Data-model-co-driven approach for real-time reconstruction of unmeasured responses via multi-fidelity surrogate modeling

The data-model-co-driven approach offers a promising solution for real-time reconstruction of structural responses without prior measurements, exemplified by unmeasured responses. However, the accuracy of existing co-driven approaches is highly dependent on the fidelity of the Finite Element (FE) model, which limits their practical applicability in Structural Health Monitoring (SHM). To tackle this issue, this study proposes a co-driven framework based on multi-fidelity surrogate modeling. In the proposed framework, a Low-Fidelity (LF) model substitutes the FE model to estimate the full-field structural responses, while a High-Fidelity (HF) model learns the error mechanism between FE-simulated and measured data and subsequently improves the accuracy of all LF outputs. The proposed approach is validated through the full-field reconstruction of cable forces in a cable net structure. For three cable force locations assumed to be unmeasured, an investigation into the influence of the correlation between LF outputs and measurements on reconstruction accuracy reveals that the proposed approach is not strictly dependent on the fidelity of the FE-simulated data. Meanwhile, the proposed approach achieves a maximum Root Mean Square Error (RMSE) of 17.93 kN at these measurement points, compared with RMSE values exceeding 38 kN obtained using an existing co-driven approach. Collectively, these findings demonstrate that the proposed approach enables real-time reconstruction of unmeasured structural responses with acceptable accuracy, while reducing its dependence on the fidelity of the FE model.