Modeling of tensile index using uncertain data sets

Abstract The objective of this investigation is to analyze and model tensile index. Two approaches are used, one based on training and validation data, while the other novel approach tests models using all possible combinations of data points. This approach is focused on small data sets which have here been obtained from nineteen pulp samples at different refining conditions in a full-scale TMP production line with a CD-76 refiner as a primary stage. From each pulp sample twenty handsheet strips for tensile index measurements were performed. Initially, specific energy and the external variables (dilution water feed rates and plate gaps) are used as predictors in a modeling approach based on an adjusted {R²} approach. Thereafter, the resulting models are compared with a combination of specific energy and internal variables (primarily consistencies) obtained from temperature measurements inside the refining zones using a soft sensor concept. It is found that specific energy and internal variables as predictors outperform the external variables when estimating tensile index.