Control-Theoretical Perspective in Feedback-Based Systems Testing

Sammanfattning: Self-Adaptive Systems (SAS) and Cyber-Physical Systems (CPS) have received significant attention in recent computer engineering research. This is due to their ability to improve the level of autonomy of engineering artefacts. In both cases, this autonomy increase is achieved through feedback. Feedback is the iteration of sens- ing and actuation to respectively acquire knowledge about the current state of said artefacts and steer them toward a desired state or behaviour. In this thesis we dis- cuss the challenges that the introduction of feedback poses on the verification and validation process for such systems, more specifically, on their testing. We highlight three types of new challenges with respect to traditional software testing: alteration of testing input and output definition, and intertwining of components with different nature. Said challenges affect the ways we can define different elements of the test- ing process: coverage criteria, testing set-ups, test-case generation strategies, and oracles in the testing process. This thesis consists of a collection of three papers and contributes to the definition of each of the mentioned testing elements. In terms of coverage criteria for SAS, Paper I proposes the casting of the testing problem, to a semi-infinite optimisation problem. This allows to leverage the Scenario Theory from the field of robust control, and provide a worst-case probabilistic bound on a given performance metric of the system under test. For what concerns the definition of testing set-ups for control-based CPS, Paper II investigates the implications of the use of different abstractions (i.e., the use of implemented or emulated compo- nents) on the significance of the testing. The paper provides evidence that confutes the common assumption present in previous literature on the existence of a hierar- chy among commonly used testing set-ups. Finally, regarding the test-case gener- ation and oracle definition, Paper III defines the problem of stress testing control- based CPS software. We contribute to the generation and identification of stress test cases for such software by proposing a novel test case parametrisation. Leveraging the proposed parametrisation we define metamorphic relations on the expected be- haviour of the system under test. We use said relations for the development of stress testing approach and sanity checks on the testing results.