Artificial neural networks and metal forming optimization using genetic algorithms

Abstract

Nowadays computer simulations of metal forming processes using the finite element method are considered an essential tool, avoiding the use of costly trial-and-error methods. Given a set of input data, the computation of metal forming process evolution and its final results is defined as a direct problem. Optimization problems can be formulated as inverse problems. The aim of an inverse problem is to determine one or more of the direct problem input data, leading to a given desired result. Evolutionary genetic algorithms have been proposed aiming to solve optimization problems. These evolutionary methods can be computer time consuming due to the large number of necessary simulations with different input parameters. Introducing an Artificial Neural Network (ANN) the computer time spent on metal forming simulations can be significantly reduced. In this paper, we intend firstly to present an ANN model that can be trained using computer simulations of metal forming processes and the results stored; later the stored results can be used for predicting metal forming simulation outputs. Secondly, considering the ANN results, a genetic algorithm will be implemented in order to optimize a metal forming process example.