This paper proposes a defect detection system for autonomous driving systems that integrates Digital Twin technology with System-Theoretic Process Analysis (STPA). Addressing common issues such as perceptual distortion, decision-making errors, and controller reliability degradation in complex environments, the system enables the identification and early warning of potential defects by establishing a bidirectional mapping between the physical system and its virtual model. The study first constructs a four-layer integrated architecture comprising the physical entity layer, data acquisition layer, digital twin layer, and analysis service layer, transforming STPA's static safety analysis results into executable dynamic test cases. Using an Adaptive Cruise Control (ACC) system as a case study, the research systematically conducts system-level hazard identification, safety constraint definition, hierarchical control structure modeling, and unsafe control behavior analysis, thereby constructing a complete defect propagation path. Finally, an algorithmic framework incorporating data validity, accuracy, and temporal characteristic detection is developed and validated using real-world road test data. Experimental results demonstrate that the proposed framework effectively monitors defect propagation from the component to the system level, significantly enhancing the safety assurance capabilities of autonomous driving systems.
Zhang et al. (Wed,) studied this question.