Data-driven subsystem synthesis method for multibody dynamics

Owing to their increasing size and complexity, applying machine learning to multibody dynamic systems becomes challenging owing to training data scarcity and prohibitive computational costs. To overcome these challenges, we employ the subsystem synthesis method that divides the system into components that can be analyzed independently for efficient synthesis into a complete system model. Specifically, we propose a data-driven subsystem synthesis method, in which each subsystem is replaced with a subsystem-wise deep neural network metamodel trained on numerical data generated using the original subsystem synthesis method. Subsystem modeling reduces the data and training requirements for machine learning while preserving the modularity and physical interpretability of the overall multibody dynamic system. The proposed method is validated through numerical simulations of representative multibody systems, including cart-pendulum, planar half-car, and three-dimensional full-car models. The results confirm that the proposed method provides computational efficiency while maintaining a certain level of accuracy, demonstrating its potential for scalable and reusable simulation of multibody systems.