Keywords: computer experiments, computer-aided stent design, multi-functional response regression, robust parameter design, Pareto frontier
Careful design of coronary stents, which are used to increase the lumen diameter of narrowed artery, is crucial to ensure minimum vessel damage during the implant process and maximum effectiveness of the implanted stent. Stent design is inherently a multiple functional target problem since a stent needs to be expanded uniformly over time to different target vessel radii for different implantation cases. This paper considers an interesting problem of robust design of a coronary stent with multiple functional outputs and multiple target functions based on a time-consuming finite element (FE) simulator. In this paper, a multi-functional output GP model is employed. The separable prior mean and covariance functions, and the Cartesian product structure of the data are exploited to simplify computations. The multiple-target robust design problem is solved by optimizing an expected integrated quadratic loss criterion, which is estimated based on the multi-functional output GP emulator. The multi-functional output GP model gives better performance than a collection of independent GP models in quantification of prediction uncertainty. The optimal control factor setting (polymer blend) for the stent, and optimal profile of the signal factor (maximum balloon radial displacement) versus target are obtained with low computational cost.