Co-Design of Resilient Timing-Constrained Cyber-Physical Systems

Liang, Hengyi

doi:https://doi.org/10.21985/n2-vgsk-x698

Work

Co-Design of Resilient Timing-Constrained Cyber-Physical Systems

Public

Download PDF

Cyber-physical systems (CPS), as a multidisciplinary area, have been widely adopted in our daily life and attract experts from various fields. CPS aims to achieve real-time and resilient connection with physical world through integration of computation, communication and control technologies. Many CPS systems, such as automotive, avionics, and industrial system, operate under harsh and continuously changing environment. As those systems are safety-critical, the system's resiliency, including extensibility, security, stability and fault-tolerance capability of the systems, must be considered during early design stage. However, addressing these issues is increasingly challenging due to (1) the steadily-growing complexity of system functionalities, (2) limited computation resources, (3) stringent timing constraints and (4) highly dynamic environment perturbations.Moreover, the addition of resiliency-related techniques, e.g. message monitoring, control adaptation, error detection and recovery, etc., may further complicate the design, verification and validation of CPS systems. While there have been plenty of works on CPS resiliency mechanisms, focusing on communication encryption, authentication, error detection etc., only a few works have explicitly considered the integration of these techniques with timing-constraints. Moreover, traditional hard timing-constraint systems are insufficient to capture system timing requirements as most functions can tolerate a certain degree of deadline misses. In this dissertation, we demonstrate our approach to tackle the design for resilient automotive systems and address the impact of external perturbations to autonomous driving systems. A holistic optimization framework is proposed to optimize system resiliency for both hard real-time systems and weakly-hard real-time systems. The optimization framework contains system modeling, analysis and multi-objective exploration across software and hardware layers. Industrial applications and synthetic cases are used to demonstrate the effectiveness of our approach. At the application level, we conduct an end-to-end analysis of an automated lane centering (ALC) system and identify how external perturbations can affect the perception module and propagate the whole ALC pipeline. We also propose an adaptive planning strategy that leverage uncertainty information. We evaluate the proposed adaptation strategy through a production-grade simulator.

Creator