Learning and Generation of BEM adaptive Meshes by Using Neural Network.

In this study, the hr-adaptive meshing method for boundary element (BEM) analysis is developed and implemented into the expert system for BEM analysis developed by the authors in a previous paper. The adaptive meshes obtained by this system are made to be learm by using a multilayered neural network with an automatic reduction of association units. It is found that it becomes possible to generate adaptive meshes for unlearning problems by using the neural network. This method is expected to be very useful particularly for the BEM analyses of problems with the shape slightly changing in the optimal deisgn.