Multiscale Modeling of Textile Composite Structures Using Mechanics of Structure Genome and Machine Learning

Xin Liu, Purdue University

Abstract

Textile composites have been widely used due to the excellent mechanical performance and lower manufacturing costs, but the accurate prediction of the mechanical behaviors of textile composites is still very challenging due to the complexity of the microstructures and boundary conditions. Moreover, there is an unprecedented amount of design options of different textile composites. Therefore, a highly efficient yet accurate approach, which can predict the macroscopic structural performance considering different geometries and materials at subscales, is urgently needed for the structural design using textile composites. Mechanics of structure genome (MSG) is used to perform multiscale modeling to predict various performances of textile composite materials and structures. A twostep approach is proposed based on the MSG solid model to compute the elastic properties of different two-dimensional (2D) and three-dimensional (3D) woven composites. The first step computes the effective properties of yarns at the microscale based on the fiber and matric properties. The effective properties of yarns and matrix are then used at the mesoscale to compute the properties of woven composites in the second step. The MSG plate and beam models are applied to thin and slender textile composites, which predict both the structural responses and local stress field. In addition, the MSG theory is extended to consider the pointwise temperature loads by modifying the variational statement of the Helmholtz free energy. Instead of using coefficients of thermal expansions (CTEs), the plate and beam thermal stress resultants derived from the MSG plate and beam models are used to capture the thermal-induced behaviors in thin and slender textile composite structures. More-over, the MSG theory is developed to consider the viscoelastic behaviors of textile composites based on the quasi-elastic approach. Furthermore, a meso-micro scale coupled model is proposed to study the initial failure of textile composites based on the MSG models which avoids assuming a specific failure criterion for yarns. The MSG plate model uses plate stress resultants to describe the initial failure strength that can capture the stress gradient along the thickness in the thin-ply textile composites. The above developments of MSG theory are validated using high-fidelity 3D finite element analysis (FEA) or experimental data. The results show that MSG achieves the same accuracy of 3D FEA with a significantly improved efficiency. Taking advantage of the advanced machine learning model, a new yarn failure criterion is constructed based on a deep neural network (DNN) model. A series of microscale failure analysis based on the MSG solid model is performed to provide the training data for the DNN model. The DNN-based failure criterion as well as other traditional failure criteria are used in the mesoscale initial failure analysis of a plain woven composite. The results show that the DNN yarn failure criterion gives a better accuracy than the traditional failure criteria. In addition, the trained model can be used to perform other computational expensive simulations such as predicting the failure envelopes and the progressive failure analysis.

Degree

Ph.D.

Advisors

Yu, Purdue University.

Subject Area

Mechanics|Artificial intelligence|Computer science|Genetics|Marketing|Materials science|Mathematics|Textile Research

COinS