Session: 01-03 -01: Advanced Manufacturing and Process–Structure Relationships in Aerospace Structures

Paper Number: 190613

190613 - An Integrated Lstm-Fea Modeling Method to Predict Curing Induced Deformation of Thermosetting Cfrp Structures 

Carbon fiber reinforced polymers (CFRPs), particularly thermosetting resin-based composites, are extensively employed in aerospace structural components owing to their exceptional specific strength and adaptable design capabilities. However, the curing process of these composites is significantly hindered by the generation of substantial residual stress, which stem from intrinsic material heterogeneity and anisotropy. Such stress inevitably induces severe curing-induced deformation (CID), such as warpage and spring-in, thereby compromising assembly precision, geometric integrity and ultimately structural reliability of the components. Although finite element analysis (FEA) has been extensively developed as a computational approach to simulate CID and mitigate the reliance on resource-intensive experimental trials, its practical application to complex three-dimensional (3D) geometries and curing processes governed by coupled chemical-thermal-mechanical response of the composites remain computationally infeasible, especially for the non-linear constitutive behaviors and non-uniform structural features.

Recent studies have leveraged artificial intelligence (AI) to accelerate FEA by predicting residual stress and deformation patterns using macroscopic data such as displacement, or graphical input such as strain or stress fields. However, these methods exhibit inherent limitation in capturing mesoscale material response, thereby restricting their applicability to diverse geometric configurations and boundary conditions. Moreover, the isolated coding environments​ of the AI platforms and FEA solvers impede their seamless integration, consequently affecting prediction accuracy and computational robustness. Another method, self-consistent clustering analysis (SCA), is based on the reduced-order technique and demonstrates theoretical potential for bridging the scales solved by the AI and FEA techniques. However, SCA struggles with the convergence rate and inability to precisely characterize intricate parameter interaction inherent in the curing processes.

To address these challenges, we established an innovative multi-scale LSTM-FEA integrated modeling method. It leverages the Long Short-Term Memory (LSTM) network’s ability to capture the history-dependent relaxation behavior of the thermoset CFRPs. Training data of relaxation stiffness was obtained from the validated mesoscale woven representative volume element (RVE) modeling, minimizing the need for time-consuming stress-relaxation experiments. This LSTM framework predicts all viscoelastic stiffness tensor components across diverse processing conditions, such as various yarn orientations, degrees of curing, temperatures and relaxation durations. The underlying mathematical formulation of the trained LSTM algorithm was transcribed into Fortran code and implemented into Abaqus through a customized UMAT subroutine. This integration enables seamless transmission of the current analysis step's updated thermal and chemical properties, which are stored as state-dependent variables (SDVs), into the LSTM model, facilitating accurate prediction of corresponding viscoelastic response.

This LSTM-FEA integration breaks the technical barrier between AI tools and FEA software, automating the iterative calculation of time-temperature-DOC-dependent relaxation moduli and residual stress across local integration points. Through seamless coupling of the mesoscopic material response with macroscopic structural analysis, the established method effectively addresses the persistent challenges of limited generalizability and computational inefficiency that have long hindered the existing approaches. Furthermore, this method with direct integration enables efficient and accurate simulation of complex non-uniform 3D CFRP composite structures under diverse manufacturing and service scenarios, demonstrating substantial potential for optimizing their process parameters and enhancing their reliability in high-performance engineering applications.

Presenting Author: Weizhao Zhang The Chinese University of Hong Kong

Presenting Author Biography: Professor Weizhao Zhang received the Bachelor of Engineering degree in the Department of Mechanical Engineering at Tsinghua University in 2014, and the Ph.D. degree in the Department of Mechanical Engineering at Northwestern University in 2019. Afterwards, he joined the Chinese University of Hong Kong as an Assistant Professor to continue his research on composite materials, with particular focus on the carbon fiber reinforced polymers (CFRPs). His research interest mainly concentrates on advanced process and performance modeling for CFRPs and development of innovative CFRPs with enhanced performance. During his research period, Professor Zhang has obtained grants from the Research Grants Council, Innovation Technology Commission and Shun Hing Institute of Advanced Engineering in Hong Kong, as well as the Commercial Aircraft Corporation of China, to complete computational mechanics tools for systematic process and performance analysis of CFRP parts in large-volume industrial applications, and develop 3D CFRP structural batteries and high-impact-resist CFRP materials. Up-to-date, research works by Professor Zhang and his group have led to nearly 50 research publications, 1 patent granted, 3 patents applied and 1 software copyright granted. The innovative non-orthogonal material model for woven CFRP prepregs during preforming has been included in the built-in library of the commercial finite element analysis software LS-DYNA, the advanced curing modeling method for unidirectional CFRP prepregs has been selected by the Commercial Aircraft Corporation of China for virtual and efficient material characterization and process parameter optimization, and the manufacturing approaches for CFRP structural batteries and high-impact-resist CFRPs have been employed to develop the scientific payload for the China’s Chang’E 8 lunar exploration mission. Besides, Professor Zhang is the associate editor for the Journal of Advanced Manufacturing Science and Technology, and the scientific committee member for multiple academic organizations and activities, such as the North American Manufacturing Research Institution of SME and the International Conference on Textile Composites.