Application On Optimal Prediction Of Mooney Viscosity Of Chloroprene Rubber By Neural Network Control

Mooney viscosity is an important quality indicator of chloroprene rubber. Although chloroprene rubber has been produced for many years in China, the Mooney viscosity of chloroprene rubber is still unstable resulted by the difficulties of real-time control and the measuring lags. So the solution to predict the money viscosity is highly desirable.In this article, a soft-sensing model for Mooney viscosity measurement is built. The study is based on a SN12 X chloroprene rubber produced in a local chemical plant. The model is based on the technology of soft-sensing and the theory of Neural-Networks. The programming tool is Neural Network Toolbox of MATLAB. The establishment of the model contains four stages. Firstly, the mechanism of the chloroprene rubber and collected sample data are analyzed. Secondly, preprocess and analyze principal component to determine instrumental variables. Thirdly, establish soft-sensing model of the Mooney viscosity of the chloroprene rubber based on Neural-Networks. Finally, the model is optimized and validated.The determination of the instrumental variables and the establishment of the model are mainly discussed in this paper. The scree plot and the loading plot are used for data analysis to determine instrumental variables. Three kinds of methods, including the entire-element method, the combination-element method and the main element method, were proposed to deal with different types of input variables and the optimum model is chosen by comparing multiple accuracy indicators.The main element method is found to be the best according to the results in continuous training and simulation of the three networks. The model is switched from a single-layer structure to a double-layer structure in order to solve the problems appearing in the predicting process. The BP neural-network is proved to be the best according the experiments. The maximum relative error between the actual Mooney viscosity of the chloroprene rubber and the prediction of the model meet the requirements in real production environment.Finally, the trained model has been used to predict the Mooney viscosity of the chloroprene rubber on the plant for one month. It proves that the model functions properly in real production environment.