Optimal Design of an Electromagnet Using Machine Learning Methods

Machine-learning methods are applied for the design of various electromagnetic systems, enabling efficient design work. The development of the design of electromagnetic systems using machine-learning methods, including optimal design methods, is an important and relevant issue. The replacement scheme of the magnetic circuit of the investigated electromagnet and its mathematical model, the implementation methods of design stages including the solution of direct and inverse problems of the magnetic circuit, the design of a control winding, etc., and an automated design system developed on their basis are presented. The problem of optimal design of an electromagnet is formulated, and the objective functions and limits of variables' values are presented. The automated design system has performed 1,000,000 designs of the electromagnet, and the results have been used to create a database that has been trained by using machine-learning algorithms. An optimal design method of the electromagnet is proposed, where a simple genetic algorithm is applied to the trained database. Moreover, two single-objective optimization problems are solved: one determining the minimum cost of the electromagnet's active materials (control winding wire and steel), and the other one determining the minimum power of the control winding. A multi-objective optimization problem is also solved, where the generalized objective function is obtained by using the ideal point method. The optimal design method implemented for the investigated electromagnet can also be successfully applied to the optimal design of other electromagnetic systems.