Integrating Machine Learning Models as Functional Components within a Systems Engineering Design Using SysML for System Simulations

Model-Based Systems Engineering (MBSE) provides a structured approach to designing and analyzing complex systems using SysML-based models. However, traditional MBSE methodologies face challenges in integrating real-time data for simulations, particularly when direct connections to physical systems are impractical due to privacy, legal, or logistical constraints. To address this limitation, this work explores the integration of machine learning-based predictive models (PM) as functional components within SysML-based system simulations to enable predictive analytics without requiring real-time data from physical sources. The methodology proposed in this paper was applied to an illustrative example of an automated insulin delivery system, where the PMs were trained on the physical system's data and hosted on a local server. The system model (SM) designed in Magic System of Systems Architect (Magic SoS) tapped into the server via its simulation toolkit to receive the predictive values it needed to run system simulations. The results demonstrate a degree of success where PMs can replace real-time system data in system simulations while maintaining high accuracy. This research contributes to bridging the gap between MBSE and data-driven modeling, demonstrating that PMs can enable dynamic, data-driven analysis without requiring real-time physical system connectivity and can act as surrogates for physical system data within SysML-based simulations. This paper is intended as a resource for systems engineering practitioners.