The optimisation of electromagnetic devices using a combined finite element/neural network approach with on‐line training

• J. Seguin ^[1] ; F. Dandurand ^[1] ; D.A. Lowther ^[1] ; J.K. Sykulski ^[2]
  1. [1] McGill University

     McGill University

     Canadá

     
  2. [2] Electrical Engineering Department, Southampton University, Reino Unido
• Localización: Compel: International journal for computation and mathematics in electrical and electronic engineering, ISSN 0332-1649, Vol. 18, Nº 3, 1999, págs. 266-274
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • The paper presents a novel method of utilising neural networks for optimisation systems. First, a conventional magnetic circuit model of the device is developed to create a set of sensitivity rules which guide the optimisation. The rules are coded in a knowledge-based neural network. Second, an error network is developed to correct the approximations inherent in the magnetic circuit approach and this combines with the first network to generate realistic outputs. Finally, the error network can be trained on-line with a finite element system. Over time, the network replaces the finite element analysis, thus speeding up the optimisation process.