OTIMIZAÇÃO DA ANÁLISE DE RESISTÊNCIA À TRAÇÃO DA FIBRA DE CELULOSE POR MODELO PREDITIVO

Abstract

In pulp production, monitoring several parameters that guide industrial decision making is essential. In this context, numerous analyses are carried out in a quality laboratory, some of which have reduced frequency due to factors such as a high number of procedural steps, the need for a stable environment, and the long time required to obtain results. Among these tests, the tensile strength of cellulose fibers stands out as one of the most important, since in a scenario where fluctuations in the short-fiber market price are constant, achieving optimal parameter conditions in production is advantageous, as it directly influences pulp productivity on the paper machine. Additionally, the strong correlation between tensile strength and fiber morphology data indicates that developing a predictive model is an effective optimization tool, especially considering the short time required to determine anatomical characteristics. Thus, the aim of this study was to use laboratory fiber morphology data to predict tensile index results through the development of a multiple linear regression model. Python was used as the tool for statistical data analysis, with scikit-learn as the main library. The variables selected for model configuration were Mass-Weighted Average Fiber Length, Fibrillation Ratio, Curl, and Fines Ratio. The final model achieved a MAPE of 2,89% and an RMSE of 2,49 N·m/g, demonstrating strong predictive capability, which was confirmed by statistical significance tests. Therefore, the model can be applied as a tool in the analytical routine, reducing the estimated analysis time from 11 hours to only 5 hours.