Investigation of blast actions on masonry structurers using non-linear finite element analysis and deep learning techniques.

Abstract

Over the past decades, the investigation of masonry behaviour under in-plane and out-of-plane response has been an improving area of research. This thesis presents the findings from nonlinear finite element analysis of masonry walls. The blast loading’s effect on masonry structures has been investigated as part of this study using finite element analysis. Low-cost housing within the close proximity of mining operations have been heavily impacted by the effect of blast loading and this research intends at investigating the effect of blast loading on masonry walls. Non-linear dynamic finite element analysis was used to create the structural model. Regarding modelling approach that is adopted in this research, the masonry units are taken as continuum elements and the mortar joints as interface elements. The simplified micro modelling is used by defining the block-to-block interaction properties (mortar is not modelled). Through the use of friction interfaces and unilateral contact, zero tensile resistance between the joints is achieved. Utilizing the concrete damage plasticity model, the prediction of tensile and compressive failure of blocks was achieved. In terms of blast application, this research considered the surface blast and the different blast loading factors are considered which included the effect of standoff distance, explosive type and weight. The explicit dynamic solver is used in this research which adopts very small-time steps, that makes it appropriate for blast loading. Moreover, the small-time increments offered by the explicit solver are suitable for solving complex contact problems. The first study investigated the behaviour of concrete brick walls under blast loading with varying standoff distance of 20m, 50m and 100m respectively. As part of the parametric investigating, the blast weight varied from 100kg TNT, 200kg TNT and 1150kg TNT for solid walls. For walls with opening, blast weights were increased to 2000kg TNT, and 3500kg TNT. According to the results, the effect of openings on walls proved to reduce the severity of damage on the walls. The orientation of bricks in masonry wall construction is regarded as one of the aspects that architects consider vital for the aesthetics of the walls. However, the influence of brick patterns has not yet been thoroughly investigated. The second aspect of this research investigated the different bonding patterns which included the stacker bond, stretcher bond and English bond. It was observed that the failure modes under out-of-plane response among these different bonds did not differ significantly, however, the different bonding pattern proved to have a significant influence on the response of the wall under horizontal loading (in-plane loading). The introduction of machine learning methods for a fast-accurate prediction of the masonry wall response is achieved in this thesis as one of the innovations of this research. When compared to conventional numerical simulations, the machine learning methods offers a significantly reduced computational cost. Machine Learning (ML) approaches in the field of structural engineering is one aspect that is still being investigated by various researchers, however, the blast effect on masonry using these approaches is still not fully developed. As part of the deep learning technique, this study investigated the adoption of Artificial Neural Network (ANN). The dataset for ANN included the numerically generated results from commercial finite element software (ABAQUS) using varying blast weights and standoff distances. This process involved linking Python coding and MATLAB programming code to automatically generate these results without having to manually open the commercial finite element software. The accuracy levels that were obtained from the ANN models were in the acceptable range.