Development of intelligent 3D solid modeler based on artificial intelligent technique

Abstract

As one of model representation schemes, the usage of solid model has been started since the early 1970s on representing correct engineering drawings. It is due to its characteristics that is unambiguous, complete and contains its own boundary. Since then, it becomes one of the important research fields and extensively used in many industries mostly in the areas of engineering design, architecture, and manufacturing. This thesis focused on two categories of research on solid model; reconstruction and representation. Since many researches are focused on reconstruction of multiple view images based on mathematical modeling and geometrical analysis, this research attempts to devise techniques or algorithms that are suitable for single view image and single sketch analysis. For that purpose, a new framework for solid model reconstruction and representation from given single view image in form of regular two-dimensional line drawing is presented. Affine transformation was used in the pre-processing stage of the framework that is in the data preparation and definition. The framework consists of neural network models for the reconstruction and the hybrid computing algorithm as representation scheme of the reconstructed solid. The reconstruction contains two categories namely deriving depth values and deriving hidden point while the representation is the combination of neural network and mathematical model. Four contributions presented in this thesis were a new framework for solid model reconstruction and representation, a new experimental data design in development of the neural network models, neural network models for solid model reconstruction, and hybrid algorithm in representing solid model. The neural network models and the hybrid algorithm have been tested and validated on three solid models namely cube, L-block and stair by using Matlab 7.14 software. The framework can be used as an alternative on the development of a sketch interpreter. This is to avoid the use of mathematical modeling in the reconstruction process and to combine neural network and mathematical modeling in representing solid model.