Using Artificial Neural Networks to Predict Process and Quality Parameters inSteel Strip Production

摘要

In this work the prediction of the mechanical properties of steel sheets and theforce in temper rolling are studied by using MLP neural networks. The necessary process data was obtained from Rautaruukki Strip Products. The chosen approach is justified by reviewing related reports in the field of industrial ANN applications. The process and data are described briefly. The results are described and discussed separately for mechanical properties, temper rolling force and the applicability of MLP models. The results show a good applicability of MLPs to predict the mechanical properties and temper rolling force. The MLP modelling is also used for identifying the sources of deviation in the mechanical properties of steel sheets. It is demonstrated that the lack of variance in some variables prevents the MLP from creating physically sound dependency between the variables and ouputs. A method for detecting input vectors with unfamiliar variable combinations that can produce unpredictable outputs is proposed and verified with examples. The conclusions of the study is that MLPs are well suited for predicting the mechanical properties and temper rolling force in steel strip production. (Copyright (c) Helsinki University of Technology).