Using Artificial Intelligence Techniques to Predict the Behaviour of Masonry Panels

Laboratory experimental data is often erroneous. This error is more apparent in data obtained from testing of anisotropic composite materials such as masonry wall panels. In this paper data colleted from the laboratory tests of masonry panels is presented. Methodologies for reducing (correcting) error in laboratory tested data are discussed. The concept of stiffness/strength corrector to model the variation in masonry properties in laterally loaded masonry panels was introduced by Zhou [1] and Rafiq et al [2] to model variation in masonry properties. A cellular automata (CA) technique was used to model the boundary effect and establish stiffness/strength corrector values for unseen panels, using zone similarity techniques introduced by Zhou et al [3] These stiffness/strength correctors are then used in a non-linear finite element analysis (FEA) to predict the failure load and failure pattern of these unseen panels. This paper demonstrates that methodologies for reducing error in experimental data can further improve the predicted failure load of the panels.