Cross-Layer Design and Adaptation of Safety-Critical Cyber-Physical Systems

Wang, Zhilu

doi:https://doi.org/10.21985/n2-ab7t-nc88

Work

Cross-Layer Design and Adaptation of Safety-Critical Cyber-Physical Systems

Public

Download PDF

Download All Files (.zip)

With growing system complexity and closer cyber-physical interaction, there are stronger needs for cyber-physical systems to adapt to the dynamic environment and improve their runtime performance. However, especially for safety-critical systems, the ability of such adaptation and improvement is often restricted by multiple factors, such as limited resources, stringent timing constraints, and safety or other functionality requirements. To tackle this challenge, we propose several algorithms and methods across two layers of the cyber-physical systems -- architecture layer and function layer. These solutions are in three categories. On the architecture layer, we leverage the weakly-hard paradigm, which provides the scheduling flexibility by allowing a bounded number of deadline misses or skips of execution. We propose a general event-based weakly-hard schedulability analysis algorithm that can be leveraged to efficiently provide timing analysis under different weakly-hard scenarios. On the function layer, we focus on the safety-critical systems, where algorithms are proposed to verify system stability and safety under the weakly-hard paradigm. On the other hand, learning-based methods become more and more popular in modern control systems for performance improvement. But there is often a lack of formal guarantees of the safety in such systems. We propose an efficient neural network global robustness certification algorithm. The global robustness can be leverage to formally verify the safety of a control system that relies on advanced sensors and deep neural networks for perceiving the environment. The last category is across the architecture and function layers. We first demonstrate that an cross-layer design of weakly-hard systems can provide better resource utilization and control performance under limit computation resource. Then, leveraging the scheduling flexibility of weakly-hard paradigm, and the safety verification techniques on the function layer, a runtime adaptation framework is proposed to improve system adaptability by allowing proactive skipping of task executions and re-allocating resources to other tasks for their performance improvement.

Creator