Authors: Fayiz Y. Abu Khadra; Jaber E. Abu Qudeiri
Addresses: Faculty of Engineering, King Abdulaziz University, P. Box 344, Riyadh 21911, Saudi Arabia ' Advanced Manufacturing Institute, Faculty of Engineering, King Saud University, P. Box 800, Riyadh 11421, Saudi Arabia
Abstract: In this paper, two metamodelling techniques namely, the neural network and the response surface methodology are used and compared to approximate a multidimensional function to predict the springback amount of metallic sheets in the bending process. The training data required to train the two metamodelling techniques were generated using a verified non-linear finite element algorithm developed in this research. The algorithm is based on the updated Lagrangian formulation, which takes into consideration geometrical, material non-linearity, and contact. A neural network algorithm based on the back propagation algorithm has been developed. This research utilises computer generated D-optimal designs to select training examples for both metamodelling techniques so that a comparison between the two techniques can be considered as fair. Results from this research showed that the neural network metamodels outperform the response surface metamodels.
Keywords: metamodelling; springback; neural networks; D-optimal designs; response surface methodology; RSM; bending process; finite element method; FEM.
International Journal of Computational Materials Science and Surface Engineering, 2013 Vol.5 No.2, pp.85 - 101
Received: 20 Dec 2011
Accepted: 22 Jul 2012
Published online: 08 Apr 2013 *