Multibody dynamics simulations (MBSs) are commonly used to analyze the dynamic behavior of complex mechanical systems as for instance boogies or rail vehicles. In particular, the complex mechanical system is modeled as composition of mass-spring sub-systems. This results in a system of differential equations which has to be solved. A large number of degrees of freedom (DOFs) and constraints involved in the complex mathematical model makes the MBS analysis computationally expensive. Suitable solvers have to be selected to obtain sufficiently accurate results.

It is of interest to explore new methods that can solve or even avoid the challenges of solving these complex MBS systems. In recent years, machine learning (ML) methods (probabilistic graphical models, Kalman filter, particle filter, deep learning) have been introduced to replace the original MBS model, reduce the calculation amount, avoid solver problems or to deal with noise interferences effectively in real applications. One of the main research aims is to provide reliable well-performing models at low computational costs to estimate the behavior of multibody system.