Date of Award

Summer 2022

Project Type


Program or Major

Civil Engineering

Degree Name

Master of Science

First Advisor

Yashar E Azam

Second Advisor

Erin S Bell

Third Advisor

Fei Han


This study investigates the efficiency of Convolutional Neural Networks (CNNs) to detect structural damage from measured structural response. In this regard, strain gages were used to measure structural response from live loads. Data used during experimentation was gathered through the testing of a full-scale concrete bridge mockup. The test captured the response of the mock-up bridge utilizing 24 transversely oriented resistance-based strain gages under similar loading conditions, but with different levels of damage induced. There are three levels of damage the bridge was subjected to; crash-induced damage to the barrier, transverse cut on the entire barrier, and transverse cut along the deck. Live load testing was done for undamaged and damaged conditions by running vehicles at various speeds over the deck and recording the response overall strain gages. Previous studies have compared the effectiveness of Singular Value Decomposition (SVD) and Independent Component Analysis (ICA) to apply a novelty index to the same data to detect damage. This study seeks to expand on this investigation by utilizing the data as a full snapshot matrix converted into a 2D greyscale image to classify frames as Damaged or Undamaged through a Supervised approach. The supervised CNN uses two convolutional layers with dropout and a fully connected output layer and reached an accuracy of 95% with only 30% of the total dataset used as training data. This study shows that CNNs provide a robust way of detecting damage from full data snapshots represented via greyscale images.