An Investigation of Surrogate-Assisted, Gradient-Based Methods for Stellarator Breeder Geometry Optimization

The stellarator blanket design problem is characterized by many degrees of freedom in which the designer is tasked with fitting a series of layered components between the plasma and the magnets, usually resulting in components of a nonuniform geometry to meet the performance goals. For such a problem, numerical optimization methods are well suited to efficiently and autonomously explore the design space and find an optimized design. This preliminary study investigates the application of surrogate-assisted, gradient-based methods to stellarator breeder geometry design.The investigation includes the development of an accurate surrogate model to predict design point gradients and its incorporation into gradient-based algorithms. The developed methodology is successfully demonstrated for the WISTELL-D stellarator configuration, reducing blanket material costs below that of a reference while meeting a target tritium breeding ratio, and a detailed analysis of the optimization runs is presented. Future work will expand this methodology to incorporate control of the shielding components and shielding-relevant metrics.