Machine learning and fractal analysis of microstructure–property relationships in Zr-30Ti-xCu (x=0,1,3,5 wt%) biomedical alloys

We prepared Zr-30Ti-xCu (x = 0, 1, 3, 5 wt%) alloys using three different preparation methods (Homogenization treatment at1020 ℃ for 20 min; homogenization treatment at 1020 ℃ for 20 min + annealing treatment at 500 ℃ for 60 min; homogenizationtreatment at 1020 ℃ for 20 min + annealing treatment at 650 ℃ for 30 min) and characterized their microstructures, testing theirmechanical properties, biocompatibility, and antibacterial performance. Gradient boosting trees (XGBoost) and Shapley value(SHAP) explanatory framework from machine learning methods were adopted to analyze the experimental data. Meanwhile,fractal dimension and local binary pattern (LBP) texture descriptors were used to quantify the microstructure complexity. Theresults indicate that high fractal dimensions correspond to complex grain boundaries and complex Zr2Cu precipitate distributions,which can effectively enhance cell adhesion and antibacterial rate. Through machine learning, it was discovered that nanoscaleZr2Cu and amorphous structures distributed along grain boundaries are the main factors affecting cell proliferation and antibacterialperformance. These specific amorphous structures provide larger surface area for cell adhesion and fast channels for Cu iondiffusion. Considering all factors, it was found that Zr-30Ti-1Cu with homogenization treatment at 1020 ℃ for 20 min + annealingtreatment at 650 ℃ for 30 min meets all performance criteria, with a Day 1 antibacterial rate as high as 78%, making it a potentialbiomedical material.