Forward evaluation and inverse design of spiral-type TMF contacts considering geometric effects based on deep operator networks

Abstract

The design of spiral-type vacuum interrupter contacts demands balancing geometric and material parameters, yet traditional numerical methods are computationally prohibitive for high-throughput optimization. This study proposes the Geometry-Material-Arc Deep Operator Network framework to accelerate forward evaluation and robust inverse design. Trained on an electromagnetic-erosion coupled model, the Fourier-enhanced surrogate achieves high precision, with average relative errors restricted to 0.37% for surface temperature and 0.09% for melt depth. Furthermore, a dual-uncertainty inverse design method utilizing deep ensembles and Monte Carlo sampling is proposed. A global scan of 125 000 conditions is completed in under 12 s, and reliability under the 99% worst-case input noise boundary is strictly ensured. This framework significantly improves design efficiency and provides a practical methodology for vacuum interrupter development.