Model-driven Analysis and Verification of Automotive Embedded Systems

Sammanfattning: Modern vehicles are equipped with electrical and electronic systems that implement highly complex functions, such as anti-lock braking, cruise control, etc. To realize and integrate such complex embedded systems, the automotive development process requires an updated methodology that takes into consideration the system’s intricate features and examines both their functional and extra-functional requirements. Early design artifacts like architectural models represent convenient abstractions for reasoning about the system’s structure and functionality. In this context, the EAST-ADL language has been developed as a domain-specific architectural language that targets the automotive industry and is aligned with the AUTOSAR automotive standard. To fully enjoy the benefits of these abstract system descriptions, architectural models need to be integrated into a model-driven development framework that enables also verification by, e.g., model checking and model-based testing. One major drawback in developing such a framework lies in the fact that architectural models, while capturing the system’s structure and inter-component communication, often lack direct means to represent the desired internal behavior of the system in a semantically well-defined way. To overcome this, one needs to provide means of integrating both structural as well as behavioral information, desirably within the same framework backed by formal semantics, in order to enable the model’s formal verification.In this thesis, we propose a tool-supported integrated formal modeling and verification framework tailored for automotive embedded systems that are originally described in the EAST-ADL architectural language. To achieve this, we first provide formal semantics to the architectural model and its behavior by proposing an equivalent formal description as a network of timed automata. This enables us to analyze the resulting network of timed automata formally by model checking, using both the UPPAAL PORT and UPPAAL SMC model checkers. UPPAAL PORT is providing efficient component-aware verification via the partial order reduction technique, while UPPAAL SMC is extending UPPAAL with statistical model-checking capabilities via probabilistic algorithms. We focus the analysis on functional and timing requirements, but also on the system’s resource usage with respect to different resources specified in the model, such as memory and energy. In an attempt to narrow the gap between the original architectural model and the eventual system implementation, we define an executable semantics of the UPPAAL PORT components that guarantees that the implementation preserves the invariant properties of the model. Assuming a system implementation that conforms to the formal model, we investigate how to provide test cases suitable for the eventual verification of such implementation, by exploiting the model checker’s ability to generate witness traces for reachability verification. Such a witness trace represents a execution of the system from its initial state to the goal state encoded by the reachability property, and becomes our abstract test case. By pairing the automated model-based test case generator with an automatic transformation from the abstract test cases to Python scripts, we enable the execution of the generated Python scripts on the system under test, which ends up in pass/fail testing verdicts. Dependency analysis is a method that is able to identify crucial intra- and inter-component dependencies early in the system’s development life cycle, if applied on architectural models. In this thesis, we also investigate how such dependencies, resulting from applying dependency analysis on EAST-ADL models, can be exploited during formal verification in order to reduce the verified state-spaces during model checking. The framework is supported by the ViTAL tool and its applicability is shown on an automotive industrial prototype, namely a Brake-by-Wire system.