Tolerance-aware robust ranking of ventricular assist device rotor designs using strict feasibility and...

Manufacturing variability can alter the hydraulic and hemocompatibility performance of a ventricular assist device (VAD) rotor even when the nominal design is computationally optimal. This study presents a surrogate-assisted robust-design workflow for a rotary VAD using a CFD-derived design-of-experiments database of 4809 rotor geometries defined by six input variables and seven performance outputs. Output-specific surrogate models were selected from linear regression, quadratic response-surface models, support-vector regression, random forests, gradient boosting, and multilayer perceptrons using five-fold cross-validated R², yielding values from 0.7367 to 0.9966. The 16 best nominal designs were then propagated through 2500 surrogate-evaluated Monte Carlo realizations under four manufacturing-tolerance scenarios ranging from mild to extreme. Nominal ranking was based on the geometric mean of normalized desirability functions, whereas robust ranking combined scaled nominal desirability, feasibility probability, percentile-based desirability, and a degradation penalty. D1297 was the nominal optimum (D = 0.8125), but D717 became the preferred robust design under mild and moderate tolerances, with robust scores of 0.9031 and 0.8741, respectively. Under severe and extreme perturbations, D174 emerged as the leading candidate because it retained the best conservative percentile behavior and the smallest degradation while preserving full feasibility. Local sensitivity analysis identified clearance gap and rotor length as the dominant robustness drivers. The results show that production-oriented VAD design selection can differ materially from nominal optimization and that tolerance-aware ranking provides a surrogate-assisted computational screening basis for prioritizing rotor candidates before direct CFD confirmation, prototype fabrication, dimensional metrology, and experimental hemocompatibility testing.