A machine learning-based survey strategy for the safety assessment of masonry churches based on prior damage

Among heritage structures, masonry churches are one of the most vulnerable building types to earthquakes. Observations of seismic damage reveal that these structures typically fail through various localised mechanisms rather than a global structural collapse. In post-seismic emergency scenarios, accurately linking observed damage to the associated failure mechanisms is crucial for planning practical retrofitting priorities and reducing risk. However, this process is currently performed manually and is prone to individual biases and inconsistencies. Additionally, the often limited knowledge available about the structure makes the exact identification of failure mechanisms difficult. To address these challenges, a digital tool capable of identifying the likely failure mechanisms based on observed crack patterns is proposed. By utilising a decision-tree-based algorithm, the most probable failure mechanisms are automatically estimated based on the available crack pattern information. In cases of limited knowledge, the algorithm can also suggest targeted survey actions to improve the accuracy of the assessment, thereby maximising the effectiveness of post-seismic evaluations