Deep learning-based seismic damage assessment for structural health monitoring (SHM) faces the dual constraints of training-data scarcity, reflecting both the cumulative effort of generating diverse simulation benchmarks at portfolio scale and the limited availability of full-scale experimental records, and sensor noise. This paper systematically investigates whether the implicit regularization imposed by the unitary constraints of parameterized quantum circuits (PQCs) can alleviate these limitations. PQCs are integrated into three distinct stages of a graph neural network (GNN) pipeline (feature encoding, ensemble combination, and message passing) and evaluated in member-level seismic damage prediction of low- to mid-rise, two-dimensional planar reinforced concrete moment-resisting frames; three-dimensional effects (torsion, bidirectional excitation, out-of-plane behavior) are outside the present scope. A dataset of 2200 nonlinear time-history analyses was constructed by applying 44 real ground motions to 50 buildings, and eight models (three classical, two quantum-inspired, three PQC-based) were benchmarked. Under data-scarce conditions, QE-GNN achieved the lowest RMSE across all training fractions from 20% to 80%, with statistically significant superiority at 20–60% ( ). Under a unified inference protocol, QM-GNN achieved the lowest noise-induced degradation rate ( at ); the three PQC models occupied three of the top five positions, with the classical GATv2 placing 2nd, and all three PQC models surpassed MLP in absolute RMSE at realistic noise levels ( ). Ablation experiments replacing PQCs with parameter-equivalent classical MLPs showed consistent directional advantages of the PQC versions across all three models, providing preliminary evidence of a contribution from the quantum circuit structure. These advantages manifested specifically under realistic SHM conditions, while PQC models maintained comparable baseline performance under ideal conditions. Per-range analysis further confirmed QE-GNN’s advantage in the engineering-critical intermediate damage regime. All quantum circuits were executed on classical simulators; no claim of quantum computational advantage is made.
Lee et al. (Fri,) studied this question.