Study on the Machine Learning Model Optimization Based on Dynamic Compression Strength of Ti-Zr-Nb Solid Solution Alloys

The Ti-Zr-Nb solid solution alloys possess great application value in the fields of blast and fragmentation warhead and shaped warhead due to its excellent strength, plasticity and impact energy release characteristics. In order to achieve accurate prediction of dynamic mechanical properties of Ti-Zr-Nb solid solution alloys and provide support to composition optimization of warhead materials, 56 Ti-Zr-Nb alloys were prepared by powder metallurgy and the dynamic compression strength was tested. Furthermore, optimization of machine learning models and selection of key features for the prediction of dynamic compression strength were carried out. The results show that the prediction error of optimized model can achieve less than 8%, and three key features can be selected and ordered as: Δχ>G>δ_G. The optimized model can be used to design new alloys with higher dynamic compression strength successfully, being 3100 MPa and higher than other similar alloys.