Geometrical digital twins of masonry structures for documentation and structural assessment using machine learning
Dimitrios Loverdos, Vasilis Sarhosis
Abstract
The generation of numerical models for masonry structures is a timely and costly procedure since it requires the discretization of a large quantity of smaller particles. Similarly, traditional visual inspection involves the cautious consideration of each element on a masonry construction. In both cases, each brick element needs to be considered individually. The work presented in this document aims to alleviate the issues arising from documenting individual masonry units and cracks on a structure using computer vision and convolutional neural networks (CNN). In particular, for the first time a dynamic workflow has been developed in which masonry units and cracks in masonry structures are automatically detected and used for the development of a complete geometric digital twin. The outcome is a collection of space coordinates and geometrical objects that represent the masonry fabric entity and allow the comprehension of the object for documentation and structural assessment. This interoperability between architectural, structural, and structural analysis models paves the way to use engineering to create a smarter, safer, and more sustainable future for our existing infrastructures.