Affiliation:
1. School of Engineering, Faculty of Science Engineering and Built Environment, Deakin University, Geelong, Australia
Abstract
Data-driven calibration techniques, consisting of theory-guided feed-forward neural networks with long short-term memory, have previously been developed to find suitable input parameters for the finite element simulation of progressive damage in fibre-reinforced composites subjected to compact tension and compact compression tests. The results of these machine learning-assisted calibration approaches are assessed in a range of virtual open-hole strength tests under tensile and compressive loadings as well as in low velocity impact tests. It is demonstrated that the calibrated material models with bi-linear softening are able to simulate the structural response qualitatively and quantitatively with a maximum error of 9[Formula: see text] with regards to experimentally measured open-hole strength values. Furthermore, the highly efficient models enable the virtual analysis of size effects as well as accurate force simulations in quasi-isotropic laminates under impact loading.
Publisher
World Scientific Pub Co Pte Ltd
Subject
Computational Mathematics,Computer Science (miscellaneous)
Cited by
2 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献