Previsão da evolução da resistência de um aço em uma sequência de passes durante a laminação a quente utilizando lógia fuzzy

Abstract

The prediction of flow stress and rolling force in steels during the hot rolling process remains a recurrent challenge in modern steelmaking, as it involves complex thermomechanical phenomena such as recrystallization, strain acumulation, and carbonitride precipitation. Classical empirical models, although widely used, show limitations in representing the nonlinear behavior of the mean flow stress (MFS) and the applied rolling force under real industrial conditions. In this context, the present work introduces the development of a prediction model based on fuzzy logic, structured in a tree format (Fuzzy Inference System Tree – FIST), to estimate the MFS and rolling force during the hot rolling of a niobium microalloyed steel plates. The study used industrial data obtained from a steel plant, considering four final thicknesses (12, 16, 20, and 30 mm), processed in eight successive finishing passes. The fuzzy model was implemented in MATLAB, employing triangular membership functions, Sugeno-type rules, and supervised training using the Particle Swarm Optimization (PSO) and Pattern Search algorithms. The results were compared with well-established empirical models, such as those proposed by Misaka and Poliak, showing that the FIST achieved superior predictive performance, especially under conditions of strain accumulation and below the non-recrystallization temperature (Tnr). A critical analysis of the results demonstrated that the fuzzy model can capture the nonlinearities of the process and integrate metallurgical mechanisms such as dynamic (DRX), static (SRX), and metadynamic (MDRX) recrystallization while maintaining high interpretability. The application of the FIST model represents a promising alternative for predicting MFS and rolling force in industrial plate rolling processes, contributing to the optimization of operational parameters and the improvement of final steel properties.