Comparing Finite Element Predictions and Image-Based Measurements of Strain in a Nitinol Medical Device: A Verification, Validation, and Uncertainty Quantification (VVUQ) Study
Comparing Finite Element Predictions and Image-Based Measurements of Strain in a Nitinol Medical Device: A Verification, Validation, and Uncertainty Quantification (VVUQ) Study
Thursday, May 16, 2019: 2:15 PM
K2 (Bodenseeforum Konstanz)
Engineers typically use strain-based fatigue analyses to assess the fatigue safety of nitinol medical devices. Given the challenge of measuring strain directly on small cardiovascular devices, analysts currently rely on finite element analysis (FEA) predictions of strain coupled with surrogate force-displacement validation evidence. However, force is an integrated quantity whereas strain is inherently a local quantity. Here, we perform FEA simulations and acquire micro-scale digital image correlation (DIC) measurements of strain for a generic nitinol inferior vena cava filter subcomponent. First, tensile and compressive properties of raw SE508 nitinol are characterized using uniaxial tests. Specimen geometry is also characterized using optical methods. Speckle patterns are then applied to device coupons by coating them with finely-ground carbon powder, and a digital optical microscope is used to acquire images of the struts during fixed-free cantilever beam motion emulating physiological loading conditions. Surface strains are extracted from the images using the open-source DIC software Ncorr-C++. Stochastic simulations mimicking the DIC experiments are performed in ABAQUS (R2016x) using Latin hypercube sampling of key input parameters. Finally, experimental and computational strains, with their uncertainties, are quantitatively compared using the modified area validation metric (MAVM). Agreement between FEA predictions and DIC measurements of strain is relatively good at small tip displacements, but differences increase following austenitic-martensitic phase transformation (MAVM≈0.1% strain for peak FEA strains <1% versus MAVM≈1.5% strain for peak FEA strains ≈4%). Overall, DIC strain measurements reveal the usefulness and limitations of current nitinol modeling and enable rigorous quantification of model error.