deconfig: U
A∈F
Q(A) → P (U
A∈F
Q(A))
deconfig (sd) = X ⇔ ∃
1
X : config (sd) •
∀s • ( s∈X ∧ sd∉χ*(s) ⇒ I q
0
(φ
i
(s)) ⊆ X )

72 Towards Automatic Transformation from UML Model to FSM Model for Web Applications
Definition7. Given a state diagram with one of its
transitions t, Uexit is the uppermost one among the states
of the set exit = {exit
i
| || | ∀ ∀∀ ∀j: N• •• • src
j
(t) ∈ ∈∈ ∈χ χχ χ*(exit
i
) ∧ ∧∧ ∧ dst
j
(t)
∉
∉∉
∉ χ
χχ
χ*(exit
i
)}, Uenter is the uppermost one among the
states of the set enter = {enter
i
|
||
| ∀
∀∀
∀j: N •
••
• src
j
(t)∉
∉∉
∉
χ χχ χ*(enter
i
) ∧ ∧∧ ∧ target
j
(t) ∈ ∈∈ ∈ χ χχ χ*(enter
i
)}.
States of the resulting FSM model are configurations
each represent a set of states of the original state diagram
which are active at present. Therefore, transitions
involved are running from one configuration to another,
which leads to the fact that each transition of the state
diagram may correspond to several transitions within
target FSM model according to the number of
configurations the source state of the original transition
belongs to. Suppose confTranSet is the transition set of
the resulting FSM, the algorithm for obtaining the set is
specified below:
for each transition t
if EGA(t) = ∅
TempSet = I q (φ
i
(src (t)))
for each q
i
∈ TempSet
config
i
= config (q
i
)
ConfSet = I config
i
DefConf = deconfig ( dst (t))
if t is a join
for each s
i
∈ src (t)
config
i
= config (s
i
)
ConfSet = I config
i
DefConf = deconfig ( dst (t ))
if t is a fork ∧ dst (t) > 1
ConfSet = config ( src (t))
defDst = U (deconfig ( dst
i
(t)) ∩ χ
*
( dst
i
(t)))
NdefDst = I (deconfig ( dst
i
(t)) \ χ
*
( dst
i
(t)))
DefConf = defDst ∪ NdefDst
else
ConfSet = config ( src (t))
DefConf = deconfig ( dst (t))
while ( ConfSet is not empty )
get a souconf ∈ ConfSet
tarconf = ( souconf \ χ
*
( Uexit (t)) ∪
(χ
*
( Uenter (t) ∩ DefConf )
source (t ′ ′′ ′ ) = souconf
target (t ′ ′′ ′ ) = tarconf
label (t ′ ′′ ′ ) = EGA (t)
confTranSet = confTranSet
∪
{t ′ ′′ ′ }
confSet = confSet\{souconf}
Then the state set can be generated by filling up with
states related to each element of the transition set
confTranSet. The initial and accept state of the resulting
FSM model InitState and AccState can also be
determined.
InitState = deconfig ( q
0
(φ
root
))
AccState = config ( q (φ
root
))
This is the process during which state set of the
original state diagram are mapping into that of the
resulting FSM model. But there’re some exceptions.
History states are not involved in the algorithm due to
their different semantics with other common states; we
handle them in a special way.
For each history state h referring to certain OR-state
Ors with a state set HS composed of all its nonorthogonal
substates, we build relations of the target states of
transitions leading out of state Ors with each hs
i
(hs
i
∈ ∈∈ ∈HS). Relations, represented by transitions, should
be established in pairs, indicating returning to the same
state that was last active when leaving the enclosing
OR-state. Suppose the target state of the transition
leading out of Ors is Htar, and the label of the transition
is denoted as l, for each hs
i
(hs
i
∈
∈∈
∈HS), a new transition
labeled “back (hs
i
)” is created with Htar and hs
i
as its
source and target state. With a transition set obtained by
the method above, the problem is then turning into the
transformation from each element of the set to its
counterparts of the resulting FSM model. Meanwhile,
existing transitions of the newly established FSM model
which labeled l should be modified. Suppose t is a
transition of the resulting FSM model labeled l, then
label′ (t) = label (t) + s (s∈source (t)).
Till now, a FSM model carrying the same semantics
with the original state diagram is constructed and
completed.
4. An Example: Software Download
An example of state diagram is shown in Figure 1, which
models a web application for software download. The
life cycle of the web application starts from its main page
(MP), then turns to download or search module according
to the choice of users. When entering the download
module, two entities will be triggered: a web page for
illustrating the usage of the software about to download
by a video clip, a dialog box for download operation.
According to the transformation mechanism we
proposed, the topological structure of the state diagram
should be captured first by constructing a HFSM model.
The resulting HFSM model can be generated as follows:
A
1
: ( { S1, S2, S3, S4 }, { l
1
, l
2
, l
3
, l
4
}, { ( S1, l
1
) →
S2, ( S1, l
2
) → S3, ( S2, l
3
) → S4, ( S3, l
4
) →
S4 }, S1, S4 )
A
2
: ( { S5, S6 }, { l
5
}, { ( S5, l
5
) → S6 }, S5, S6 )
A
3
: ( { S7, S8, S9 }, { l
6
, l
7
}, { ( S7, l
6
) → S8, ( S8, l
7
)
→ S9 }, S7, S9 )
A
4
: ( { S10, S11, S12 }, { l
8
, l
9
}, { ( S10, l
8
) → S11,
( S11, l
9
) → S12 }, S10, S12 )
φ: φ
root
= { A
1
}, φ (S2) = { A
2
, A
3
}, φ (S3) = { A
4
}, φ
(S1) = φ (S4) = … = φ (S13) = ∅
F = ({ A
1
,…, A
4
}, φ )
Then, each transition of the exampled state diagram
could be transformed into several transitions of the
resulting FSM model by the algorithm we proposed with
the HFSM model above. The results are shown as follows
where L
i
indicates the transition of the state diagram
which labeled l
i
; Ci indicates one of the configurations of
the HFSM model.

Towards Automatic Transformation from UML Model to FSM Model for Web Applications 73
L
1
: C1 = { root, S1}, C2 = { root, S2, S5, S7 },
( C1, l
1
) → C2
L
2
: C3 = { root, S3, S10 }, ( C1, l
2
) → C3
L
3
: C4 = { root, S2, S6, S9 }, C5 = { root, S4 }, ( C4,
l
3
) → C5
L
4
: C6 = { root, S3, S11 }, C7 = { root, S3, S12 }, ( C3,
l
4
) → C5, ( C6, l
4
) → C5, ( C7, l
4
) → C5
L
5
: C8 = { root, S2, S5, S8 }, C9 = {root, S2, S5, S9 },
C10 = { root, S2, S6, S7 }, C11 = { root, S2,
S6, S8,}, C12 = {root, S2, S6, S9 }, ( C2, l
5
)
→ C10, ( C8, l
5
) → C11,
( C9, l
5
) → C12
L
6
: ( C2, l
6
) → C8, ( C10, l
6
) → C11
L
7
: ( C8, l
7
) → C9, ( C11, l
7
) → C12
L
8
: ( C3, l
8
) → C6
L
9
: ( C6, l
9
) → C7
L
10
: ( C1, l
10
) → C8, ( C1, l
10
) → C11
L
11
: C13 = { root, S13 }, ( C12, l
11
) → C13
L
12
: ( C7, l
12
) → C2
L
13
: ( C6, l
13
) → C13
We can now generate the state set of the resulting FSM
model, which is filled up with all the configurations
mentioned above: Q = { C1,…, C13}. The initial state is
C1 while the accept state is C5.
Finally, noticing there’s a history state H within the
state S3, we should add several new transitions to the
transition set of the FSM model:
( C5, “back (S10)” ) → C3, ( C5, “back (S11)” ) → C6,
( C5, “back (S12)” ) → C7
Meanwhile, transitions labeled l
4
should be modified into:
( C3, l
4
(S10)) → C5, ( C6, l
4
(S11)) → C5, ( C7, l
4
(S12)) → C5.
5. Implementation of the Transformation
Mechanism
As automatic testing is our final goal of model
transformation, the implementation of such mechanism
by computer itself is required. The method proposed in
this section can be applied to all the diagrams of UML
model, only the transformation mechanism varies when
dealing with different kinds. Since computers are unable
to understand and analyze meanings conveyed by
diagrams, texts carrying equivalent amount of information
would help. Here, we choose XMI.
5.1 XMI
XML Metadata Interchange (XMI) is a standard that
enables users to express objects using Extensible Markup
Language (XML), the universal format for representing
data on the WWW. As a bridge across the gap of objects
and XML, it provides a standard mapping from objects
defined by UML to XML, fulfilling object-oriented
feature of both UML and programming languages. In
addition, many mature tools supporting transformation
from UML diagrams to corresponding XMI files are
presented, such as argoUML. Therefore, XMI becomes
the ideal textual representation of those UML diagrams.
5.2 Implementation Method
First of all, XMI files are needed which can be easily
obtained as output of argoUML with inputs as UML
diagrams. As shown in Figure 2, when receiving the
resulting XMI, we extract semantics by recognizing
different tags which indicate the location of information
related to certain elements of the UML diagrams. Then,
data structure based on the corresponding HFSM model
could be constructed. With topological information
provided by the data structure, mapping rule for
transforming to FSM models works. Finally, resulting
models are made to be hold in XML files with schema
defined by ourselves.
A tool prototype has been developed to support our
transformation mechanism and implementation method.
It takes state diagrams carried by XMI files as inputs and
resulting FSM model carried by XML files as output.
Also, one can modify the chosen XMI file through an
edition platform provided by the tool before
transformation operation starts.
Figure 1. State diagram of a web application for software download

74 Towards Automatic Transformation from UML Model to FSM Model for Web Applications
Figure 2. Implementation process
5.3 Simulation
For the purpose of verifying the correctness of our
approach, we use the tool developed by ourselves to
simulate the example presented in the previous section.
Figure 3 shows an interface of our tool for automatic
testing for web applications. The characters in the main
frame are the textual representation of the exampled state
diagram.
After choosing transformation function of the tool, the
model will then be transformed into FSM model written
in XML language, as shown in Figure 4.
Figure 3. An interface of the tool for automatic testing
for web applications
Figure 4. FSM model written in XML language
To illustrate the resulting FSM model more clearly, our
tool implements the visualization of its textual
representation, which can be seen in Figure 5.
6. Conclusions
This paper proposes a method for transformation from
UML model to FSM model. It allows users to model a
system with the language they used to without barriers
towards automatic and efficient testing. As we focus on
the translation of state diagrams, a specific
transformation mechanism is proposed which enables
generation of corresponding FSM models with same
semantics.
Modelers create one state diagram for each object of
the system and other UML diagrams for relations
between them. Since our specific transformation
mechanism serves for every single state diagram,
synthesis of the FSM models each obtained from one of
these state diagrams should be discussed. It depends on
the information provided by other UML diagrams like
class diagrams, sequence diagrams etc. Besides, these
UML diagrams themselves need to be transformed into
FSM with meta-model we defined so as to generate target
model that covers information carried in all of the given
UML models. They could either be transformed directly
into FSM models, or to state diagrams as the first step,
which would then come into FSM models by the
mechanism we proposed. Experiments about comparison
on efficiencies of both should be hold with complete
transition mechanisms before the choice can be made.
Besides, details of elements contained in labels
including event, guard and action, as well as the action
attribute of states, are not considered in our research,
their affections to the correctness of transformation is
also a part of the future work.
Figure 5. The visualization of the model’s textual
representation
model level
HFSM model
UML
diagrams
implementation
level
XMI files
FSM models
XML files
semantic
extraction
semantic
extraction
t
opological
information
topological
information
semantic extraction
transformation
mechanism
s
e
mantic ex
tr
action
data structure for HFSM model
transformation
mechanism
save
as
save
as

Towards Automatic Transformation from UML Model to FSM Model for Web Applications 75
7. Acknowledgement
This work has been supported by National High-
Technology Research and Development Program of China
under grant No. 2007AA01Z144, Natural Science
Foundation of China under grant No. 60673115, National
Grand Basic Research Program of China under grant No.
2007CB310800, Research Program of Shanghai
Education Committee under grant No. 07ZZ06 and
Shanghai Leading Academic Discipline Project, Project
Number: J50103.
REFERENCES
[1] S. R. Dalal, A. Jain, N. Karunanithi, and B. M. Horowitz,
“Model-based testing in practice,” Proceedings of the 21st
International Conference on Software Engineering, Los
Angeles, California, United States, pp. 285–294, May
1999.
[2] H. Robinson, “Graph theory techniques in model-based
testing,” International Conference on Testing Computer
Software, 1999.
[3] W. E. McUmber and B. H. C. Cheng, “A general
framework for formalizing UML with formal languages,”
Proceeding of the 23rd international conference on
Software engineering, Toronto, Canada, pp. 433–442,
2001.
[4] J. Offutt, S. Y. Liu, A. Abdurazik, and P. Ammann,
“Generating test data from state-based specifications,”
The Journal of Software Testing, Verification, and
Reliability, pp. 25–53, 2003.
[5] C. J. Mallery, “On the feasibility of using FSM
approaches to test large web applications,” May 2005.
[6] The precise group,
http://www.cs.york.ac.uk/puml/.
[7] R. M. Borges and A. C. Mota, “Integrating UML and
formal methods,” Electronic Notes in Theoretical
Computer Science, Elsevier Science Publishers, pp.
97–112, July 2007.
[8] D. Latella, I. Majzik, and M. Massink, “Automatic
verification of a behavioral subset of UML Statechart
diagrams using the SPIN model-checker,” Formal Aspects
of Computing, pp. 637–664, 1999.
[9] I. Traore, “An outline of PVS semantics for UML
statecharts,” Journal of Universal Computer Science,
2000.
[10] M. Z. Lai and J. Y. You, “Formalize the time-extended
UML state chart with timed automata,” Computer
Applications, pp. 4–6, August 2003.
[11] E. Mikk, Y. Lakhnech, and M. Siegel, “Hierarchical
automata as model for statecharts,” Proceedings of the 3rd
Asian Computing Science Conference on Advances in
Computing Science, pp. 181–196, 1997.
[12] G. Booch, J. Rumbaugh, and I. Jacobson, “The unified
modeling language user guide,” China Machine Press,
Beijing, 2006.
[13] R. Hower, “Web site test tools and site management
tools,” Software QA and Testing Resource Center, 2002.
[14] Belinfante, L. Frantzen, and C. Schallhart, “Tools for Test
Case Generation,” Model-based Testing of Reactive
Systems, Springer LNCS 3472, Springer-Verlag, pp.
391–438, 2005.
[15] M. Utting, A. Pretschner, and B. Legeard, “A taxonomy
of model-based testing,” Technical Report 04/2006,
Department of Computer Science, The University of
Waikato (New Zealand), April 2006.
[16] I. K. El-Far and J. A. Whittaker, “Model-based software
testing,” Encyclopedia of Software Engineering,
Wiley-InterScience, Vol. 1, pp. 825–837, 2002.
[17] M. Blackburn, R. Busser, and A. Nauman, “Why
model-based test automation is different and what you
should know to get started,” in International Conference
on Practical Software Quality and Testing, 2004.
[18] B. Legeard, F. Peureux, and M. Utting, “Controlling test
case explosion in test generation from B formal models,”
The Journal of Software Testing, Verification and
Reliability, 14(2): pp. 81–103, 2004.
[19] A. Pretschner, H. Lötzbeyer, and J. Philipps, “Model
based testing in evolutionary software development,”
IEEE International Workshop on Rapid System
Prototyping 2001, pp. 155–161, 2001.
[20] T. Isakowitz, E. A. Stohr, and P. Balasubramanian,
“RMM: A methodology for structured hypermedia
design,” Communication of the ACM, Vol. 38, No. 8,
August 1995.
[21] F. Coda, C. Ghezzi, G. Vigna, and F. Garzotto, “Towards
a software engineering approach to web site
development,” Proceedings of 9th International Workshop
on Software Specification and Design, Ise-Shima, Japan,
April 16–18, 1998.
[22] H. Gellersen and M. Gaedke, “Object-oriented web
application development,” IEEE Internet Computing,
January–February 1999.
[23] J. Conallen, “Modeling web application architectures with
UML,” Communications of the ACM, Vol. 42, No. 10,
October 1999.
[24] D. C. Kung, C. H. Liu, and P. Hsia, “An object-oriented
web test model for testing web applications,” First
Asia-Pacific Conference on Quality Software, pp. 30–31,
October 2000.
[25] F. Ricca and P. Tonella, “Analysis and testing of web
applications,” Proceedings of the 23rd International Conference
on Software Engineering, pp.12–19, May 2001.