Modelo de previsão de propriedades mecânicas de perfis estruturais laminados a quente: uma abordagem em redes neurais artificiais

Abstract

The main objective of this work is to develop an empirical mathematical model used to predict the mechanical properties of hot rolled steel sections. This model is based in some production variables, including the rolling process and the steel chemical composition, using, as source, the production data bank. The methodology chosen is the Artificial Neural Net (ANNs) which is capable to predict results with good accuracy and has the important characteristic of learning and updating of its internal structure. The last characteristic allows the model to reflect any change occurred in the studied process. Some statistics tools were used aiming to help the model development: Variance Analyses, Linear Regression Analyses, Multiple Linear Regression Analyses and Scatter Plot. These tools helped to comprehend and interpret the data variability. They were used to define the ANN Architecture. It was observed that the results from final model were according to the expected and consistent with the metallurgical trends. The results were superior to that obtained from tradicional statistical methods. It is possible to verify the effect of each variable in isolation. It was verified that que data dispersion is an important aspect for the success of any prediction model. The lower the data variability, the better the model prediction performance. The model will allow that the chemical composition design will be performed with high accuracy, aiming reduced production costs and it will improve the comprehension of the effect of each process variable on the steel products mechanical properties. With the development of the Simulator System it will be possible to visualise the simultaneous effect of two variable process on the steel products mechanical properties.