Comparative Study of Mooney Viscosity Prediction Models for Rubber Compounds based on ANFIS with Different Architectures

Abstract

Mooney viscosity is an important parameter in rubber compound industry because it is one of the processing windows and key properties of a rubber compound. As dynamic behaviors of rubber compounds are nonlinear and rubber product manufacturing process affects dynamic behaviors, an exact model for predicting Mooney viscosity has not been found. This paper presents the prediction models based on Adaptive Neuro-Fuzzy Inference Systems (ANFIS) for rubber compounds with different architectures and the effects of changes of certain parameters in each model on prediction performance. The database is collected from the historical data of manufacturing and then cleansed by removing errors in process and out-of-spec values. Both premise and consequent parameters of rules are created using the parameter initialization algorithm. The effects of different numbers of inputs and epochs, different input variables, and different interpretation methods are investigated. The simulation results show that the minimum value of RMSE for data testing is obtained by using the parameters initialization algorithm with 100 epochs, 3 inputs and OR interpretation method. Moreover, the lower number of epochs indicates the faster processing of the model. It is expected that the Mooney viscosity can be predicted and shown immediately at the end of mixing process.