Data-driven Prediction Of Tissue Paper Physical Properties And Optimization Of Its Beating Process

Tissue paper is closely related to life,and correspondingly has more rigid standards and precision manufacturing technology.This dissertation takes tissue paper as the research object.Collecting data from raw materials to products properties within various key units of the production line of typical tissue paper enterprises.After data preprocessing,the tissue paper beating degree prediction model and physical properties prediction model were proposed.The model optimizing the beating process was proposed to meet the quality requirements and minimum cost of the original paper.Firstly,based on the problem that the beating degree of the pulp cannot be monitored online,a beating degree model is established based on the Gradient Boosting Decision Trees(GBDT)algorithm.With sampling and obtaining the fiber morphology and the beating degree data under different pulping combinations and different refining processes.After data preprocessing,a training data set is obtained.Cross-validation is used to test the accuracy of the beating degree model based on GBDT algorithm.The root-mean-square error(RMSE)of the model was 0.99,and the mean-relative error(MRE)was 3.96%,and the mean-absolute error was 0.78(°SR).The results show that the model has good accuracy.Under the same precision,compared with the Support Vector Machine Regression(SVR)algorithm,the GBDT algorithm operation time is one-fiftieth of the SVR in the same computer environment,that the GBDT operation speed is fast.The tree structure of GBDT model is analyzed and the relative importance of the 11 input variables is calculated.The relative importance of the specific energy consumption,flow rate and the rate of fibrillation is higher.They have a greater impact on beating degree and require a major monitoring in production.Secondly,in order to solve the problem of online prediction of the physical characteristics of household paper,the principal component analysis method was used to reduce the dimensionality of the data set,and the machine learning technique was used to establish the predictive models of average tensile strength,the elongation(Machine direction,MD),the elongation(Cross direction,CD)and water absorption.The industrial data was used for verification.The MRE of average tensile strength was 11.20%,the MRE of elongation(MD)was 10.48%,and the MRE of elongation(CD)was 8.23%,and the MRE of water absorption was 6.01%.It shows that the average tensile strength,elongation direction(MD)and water absorption have good model accuracy,and R~2 was higher than 0.8.The Lasso algorithm was used to analyze the importance of the model input variables(109 in total).The most notable of these is that the speed of the coil and the speed of the vacuum cage are significantly affecte the elongation(MD)and should be monitored.Finally,based on the established physical properties of the paper,an optimization model for the pulping process of tissue paper was established.The two objectives of minimizing are the cost and satisfying the quality requirements.Based on the evalution algorithm,the solution including slurry ratio and refiner's process parameters is found.Taking a certain type of tissue paper as a case,the results show that the papermaking process optimization model based on the fast non-dominated sorting algorithm can solve the paper pulping process with lower cost and quality requirements.