Stiffness prediction and analysis of needled C/SiC composites considering defects: Based on the GMC and IPSO‐FNN

Abstract The complexity of manufacturing defects imposes constraints on the precise prediction of elastic modulus in needled‐ceramic matrix composites. This study presented a novel computational framework that synergistically combined a multiscale homogenization method with an improved particle swarm optimization neural network to predict the stiffness of the composites. The framework developed has successfully captured the inherent nonlinear correlation between the multiscale microstructural characteristics of the composites and their macroscopic stiffness. A fully connected neural network was employed, and hyperparameters were optimized by using an improved particle swarm optimization algorithm. Microstructure parameters obtained from microscopy were utilized to develop the multiscale generalized method of cells. Through the derivation of the strain transformation matrix at different scales, a database comprising manufacturing defect parameters and elastic properties was established. Structural parameters and operating temperature were identified as the input parameters for the neural network, whereas the predicted elastic properties were utilized as the corresponding outputs. Implementation of the neural network training and validation procedures was executed through MATLAB, with mechanical tests conducted to validate the accuracy of the simulated and predicted elastic properties. The results demonstrated that this framework could effectively predict the stiffness of the composites, achieving high accuracy and computational efficiency.