Bridge Health Monitoring (BHM) systems generate nonstationary time-series data that pose challenges for accurate structural state prediction. This study proposes a novel neural network-based method for predicting bridge states, the Temporal Multi-Scale Convolutional Neural Network (T-MSCNN) to enhance the prediction of dynamic strain parameters, which directly reflect structural stress states. The T-MSCNN integrates multi-scale convolutional layers for local feature extraction and gated recurrent units (GRUs), by using the Convolutional Neural Network (CNN) model and the Gated Recurrent Unit (GRU) model to address the intricacies of the nonstationary BHM data. Validated with real strain data from the Qijiang River Bridge in China, the model demonstrated superior performance over traditional models (HA, ARIMA, SVR) and standalone deep learning models (CNN, GRU), achieving reductions in prediction error by Root Mean Square Error (RMSE) method—up to 77.7%, compared to the ARIMA model and consistently improving even over the strong GRU baseline. The perturbation analysis confirms its robustness under noise interference. The T-MSCNN provides a reliable data-driven framework for structural health diagnostics, with potential applicability to other fields involving nonlinear spatiotemporal data analysis.

Welding repair is often unavoidable when steel structures in seismic regions are damaged, yet how it affects the cyclic performance of ultra-high-strength steel (UHSS) members is still not well understood. In this study, the hysteretic behaviour and failure modes of Q1100 UHSS welded box-section beam-columns with extended flanges were examined experimentally, considering several weld-repair histories and strengthening measures. Ten test schemes from the SA, SB and SC series were subjected to a constant axial compression together with quasi-static cyclic lateral loading. The variables covered the number of repair passes (0–3), the repair technique, the axial load ratio (0.057–0.216), the steel grade (Q1100 versus Q345), the boundary restraint and the cross-sectional geometry. The Q1100 specimens failed in a fracture-dominated mode that began in the lower plate/weld region, whereas the Q345 specimen failed by buckling followed by tearing. In the SA and SB series, repeated repair tended to reduce the hysteretic energy dissipation and stiffness and to give asymmetric responses; these trends are consistent with cumulative welding damage, although the damage itself was not characterized. Appropriate strengthening combined with lateral restraint either restored or improved the deformation capacity, and the ultimate drift ratios lay between 2% and 9%. The skeleton curves ranged from an almost linear-elastic shape to an S-shape, depending on ductility, axial load and restraint. Extended flanges improved the lateral resistance, while unequal flange widths produced an asymmetric response. When checked against JGJ/T 483-2020, several Q1100 specimens satisfied the strength criterion but exceeded the in-plane or out-of-plane stability limits, which suggests that stability governed a number of the repaired schemes. No metallographic, hardness, residual-stress or fractographic analysis was carried out, so the damage mechanisms proposed here remain interpretive. Even so, these observations offer preliminary evidence for post-earthquake repair strategies and for design checks of Q1100 UHSS members of the type tested.