Intelligent optimization design of specimen on sheet materials for Split Hopkinson Tensile Bar tests

Abstract

The Split Hopkinson Tensile Bar (SHTB) is one of the most widely used methods to study the high strain rate behavior of materials. For these experiments usually dogbone-shaped sheet specimens are used. However, there's no agreement on the exact dimensions. In the present study, mechanism of the influence of specimen responses on accuracy of SHTB experiments was investigated with finite element program ABAQUS (Explicit). Indicators which can evaluate the measurement accuracy of specimens are proposed based on this. Orthogonal test is designed to establish the sample database for back-propagation (BP) neural network, which is adopted to fit the non-linear mapping from structure parameters to accuracy indicators of specimen. Optimal design of structure for sheet specimen is obtained with Genetic Algorithm (GA) according to the fitness of individual determined by trained and qualified BP neural network. At last, numerical simulations are adopted to verify the validity of the optimal structure for sheet specimen. The result of this study can provide recommendations for specimen design and data reliability analysis in Split Hopkinson Tensile experiments.