Data-Driven Hybrid Modeling for Digital Twin of Large-Scale Structures With Local Nonlinearities

Abstract Digital twin (DT) modeling technology is the core for accurately portraying physical entities. It provides decision-makers and managers with real-time monitoring, simulation, and optimization capabilities, thus enhancing their understanding and control over complex systems. However, DT modeling techniques for local nonlinear contact structures in structural health monitoring have yet to be thoroughly investigated since repetition and redundancy in simulation processes in existing approaches. To address these issues, we propose a novel approach, called the data-driven hybrid modeling (DDHM) method, which can effectively settle contact nonlinear dynamic problems in structural health monitoring. This approach leverages a nonlinear force prediction model, modal reduction, and kernel functions to represent and analyze nonlinear dynamic structural behaviors efficiently. The DDHM method combines physics-based principles with data-driven modeling approaches to connect the physical and digital worlds and facilitate accurate and efficient analysis of intricate structural systems. To assess its effectiveness, the method is tested on two numerical examples: flat plates and telescopic boom. The findings demonstrate that the DDHM method achieves a lower online computational cost and satisfactory accuracy compared to both the finite element method (FEM) and traditional reduced-order models, thereby improving the computational efficiency in digital twin modeling of large-scale nonlinear structures.