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

Artificial intelligence (AI) is used to interpret vibration and dynamic-response measurements for bridge nondestructive testing (NDT) and structural health monitoring (SHM), but accuracy on a dataset does not show that a method is ready for use on real bridges. Searches of Scopus, Web of Science, and IEEE Xplore in August 2026 and a recall audit yielded 794 journal studies, each coded for diagnostic capability, validation realism, damage authenticity, generalisation, temporal maturity, engineering-use maturity, and five trustworthiness attributes. Enabling inference and detection account for 67.8% of studies and prognosis for 0.6%. Numerical validation is the largest category (45.5%); in-service validation covers 30.7%. Confirmed full-scale physical change appears in 61 studies, but only six (0.8%) combine it with in-service monitoring. Transfer to an unseen bridge appears in 16 studies (2.0%), year-scale evidence in 20 (2.5%), and maintenance decision support in one (0.1%); 65.2% show none of the trustworthiness attributes. These patterns held across ten sensitivity scenarios, including full-text verification of 52 studies. Agreement between the original screening and an independent human screening of 350 records was moderate (κ = 0.43), so some eligible studies may be missing. Diagnostic capability is broad, but the evidence needed for deployment remains scarce.