Efficient Multiphysics Circuit Simulation for Transformer Optimization Using Hybrid Dowell Artificial Neural Network

Abstract

This paper introduces a practical approach for transformer design optimization using a novel hybrid Dowell artificial neural network (HDANN) model. This model is a highly efficient and accurate method to estimate leakage inductance, which is an important parameter that affects the performance of transformers and power converters. The model combines the conventional hybrid Dowell's model, which uses analytical equations, and an artificial neural network, which uses machine learning techniques. It is integrated into an optimization program to optimize the design of a transformer in terms of its size and loss. We investigated the HDANN design scope by testing various transformer conditions. The results provided an in depth understanding of the capabilities and limitations of the HDANN model for transformer design. By understanding the HDANN design scope, the optimization program was implemented in a multidomain circuit simulation, which includes electric and magnetic circuits. This allows a high-speed co-simulation using the optimized transformer design that considers the geometry and material characteristics for the desired circuit specification. It showed that the HDANN offers significant advantages over existing design optimization methods, including improved ease of application, accuracy, and efficiency. The effectiveness of the optimization using the HDANN with the defined geometric parameters was demonstrated by the circuit analysis results of a phase-shift full-bridge converter. Summarizing, the proposed method can potentially revolutionize how transformers are designed and implemented for various applications, leading to increased design reliability as well as reduced power loss and size.