Artificial Intelligence-Driven Real-Time Restoration of a Cultural Heritage Analog Digital Twin Model

The preservation of cultural heritage structures is increasingly challenged by environmental degradation, aging, and human activities. This study presents an artificial intelligence (AI)-driven framework for real-time monitoring, crack detection, and restoration planning within a cultural heritage digital twin environment. The proposed system integrates image processing and predictive modeling techniques to identify structural deterioration and support timely restoration decisions while minimizing risks to heritage assets. The framework is trained and evaluated using the DHMuralVersion1 dataset, which combines high-quality images from four original sources to provide a comprehensive representation of heritage surface conditions. Experimental results demonstrate strong predictive performance, achieving a Mean Absolute Percentage Error (MAPE) of 1.39 in crack prediction, indicating high accuracy in detecting and forecasting crack development. The model effectively supports continuous condition assessment and restoration planning for heritage conservation. However, limitations were observed in binary mask generation, where segmentation errors occasionally occurred when applied to diverse datasets. Future work will focus on improving mask generation techniques, expanding the dataset with more varied heritage images, and incorporating environmental factors influencing deterioration. These enhancements are expected to improve model robustness and broaden its applicability in cultural heritage conservation and restoration.