Efficient Design Method for a forward converter transformer based on KNN–GRU–DNN Model

Abstract

Nowadays, many industries need many power topologies and the various applications require various required characteristics. To satisfy each required characteristic, it is needed to design a power topology that is proper to the required characteristics. The power topology design needs a multi-physics(electric, magnetic, thermal, etc) design. The electric and magnetic designs are important to the isolated power topology and use mathematical modeling. The electric design uses mathematical modeling on the power topology circuit analysis and the magnetic design uses mathematical modeling on the magneto-equivalent circuit which analyzes transformer characteristics. The results of electric and magnetic design are required characteristics of each and the feedback processes are required to satisfy electric and magnetic requirements. The iterative feedback decreases design efficiency and has a high dependence on the designer’s experience. This dissertation proposes a forward converter design method that adopts artificial intelligence and increases design efficiency. From adopting artificial intelligence(KNN–GRU–DNN model), the iterative feedback processes between electric and magnetic design are not essential and the power topology design can be simplified. Not processing feed back loop, the proposed method can define the transformer. Defining the transformer core, the loss of the transformer(i.e., eddy current loss, hysteresis loss, and transformer winding conduction loss) can be calculated. From this, the number of design result increase which mean a more accurate design is possible. From the simplified design method, the accessibility of the design process can be increased. The design estimation accuracy of the proposed AI model was based on Google colaboratory validation. Once the learning process is completed, the proposed AI-based transformer design can obtain various isolated-converter topology designs without further repeated training procedures. To verify the proposed power topology design method result, this dissertation uses MATLAB R2020a, Ansys Electronics desktop 2018.2, hardware-in-the-loop (HIL) experiment using OPAL-RT, and experiment with the design method result. MATLAB is used to power topology circuit analysis, ANSYS is used FEM to analyze transformer characteristics, and the Hardware in the loop experiment is used to verify design results in the real environment.