Original Article
  • 딥러닝 기법을 적용한 평직 CFRP 복합재의 유효물성 예측
  • Yuseon Lee*, **, Dong-Hyeop Kim***, Sang-Woo Kim*, **, ****†

  • * Department of Aerospace and Mechanical Engineering, Korea Aerospace University
    ** Department of Smart Air Mobility, Korea Aerospace University
    *** LIG Defense&Aerospace
    **** Department of Aeronautical and Astronautical Engineering, Korea Aerospace University

  • Deep Learning–Based Prediction of Effective Properties of CFRP Plain-Woven Composites
  • 이유선*, ** · 김동협*** · 김상우*, **, ****†

  • This article is an open access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

A deep neural network(DNN) based methodology is proposed to predict the effective thermo-mechanical properties of plain-weave CFRP composites. Ten distinct effective properties are evaluated by selecting three primary geometric parameters as design variables to account for variations in textile architecture. Representative volume element(RVE) modeling was performed using TexGen software, with periodic boundary conditions(PBCs) applied via the EasyPBC plugin. Effective thermo-mechanical analyses were subsequently conducted using the commercial finite element analysis(FEA) software, ABAQUS. The entire data generation process was automated to construct a comprehensive database of 728 datasets. Model performance was evaluated using 164 validation samples, which account for 20% of the total dataset. The evaluation yielded a Mean Relative Error(MRE) of 0.79% and a coefficient of determination(R2) of 0.9991, confirming the model's reliability as a surrogate. While FEA-based homogenization required approximately 63.1 s, the DNN model completed predictions in only 14 ms, confirming a significant leap in computational efficiency. These results verify that the proposed DNN model effectively predicts effective thermo-mechanical properties in response to variations in weaving parameters.


본 논문에서는 심층 신경망(deep neural network, DNN)을 활용하여 평직 CFRP 복합재의 열·기계적 유효물성을 예측하는 방법론을 제시하였다. 직물 구조의 변화에 따른 10가지의 유효물성을 예측하기 위해 주요 기하학적 변수 3가지를 설계 변수로 선정하였다. TexGen 소프트웨어를 활용하여 RVE(representative volume element) 기반 모델링을 수행하고 Easy PBC 플러그인을 통해 주기 경계 조건을 부여하였다. 이후 상용 유한요소해석(finite element analysis, FEA) 프로그램인 ABAQUS를 활용하여 유효 열·기계적 해석을 수행하였다. 모든 데이터 생성 과정은 자동화하였으며, 총 728개의 데이터 세트를 구축하고 이를 DNN 모델 학습에 사용하였다. 전체 데이터의 20%인 164개의 검증 데이터를 활용한 모델 성능 평가 결과 평균 상대 오차는 0.79%, 결정계수(R2)는 0.9991로 나타나 유효물성 예측을 위한 신뢰성 있는 대리모델로 활용 가능함을 보여주었다. 유한요소해석이 약 63.1 s 소요되는 것에 비해 DNN 모델을 약 14 ms 내에 예측을 완료하여 수치 해석 대비 비약적인 계산 효율성을 확인하였다. 이를 통해 제안된 DNN 모델이 직조 변수의 변화에 따른 유효 열·기계적 물성을 효과적으로 예측할 수 있음을 확인하였다.


Keywords: 탄소섬유강화플라스틱(Carbon fiber reinforced plastics), 유효물성(Effective property), 유한요소해석(Finite element analysis), 딥러닝(Deep learning), 심층신경망(Deep neural network)

This Article

Correspondence to

  • Sang-Woo Kim
  • * Department of Aerospace and Mechanical Engineering, Korea Aerospace University
    ** Department of Smart Air Mobility, Korea Aerospace University
    **** Department of Aeronautical and Astronautical Engineering, Korea Aerospace University

  • E-mail: swkim@kau.ac.kr