Test profiling for usage models by deriving metrics from component-dependency-models

In industry, embedded systems become complex multiple functionality devices. Research and industry have investigated the issue, how to derive test cases to cover test requirements within minimal time. For the automation of all activities related to testing tool support is available. Yet, one question is getting increasingly difficult: How to derive good, i.e. significant test cases, embracing requirements from an external perspective and incorporating knowledge about the internal design and architecture of the system under test (SUT). In this paper we present the use of internal dependences and implementation metrics in combination with an exterior view on the basis of requirements with the aim to profile the test model for the generation of good, i.e. significant test cases. The incorporation of different sources of knowledge of the SUT into test generation reduces the effort to determine test cases for different test goals independently while recreating the knowledge of the SUT in each test process step. We applied the method to a real implementation of automotive functionality in the thermal control unit of a German OEM, that was designed in Matlab Simulink and architectured in AUTOSAR 3.2. The existing test suites were linked with the internal dependences of the Matlab Code, as to assess the reachability of the existing test cases already on the implementation model. In a next step, model metrics were calculated for the implementation modules. They were set into relation with the test cases and code coverage. In doing so, we profiled the test cases and assessed the significance and expressiveness of the model metrics.