Real‐time detection and measurement of cracks in mortars containing waste <scp>PVC</scp> exposed to high temperatures using deep learning‐based <scp>YOLO</scp> models

Abstract Thermal expansion and shrinkage of mortar and concrete surfaces under the influence of high temperatures lead to fine and irregular cracks on the surface, weakening the internal structure of mortar and adversely affecting the long‐term performance. Such cracks allow water carrying chemicals to penetrate deep into the mortar, deteriorating the internal structure of mortar. Therefore, the detection of the cracks rapidly and accurately is critical for the safety of the structure. This study aims to detect and characterize cracks occurring on the surfaces of waste PVC containing mortars exposed to high temperatures using deep learning models. For this purpose, YOLOv5, YOLOv8, and YOLOv11 models, which are popular You Only Look Once (YOLO) models in real‐time applications, were used as deep learning models. The dataset consists of 5027 images captured from mortar surfaces exposed to high temperature at 800°C, and this dataset is split into 80% training, 10% validation, and 10% test sets. Using these datasets, all sub‐models (n, s, m, l, x) of YOLOv5, YOLOv8, and YOLOv11 models were trained separately. In addition, crack area, perimeter, and width of cracks were measured from the actual and predicted images using pixel‐millimeter transformation. The results showed that the x sub‐models were the most accurate YOLO models, and the model with the highest accuracy for these models was YOLOv11x with a mean Intersection‐over‐Union (mIoU) of 81.2%. Among the YOLO models, YOLOv5x was the best model for accuracy with 0.021 mm 2 area, 0.831 mm perimeter, and 0.0008 mm width differences between actual and predicted images. In the study, the average crack width measured from the mortar surface was 0.058 mm. In addition, the YOLOv5n model is the best model for real‐time applications with a segmentation speed of 27 ms.