A neural networks inversion-based algorithm for multiobjective design of a high-field superconducting dipole magnet

In this paper, an original algorithm to solve multiobjective design problems, which makes use of a neural network (NN) inversion method, is presented. The proposed approach allows us to explore the solutions directly in the objectives space, rather than in the parameters space, with a great saving of computation time in the reconstruction of the Pareto front. A multilayer perceptron NN is first trained to solve the analysis design problem. The inversion of the neural model allows us to obtain the design parameters, starting from the desired requirements on all the conflicting multiple objectives. The performance of the method is demonstrated by its application to the design of a high-field superconducting dipole magnet, where a tradeoff between the superconductors volumes is required in order to obtain a prescribed magnetic field value in the dipole axis.