Can AI help us inspect masonry structures without tearing them apart?
The Reincarnate “AI-Based Non-Destructive Crack Detection in Masonry Structures” demonstration led by our coordinator TU Berlin explores how artificial intelligence can support non-destructive inspection through automated crack detection. Focusing on image-based analysis, the study investigates how machine learning can improve the consistency and scalability of structural condition assessments, reducing reliance on manual visual inspections.
Check the one pager
The approach combines computer vision techniques with both traditional machine learning and deep learning models. Using a custom dataset of masonry surface images, researchers manually annotated crack patterns and trained different models to identify and segment structural defects. While Random Forest and XGBoost were tested on handcrafted image features, convolutional neural networks (CNNs) and UNet architectures were used to learn crack geometries directly from image data.
Check the three pager
The results show a clear performance advantage for deep learning approaches. CNN-based classification models achieved crack detection accuracies of up to 89%, while the UNet segmentation model reached an Intersection over Union score of approximately 0.81 and a Dice coefficient of around 0.89. These results demonstrate the system’s ability to accurately delineate fine crack patterns and distinguish damaged from non-damaged surfaces with a high level of reliability.
By enabling reproducible and data-driven condition assessments, the demonstration contributes to more informed maintenance planning and lifecycle management of existing buildings. The methodology also provides a foundation for integrating AI-supported inspection into future digital construction workflows within the Reincarnate framework.
Read the full report