Towards Artifacts-Based Behavior Modeling of Cyber-Physical Systems for Building Digital Twins

Behavior modeling of complex systems is a critical aspect in developing digital systems, such as Digital Twins (DT). It enables accurate representation of intricate dynamics and interactions, facilitating predictive capabilities that are essential for forecasting future states and behaviors. This paper presents a methodology for building predictive DTs of complex systems, addressing the critical challenge of behavior modeling. Our approach integrates data-driven techniques, including machine learning (ML), with physics-based or mathematical models, leveraging both real-time and historical data. We argue that this hybrid modeling strategy is essential for creating DTs capable of accurate prediction and “sliding forward in time” to visualize future system states. We propose a customizable framework that supports the development of a predictive maintenance module, covering physics/mathematics-based modeling, behavior learning, and uncertainty quantification. This framework is composed of programmable components derived from identified modeling artifacts, allowing for the seamless integration of diverse modeling techniques. Another key contribution of this work is the development of concrete modeling principles for designing predictive DTs, such as the principle of data-driven model calibration informed by physical constraints. We validate the effectiveness of our framework and principles through a case study that demonstrates the creation of a battery DT prototype. This is done for the battery of a laptop, utilizing real-time data collection and a discharge model to project battery discharge behavior. Our experiment highlights the potential of our approach for real-world applications in cyber-physical systems (CPS) by enabling proactive maintenance and improved system performance. Future work will extend this framework into a more sophisticated DT development environment following the artifactbased design presented here.