An innovative image processing-based framework for the numerical modelling of cracked masonry structures

Abstract

A vital aspect when modelling the mechanical behaviour of existing masonry structures is the accuracy in which the geometry of the real structure is transferred in the numerical model. Commonly, the geometry of masonry is captured with traditional techniques (e.g. visual inspection and manual surveying methods), which are labour intensive and error-prone. Over the last ten years, advances in photogrammetry and image processing have started to change the building industry since it is possible to capture rapidly and remotely digital records of objects and features. Although limited work exists in detecting distinct features from masonry structures, up to now there is no automated procedure leading from image-based recording to their numerical modelling. To address this, an innovative framework, based on image-processing, has been developed that automatically extracts geometrical features from masonry structures (i.e. masonry units, mortar, existing cracks and pathologies, etc.) and generate the geometry for their advanced numerical modelling. The proposed watershed-based algorithm initially deconstructs the features of the segmentation, then reconstructs them in the form of shared vertices and edges, and finally converts them to scalable polylines. The polylines extracted are simplified using a contour generalisation procedure. The geometry of the masonry elements is further modified to facilitate the transition to a numerical modelling environment. The proposed framework is tested by comparing the numerical analysis results of an undamaged and a damaged masonry structures, using models generated through manual and the proposed algorithmic approaches. Although the methodology is demonstrated here for use in discrete element modelling, it can be applied to other computational approaches based on the simplified and detailed micro-modelling approach for evaluating the structural behaviour of masonry structures.