Neural Network-Based Multi-Objective Optimization of Adjustable Drawbead Movement for Deep Drawing of Tailor-Welded Blanks

Abstract

To improve the formability in the deep drawing of tailor-welded blanks, an adjustable drawbead was introduced. Drawbead movement was obtained using the multi-objective optimization of the conflicting objective functions of the fracture and centerline deviation simultaneously. Finite element simulations of the deep drawing processes were conducted to generate observations for optimization. The response surface method and artificial neural network were used to determine the relationship between variables and objective functions; the procedure was applied to a circular cup drawing of the tailor-welded dual-phase steel blank. The results showed that the artificial neural network had better prediction capability and accuracy than the response surface method. Additionally, the non-dominated sorting-based genetic algorithm (NSGA-II) could effectively determine the optima. The adjustable drawbead with the optimized movement was confirmed as an efficient and effective solution for improving the formability of the deep drawing of tailor-welded blanks.