Application of neural networks for predicting convective impact heat transfer coefficients in post-rolling steel plate cooling

Abstract

The water jet impingement boiling heat transfer coefficient (HTC) during controlled cooling after rolling is a critical parameter governing microstructural evolution and temperature distribution in steel plates. Conventional experimental measurement and numerical simulation approaches show limited applicability and high computational cost, which restricts their use for real-time industrial prediction. To overcome these limitations, this study adopts neural network–based methods and develops backpropagation (BP) and long short-term memory (LSTM) models to construct HTC prediction models under diverse operating conditions. Rapid prediction capability is achieved, and the engineering applicability of the proposed models is systematically evaluated. The results show that the LSTM model outperforms the BP model in representing the nonlinear coupling between operating parameters and HTC, as well as process-dependent characteristics. Specifically, the LSTM model yields an average error of 0.20 with a standard deviation of 29.77, whereas the BP model shows an average error of −3.87 with a standard deviation of 77.91. Moreover, the LSTM model maintains high prediction accuracy in the critical boiling heat transfer regime. The predicted workpiece temperature evolution shows strong agreement with experimental measurements, which supports the reliability of the neural network–based approach. Overall, this study shows that neural network–driven HTC models can deliver millisecond-level response times while maintaining high accuracy, providing an efficient and practical solution for real-time prediction in controlled cooling processes after steel plate rolling.