AI-Enabled Bridge Nondestructive Testing Using Vibration and Dynamic Response: An Investigation of Diagnostic Capability and Evidence Maturity

Artificial intelligence (AI) increasingly interprets vibration and dynamic-response measurements for bridge nondestructive testing (NDT) and structural health monitoring (SHM), yet predictive performance within a dataset does not establish diagnostic or operational maturity. This investigation searched Scopus, Web of Science Core Collection, and IEEE Xplore through 23 August 2026, screening 3,412 records and retaining 727 peer-reviewed studies as the systematic core. Unlike prior reviews organised by algorithm type, each study was coded at the level of the individual validation case for diagnostic capability (D0-D4), validation realism (R1-R4), damage authenticity (A0-A4), generalisation (G0-G3), temporal maturity (T0-T2), engineering-use maturity (M0-M3), and five independent trustworthiness attributes (E, U, X, P, F). The corpus is dominated by enabling inference and detection (D0-D1, 67.4%), whereas prognosis (D4) appears in only 6 studies (0.8%). Numerical validation predominates (R1, 48.1%), confirmed full-scale physical damage occurs in 61 studies (8.4%), and only 8 (1.1%) combine operational monitoring with confirmed damage. Cross-bridge transfer (2.1%), long-term monitoring (2.2%), decision support (0.4%), and explicit trustworthiness (none of the five attributes in 67.0%) are all rare, and only one study combines operational validation, confirmed damage, and verified transfer. The principal gap is therefore not AI capability but the scarcity of evidence where capability, realistic validation, authentic damage, transferability, durability, trustworthiness, and decision relevance converge.