Model-driven development (MDD) can help verify the accuracy of models and generate source codes, which allows a programmer to minimize the development time required to evaluate the software so that he or she can focus on the modeling process. Thus, modeling should be taught with MDD because it allows students to acquire modeling skills in a short period of time. We conducted a course to teach UML modeling to two groups. The rst group used the MDD tool, while the second did not. We elucidate the advantages of each case with and without the use of the MDD tool. Based on our results, we propose the effective use of MDD tools in UML modeling education.
Object-oriented modeling is widely used during embedded software development
and is taught in many universities. Modeling ensures good quality and
productivity during software engineering [
However, it is difficult to teach modeling. In the early stages of the learning process, students have questions such as \Why do we need modeling"? \How do we model"? and \Is this model equivalent to the speci cation"?
MDD is used to verify the accuracy of models and generate source code. This allows a programmer to minimize development time so that he or she can focus on the modeling process. Therefore, modeling should be taught with MDD, because it allows students to acquire modeling skills in a short period of time.
Previous studies based on the research on MDD and education divided into
three categories: studies on MDD education [
One issue that arises when we use the MDD method for modeling education
is that students tend to neglect the quality of the model because they focus on
completing the functional aspects that can be evaluated with the MDD [
The purpose of this study is to demonstrate the effectiveness of MDD tools for modeling. From these results, we propose the effective use of MDD tools in UML modeling education. Note that, in this paper, tools with the ability to automatically generate codes from UML models are called MDD tools.
In this paper, we discuss the following research questions.
When using MDD tools in modeling education for novices, RQ1 what is the difference in the process of creating the model? RQ2 what is the difference in model quality?
Further, model quality is of three types (syntactic, semantic, and pragmatic)
[
We conducted a course to educate two groups in UML modeling. The rst group members used the MDD tool, while the second group did not. We clarify the advantages of each case with and without the use of MDD tools. From these results, we propose the effective use of MDD tools in UML modeling education. 2
There are some modeling education studies using BridgePoint as an MDD tool
[
This platform allows making class diagrams and state machine diagrams. A class diagram consists of classes and relations, while state machine diagrams consist of states (including an initial state), event transmission states, events, and actions.
In order to collect the change history of the model, a model repository is created. By storing models in the model repository when a new or additional changed model is created, learners can see the past versions of models later. The model repository is an additional function of the MDD tool. 3 3.1
We performed experiments on subjects divided into two groups. The rst group used the MDD tool (experimental group) and is called the with-MDD group, and the second group did not use it (control group) and is called the without-MDD group.
For the with-MDD group, we prepared an MDD tool that can generate code, and group members were allowed to check their operations at any time. In contrast, in the without-MDD group, we prepared the same tool, but group members were allowed to check operations only once at end of the exercise. 3.2
The number of subjects was 12, and the subjects were well-versed with JAVA and UML notation. The educational items used in the experiment are listed in Table 1. The integrated exercise was aimed at developing an auto transport robot for a ctitious transportation company. The development objective was to develop a robot vehicle using LEGO Mindstorms NXT (Figure 1). The automated operations were transportation, forwarding, and out-of-service. These three operations were affected by the presence or absence of delivery destinations or cargoes. A wall detector (sonar sensor) monitored the delivery destinations while a bumper (touch sensor) detected the forwarding destination. The task of the robot was to trace a black line along a course using a line monitor (light sensor) to make stops at the delivery destination, forwarding destination, or the garage. The robot had to behave appropriately at each point and deliver the cargo.
Typical examples of class diagrams of the with- and without-MDD groups are shown in Fig. 2 and Fig. 3.
Subjects of the with- and without-MDD groups were respectively labeled A-F. Table 2 summarizes the number and kind of class categories.
In the without-MDD group, the subjects had classes related to only services (examples: transportation, forwarding, and out-of-service) with the exception of a class for controlling the whole system. In contrast, the with-MDD group had numerous classes related to functions (examples: line trace, edge change, and rotation), and there were four class diagrams which did not include a class related to services.
When subjects created the integrated exercise model, four subjects reused the basic exercise model in the with-MDD group; however, no subjects reused models in the without-MDD group. In order to analyze the process of creating the model, exercise time divided into three: rst, middle, and last. Table 3 illustrates the number of classes added, modi ed, and deleted at the time of the rst, middle, and last. In the without-MDD group, the proportion of the number of added and deleted classes reduced in the last, while the proportion of the number of states added and modi ed increased in the middle and last. Thus, it is assumed that subjects xed the class structure by adding and deleting classes at rst and then designed the behavior model by adding and modifying states. On the other hand, in the with-MDD group, subjects also added and deleted classes in the last. Therefore, the with-MDD group changed the class structure until they reached the last. In the with-MDD group, subjects tended to modify a functional model and then created a static structure of the system. In contrast, in the without-MDD group, subjects created a new structure model and then created the behavior model. Therefore, when the subjects did not use MDD, they tended to create a model by implementing a top{down approach; however, when they used MDD, they did so using a bottom{up approach. 4.2
Class diagrams Table 4 shows the number of errors in the class diagrams. Each subject created one class, and therefore, the maximum number of errors is six.
In six of the seven items, there were many errors in the with-MDD group. All members of the without-MDD group committed the error \One class has several responsibilities." The model quality of the classes was high for the withoutMDD group. However, no members of the without-MDD group could separate the functions of the class and no member created a function class.
We think that if the models were not executable, the subjects would have focused on operation models at a higher level of abstraction. It is difficult to model functions that have a low level of abstraction such as a running style. State machine diagrams Typical examples of state machine diagrams of the with- and without-MDD groups are shown in Fig. 4. Table 5 shows the number of errors in the state machine diagrams.
The with-MDD group had \No states names" and \State names inappropriate" errors such as a or b.
We evaluated the achievement rate for the integrated exercise as an
evaluation function for measuring the semantic quality. Table 6 lists the achievement
rates for the integrated exercise. The with-MDD group had a high achievement
rate because they were able to test the models.
Currently, many educational modeling courses apply a top{down approach. In
contrast, successful MDD practice tends to be driven from the bottom{up
approach. Therefore, Whittle et al. suggested that modeling should be taught with
a bottom-up approach rather than a top{down one [
We conducted a course to educate two groups in UML modeling. Subjects from one group used the MDD tool, while those from the other group did not. We veri ed the advantages of each case, with and without the use of the MDD tool. The results showed that when subjects did not use MDD, they tended to create models using a top{down approach and gave appropriate names to classes and states, and that when they used MDD, they tended to create models using a bottom{up approach and achieved a high achievement rate in the exercise.
Based on these results, we believe that by limiting the number of times operations can be checked and by automatic code generation, a change in the approach (top{down or bottom{up) to development of models can be encouraged. In the future, we intend to clarify the number and timing of optimal operation checks.