A hybrid data-driven model order reduction strategy for flexible multibody systems considering impact and friction. (March 2022)

• Record Type: Journal Article
• Authors: Song, Ningning; Peng, Haijun; Kan, Ziyun
• Is Part Of: Mechanism and machine theory. Volume 169(2022)
• Publication Date: 2022-03
• Subjects: Hybrid model order reduction -- Impact and friction -- Flexible multibody systems -- Nonsmooth contact method -- Parameter interpolation -- Data-driven method; Machine theory -- Periodicals; Machinery -- Periodicals; Machines -- Périodiques; Génie mécanique -- Périodiques; Machine theory; Machinery; Periodicals; 621.81
• DOI: 10.1016/j.mechmachtheory.2021.104649
• Languages: English
• ISSNs: 0094-114X

Abstract

Highlights: A novel hybrid model order reduction method is proposed for contact flexible multibody system. A proper symplectic decomposition strategy is proposed for flexible multibody system. The generalization ability of reduced-order model is studied. Abstract: The numerical simulation of flexible multibody systems (FMBSs) with contact is important. Meanwhile, the contact problem is always a key and difficult point for FMBSs; and when impact and friction affect FMBSs, the computational efficiency will dramatically decline. In this work, the nonsmooth contact method (NSCM) for FMBSs is briefly introduced. Moreover, in order to save the computational costs of a contact FMBS, a novel hybrid model order reduction (MOR) strategy based on the idea of data-driven method is proposed. For the hybrid MOR strategy, the reduced-order models (ROMs) based on the NSCM in smooth force part and nonsmooth force part are derived using proper symplectic decomposition and proper orthogonal decomposition, respectively. In addition, in order to adapt the ROMs to parameter variation, the parameter interpolation method on the tangent space of Grassmann manifold is studied. Finally, several numerical examples are tested using the proposed methodology, the results show that compared with the original model, the ROM can save 65.89% of the computing time, and the root-mean-square error between the ROM and original model can reach the order of 10 −12 .