Co-simulation of dynamic systems in parallel and serial model configurations

Abstract

Recent advancements in simulation software and computation hardware make it realizable to simulate complex dynamic systems comprised of multiple submodels developed in different modeling languages. The so-called co-simulation enables one to study various aspects of a complex dynamic system with heterogeneous submodels in a cost-effective manner. Among several different model configurations for co-simulation, synchronized parallel configuration is regarded to expedite the simulation process by simulating multiple submodels concurrently on a multicore processor. In this paper, computational accuracies as well as computation time are studied for three different co-simulation frameworks: integrated, serial, and parallel. For this purpose, analytical evaluations of the three different methods are made using the explicit Euler method and then they are applied to two-DOF mass-spring systems. The results show that while the parallel simulation configuration produces the same accurate results as the integrated configuration, results of the serial configuration show a slight deviation. It is also shown that the computation time can be reduced by running simulation in the parallel configuration. Therefore, it can be concluded that the synchronized parallel simulation methodology is the best for both simulation accuracy and timeefficiency.