MatArc-DeepONet: a surrogate model for fast erosion prediction and inverse design of TMF contacts material in vacuum interrupters

Abstract The interruption reliability of high-capacity vacuum interrupters (VIs) is directly determined by the erosion conditions of transverse magnetic field (TMF) contacts. Traditional designs heavily rely on trial-and-error experiments and time-consuming numerical simulations, making it difficult to support the rapid exploration of a wide range of parameters. To this end, a data-driven surrogate model based on a deep operator network (DeepONet), named Material-Arc DeepONet (MatArc-DeepONet), is proposed in this paper for the fast and high-accuracy prediction of the transient erosion characteristics of TMF contacts. Firstly, a physical model for the transient TMF contact erosion is established, and a high-quality dataset containing 5000 cases is generated within the design space of the peak current and Cr alloy ratio. Secondly, to address the periodic high-frequency heat pulses generated by the high-speed moving arc, a Fourier feature mapping mechanism is introduced into the model. Tests show that the average relative errors of the model for the maximum melt pool depth and the maximum surface temperature are only 0.46% and 0.08%, respectively. Moreover, the calculation time for a single case is significantly shortened from 41.2 s by the traditional solver to 0.1 ms by the network forward pass. Finally, relying on the highly efficient inference capability of the model, an inverse design mapping of contact materials is constructed by setting the maximum melt pool depth after a single interruption as the safety threshold. This method provides a framework for the optimal selection, cost reduction, and efficiency enhancement of VI contact material.