Statistical Determination of Johnson-Cook Model Parameters for Porous Materials by Machine Learning and Particle Swarm Optimization Algorithm

This study developed a reliable methodology combined by the machine learning technique, that is, multi-task Gaussian process regression (MTGPR) and particle swarm optimization (PSO) algorithm to identify the Johnson-Cook (JC) constants for porous structures rapidly, meanwhile reducing the experimental consumption. Ti6Al4V (TC4) porous samples with three different porosities were designed based on lattice Weaire–Phelan (LWP) structure and manufactured by selective laser melting technology. Uniaxial compressive tests of the built LWP samples were carried out to measure their mechanical properties. The resultant JC constants derived from MTGPR-PSO model were imported into the compressive finite element analyses (FEAs) of LWP structures with various porosities. It proved that the simulated mechanical behaviors presented an explicit agreement with the experimental results. Furthermore, other JC models of TC4 porous material from previous reports were utilized to perform the compressive FEAs of LWP samples, indicating that the JC models could not be shared among the porous structures fabricated by different manufacturing processes or with various lattice structures. Hence, an effective method like the MTGPR-PSO model was urgently required to identify the JC constants of porous samples customized by additive manufacturing.