چكيده به لاتين
Solvent extraction is a separation process that is widely used in the production of high purity materials for nuclear applications and high technology. This process mainly involves selectively extracting the preferred metal from its aqueous solution to an organic phase. Process development, analysis, optimization and control of elements in this method is a intricate complicated task. Computer simulation provides useful tools in this area. The use of artificial neural networks to simulate equilibrium data processes such as liquid-liquid extraction in elements can be a useful and efficient tool. In this study, first, different experiments to investigate the effect of different parameters such as pH of solution, extractor concentration ("C" _"e" ), solution type (nitric or sulfuric), extractor type (D2EHPA or PC88A), temperature, extractor ratio the Samarium and lutetium extraction efficiency has been performed (in mixing extractors) and then the neural network simulation has been used to optimize the efficiency of samarium and lutetium extraction, which used experimental data to train and test the designed neural network. Based on preliminary experiments, six variables including pH of solution, extractor concentration, solution type, extractor type, temperature and ratio of extractors as neural network input parameters for samarium and neural network output samarium extraction efficiency four variables were considered, including solution pH of solution , extractor concentration, solution type and extractor type for lutetium neural network input and neural network output lutetium extraction efficiency was selected. In this study, neural network backpropagation model and Levenberg–Marquardt training function were used. For samarium neural network, three hidden layers with logsig transfer function and a purelin transfer function was used in the output layer and for lutetium network, two hidden layers with transfer function were used logsig and a pureline transfer function were used in the output layer. Based on the results, it was observed that the neural network predictions are very close to the experimental results and the error rate of the neural network is low and the results are acceptable, Also, the regression rate for neural network test was obtained for samarium is R^2= 0.99 and the mean error squares were MSE = 0.355, and for lutetium was R^2= 0.99 and MSE = 5.05.
