Preform tool shape optimization and redesign based on neural network response surface methodology

Preform tool shape optimization using response surface method (RSM) was developed in this work. Neural network approximation model was employed for response surface construction in order to overcome the limitation of quadratic polynomial model in solving non-linear problems. A two-step axisymmetric forging problem was studied as an example using proposed method. Optimum was achieved by using pattern search optimization method to search response surface describing relationship between preform shape and die cavity fill ratio. In addition to that, with respect to the complexity of the optimum solution, the knowledge-based redesign concept was proposed. Simplified preform shape description model was built based on the knowledge extracted from previous optimization and additional shape optimization in terms of a new optimization objective was conducted to obtain a better redesign preform shape. Finally, comparison was made between the original optimal shape and redesigned one; better result was achieved by using the concept proposed.