Modular neural network modeling of compressive strength of high-alumina bricks by using tangent function

In this study, a feed forward modular neural network consists of four layer is presented to predict the compressive strength of the high-alumina bricks. Model hierarchy consists of 12 parameters that affect the production directly. The inputs of the model consist of chemical composition (SiO2%, Al2O3%, TiO2%, Fe2O3%, CaO%, MgO%, Na2O% and K2O%), brick volume, sintering temperature, bulk density and apparent porosity and the output of the model is compressive strength. The transfer function used in the each neural mode was tangent function. Errors of the training feedback to the system by back propagation algorithm (BPA). Because of the limited data range used for training, the prediction results and the level of error 2.78% were satisfactory only within the same range. The ability of the model developed has proven that prediction of the compressive strength of high-alumina bricks established successfully. Sensitivity analysis indicated that bulk density and apparent porosity were the most significant inputs affecting compressive strength within the limits of the model whereas sintering temperature and brick volume were found to be relatively significant in the model.