Predicting core losses in power converters based on open source database

Abstract

Core losses are closely related to flux densities, frequencies, temperatures, exciting waveforms, and DC biases. Hysteresis losses and dynamic losses are considered the main components of core losses. A method for predicting core losses based on an open-source database and a data-driven physical model is proposed. To a certain predicted point, two paths in the procedure can be executed automatically. First, if there are two points in the database with similar working conditions to the predicted point, core loss power density can be directly predicted based on separate equations. Secondly, if loss power density for specific waveforms in the database are missing, hysteresis loss power density and dynamic loss power density under sinusoidal excitations can be utilized to estimate core loss power density for specific waveforms. Based on the proposed method, the impact of temperature on hysteresis loss power density and dynamic loss power density can also be individually fitted and analysed. The proposed concepts demonstrate good flexibility in handling various complex waveforms in power electronics and have been accurately verified using the loss data of N49 and N87 in the MagNet database.