Sparse-measurement-constrained full-field stress updating of simulation-trained neural operators for aerospace structures

Accurate full-field stress prediction for real aerospace load-bearing structures is challenging because finite element (FE) simulations provide dense but biased stress fields, whereas experimental measurements are reliable but sparse. This study proposes a sparse-measurement-constrained full-field stress updating framework for simulation-trained neural operators. A fixed-mesh Fourier neural operator is first trained on FE-generated von Mises stress fields to provide a simulation-informed full-field prior. Sparse sensor measurements are then used to diagnose structured simulation-to-reality discrepancies and guide lightweight updating without retraining the surrogate. For a single-rod component with load-level-dependent discrepancy, a gated affine update selectively corrects low-load errors while preserving reliable medium- and high-load predictions. For a double-rod connected component with spatially heterogeneous discrepancy, a regional inverse distance weighting update propagates sensor residuals to the full field according to structural topology. Validation on two aerospace engine-frame components shows that the proposed updates improve sensor-level consistency and preserve physically meaningful stress-field patterns. In leave-one-sensor-out validation, the Case-1 MAPE is reduced from 6.109% to 4.113%, and the Case-2 MAPE is reduced from 24.328% to 18.022%. The framework provides a practical route for measurement-constrained digital-twin stress evaluation under sparse sensing conditions.