Data-Driven Model Order Reduction for real-time multibody simulations

Currently, multibody models are the state of the art to describe the dynamics of complex mechanical systems. However, due to the nonlinear characteristics and the relatively large number of Degrees of Freedom (DoFs), multibody simulations typically cannot be run in real-time. In this work, a novel data-driven local Model Order Reduction (MOR) approach is proposed, in order to perform most of the computations offline and obtain online real-time capabilities. Using displacement data either from experiments or simulations, a local Principal Component Analysis (PCA) is performed. The obtained Reduced Order Bases (ROBs) permit to locally reduce the system to a linear model and to constrain its motion without the necessity of additional equations. Thus, the dynamics can be linearly approximated with a reduced number of DoFs in a system of Ordinary Differential Equations (ODEs) that can be solved in real-time. An application example of the technique shows its increased velocity with respect to the solution of a full multibody simulation while maintaining results accuracy.status: accepte