Abstract

This dataset, developed as part of the Horizon 2020 HOLAHERIS project, contains data, models and results related to a machine learning predictor for residual drift capacity in damaged masonry walls based on mechanically consistent crack patterns induced by external actions (earthquake-like loads and differential settlements). The machine learning predictor is trained through 100 crack patterns generated by an accurate block-based numerical model, and the related residual displacement capacity estimated by means of the numerical model within a pushover analysis framework. 12 additional cases, with different geometries, textures, axial load ratios and sizes, are also used to validate the approach a posteriori. Good predictions on masonry piers with features different from those used in the training data support the generalization potential of the proposed method. Accordingly, the training data set could be straightforwardly enlarged also by using numerical models for masonry (e.g., utilized in other research groups). The machine learning predictor is implemented within a Python code which is released in this dataset, together with the numerical models developed and examples of extractions of pushover curves and crack width cumulative distributions. All input and output data are collected within the same Python code.