Electromagnetic design of MRI superconducting magnet based on novel hybrid optimization methods

Abstract

This paper proposes a novel hybrid optimization algorithm that combines linear programming (LP), genetic algorithm (GA), and nonlinear programming (NLP) to achieve the optimal design of highly homogeneous superconducting magnets for MRI systems. Initially, the predetermined rectangular region is divided into an array of superconductor coils. Then, linear programming is utilized to minimize the consumption of superconducting conductors as the objective function and to obtain the nonzero current regions by considering the field peak-to-peak uniformity in the diameter of the spherical volume (DSV) and the range of the 5 Gauss stray field as constraints. Subsequently, the genetic algorithm is employed to convert the nonzero current regions into coils with rectangular cross-sections. Finally, the NLP is applied to adjust the position of each coil to obtain the magnet criteria. An illustrative example is provided: an actively shielded MRI superconducting magnet with a center field strength of 1.5 T. The effectiveness of this optimization method is demonstrated through the design, electromagnetic analysis, and stress analysis conducted on this example.