Pareto optimal design space exploration of cyber-physical systems

Abstract

Cyber-physical systems (CPS) integrate a variety of engineering areas such as control, mechanical, and computer engineering in a holistic design effort. While interdependencies between the different disciplines are key attributes of CPS design science, little is known about the impact of design decisions of the cyber part on the overall performance qualities of the system. To investigate these dependencies, this paper proposes a simulation-based Design Space Exploration (DSE) framework that considers detailed cyber system parameters such as cache size, bus width, and voltage levels in addition to physical and control parameters of the CPS. We propose a DSE algorithm that explores the parameter configurations of the cyber-physical sub-system in order to approximate the Pareto-optimal design points with respect to design objectives such as energy consumption and control stability. For validation, we have successfully applied the proposed framework to an inverted-pendulum application. Here, our holistic evaluation of the Pareto-optimal points reveals the presence of non-trivial trade-offs that are imposed by the control, physical, and detailed cyber parameters. For instance, the identified energy and control optimal design points comprise configurations with a wide range of CPU speeds, sampling rates, and cache configurations following non-trivial zigzag patterns. The proposed framework could identify and manage these trade-offs and, as a result, is an imperative first step to automate the search for superior cyber-physical system configurations.