Fast and Accurate Inductor Design Using Feedforward Neural Networks Trained By Quasi-Newton Method

This paper describes the development of feedforward neural networks (FFNNs) for the inverse problem in the accurate air and iron-core inductors design. Utilizing the finite element method (FEA) for designing an inductor, a dataset of the mapping from design parameters to corresponding inductance can be obtained. This dataset is later used to train an FFNN forming an inverse mapping from the desired inductance to the corresponding design parameters of the inductor. In this study, the quasi-Newton method is used to update the weights and biases of FFNNs as it can result in perfect convergence by minimizing a non-linear data error function. The simulation results show that the FFNNs can appropriately infer the design parameters of inductors to achieve the desired inductance. The proposed method in this study can be also applied to the fast and accurate design of other types of electromagnetic devices including power transformers and electric motors.