Pruning Architectural Models of Automotive Embedded Systems via Dependency Analysis

Abstract

Dependency analysis techniques are widely used to understand software implementations, and reduce their verification efforts. Recently, architectural languages have started to be integrated in the development of complex embedded systems. Such languages provide early development artifacts, which can be used to specify the structure and functionality of a system, and can be also analyzed in order to provide early information regarding the system's correctness. By performing dependency analysis on architectural languages, crucial dependencies can surface earlier in the life cycle. Once computed, these dependencies can be used to prune the architectural models in an attempt to reduce the early design-stage verification efforts. In this paper, we propose a dependency analysis-based technique that can be applied to prune models in EAST-ADL, an architectural description language tailored to automotive systems development. To achieve correct pruning, we investigate the types of dependencies that can appear in an architectural model, and how these dependencies create dependency chains within the model. Next, we investigate how such dependency chains can be exploited in formal verification in order to reduce the verified state-spaces during model-checking. Assuming a given requirement, our pruning method entails that only the relevant dependency chains are examined during EAST-ADL model-checking against that particular requirement. We validate our analysis results by comparing them to those obtained by applying an analytical approach for end-to-end timing analysis in EAST-ADL models. The methodology is illustrated on a Brake-by-Wire industrial system.