Inverse analysis of defects in CFRP specimen with graph neural network using stress distribution of homogenized FEM
Please login to view abstract download link
In this study, for a complex-shaped carbon-fiber-reinforced plastic (CFRP), a machine learning model is used to predict the three-dimensional information of defects from the sum of principal stresses at the surface (DSPSS) calculated from the finite element method (FEM). The reason for using DSPSS in this study is that it is intended to perform inverse estimation of defects using infrared stress measurements. CFRP specimens are made in different microstructures and lamination patterns. Here, a prosthetic leg with a curved surface is used as the analysis model. A CFRP prosthetic leg is characterized by the lamination of prepregs with multiple microstructures, such as plain weave, twill weave, and unidirectional in practical use. Homogenized finite element analysis is performed on the aforementioned microstructures used in the CFRP prosthetic leg, and a macroscopic finite element analysis is conducted to accurately reproduce DSPSS obtained from infrared stress measurements. A graph neural network is used to predict the three-dimensional structure of the defects inside the prosthetic leg based on the DSPSS of the prosthetic leg analyzed by FEM. As defects, contamination caused by the failure to remove the protective sheet of the prepreg during the manufacturing process of the CFRP structure is considered.
