Interactive visualisation: A support for system identification

Due to the complex and composite nature of the material constituents that the masonry is composed of, accurate analytical modelling of masonry wall panels has proved to be extremely difficult. In practice, engineers are using empirical models such as the yield line or strip analyses methods to predict the response of masonry panels to lateral loads. In research, specialised finite element analysis (FEA) tools have been used, which is mostly limited to a smeared material model over the entire surface of the panel. This is an over simplification of the true behaviour of the masonry as a composite material. In order to overcome this problem, previous research at the University of Plymouth resulted in the development of a numerical model updating process in which ‘corrector factors’ are derived by minimising the error between the theoretical and analytical load deflection relationships over the entire surface of the panel. These corrector factors are used to modify the flexure rigidity of various zones within the panel for use in a non-linear FEA process. This technique has proved to be effective in predicting the response of the masonry panels to lateral loads, which was comparable with the experimental results. However, two key drawbacks of this method were the complexity of deriving corrector factors for panels of different sizes and for panels with and without openings. Additionally, due to the inverse problem nature of the model updating process, a number of scenarios could result in similar error. Initial investigation into the pattern of corrector factors revealed that panel boundaries had a major influence on the behaviour of the panels. Hence, this led the research team to investigate modelling panel boundaries as springs without changing the material properties. The authors used an interactive visualisation clustering genetic algorithm tool (IVCGA). The search engine of the IVCGA uses the GA to derive values for spring constants along the boundaries of the panel and the visualisation tools within the IVCGA identifies clusters of good solutions from the data generated by the GA. This process is then repeated for each objective independently and a set of clusters that are mutually inclusive for all objectives are identified. The paper demonstrates that further exploration of solutions inside this mutually inclusive region can lead to identification of solutions that perform much better than the previous approaches and this could result in finding the correct model for the problem. This new approach of has successfully overcome the drawbacks of the previous research.