On the Use of Simulation-Driven Machine Learning for Real-Time Damage Prognosis of Masonry Walls
Please login to view abstract download link
This study introduces a simulation-driven machine learning approach aimed at predicting the residual displacement capacity and stress increase in damaged masonry walls, with the crack pattern serving as the sole input. The method employs an accurate block-based numerical model to simulate damage in masonry walls under earthquake-type loads and settlements. The cumulative distribution of crack widths in the damaged wall is extracted and transformed into a discrete crack width exceedance curve. Each curve is then correlated with the residual displacement capacity and stress increase in critical regions of the wall, using as reference an undamaged state. The resulting simulation-generated dataset is used to train and validate machine learning models, including convolutional neural networks. This approach offers a cost-effective, real-time damage prognosis tool for masonry walls and structures under both static and seismic loading conditions.