Lecture Notes in Computer Science 3472

13.5 Summary
13 Real-Time and Hybrid Systems Testing 385
In this chapter, we have discussed test case generation for real-time and hybrid
systems. As we have seen in the beginning, test automation for real-time and
hybrid systems differs from that for untimed systems. The time domain has to
be taken into account as well as the dense value domain for hybrid systems,
e.g. for test evaluation. The model of the SUT must abstract from the real
behavior of the SUT to allow sensible test case generation. In contrast, the test
driver concretizes the generated inputs to feed the SUT. Vice versa, the concrete
outputs of the SUT are abstracted again so that test evaluation can be done with
respect to the abstract model.
Process Algebra vs. Automata Different modeling techniques can be used for this
purpose. We have discussed ACSR as a process algebra and different variants of
automata that already have been used for test case generation. A process algebra
is capable of modeling a system consisting of processes that communicate with
each other via events. Concurrency can be modeled explicitly as there are operators
for interleaving and parallel execution with synchronization. Time is considered
discrete. One problem of process algebras is that this modeling technique
has no support in industry where graphical modeling is preferred. Nevertheless,
testing based on process algebra has been proven useful and practicable.
In contrast, automata based modeling has more support as this is very popular,
e.g. with respect to UML where state machines are used as automata
variants. Timed automata introduce either discrete or dense time to be capable
of modeling timing constraints. Graphical models are in principle easy to understand
as control flow can be captured at one sight. However, large models become
intractable in graphical representation. Hierarchy or different abstraction levels
must be used to better this situation.
Discrete vs. Dense Time With respect to time, this is either modeled discrete or
dense. In the first case, a timer is a counter that is incremented continuously. As
the computer itself is working discrete, this seems appropriate. In contrast, time
is naturally dense so modeling based on dense time is closer to reality, e.g. when
analog sensors and actuators are used. For real-time systems, it must be chosen
if the model represents natural time or computer time beforehand. For hybrid
systems, time must be modeled dense as the continuous parts of the system rely
on this.
Discrete vs. Dense Values Another aspect is the value domain that has to be
tested. This is considered discrete for real-time systems. We have seen that test
case generation builds up on methods for untimed systems and enhances the
generated test cases with timing information. Discrete values for inputs and
outputs to respectively from the SUT can be easily generated and evaluated.
Time points and intervals for generating inputs and outputs are chosen with
respect to the selected test case generation method. For hybrid systems, this
situation differs as we have dense-valued variables. Until now, nobody has tackled