Artificial Neural NetworkModeling of Guar Gum Apparent Viscosity

The precise determination of apparent viscosity of guar gum solutions will help the mud engineer to better evaluate its behavior under diverse conditions. Therefore, it is essential to find a way to determine apparent viscosity at different situations. In this study, two empirical models compared to artificial neural network were applied to predict apparent viscosity values of guar gum solutions. At both empirical models, the apparent viscosity was considered as a function of concentration, temperature and shear rate. The results showed that the models have appropriate accuracy to estimate the apparent viscosity of guar gum solutions, whereas the coefficient of determination (R2) for both models obtained 0.993. But, both models had the limitation of initial guess for determination of equation constants. Besides, to determine the apparent viscosity, artificial neural network was applied using multilayer perceptron (MLP) and Levenberg- Marquardt learning algorithm. The architecture of neural network was designed as 3:4:1, whereas 3, 4 and 1 are representatives of input parameters, the optimum neuron numbers in hidden layer and output parameter which is the apparent viscosity, respectively. Two activation functions (logsig and tan-sig) were separately applied into hidden layer and finally the best function was selected. The whole data were divided into three parts including 70 % training (330 data), 15 % validation (69 data) and 15 % testing (69 data). In the end, R2 values of training (0.9993), validation (0.9959) and testing (0.9977) data were determined so that the best activation function (log-sig) was used in the hidden layer of neural network.