Integrated Multi-Sensor Data for Urban Infrastructure Health Monitoring Based on Digital Twin Structural Degradation Modeling

In response to the problem of degraded modeling accuracy caused by the mismatch between multi-source monitoring data and the spatiotemporal scale of digital twins, this paper proposes a digital twin degradation modeling method that integrates multi-sensor tensor representation and physical constraint online evolution: a cross modal tensor is constructed by grouping response mechanisms and anchoring resampling degradation events, and a latent space representation with degradation order relationship is established through dual-channel contrastive encoding and monotonic regularization; The twin state is layered into slowly varying degradation parameters and rapidly varying response states, guided by a mapping network prior, updated with physical residuals in a closed-loop manner, and introduced adaptive forgetting factors and trusted domain constraints; Furthermore, the degradation is described using a time-varying drift generalized stochastic process, and sufficient conditions for identifiability are given. The degradation model parameters are extended to the state space for joint filtering and verification of prediction consistency. The distance between anchor points in the acceleration phase is compressed to about one-third of the steady phase, and the hidden space trajectory advances unidirectionally. Two types of validation-case experiments show that the separation degree and monotonicity of the fused representation surpass the optimal single-mode baseline, and the tracking error is significantly lower than that of the fixed twin; The parameter estimation error of the complete method has been reduced to 48% of the baseline, and the root mean square error of life prediction has been reduced to 52%. The prediction advantage relative to non-twin reference is about 2.3 times that of its parameter estimation advantage; The attribution shows that the contributions of the alignment, following, and self-consistency stages are 0.52, 0.28, and 0.11, respectively. The interaction effect is positive at 0.09. The instantaneous contribution of the acceleration stage of the tracking stage increases from 0.06 to 0.62 and synchronously rises with the alignment stage, verifying that the overall gain exceeds simple superposition.