Exploration of the Cyber-Physical Design Space

Abstract:

Generally, a cyber-physical-system (CPS) is one that combines computational and physical entities in a unified design effort. The design of CPSs needs good understanding of both subsystems, as small changes in the physical subsystem (PS) or the cyber subsystem (CS) may have significant consequences with respect to the overall system performance. For example, it is well known that the weight and size of physical objects like pendulums directly influence the performance requirements for the CS – but also that scheduling decisions on the CS may affect the timing jitter of an otherwise correct control application so that the control process fails. Therefore, a good CPS design methodology must carefully model and account for physical attributes of a system. One goal of the Design Science for Cyber Physical Systems project is the development of a design space exploration (DSE) framework that considers the variabilities of the PS and the required adaptations of the CA. The basic idea is to derive parameterizable models of the PS, the CA and the CS. The parameterizable models then can be instantiated by the DSE tool that, for each design point, invokes an executable simulation model to evaluate the quality of the CPS design. After an evaluation of the designs, a system configuration is proposed for implementation, containing parameters for the physical and the cyber subsystems. Our design flow is facilitated by the following main ideas:

1) We ensure rigorous parametric description following the equation derivation approach of the physical subsystem,
2) Settings of the CA are determined as a function of the parameters of the PS and the CS for every iteration of the design space exploration, which eliminates time consuming search for suitable configurations,
3) The DSE tool instantiates parameterized executable models of the CPS, which can be parallelized for the search of superior system configurations.

The encapsulated control knowledge and the efficient simulation enable CPS design analysis, available for system engineers even without strong control background. Based on small-scale examples, like the falling ball or the inverted pendulum, we investigate the required steps, necessary to prepare the models. Apply in a systematic DSE the models facilitate the anlysis of non-trivial cyber-physical design trade-offs, for instance between control quality and energy consumption. On the poster we show our approach for a rotary inverted pendulum system, for which we explored the design space that included a selectable length of the pendulum and a flexible sampling time in the CS. The results, which were validated in a real-life setup, show that our approach can automatically identify superior cyber-physical configurations which would have been ignored using existing design space exploration techniques.