Abstract: The factors influencing drawbeads force are firstly analyzed making use of gray relational analysis, and the main factors are obtained. Making use of Latin hypercube, the main factors are sampled. The box forming is simulated with DYNAFORM, and the sample objective data are obtained. The thinning and thickening and major strain are selected as input parameters, and drawbeads geometry parameters are selected as output objective. The inverse model of drawbeads geometry parameters is established. The back propagation(BP) network weights are optimized with genetic algorithm(GA). Compared with the predictive values by BP, the parameters values by GA-BP are more accuracy, and its accuracy can be greatly improved. It showed the GA-BP mapping can greatly improve the predictive capability. Based on the GA-BP, the nonlinear function of the forming thickening and drawbeads geometry parameters is obtained making use of the optimized weights, and the function is optimized with GA. Finally the optimum geometrical parameters of drawbeads are obtained. The numerical simulations of box before optimization and after optimization are compared. The results show the optimized drawbeads can greatly improve the formability of sheet metal forming.

Key words: Back propagation neural network, Drawbeads, Genetic algorithm, Gray relational analysis, Inverse and optimization