Co-authors​ :

 Siyang WU (CHINA), Licheng GUO (CHINA) 

3D woven composites (3DWCs) exhibit a great deal of attractive mechanical performance, so they are widely applied in the manufacturing of advanced industrial equipment. The 3DWCs possess highly complex mesoscale weave architecture, composed of interwoven fiber tows arranged in various directions and resin matrix. Moreover, the constituent materials of 3DWCs exhibit highly non-linear mechanical behavior, necessitating the use of appropriate constitutive models for characterization. Numerous studies have shown that performing direct numerical simulation (DNS) on the 3DWC requires tremendous computational resources, especially for damage problems. To improve computational efficiency for the damage problems of 3DWCs, a data mining-based self-consistent clustering analysis (SCA) framework, combined with a mesoscale damage model, is developed to investigate the damage evolution behavior of the 3DWC. The SCA method includes two stages: offline database compression and online prediction. In the offline stage, a novel offline clustering scheme based on the local material orientation and the strain concentration tensor is proposed and executed sequentially for each type of fiber tows due to the fluctuation characteristics of binder tows and weft tows. In the online stage, the cluster-based non-linear Lippmann–Schwinger equations are solved. Meanwhile, a mesoscale damage model is developed to characterize the damage initiation and performance degradation of the constituents. A mesoscopic cluster-based RVE model considering actual fluctuation paths and cross-section dimensions of fiber yarns, is established to characterize the mechanical performance of the 3DWC. For the elasticity problem, the results predicted by the SCA method are compared with those of the finite element analysis (FEA) method, which shows that the SCA method can effectively predict stress distribution and homogenization performance. For the damage problem, the stress-strain curves and damage evolution courses predicted by the SCA method with different cluster schemes are compared with those of the FEA method and experimental results, which reveals that the SCA method can reliably predict the non-linear mechanical behavior of the 3DWC, and the solution accuracy of SCA method improves as the number of clusters increases. Of particular importance is the ability of SCA method to improve computational efficiency by 425 times compared to the FEA method under the same constitutive model. The SCA method will provide a crucial analytical approach for multiscale damage analysis and structural optimization of 3DWC structures in the future.