Modelling of thermal shrinkage of seamless steel pipes using artificial neural networks (ANN) focussing on the influence of the ANN architecture

This paper presents two main novel findings. (1) The first finding is the development of an artificial neural network (ANN) model for thermal shrinkage of seamless steel pipes, which represents a new application for ANNs. Mill operators need such fast and accurate models to predict the final pipe outer diameter at ambient temperature based on the hot state immediately after rolling. The goal of this work was to lower the reject rate. However, small relative changes in the diameter are currently difficult to predict by using conventional ANNs. Therefore, a more sensitive target variable and a modified ANN architecture were applied to solve this problem. Data for training and validation were obtained from measurements on a hot-rolling mill. (2) The second finding is based on an investigation performed to determine the number of hidden neurons affected the model response, considering the data used. The knowledge obtained helps to determine the most suitable number of hidden neurons and to prevent overfitting. No generally accepted solution to these problems had previously been proposed in the literature. Consequently, this paper significantly supplements current research studies that describe applications of ANNs.