Physics-informed neural network for predicting hot-rolled steel temperatures during heating process

Abstract

The heating process in hot-rolled steel manufacture is a key step for product quality. It is desired that the steel slabs can be heated to the target temperature in the furnace with good temperature uniformity but low energy consumption. Due to the extreme conditions inside the furnace, the internal temperatures of steel slabs cannot be directly monitored. The existing methods use key thermal parameters to numerically analyze and simulate the hot-rolled steel temperatures during the heating process. However, these conventional methods require substantially intensive calculations and thus cannot be applied to the real-time simulation of hot-rolled steel temperatures in the furnace. This paper presents a novel predictive model that is developed based on a physics-informed neural network for prediction of the hot-rolled steel temperatures. It embeds thermophysical information in the neural network, in order to efficiently predict the internal temperatures of steel slabs during the heating process. The temperature predictions derived from the physics-informed neural network-based model were compared with the results obtained from an experiment and the conventional numerical method. The comparisons generally indicate that the physics-informed neural network-based model yields accurate temperature predictions, but with greatly improved computational efficiency over the conventional numerical method.