Scholar: Zaid Alnimri
When: May 20, 2026, at 1:00PM
Where: KH A549
Zoom: https://unl.zoom.us/j/95279199519
Title: SCOUR DETECTION AND REGIONALIZATION USING SUPERSTRUCTURE DATA
Abstract: Scour is a leading cause of bridge failure worldwide. Traditional inspection methods are costly and impractical, while vibration-based approaches often struggle in short, stiff bridges and cannot clearly distinguish between scour and superstructure damage. This study presents a strain-based, unsupervised framework that uses only superstructure data to detect damage, identify the side of the bridge where it occurs, and classify its origin. Principal Component Analysis (PCA) covariance features were used for damage detection, reconstruction error for regionalization, and principal component re-ordering through a Gated Recurrent Unit (GRU)–Autoencoder framework to distinguish between superstructure and substructure (scour) damage. The approach was validated using field tests on two full-scale single-span bridges and finite element models of existing two-span and three-span bridges to study the effect of continuity, representing scour-critical bridges in Nebraska, subjected to controlled superstructure damage, simulated scour, and approach slab degradation. A Convolutional Neural Network (CNN)–Transformer–Autoencoder was also proposed and validated on the same bridges for comparison.