An automated platform to detect, assess, and quantify deterioration in concrete structures

This study was motivated by the critical need for accurate and fast damage detection in concrete structures, benefiting from significant advantages offered by automated condition assessment strategies compared to manual ones. To move toward reducing inspection time, cost, and human error, the current study developed a deep convolutional neural network model tailored for detecting and quantifying deterioration in concrete structures. The model improved existing architectures by accommodating representative image resolutions and implementing a region-growing algorithm for precise defect quantification. To provide a holistic platform, this study established additional transfer learning and fine-tuning steps. Results showed the platform's capability to detect cracks as narrow as 0.5 mm, while defect dimensions were accurately quantified with errors under 3 %. The damage assessments were then linked to industry-level standards for structural inspections, providing condition states for various defects. This led to an end-to-end workflow for automated condition assessment, facilitating data-enabled maintenance and repair actions.