Two-stage deep learning for the modeling and prediction of fiber volume fraction in soft-hard hybrid braided preform during compaction process

Abstract

To improve the Fiber Volume Fraction (FVF) of carbon/carbon (C/C) composite soft-hard hybrid braided preforms and reduce the interlayer deviation of fiber content, the traditional single compaction process is optimized. To address this challenge, this paper proposes a two-stage deep learning architecture to enhance the prediction accuracy for preform FVF. In the first stage, a neural network is used to construct a multiple regression model (NNMR) that incorporates multiple process parameters. Based on an investigation into the compaction mechanism of preforms, the influence of multiple process parameters on the FVF is analyzed. Specifically, the NNMR is employed to learn the nonlinear relationships between these parameters and the preform FVF. For the second stage, a Nonlinear Model Predictive Control (NMPC) framework is employed, where the fitted process parameters serve as model inputs and the preform FVF as the model output. By constraining the target range, dynamic and refined adjustments are performed to obtain the optimal combination of process parameters. Notably, this method can overcome the limitations of traditional compaction methods and achieve efficient, accurate, and reliable prediction of preform FVF, thereby providing theoretical support for preform compaction processes.