Development of Optimized Section-Geometries for 3d Printed Structures Using Axial Reactions

Abstract

This thesis presents the details of finite element (FE) modelling of three-dimensional printed concrete (3DPC) walls, parametric FE analysis of 3DPC walls, and various machine learning (ML) models to predict the ultimate axial load (fmax) and the displacement corresponding to the ultimate axial load (u-fmax) of 3DPC walls. Six different ML algorithms, namely random forest regressor (RFR), extra trees regressor (ETR), gradient boosting regressor (GBR), hist gradient boosting (HGB), light gradient boosting machine (LightGBM), and extreme gradient boosting (XGBoost), were used. After that, the average prediction (AvPred) value was used to obtain the accuracy rate of the arithmetic mean of the predicted outputs. Furthermore, a comprehensive database of 3DPC walls was compiled by selecting a set of physical and mechanical properties from the literature. Then, FE models were created using the results of these tests and fmax and u-fmax values were found for these walls using explicit dynamic analysis in ABAQUS. Thus, FE models were validated using the results of experimental tests and 61800 3DPC walls with various geometries, and five distinct cross-sections were evaluated. After this process, the obtained data was trained, and an ML tool was created to automatically calculate fmax and u-fmax values according to the concrete class, dimensions and cross-sectional area models of the wall in order to be helpful to the users. In addition, two different 3DPC buildings were designed by considering these five different wall models, and fmax and u-fmax values were calculated using the proposed ML tool. Eventually, the environmental, structural and economic effects of 3DPC buildings were examined. It was noted that among the developed wall models, the most suitable model for high wall thickness varies according to the structural framing plan, but the most suitable model for low wall thickness is the carved model.