Developing a deep learning algorithm that can perform three dimensional object identification and real time finite element analysis by volumetric pixelation method

Finite element analysis is a simulation method which is designed in a computer environment to determine the physical behavior of structures and solved with certain formulations. Depending on the size or complexity of the problem, preparation and solution time increases. The requirements for finite element analysis are geometry and boundary conditions. The FEA is difficult and tiring because the preparatory phase requires too much input. In addition, advanced engineering knowledge is required to perform FEA. The increase in the use of three-dimensional printers has increased the need for FEA. With the development of the processing power of mobile phones and tablets, an infrastructure was established to enable the feasibility of finite element analysis with telephone camera without the need for input. Deep learning algorithms are designed and trained in order to perform simple analysis automatically. The solutions of artificial intelligence in all areas of our lives in the field of finite element analysis have been investigated and different algorithms have been written for the necessary methods. In this study, an algorithm that identifies the material to be analyzed and classifies its material has been created by the image processing method and has worked with high success rate. A customized convolutional neural network was designed to classify three-dimensional geometries. In order to teach three dimensional geometries to neural networks, solid pixelation method was used and neural network was successfully trained. In order to accelerate the finite element analysis and classify the geometries and provide input to the FEA, 3888 different analysis results were trained in artificial neural networks and accuracy rates were determined. Finite element analysis showed that the problems with a solution time exceeding 48 hours yielded 91% accuracy in 10 seconds with artificial neural network.