我要吃瓜

Citation

Bogomolov S, Gomes C, Isasa C, Soudjani S, Stankaitis P & Wright T (2024) Reachability analysis of FMI models using data-driven dynamic sensitivity. SIMULATION. https://doi.org/10.1177/00375497241261409

Abstract
Digital twin is a technology that facilitates a real-time coupling of a cyber–physical system and its virtual representation. The technology is applicable to a variety of domains and facilitates more intelligent and dependable system design and operation, but it relies heavily on the existence of digital models that can be depended upon. In realistic systems, there is no single monolithic digital model of the system. Instead, the system is broken into subsystems, with models exported from different tools corresponding to each subsystem. In this paper, we focus on techniques that can be used for a black-box model, such as the ones implementing the Functional Mock-up Interface (FMI) standard, formal analysis, and verification. We propose two techniques for simulation-based reachability analysis of models. The first one is based on system dynamics, while the second one utilizes dynamic sensitivity analysis to improve the quality of the results. Our techniques employ simulations to obtain the model’s sensitivity with respect to the initial state (or model’s Lipschitz constant) which is then used to compute reachable states of the system. The approaches also provide probabilistic guarantees on the accuracy of the computed reachable sets that are based on simulations. Each technique requires different levels of information about the black-box system, allowing the readers to select the best technique according to the capabilities of the models. The validation experiments have demonstrated that our proposed algorithms compute accurate reachable sets of stable and unstable linear systems. The approach based on dynamic sensitivity provides an accurate and, with respect to system dimensions, more scalable approach, while the sampling-based method allows a flexible trade-off between accuracy and runtime cost. The validation results also show that our approaches are promising even when applied to nonlinear systems, especially, when applied to larger and more complex systems. The reproducibility package with code and data can be found at https://github.com/twright/FMI-Reachability-Reproducibility.

Keywords
Reachability analysis; digital twins; Functional Mock-up Interface; dynamic sensitivity equations; Lipschitz constant

Journal
SIMULATION

• Reachability analysis of FMI models using data-driven dynamic sensitivity