Data-driven prediction of subsystem dynamics in the explicit co-simulation of multibody systems

Co-simulation is an effective way to predict the dynamics of complex engineering setups, in which the main system is divided into subsystems. Each subsystem is integrated by a particular solver, and the coordination of these simulation units is coordinated by means of the exchange of limited amounts of information at specific points in time. This enables the implementation of modular computing environments, but it often introduces numerical errors in the solution that deteriorate the quality of the results and may eventually lead to the instability of the integration. Implicit co-simulation schemes remove these errors by iterating over each step. In explicit co-simulation setups this is not possible and alternative solutions are necessary. This work presents a data-driven approach to predict subsystem dynamics in explicit co-simulation setups, aimed at mitigating the energy errors introduced by the discrete-time co-simulation interface. The proposed solution only uses information contained in the coupling variables exchanged between subsystems and does not need knowledge of their internal details. The method has been tested with nonlinear and multirate benchmark problems. Results confirmed the ability of the proposed solution to remove numerical errors caused by the coupling interface and improve the accuracy and stability of the co-simulation of the overall system dynamics.