DESIGNING AN ELECTROMAGNET WITH A STRAIGHT ARM USING MACHINE LEARNING METHODS

Abstract

The studied electromagnet is used in switching devices as a drive, in high-speed trains on a magnetic cushion as a magnetic suspension, etc. In the previous works of the authors, an electromagnet with a forward arm was studied, in particular, a mathematical model of the system was developed, the forward and reverse problems of the magnetic circuit were formulated and solved, the methods of calculating the control coil and the methodology of optimal design of this electromagnet using the genetic algorithm were described, an automatic system was developed, including all stages of design of an electromagnet by the classical method: setting the input values of the parameters (electromagnetic force between the poles of the cores and the armature, magnetic induction in the air gap, ambient temperature, etc.), determining the dimensions and choosing the materials of the magnetic wire and coil, calculating the magnetic flux, electromagnetic force, etc. The problem of this research is to design a database and to train them using machine learning algorithms, to estimate and analyze the predicted values of electromagnet parameters. The database, compiled with the calculation data obtained as a result of the design of an electromagnet with a straight arm, includes the results of one million design options. The training of the compiled database was carried out using the algorithms of the machine learning package Regression Learner in the MatLab program environment: Regression Trees (Fine Tree, Medium Tree), Linear Regression Models (Linear, Interactions Linear, Robust Linear), Support Vector Machines - SVM (Linear), Neural Network (Narrow, Bilayered, Trilayered). The results of training and the characteristics of forecasting are presented and analyzed in the work.