Automatic Description of Rubble Masonry Geometries by Machine Learning Based Approach

The geometrical description of the components of rubble masonries constitutes a key-point in the definition of their mechanical response. A variational autoencoder (VAE) is proposed as a tool for the automatic description and generation of rubble masonry geometries. The encoder and the decoder forming the VAE are implemented by defining two convolutional neural networks trained by using binary images extracted from a publicly available masonry database.