基于深度学习的复合材料微结构-性能的构效关系研究OA

The"material-structure integration"characteristic of composites results in significant uncertainty in their microstructure,posing challenges for their reliable application in high-safety-standard structures such as those in aerospace engineering.Based on material informatics approach,this paper develops a data-driven predictive model.To this end,the microstructure of unidirectional carbon fiber composites is generated using a random collision algorithm,and the corresponding elastic and shear moduli were calculated by the FEM of micromechanics.Then,a reduced-order data representation of the microstructure was determined using two-point spatial correlation functions and principal component analysis.Finally,based on the database of the digital expression of the microstructure and the corresponding transverse mechanical properties,the machine learning model is used to capture the potential nonlinear relationship between the microstructure and the transverse elastic properties,so that the transverse mechanical properties of composites can be quickly predicted through the microstructure.Furthermore,through the characteristic analysis on the digital expression of the microstructure,the quantitative microstructural characteristics that have a great impact on the mechanical properties of composites were obtained.Analysis of digital microstructural representations reveals that,given a fixed fiber volume fraction,the average fiber spacing is a critical quantitative feature affecting the mechanical behavior of composites.It is of great significance for both forward performance prediction and inverse design optimization.