Automated Support for the Architecting of Distributed Embedded Systems : Methods and Analysis for Industrial Adoption

Sammanfattning: The architecture design complexity of modern embedded systems, such as in the automotive domain, is growing due to the rapidly evolved functionalities, the increasing amount of interactions between functions and computation nodes, and the stringent extra-functional requirements. Architecture design is crucial since it affects nontrivial system properties such as safety, cost, performance of functionalities and also the development time. An important enabler to deal with this complexity is to provide computer aided architecture design. This thesis focuses on such support for Design Space Exploration (DSE), relying on a model-based design (MBD) environment.The goal of this thesis is to improve the industrial adoption of DSE methods to facilitate the architecture design of distributed embedded systems in the automotive industry. The main contributions of this thesis are as follows: (1) Applying architecture recovery in the automotive industry to extract architecture models from legacy ECU source code. The recovered architecture models can be used to facilitate system understanding, to verify the software implementation against its specification and also to enable DSE for architecture design. (2) A systematic gap analysis was conducted between the state-of-the-art DSE methods and the industrial needs, through literature studies and interviews with experienced system architects. Identified gaps are analyzed from the following perspectives: Architecting scenarios, architectural decisions, quality attributes, cost model, procurement strategy, system variability and functional safety. (3) A new problem formulation was proposed to reduce the design space by utilizing the features of evolutionary architecting and the AUTOSAR layered architecture. (4) In order to enhance the flexibility of the DSE methods by enabling the customizability of the architectural constraints, an automatic transformation method is proposed to translate formally described architectural constraints into the corresponding mixed integer linear programming(MILP) constraints, commonly used for DSE. (5) This thesis also investigates the potential impacts of vehicular communication on the future architecture of automotive embedded systems from the timing perspective through a case study to enable a commercial truck with cooperative driving functionalities. The receiving bias problem was identified during the case study and effective architectural solutions were proposed. The case study also showed that the adoption of vehicular communication would not have significant architectural impacts in terms of timing.