abstract
- The present research involves the development of a computational framework for numerical analysis of large-scale masonry structures. The main challenges in the analysis of existing structures of historic or strategic importance are the lack of information on the macroscale behaviour of masonry and the high computational cost of mesoscale analysis. The former requires extensive experimental programs that are virtually impossible without significantly affecting the structural integrity of the existing buildings. The latter, i.e. the mesoscale approach, requires only information on the behaviour of masonry constituents; however, it cannot be applied to analysis of structures that span hundreds of meters due to the required computational effort. To address these challenges, first a simplified mesoscale framework was developed incorporating an embedded discontinuity approach to model discrete crack initiation and propagation through masonry constituents. The approach enables the use of a simple unstructured finite element mesh and is computationally accurate and efficient. The proposed framework was validated against various experimental investigations on small-scale masonry assemblages. The second part of the research involved the development of a macroscale formulation for modelling masonry as a continuum with an underlying microstructure that exhibits anisotropic deformation and orientation-dependent strength characteristics. The identification of the constitutive parameters for the macroscale model was accomplished by simulating a series of biaxial and uniaxial tension-compression tests on masonry panels at different orientation of bed joints. Since the macroscale strength properties are highly dependent on the arrangement of masonry constituents and their individual strength properties, changing any of these parameters requires a repetition of the entire numerical procedure. To address this challenge, a series of artificial neural networks was developed that can predict the macroscale strength and deformation properties of masonry based on the mechanical properties of constituents. In addition, another neural network was developed to assess the average orientation of macrocracks at the onset of failure at the macroscale. The results of the developed macroscale framework were compared with the mesoscale approach for analysis of a large-scale masonry wall.