关键词: 应力分布, 弹道冲击, 聚类算法, 计算机视觉, 有限元分析

Abstract:

In order to investigate the performance of different clustering algorithms in processing the finite element resultant stress cloud, a single-layer ballistic impact finite element model of Twaron® plain fabric was established in the study, and four different clustering algorithms, namely, k-means, GMM, Mean-shift, and DBSCAN, were used to cluster the stress cloud and analyze the results comparatively with the resultant images of the stress distributions as an example. The results show that: Mean-shift and DBSCAN clustering algorithms are not suitable for processing a large number of finite element stress maps; k-means and GMM clustering algorithms improve the efficiency of 74.24 and 172.64 times compared with the traditional manual processing, but GMM clustering algorithms produce errors when the color of the image is not clearly differentiated; k-means clustering algorithms ensure high efficiency, while the error is only 0.5 and 1.0 times, respectively; and k-means clustering algorithms are more efficient than traditional manual processing. The k-means clustering algorithm guarantees high efficiency while the error is only within 0.85%, therefore, the k-means clustering algorithm is the most suitable among these four algorithms for fast, objective and quantitative analysis of a large number of stress cloud maps. Using this method, it was measured that the area of the stress interval of single-layer Twaron® plain fabrics decreased with a stress area change rate of 5.61×107 mm2/s for 0 ~600 MPa, and the area of the stress interval increased with a stress area change rate of 5.27×107 mm2/s for 600 ~1200 MPa within 1~15 μs of the impacts.

Key words: stress distribution, ballistic impact, clustering algorithms, computer vision, finite element analysis