Predicting Stress and Damage in Carbon Fiber-Reinforced Composites Deformation Process using Composite U-Net Surrogate Model

TL;DR

The paper tackles the high computational cost of traditional FEM/IGFEM for predicting stress and damage evolution in CFRC under deformation. It introduces Composite-Net, a modular, auto-regressive, multi-network surrogate built from Damage-Net, UTS-Net, and Necking-Net that jointly captures microstructural and macro-scale responses throughout the loading history. The approach achieves high fidelity, with RMSEs for stress below $<15\mathrm{MPa}$ and damage below $<40\%$ in most cases, and delivers a substantial speed-up of over $60\times$ compared to IGFEM. By enabling crack initiation, propagation, and post-UTS behavior to be predicted efficiently, the framework holds promise for accelerated design, analysis, and reliability assessment of CFRC structures in aerospace and automotive applications. The work also sets the stage for extensions to 3D geometries and a broader set of loading conditions using physics-informed refinements and transfer learning.

Abstract

Carbon fiber-reinforced composites (CFRC) are pivotal in advanced engineering applications due to their exceptional mechanical properties. A deep understanding of CFRC behavior under mechanical loading is essential for optimizing performance in demanding applications such as aerospace structures. While traditional Finite Element Method (FEM) simulations, including advanced techniques like Interface-enriched Generalized FEM (IGFEM), offer valuable insights, they can struggle with computational efficiency. Existing data-driven surrogate models partially address these challenges by predicting propagated damage or stress-strain behavior but fail to comprehensively capture the evolution of stress and damage throughout the entire deformation history, including crack initiation and propagation. This study proposes a novel auto-regressive composite U-Net deep learning model to simultaneously predict stress and damage fields during CFRC deformation. By leveraging the U-Net architecture's ability to capture spatial features and integrate macro- and micro-scale phenomena, the proposed model overcomes key limitations of prior approaches. The model achieves high accuracy in predicting evolution of stress and damage distribution within the microstructure of a CFRC under unidirectional strain, offering a speed-up of over 60 times compared to IGFEM.

Figures (17)

• Figure 1: (A) The schematic of a microstructural representation of the composite under uniaxial applied displacement with the applied boundary conditions. (B) The obtained stress-strain response: The initial state (1), the UTS (2) indicating peak material strength, and the final stage representing material failure (3). The contours present the von Mises stress and damage distribution within the microstructural representation of the composite at different strain levels. SEPASDAR_FEM
• Figure 2: Schematic representation of Composite-Net, consisting of three interconnected neural networks: Damage-Net (yellow) for mapping microstructure to the final damage pattern, UTS-Net (gray) for predicting the response from initial condition to UTS, and Necking-Net (green) for modeling the transition from UTS to $\epsilon_f$. $x_{\epsilon}$ and $N_{\epsilon}$ are the input features of the UTS-Net and Necking-Net, respectively; $x_{\epsilon+d\epsilon}$ and $N_{\epsilon+d\epsilon}$ are the output features of the UTS-Net and Necking-Net, respectively;
• Figure 3: Architecture of U-Net used in Damage-Net, UTS-Net and Necking-Net (Left). The simplified schematic of the U-Net architecture (Right).
• Figure 4: Schematic representation of the Damage-Net for mapping microstructure to the final damage pattern.
• Figure 5: Schematic representation of UTS-Net for predicting the von Mises stress and damage response from initial condition to UTS.