Quantitative Analysis of Crack Growth and Severity in Reinforced Concrete Structures Using Deep Learning and Computer Vision

The accurate and objective assessment of reinforced concrete structures is paramount for maintaining structural integrity and optimizing long-term maintenance planning. This study introduces a unified deep learning and computer vision framework designed for the automated detection, classification, and standards-aligned quantitative analysis of concrete cracks. The methodology begins with the automated categorization of an approximately 7,000-image concrete surface dataset into seven specific defect types including Thermal, Serviceability, and Strength Failure Cracks based on geometric metrics like crack length and width. This automated pre-classification step successfully mitigates the subjectivity and inconsistency associated with traditional manual labeling, providing a robust foundation for model training. A Convolutional Neural Network (CNN), implemented using Python, TensorFlow, and Keras, was trained over 50 epochs to detect and classify these categorized defects. The model achieved a final classification accuracy of 91.1%, demonstrating strong generalization and outperforming models trained on unrefined datasets. Following detection, a quantitative damage measurement module utilizes Otsu thresholding, morphological filtering, and skeletonization to precisely extract geometric parameters. Automated functions estimated key crack metrics, including length (5–180mm) and width (0.2–4.5mm), and surface deterioration percentage. These measurements are used to assign a severity grade (minor, moderate, or severe), aligned with established ACI 224R-01 and ACI 318-19 guidelines. Visualization techniques, such as severity-based color coding and multi-panel views, enhance the interpretability and validate both the detection accuracy and measurement reliability. By integrating automated data refinement, CNN-based recognition, and objective standards-aligned quantitative assessment, this framework provides a scalable and reliable tool for real-time structural health monitoring.