Prediction of properties of rubber by using artificial neural networks

Abstract

Different rubber formulations were designed using nitrile rubber and a mixed crosslinking system consisting of sulfur/accelerator and electron beam radiation. Based on the experimental results, an artificial neural network (ANN) was constructed to simulate the mechanical properties and volume fraction of rubber. The ANN could predict accurately the above properties for a series of nitrile rubber compounds. However, the number of training data played a key role in the ANN predictive quality. In addition, the more complex the nonlinear relation between input and output was, the larger was the number of training dataset required. The predicted results were further validated using another mathematical model. The constructed ANN was verified with a completely different styrene butadiene rubber system. The prediction was found to be extremely good.