Speaker
Description
Artificial Neural Network (ANN)-based digital twins are increasingly utilized in scientific research applications, replacing Finite Element Method (FEM) models to save time and resources. These ANN models deliver rapid and accurate results comparable to the FEM models on which they are trained. Optimization, employing metaheuristic methods like the Genetic Algorithm and Particle Swarm Optimization, for High-Temperature Superconducting Magnetic Energy Storage (HTS SMES) solenoid-type coils tends to be time-consuming due to the increased variables from multiple FEM simulations.
An ANN-based model focusing on a single Double Pancake (DP) with varying inner diameter, number of turns, and supporting layer thickness was trained to calculate magnetic flux density distribution using a solved FEM model. Subsequently, SMES optimization was conducted to minimize length while maintaining a fixed stored energy of 1 MJ and hoop stress within 250 MPa. The axial placement of these coils at varying distances was considered in the optimization process. The magnetic field was determined by the vector sum of the magnetic fields from each coil. Real coded genetic algorithms were employed for optimization. The speed and results of the optimization were compared with magnetic field calculations based on the interpolation function available in MATLAB. Parallel processing was utilized within MATLAB for faster calculation
| Submitters Country | India |
|---|
