Mining the relationship between the dynamic compression performance and basic mechanical properties of Ti20C based on machine learning methods

Abstract

The dynamic compression performance of titanium alloys is important for material design under shock, but the intrinsic relationship between them and basic mechanical properties is still unclear. In this work, based on the mechanical-property data set of Ti20C sheet (4788 pieces of data), the correlation between them was constructed through data-driven and machine-learning methods. Through the trained random-forest regression models, the Quantitative Maps were constructed, and the dynamic compression strength σD, critical fracture strain εf, as well as impact absorption energy ED, were effectively predicted, with the accuracy rates all over 86.11%. Accordingly, the zone of excellent dynamic performance and corresponding quasi-static tensile properties in Maps can be rapidly screened. Furthermore, combined with microstructure observation, it was found that nano-scale acicular secondary α-phase significantly increased σD, micro-scale secondary α-phase resulted in excellent dynamic strength and plasticity, while the equiaxed, bimodal and mixture microstructure without secondary α-phase corresponded to the high value of εf. Meanwhile, both nano-scale and micro-scale secondary α-phase contributed to the high value of ED. Finally, the generalization capabilities of models were validated. By comparing predicted values with experimental values, it was found the models realized the relatively accurate prediction of dynamic compression performance on other titanium alloys.

Keywords

• Dual-phase titanium alloys
• Dynamic compression performance
• Random forest
• Data-driven