Design Optimization of Pulse Transformers in Series-Type Hybrid Circuit Breakers Using a Neural Network Based Surrogate Model

Multi-objective optimization is often required in designing electromagnetic components, such as motors or transformers, under complicated trade-off considerations. Finite Element Method (FEM) simulations are usually conducted to achieve the optimal design with the penalty of long computation time. This paper presents a data-driven design optimization method that partially replaces FEM simulation with a deep neural network (DNN) based surrogate model. The proposed framework adopts the Non-dominated Sorting Genetic Algorithm II (NSGAII) method. Through the case study of a pulse transformer design in a series-type hybrid circuit breaker (S-HCB), we demonstrate that the new DNN- based optimization approach can significantly reduce the optimization time by sevenfold while preserving up to 98.66% of the average accuracy of the conventional FEM approach. The proposed optimization strategy is expected to benefit a wide range of engineering optimization problems that involve a large amount of computation.