Optimization of Multiple Response Process by Neural Networks Based on Desirability Concept

In this paper, a method is proposed for Multiple Response Optimization (MRO) by neural networks and uses desirability of each response for forecasting. The used neural network is a feed forward back propagation one with two hidden layers. The numbers of neurons in the hidden layers are determined using MSE criterion for training and test data. The numbers on neurons of the first layer last layer are equal to the numbers of the factors and responses, respectively. After training the network, forecasting phase are done by giving different factor levels to calculate desirability of different experiments. Then total desirability is calculated. The optimal combination is which have the greatest total desirability. Finally, a numerical example is expressed to illustrate the capability of the neural network. The results of the research shows that although determining of suitable neural network is time consuming but have more accuracy than Response Surface Methodology (RSM). Also, the obtained optimal combination from RSM is one of optimal solution by neural network.