1. Inicio
  2. Atrás
Título: Determining finite element mesh density from problem specification usng neural networks
Autores: Dyck, Derek
Fecha: 1990
Publicador: McGill University - MCGILL
Fuente:
Tipo: Electronic Thesis or Dissertation
Tema: Engineering, Electronics and Electrical.
Computer Science.
Descripción: This thesis addresses the problem of how to determine the optimum level of mesh discretization required to solve a magnetic device accurately and efficiently using finite elements. Currently, most finite element packages require user intervention to assure that the mesh density is appropriate for the device. This requires that the user be knowledgeable in finite-element analysis and magnetic device design.
The approach introduced here uses a neural network which is trained to recognize significant geometric features and material properties from the description of a magnetic device. Based on its knowledge of meshing rules the neural network computes the mesh density required for an optimum mesh of the device. The neural network acquires this knowledge from examples of "ideal" meshes.
The system requires no user intervention and can be used either independently or as a preprocessor to an adaptive mesh refinement system.
Idioma: en