Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
Rolling load prediction for thick plates via artificial neural networks
Submission Author:
Perseu Silva Soares , MG
Co-Authors:
Perseu Silva Soares, Yukio Shigaki
Presenter: Perseu Silva Soares
doi://10.26678/ABCM.COBEM2021.COB2021-1917
Abstract
Rolling is the most versatile and commonly used mechanical forming process in manufacturing. In reason of this and the great world market competition, it is imperative to produce better quality products and improve constantly the process control. This can be performed if a good rolling load prediction is carried out, that is a fundamental parameter to properly setup the rolling mill. Due to this fact, the conventional mathematical methods are not sufficient to maintain a good prediction ability for the rolling force, for this parameter prediction involves several nonlinear phenomena. A few conventional methods limitations can be avoided if an artificial neural network (ANN) model is applied to rolling load prediction. Therefore, from a comprehensive study of the fundamentals of hot rolling process, the network input variables were selected, namely: rolling speed, temperature, final thickness, width, reduction, equivalent carbon and work roll diameter. And, from the heavy plate rolling process industrial data it was possible to train a multilayer perceptron neural network with variable learning rate, developed in Python programming language. The neural network performance was compared to the Schultz model performance with the same data base and showed better capability for all three evaluation parameters (coefficient of determination, maximum absolute percentage error and the number of predictions with absolute error greater than 10%). However, Schultz model can be more easily applied and adjusted.
Keywords
Hot-rolling, rolling load, Artificial neural network (ANN), thick plates
