Study on the Applicability of Deterioration Detection Techniques for Sandstone Heritage in Multi Environment using MLP-Attention Model

Abstract

Stone cultural heritage, encompassing a broad spectrum of artifacts such as stone artworks, buildings, tools, and utensils, represents one of the most significant categories of cultural heritage. However, the conservation of these cultural heritage faces challenges from the process of deterioration. This degradation not only compromises the structural integrity of the heritage but also results in the loss of invaluable historical information. Thus, there emerges a critical demand for effective methods to detect and assess the condition of stone cultural heritage, enabling timely and precise conservation interventions. Here we obtained sandstone samples under different deteriorative environments through laboratory-simulated deterioration experiments and employed a variety of detection technologies to capture a series of changes in the deterioration detection parameters during the sandstone deterioration process. Subsequently, a deep learning model was established to correlate the detection parameters with the degree of stone deterioration. A SHAP analysis was then conducted to determine the contribution of various parameters to the degree of stone weathering under different experimental environments, providing recommendations for selecting appropriate detection technologies and indicators adapted to different deterioration environments. To further analyze the deterioration processes of the stone samples, XRD analysis was conducted to observe changes in mineral composition throughout the deterioration process. SEM images were utilized to examine the changes in micro-morphology and the internal pore structure associated with deterioration. This study provides a basis for the scientific design of deterioration detection schemes by selecting the most suitable testing technology for optimal deterioration assessment under specific environmental conditions.