A deep learning-driven mobile application for monitoring of tile degradation in heritage buildings

Heritage buildings are vulnerable to gradual deterioration, making continuous inspection and regular maintenance essential for preventive conservation. However, conventional inspection methods rely heavily on manual on-site fieldwork, specialized heritage knowledge, and skilled personnel. This study presents an artificial intelligence-powered mobile application for real-time, on-site monitoring of tile degradation in heritage buildings. The developed Android-based application integrates several workflows, including automated pixel-wise damage detection using the deep learning algorithm You Only Look Once (YOLOv11), panoramic façade damage analysis enhanced through close-up semantic segmentation, and a question-based module designed to assist inspectors in damage identification. The application was developed to detect tile defects (craters, glaze detachment, and cracks) in heritage buildings. A dataset of 4965 annotated images was used for model training and evaluation, achieving a mean average precision (mAP50) of 68.2% for bounding-box detection and 51.8% for semantic segmentation on the test dataset. The proposed framework provides an accessible tool for automated heritage tile inspection, and has been tested on several cultural heritage buildings in Portugal.