An Improved VAE-Based Quality Prediction Method with Application to a Hot Strip Rolling Mill Process

In a hot strip rolling mill (HSRM) process, the control of the crown quality of a strip steel is of paramount importance. In this paper, a distributed crown prediction model based on an improved Variational Autoencoder (VAE) framework is proposed. The model integrates a multi -layer VAE architecture to better accommodate the complex distribution of actual production data. Utilizing real production data from Ansteel's HSRM process, the model was subjected to extensive training and validation to optimize its predictive accuracy. Then in comparison with centralized VAE and VAE models with specialized prediction layers, the distributed model demonstrated exceptional predictive performance, with evaluation metrics including mean absolute error, root mean squared error and mean absolute percentage error. It can be seen from the application result that: the proposed method shows a better prediction result compared with the traditional methods.