This paper presents a method of determining the friction coefficient in metal forming using multilayer artificial neural networks based on experimental data obtained from strip drawing test. The number of input variab...This paper presents a method of determining the friction coefficient in metal forming using multilayer artificial neural networks based on experimental data obtained from strip drawing test. The number of input variables of the artificial neural network has been optimized using genetic algorithm. This process is based on surface parameters of the sheet and dies, sheet material parameters and clamping force as input parameters to train the neural network. In addition to demonstrating the fact that regression statistics model using genetic selection and intelligent problem solver are better than models without preprocessing of input data, the sensitivity analysis of the input variables has been conducted. This avoids the time-consuming testing of neurons in finding the best network architecture. The obtained results from this study have also pointed out that genetic algorithm can successfully be applied to optimize the training set and the outputs agree with experimental results. This allows reduction or elimination of expensive experimental tests to determine friction coefficient value.展开更多
文摘This paper presents a method of determining the friction coefficient in metal forming using multilayer artificial neural networks based on experimental data obtained from strip drawing test. The number of input variables of the artificial neural network has been optimized using genetic algorithm. This process is based on surface parameters of the sheet and dies, sheet material parameters and clamping force as input parameters to train the neural network. In addition to demonstrating the fact that regression statistics model using genetic selection and intelligent problem solver are better than models without preprocessing of input data, the sensitivity analysis of the input variables has been conducted. This avoids the time-consuming testing of neurons in finding the best network architecture. The obtained results from this study have also pointed out that genetic algorithm can successfully be applied to optimize the training set and the outputs agree with experimental results. This allows reduction or elimination of expensive experimental tests to determine friction coefficient value.
