Laxman KC

Date of Award

Summer 2023

Document Type

Open Access Dissertation


Civil and Environmental Engineering

First Advisor

Paul Ziehl


In South Carolina, the South Carolina Department of Transportation (SCDOT) manages 90% of the state's inventory of 9,400 bridges built in the 1950s. With more than 30% of its bridges having been under-designed and experiencing decades of service deterioration, the implications for public safety, the economy, and the transportation system are significant. To prioritize safety and address these concerns, SCDOT is in the process of conducting thorough inspections of bridges, load ratings, and exploring cost-effective and practical methods to strengthen the bridges and reduce load postings. This study is a part of the multi-year research investigation supported by the SCDOT. There are two main objectives of this study: (1) to investigate methods to strengthen the slab bridges and reduce the number of load postings, and (2) develop deep learning models for an automated inspection for crack detection, estimation of crack spacing, and crack depth estimation. Two research studies were performed to accomplish the aforementioned goals.

The innovation of the first study lies in exploring methods for strengthening one-way RC precast slabs from above, considering factors such as increase in flexural strength and ductility to reduce the number of load postings, ease of implementation, and cost-effectiveness. Experimental tests were conducted in the lab on five slabs obtained from decommissioned bridges in South Carolina by subjecting them to monotonic loading until failure. One slab served as a reference, another was strengthened with steel channel sections from above, a third was strengthened with steel plate sections from above, and the remaining two slabs were joined using near surface mounted steel rebars for strengthening. The effectiveness of the proposed methods in increasing both the flexural strength and ductility of the slabs was assessed. Additionally, the cost associated with each method of strengthening was taken into consideration during the evaluation process. The results showed that strengthening methods from above can effectively improve the flexural strength and ductility of the slabs, while being more economical and easier to implement than other methods of strengthening.

In the second study, a comprehensive framework implementing deep learning models was developed for an automated inspection for crack detection, estimation of the spacing of the cracks to obtain a condition state for the determination of condition factor for load rating, and estimation of the crack depth. A binary Convolutional Neural Network (CNN) model was trained, validated, and tested using a dataset of images to successfully detect cracks. The model demonstrated high accuracy and reliability in detecting the cracks in images captured from an RC slab that was tested in the lab. A methodology that implements the pre-trained CNN models for an automated inspection of the cracks was modified to automatically estimate the spacing of the cracks. This facilitates the determination of the condition factor based on the condition state of the slab necessary for the load rating of the bridges. Furthermore, an integrated CNN model was developed, incorporating regression models to estimate the depth of the cracks. The images used for training the model were associated with their respective crack depth values, and CNN’s feature extraction layer was utilized to extract relevant features from the images. These extracted features were then employed to train, validate, and test the model for estimating the depths of cracks. Results indicated the models are accurate and reliable for automated inspection of the cracks, determination of the condition factor, and estimation of the crack depth, which could help in evaluating the condition of bridges, consistent load rating, and choosing suitable repair methods.

Available for download on Saturday, August 31, 2024