Model-bazlı testler için modellerin otomatik iyileştirilmesi

Model-Based Testing (MBT) enables automatic generation of test cases based on models of a system. It has been successfully applied in various application domains, each of which might introduce specific challenges. In this dissertation, we introduce methods and tools for addressing some of these challenges for the consumer electronics domain. In particular, we focus on the testing of Digital TV systems as our case study. We identified the following 3 problems in this context: i) Models of the system are created based on requirement specifications, which are often incomplete and imprecise. Therefore, these models are subject to accidental omissions of certain system behavior. As a result, critical faults can be left undetected by the generated test cases. ii) Resources are extremely limited in the consumer electronics domain. It is not feasible to attain an extensive coverage of test models. iii) A product family in consumer electronics often includes hundreds of systems. The set of features can highly differ among these systems. Therefore, the MBT process and modeling must be flexible to systematically manage variability and increase the amount of reuse for test models. To tackle the first problem, we introduce an approach and tool for automatically extending test models based on a set of collected execution traces. These traces are collected during Exploratory Testing (ET) activities. Several critical faults were detected in 3 case studies after generating test cases based on extended models. These faults were not detected by the initial set of test cases. They were also missed during the ET activities. As a solution for the second problem, we iteratively update test models in 3 steps to focus the test case generation process only on execution paths that are liable to highly severe failures. We use Markov Chains as test models, in which transitions among states are annotated with probability values. First, we update these values based on usage profile. Second, we perform an update based on fault likelihood that is estimated with static code analysis. Our third update is based on error likelihood that is estimated with dynamic analysis. We generate and execute test cases according the updated values after each iteration of updates. New faults can be detected after each iteration. To address the variability problem, we document variations among tested systems explicitly with a feature model. We map optional and alternative features in the feature model to a set of states in the test model. Transition probabilities in the test model are updated according to the selected features so that the generated test cases focus only on these features. This approach facilitates the reuse of a test model for many systems.