A Method for Building Trustworthy Hybrid Performance Models for Cyber-Physical Systems of Systems

Abstract

The increasing dynamism and interconnectivity of Cyber-Physical Systems of Systems (CPSoS) emphasize the need for advanced performance modeling approaches. Performance models serve the purpose of exploring and evaluating design solutions, providing insights into system behavior during the early design phases, and managing performance risks during development. Despite the advancements in hybrid modeling within the field of performance modeling, the challenge of constructing models effectively and efficiently while ensuring their trustworthiness has not received much attention. This paper proposes a method, that is, step-by-step guidelines on using formalisms, techniques, and tools, to construct trustworthy hybrid models for analyzing and predicting stochastic timing performance of CPSoS. The method integrates design patterns for improved model design and development, alongside Bayesian calibration and statistical validation techniques to further enhance trustworthiness of the models. Although many individual techniques, formalisms and tools are available, our contribution lies in proposing a systematic method for achieving trustworthiness within the context of hybrid performance modeling and simulation. We concretize the method for cloud-based cyber-physical systems. Through two case studies, we show the efficacy of the proposed method for accurately predicting end-to-end latency of offloading imaging tasks to the cloud. The first case study showcases the application of design patterns in a practical scenario involving a cloud-based healthcare system—a collaboration with an industry partner in healthcare systems. The second case study has been implemented as a prototype to show the use of calibration and validation techniques with operational data in a laboratory context using an image-reconstruction application.

Keywords

• Bayesian calibration
• cloud-based cyber-physical systems
• design patterns
• hybrid modeling
• model validation
• trustworthy performance models