ML-augmented Ti-based microrobotic stents

Abstract

The integration of microrobotic stents into biomedical applications has the potential to revolutionize invasive procedures by enabling precise drug delivery, imaging, and vascular interventions. These interventions demand alloys with high radial stiffness for structural integrity and low density for biocompatibility. We developed a machine learning (ML)-finite element analysis (FEA) framework to optimize titanium (Ti)-based and Ti-based high-entropy alloys (Ti-HEAs) compositions using a curated database of 238 alloys. Gaussian process regression (GPR) is trained on FEA-simulated radial stiffness and constrained optimization (interior-point, sequential quadratic programming (SQP), active-set) identified high-performance candidates. The interior-point algorithm yielded the highest stiffness (483.54 kN/m) with balanced composition (Ti: 76.29 at%, Nb: 6.88%, Zr: 7.34%, Ta: 7.31%), outperforming the dataset maximum (TiSn 20, 472.49 kN/m) by 2.32% and Ti-6Al-4 V (368.96 kN/m) by 31%. All algorithms converged to at least 469 kN/m despite compositional diversity, confirming robustness. The framework enables rapid, physics-informed alloy design for next-generation biomedical microrobotics.