Springback and Geometry Prediction – Neural Networks Applied to the Air Bending Process
This paper describes the application of neural network techniques to sheet metal bending process, particularly for the prediction of springback phenomenon and bending part final geometry (final radius and bending angle). Springback is an important unwanted change in shape causing accuracy problems. Traditional and new simulation techniques (FEM) of springback minimizing are laborious trial-and-error procedures that involve long cycle times and cost increases. To reduce the trial-an-error procedure, an artificial neural network (ANN) model is developed as an approximator. A back propagation neural network model has been developed using experimental data from several tension and bending tests performed on aluminium and stainless steel. The convergence of the mean square error in training came out very well and the performance of the trained network has been tested with unseen kept back data from experiments and found to be in good agreement.
KeywordsSheet Metal Back Propagation Neural Network Sheet Metal Part Bend Radius Punch Displacement
Unable to display preview. Download preview PDF.
- 1.Vasile, D., Vasile, D., Thamodharan, M., Wolf, A.: The use of neural network in metal industry. In: IEEE Conference Record, vol. 2, pp. 1104–1108 (2000)Google Scholar
- 4.Garcia-Romeu, M.L.: Contribucion al estudio del doblado al aire de chapa. Modelo prediccion angulo de recuperacion y radio de doblado final. Th. Diss., Girona (2005)Google Scholar
- 5.Mehrotra, K., Mohan, C.K., Ranka, S.: Elements of artificial neural networks. MIT Press, Cambridge (1997)Google Scholar