Data-driven mapping-relationship mining between hardness and mechanical properties of dual-phase titanium alloys via random forest and statistical analysis

In order to accelerate the research on the property optimization of titanium alloy based on high-throughput methods, it is necessary to reveal the relationship between hardness and other mechanical properties which is still unclear. In this work, taking Ti20C alloy as research object, almost all the microstructure of dual-phase titanium alloys were covered by traversing over 100 heat treatment schemes. Then, massive experiments including microstructure characterization and performance test were conducted, obtaining 51,590 pieces of microstructure data and 3591 pieces of mechanical property data. Subsequently, based on large-scale data-driven technology, the quantitative mapping relationship between hardness and other mechanical properties was deeply discussed. The results of random forest models showed that the correlation between hardness (H) and Charpy impact energy (A_k) (or elongation, A) was hardly dependent on the microstructure types, while the relationship between H and tensile strength (R_m) (or yield strength, R_p0.2) was highly dependent on microstructure types. Specifically, combined with statistical analysis, it was found that the relationship between H and A_k (or A) were negatively linear. Interestingly, the relationship between H and strength was positively linear for equiaxed microstructure, and strength was linked to d^−1/2 (d, equivalent circle diameter) of α-grains in the form of classical Hall–Petch formula; but for other microstructures, the relationships were quadratic. Furthermore, the above rules were nearly the same in the rolling direction and transverse direction. Finally, a "four-quadrant partition map" between H and R_p0.2/R_m was established as a versatile material-screening tool, which can provide guidance for on-demand selection of titanium alloys.