Study on the vision transformer numerical regression based prediction model of the outlet induction heating set temperature for a hot-dip galvanised steel strip

Abstract

The quality of galvanising alloying for a continuous hot-dip galvanised steel strip subjected to the induction heating (IH) treatment after being delivered out of a zinc pot can be effectively improved. The temperature of an IH outlet setting strip represents a key factor affecting the alloying quality. However, traditional prediction models have low accuracy, which requires additional manual intervention and poses difficulty in improving the quality of coating alloying. To address these challenges, this study proposes the vision transformer-based algorithm for numerical data regression, named the vision transformer numerical regression (ViTNR) algorithm. In addition, using the historical datasets of IH outlet strip temperature in the hot-dip galvanising production, this study develops a ViTNR prediction model for IH outlet setting strip temperature. The proposed ViTNR model of IH outlet strip setting temperature is compared with four typical deep learning-based models, including convolution neural networks, and five shallow learning-based models, including the AdaBoost model. The comparison results show that the proposed ViTNR model can achieve higher prediction accuracy and better generalisation ability in the IH outlet strip temperature prediction application scenarios compared to the comparison models.