Concrete Fault Detection Using Deep Learning: Towards Waste Reduction in Bridge Inspection

Bridge inspections across the United States represent a significant commitment of resources, given their biannual mandate for many structures. With a notable number of aging bridges, there's an imperative need to enhance the efficiency of these inspections. This study harnessed the power of computer vision, integrating it with the principles of waste reduction from Lean manufacturing, to streamline the inspection process. We examined the efficacy of 15 distinct deep learning Convolutional Neural Network (CNN) models, including well-known architectures like Inception, VGG16, Xception, LeNet, and a tailor-made CNN. These models were applied to a data set of 40,000 high-quality images from the Concrete Images for Classification dataset. With their heavy reliance on time and resources, traditional fault detection methods often fall short. In contrast, the deployed models, with accuracy rates reaching up to 99%, showcase a significant potential in promptly identifying concrete anomalies, enhancing safety, conserving resources, and potentially saving lives. VggNet proved the fastest (3.5 h), and LeNet the slowest (100.10 h. Considering their similar performance in other metrics, VggNet stands out as the preferred model for concrete inspection. The custom CNN model also performed well in comparison.