| This thesis summarizes the methods and development process of a soft sensor based on data-driven method. Besides, an ordinary demand analysis of soft sensors and its application actuality in pulping and paper-making process are present. What’s more, detailed information about how to develop a soft sensor with Case-Based Reasoning(CBR) is given in this thesis. Considering there are some key process variables and quality indexes could not be measurement online, the purpose of this study is to explore the technology roadmap for the application of CBR technology for estimation the key variables which could not be read by online sensor and optimization the corresponding process in pulping and paper-making process.Soft sensors for estimation the outlet consistency and freeness of refined pulp in the alkaline peroxide mechanical pulping(APMP) high-consistency refining process was developed based on CBR method. It is difficult to build a good mechanism model for prediction of consistency and freeness because APMP process is multivariable and mechanism complicated. After analysis of the APMP process, correlation relationship between primary variables and the preliminary secondary variables was investigated. Principle component analysis(PCA) and multicollinearity diagnostics were also applied to detect multicollinearity problems with preliminary secondary variables to avoid redundancy. Finally, eleven secondary variables for estimation of freeness and consistency are selected. Those variables were including the power of MSD devices, flow of applied chemicals, specific energy of refiner, quantity of dilution water, gap of refiner plates, etc. The CBR soft sensor could give good prediction results. A papar mill field data was adopting for this simulation. RMSE and CV-RMSE for consistency are 0.71 and 1.72%; RMSE and CV-RMSE for freeness are 4.29 and 0.73%.After a mechanicsm analysis, the beating degree and wet weight of unrefined pulp, pulp flow, pulp consistency, power of refiner of a low-consistency refining process in stock preparation are selected as secondary variables, and soft sensors based on CBR method was developd for beating degree and wet weight predictions. In this case study, the CBR soft sensor could learn more knowledge to improve the prediction ability and need not the online correction step through case revise and case retain. The paper mill data are used to test the performance of CBR soft sensor. RMSE and CV-RMSE for beating degree are 1.3 and 4.32%; RMSE and CV-RMSE for wet weight are 0.5 and 19.09%. Besides, CBR soft sensor could give the estimation value and prediction performance still could be accepted when some secondary variable conld not read online because the instruments failed. Based on the CBR soft sensor for beating degree and wet weight, a simulation of operation optimization was applied in a refining process. The results show it could save 7 Kwh each ton air dry pulp in pulp refining process.To prediction the kappa number of unbleached pulp in a bleaching process, a soft sensor based on CBR method was developed. The soft sensor for kappa number is a experts experiences case base and could be used to detect whether the process condition is good. An optimization model was developed based on this CBR kappa number soft sensor and a search step used golden section method to reduce calculation time was obtained by analysis of the real control experience. These models are employed in D/C stage to optimize the bleaching process. When serviced in the field site, a good result was coming out. The brightness was shift 60.7 %ISO from 62.9 %ISO, and the variance was reduced.All the results show the CBR method could be application in pulping and paper-making process for estimation the process variables and optimization the process. However, this study is only an application research and more effort and attention should pay to how to build a case search index and optimize the weight of feature attribute when these models are serviced in a pulp and paper mill in the future. |
PDF Full Text Request