Practical Uncertainty Quantification for Nitinol Implant Simulations

Finite element predictions of mean and alternating strains critically influence the fatigue assessment of nitinol implants in two ways: 1) simulations of surrogate or coupon specimens (e.g., Z-, C-, or diamond geometries) are used to establish strain-life and Goodman-type relationships, and 2) simulations of full devices under physiological loading are used to estimate strains associated with anticipated worst-case use conditions. These simulation results are then combined to estimate fatigue safety factors for a device. Despite the importance of these analyses, adoption of rigorous simulation uncertainty quantification practices has met resistance due to a perceived imbalance between burden and benefit. Here, we propose a simplified uncertainty quantification approach that leverages author experience and recent literature to improve practicality. As a case study, we consider simulation of laser-cut diamond specimen struts under bending. In brief, a mesh refinement study is first performed using nominal diamond dimensions to estimate the discretization error in strain quantities of interest. A sensitivity study is then used to aggressively filter material, geometry, and boundary condition inputs to only those with the largest impact on simulation predictions. The identified key input parameters are characterized, and estimated uncertainty bounds for each are propagated to a controlled number of simulations. Predicted strain fields are finally compared to digital image correlation (DIC) measurements using conventional and novel models for nitinol superelasticity. This simplified uncertainty quantification approach frees resources for priorities like model validation and design iteration without neglecting the impact of real-world uncertainty on device mechanics.