Uncertainty-Aware Distributed Damage Diagnosis with Vehicle-Induced Long-Gauge Strain Transmissibility Functions

This chapter presents a damage diagnosis method for beam-like structures using vehicle-induced long-gauge strain transmissibility functions, leveraging their moving-load-independent properties and direct sensitivity to localized stiffness changes. Defined as the ratio of Fourier-transformed strain responses between target and reference sensors under moving loads, the long-gauge strain transmissibility functions are proven to be independent of input spectra and proportional to the bending stiffness ratio of monitored zones. To exploit these attributes, a distributed damage diagnosis strategy decomposes global structural assessment into element-level sub-tasks via a parallel-disentangled variational autoencoder-based regressor. Each sub-task links transmissibility patterns to latent representations of damage extent within a Bayesian framework, enabling concurrent damage quantification and uncertainty estimation. By training exclusively on single-element damage scenarios, the approach eliminates reliance on complex multi-damage datasets, streamlining network efficiency. A divide-and-conquer architecture further reduces labeled data requirements, as only numerical models with single-element perturbations are required. Validated through numerical study and experimental tests on continuous beams, the method demonstrates robustness in pinpointing damage locations and severities while minimizing computational cost. The integration of physics of transmissibility properties, parallelized deep learning, and Bayesian uncertainty quantification advances structural damage diagnosis by balancing interpretability, scalability, and operational practicality for engineering applications.