Application Of Pattern Recognition In Optimization And Modeling For Carbon-black Production Process

Carbon black (CB) is created from the incomplete combustion or cracking processes of hydrocarbon material (solid, liquid or gas). CB is an important padding of rubber products (mostly tires). Rubber industry can't successfully develop without the assistance or cooperation of CB industry. In allusion to the disadvantageous condition on the production of CB principally depending on experience, pattern recognition (PR) was put forward to the optimization of CB production process. We applied PR to setting up the model between technics parameters and quality targets. In addition, we have developped the optimization system for carbon black technics parameters.Firstly, several methods in pattern recognition were applied to mine the data information in carbon black production process. The methods, such as cluster analysis(CA), non-linear mapping(NLM), principal component analysis (PCA), whitening transformation-linear map (WTLM), multi-discrimination vector (MDV) and so on, were used. After the comparison for the results of these methods, we found that the optimization results from MDV were better, and two discrimination vectors nearly contain one hundred percent of the information in original samples. From the sample classification figure, local optimal area and optimization direction were found out. As the calculated results based on MDV were discussed, primary control parameters for carbon-black production were concluded, and the change tendency for each parameter in the optimization process was resolved in detail. By the combination of the results of PR and actual production, the optimal scheme was established. The results indicated that if the data-mining technology based on pattern recognition was introduced into carbon black industry, the ability of producing high quality carbon-black would be improved, and the production technology would be ameliorated. All the results have some directive function for the development of carbon black industry.Secondly, we have established the linear regression model between parameters and quality targets using these methods in PR, such as principal component regression (PCR), multiple linear regression (MLR), stepwise regression (SR) andpartial least squares regression(PLSR). For validation data set and prediction dataset, the change range of the calculated value is confined in these linear model, and the relative error between the prediction values and the actual values is more than 5.30%, which doesn't truly reflect the actual production. We have also set up the nonlinear mapping model with the artificial neural networks based on error back-propagation (ANN-BP), Radial Basis Function Networks (RBF) and ELMan neural networks. By the comparison for these models, we concluded that the average prediction error(APE), average relative prediction error(ARPE), and Sum-Squared Error (SSE), in RBF were smaller than others'. The fitting relative error in training data set is less than 1.00%, and ARPE for ISSA and DBP are 1.74% and 1.44% respectively, which satisfies the actual demand. The results show that there exists complex nonlinear in the production process of CB, the linear regression methods are not fit for setting up the prediction model, with favorable approach ability and '.earning rate, while RBF is suitable to construct the model of carbon-black production process. RBF has preferably solved the problem of setting up prediction model in carbon-black production process.In addition, we have developped the optimization system of carbon black technics parameters. We have compiled the calculation program with MATLAB. Microsoft SQL-SEVER 2000 was database, and visual C# was the program language with microsoft visual studio ?NET as development plat. The system has realized the combination between the algorithm programs based on MATLAB and visual C# program. We have carried out the design of system module and database, and exploited the interface for land, login and the input interface of data set. The system has the functions, such as the update of data, the pretreatment of data, the appl...