Artificial Neural Network Modeling for Prediction of Roll Force During Plate Rolling Process

Abstract

Accurate prediction of roll force during hot rolling process is very important for model based automation (Level-2) of plate mills. Exit thickness of plate for each pass is calculated from roll gap, mill spring, and predicted roll force. The response of gauge control hardware is highly dependant on the accuracy of prediction of roll force. Traditionally, mathematical models based on plane homogeneous plastic deformation theory are used for prediction of roll force. This method is based on many simplified assumptions which are not valid for actual industrial application. An artificial neural network (ANN)–based data driven model has been developed for prediction of roll force during plate rolling process. A very accurate data acquisition system has been installed in Plate Mill of Bhilai Steel Plant through which input and output parameters have been recorded. For a particular grade of steel, inputs to the ANN model are roll gap of previous pass, roll gap of current pass, rolling temperature, rolling speed, plate width, and pass number (6 inputs). The model output is roll force (1 output). In this article, the methodologies of development, training, and validation of ANN model has been discussed. Feed forward network has been chosen as ANN structure. Back propagation algorithm with variable learning rate and conjugate gradient optimization of cost function has been chosen as network training methodology. The model was found to be highly accurate with r-square value about 0.94.