缩进
分离式霍普金森压力棒
人工神经网络
材料科学
LS-DYNA系列
结构工程
应变率
复合材料
机械
有限元法
工程类
计算机科学
物理
人工智能
作者
Samaneh Pourolajal,Gholam Hossein Majzoobi
标识
DOI:10.1177/03093247221140981
摘要
Stress-strain curves of materials normally change with strain rate and temperature and are normally defined by material models. In this study, a new technique was developed for determining the constants of material models. This technique was based on dynamic indentation test, numerical simulation using Ls-dyna code and artificial neural network. An indenter of tapered shape was shot against the materials as the target by a gas gun. The experiments were carried out for four strain rates and four temperatures. The target was made of pure copper. The penetration depth-time and load-time histories were captured by a LVDT and a piezoelectric load-cell, respectively and the load-penetration depth curve (P-h) was obtained. This curve is characterized by five parameters which are determined for each indentation test. On the other hand, the indentation test was simulated using Ls-dyna hydrocode. From the simulations, the P-h curves were obtained using Johnson-Cook (J-C) and Zerilli-Armstrong (Z-A) material models and the characterizing parameters of the numerical P-h curves were also identified. Finally, an artificial neural network (ANN) was trained by the numerical P-h curves parameters as the input and the constants of J-C and Z-A models as the output. The trained neural network was then tested by the experimental p-h curves parameters as the input and the constants of J-C and Z-A models as the output. Moreover, a number of dynamic compression tests were performed using the well-known Split Hopkinson Bar at the same strain rates and temperatures used for indentation tests and the stress-strain curves of material were obtained. A reasonable agreement was observed between the stress-strain curves predicted by neural network and the Split Hopkinson Bar. The proposed method does not need sophisticated instrumentation and in fact, the load-time and indentation depth-time histories are directly converted to stress-strain of material using an artificial neural network.
科研通智能强力驱动
Strongly Powered by AbleSci AI