The use of Neural Network Modeling for the Selection of Thermophysical Coefficients

In this work, a simulation model of the thermal state of the work rolls at the hot rolling mill 2000 (m) of Public Joint-Stock Corporation Magnitogorsk iron and steel works (PJSC MMK) is built. A neural network was used as a tool for adjusting the thermophysical transfer coefficients of hot rolling mills. The constructed neural network is a multilayer perceptron, consisting of an input layer, two hidden layers, and an output layer. The neural network is implemented in PyTorch, a scientific computing package based on the interpreted Python programming language. To train the neural network, we used the basic data array obtained from the control system of the hot rolling mill. During industrial experiments, the temperature of the work rolls was measured using HotFind-DXT series thermal imagers developed by SDS. A neural network has been built that allows adapting the thermophysical heat transfer coefficients with an average relative error of 11.7%. A neural network will speed up and simplify the selection of thermophysical coefficients for models of the thermal state of work rolls and adequately assess the current thermal state of work rolls during the rolling process. Using a neural network, a model of the thermal state of work rolls was developed