Application of General Regression Neural Network in Forecasting the Effect of 6PPd Antioxidant and Gamma Irradiation on NBR Rubber Mechanical Properties

Abstract

Different NBR rubber compounds are designed using different concentration of 6PPD antioxidant and different gamma irradiation doses. In this study, an attempt has been made to predict the crosslinking density and mechanical properties of rubber compounds using the general regression neural network (GRNN) technique. The GRNN are trained using the experimental results first for five gamma doses as input and different number (2 or 4) of corresponding mechanical properties as the program output. Two cases are studied first for NBR rubber with 2.5 phr 6PPD antioxidant, Second for NBR rubber without 6PPD antioxidant. After training the GRNN, it is simulated to estimate the rubber mechanical properties for twenty one gamma doses (including the five measured values) as input. In the first case two conditions are carried. The outputs in the first condition are crosslinking density with hardness and tensile strength with elongation at break. While the outputs of the second condition are the four mechanical properties together. The mean absolute percentage error (MAPE) of the second condition in the first case is smaller than that of the first condition. For this reason in case of rubber without 6PPD the outputs are taken as the four mechanical properties. The simulation results takes 15 sec for estimating mechanical properties corresponding to twenty one gamma doses, while the experimental results took approximately between seven to thirty days for five gamma doses. GRNN provides excellent predictions with a high degree of correlation depending on increasing the number of mechanical properties used in the training process. Also it predicts mechanical properties for a large number of doses that cannot be measured experimentally in a very short time.