Machine learning-driven Component optimization for eco-friendly High-Voltage Circuit Breaker design

Hexafluoride (SF₆), we are developing a new generation of high-voltage circuit breakers

(HVCBs) that uses CO₂-C₄F₇N as insulating and arc-quenching gases mixtures.

In eco-friendly HVCBs, fault interruption is governed first by thermal interruption and subsequently by dielectric recovery. These processes are initiated by arcing contact separation and are strongly influenced by the geometry of the arcing zone. To capture the complex physics involved in interruption, high-fidelity (HF) simulations, such as Computational Fluid Dynamics

(CFD), were conducted. Traditionally, the optimization of components such as nozzles, diffusers, and exhaust systems relies on iterative CFD and multiphysics simulations. Although highly accurate, these simulations are computationally time consuming and therefore restrict the extent of design-space exploration.

To address these limitations, this study introduces a machine-learning (ML) driven framework for component-level optimization in eco-friendly HVCBs. HF simulation data are used to train supervised ML models capable of rapidly predicting key performance metrics across wide parameter ranges. Using this surrogate model, multiobjective optimization (MOO) can be performed within seconds, enabling efficient exploration of competing design parameters and producing densely populated Pareto fronts.