This study proposes a procedure for correcting the dynamic response of nonlinear structures using sensor data obtained at specified locations. Simulation-based time-series prediction can reduce the online computational cost; however, the predicted response inevitably includes model-discrepancy errors due to variations in the actual structure’s material properties and boundary conditions. Hence, we introduce a simulation-response updating procedure that leverages the residual between the measured and predicted displacement responses acquired during operation. A transformer-based time-series model is used to approximate the displacement response, and the residual information in the prediction interval is exploited to correct the nonlinear displacement response. The proposed approach is validated through numerical examples of a cantilever beam and box-shaped beam.
Park et al. (Thu,) studied this question.