SDVs(Software-defined Vehicles) represent one of the most advanced research and development technologies in the automotive industry, wherein vehicle control is determined and executed primarily through software. As SDVs continue to evolve, users are a...
SDVs(Software-defined Vehicles) represent one of the most advanced research and development technologies in the automotive industry, wherein vehicle control is determined and executed primarily through software. As SDVs continue to evolve, users are able to experience new values that transcend the traditional notion of automobiles as mere means of transportation, by integrating ICT(Information and Communications Technology) into vehicles through paradigms such as V2X(Vehicle-to-Everything).
However, the emergence and development of SDVs have also introduced new challenges in the domain of safety assurance. Since software constitutes the foundation of control in SDVs, the overall system functions are realized through the extensive interactions of numerous controllers. Traditional reliability-based safety engineering techniques, which analyze components rather than interactions, are increasingly inadequate to guarantee the safety of such complex systems.
In response, international standards governing automotive safety have also evolved. Beyond functional safety, as defined in ISO 26262, developers are now required to demonstrate SOTIF(Safety of the Intended Functionality) under ISO 21448. Thus, addressing the transformation of safety assurance in SDVs necessitates research that can both overcome the limitations of traditional safety engineering methods and achieve compliance with these dual standards.
This dissertation proposes a safety assurance framework for SDVs, aimed at preventing accidents and ensuring system safety in safety-critical automotive applications through a stepwise process that supports the development of both functional safety and SOTIF, spanning from system analysis to safety verification. The proposed framework restructures system-theoretic hazard analysis and hierarchical reliability-based safety analysis, grounded in a CSD(Control Structure Diagram) modeled from the perspectives of Function, Behavior, and Structure. By adopting an MBSE(Model-based Safety Engineering) approach, the framework supports safety assurance that reflects the interaction-driven characteristics of SDVs.
Furthermore, to enhance the accuracy and credibility of safety assurance results, this dissertation introduces a methodology that builds upon the results of hierarchical safety analysis to perform probabilistic safety integrity verification. System failure states and transitions identified in the safety analysis are formally synthesized into a Markov model, which is subsequently evaluated through probabilistic model checking to verify compliance with the ASIL(Automotive Safety Integrity Level). This approach mitigates potential systematic failures arising from variability in analyst judgment or methodology and enables safety analysis that accounts for uncertainties in empirical data, such as failure rates affected by thermal or aging phenomena.
The proposed framework has been applied to the EPS(Electric Power Steering) controller under development at H Corporation to evaluate its validity. The results demonstrate that the framework systematically identifies potential hazards, derives safety requirements that maintain the system in a safe state, and confirms that these requirements are successfully implemented in the design. Furthermore, the quantitative verification shows that the safety integrity goals anticipated during the analysis phase are met, thereby substantiating the assurance of SDV safety.
The contributions of this dissertation to the safety assurance of software-defined controllers are threefold: (1) preventing the omission of potential accident hazards that may arise due to the unique characteristics of SDVs, (2) establishing a safety assurance framework tailored to the automotive domain on the basis of MBSE, and (3) improving the consistency and reliability of safety analysis results through formal synthesis and probabilistic model checking.