Research on Multi-Scale High-Fidelity Digital Model and Reduced-Order Model for Digital Twin Modeling of Composite Structures
Abstract
The research purpose of this paper is to establish a multi-scale high-fidelity digital model and a reduced-order model for digital twin of composite structures. Firstly, a high-fidelity digital model is established for the composite structure using micro-meso-macro multi-scale method to accurately describe the composite structure. Then, in order to meet the real-time needs of digital twin, the reduced-order model is studied by using three training algorithms for comparative study: Bayesian Regularization (B-R), Levenberg-Marquardt (L-M), and Scaled Conjugate Gradient (SCG). B-R training algorithm with smaller error is chosen to establish a neural network reduced-order model. Finally, with the multi-scale high-fidelity digital model as a reference, the maximum error of the established reduced-order model is 5.64%, which can meet the needs for multi-scale digital twin modeling of composite structures.