Modeling and parameter optimization of the papermaking processes by using regression tree model and full factorial design

One of the major challenges in the pulp and paper industry is taking advantage of the large amount of data generated through its processes in order to develop models for optimization purposes, mainly in the papermaking, where the current practice for solving optimization problems is the error-proofing method. First, the multiple linear regression technique is applied to find the variables that affect the output pressure controlling the gap of the paper sheet between the rod sizer and spooner sections, which is the main cause of paper breaks. As a measure to determine the predictive capacity of the adjusted model, the coefficient of determination (R-2) and s values for the output pressure were considered, while the variance inflation factor was used to identify and eliminate the collinearity problem. Considering the same amount of data available by using machine learning, the regression tree was the best model based on the root mean square error (RSME) and R-2. To find the optimal operating conditions using the regression tree model as source of output pressure measurement, a full factorial design was developed. Using an alpha level of 5%, findings show that linear regression and the regression tree model found only four independent variables as significant thus, the regression tree model demonstrated a clear advantage over the linear regression model alone by improving operating conditions and demonstrating less variability in output pressure. Furthermore, in the present work, it was demonstrated that the adjusted models with good predictive capacity can be used to design noninvasive experiments and obtain