An exergy-related soft sensor modeling framework for quality prediction in multimode hot rolling process based on bidirectional minimal gated unit and broad learning system

Abstract Quality prediction plays a vital role in enhancing the product quality and economic benefits of manufacturing processes. Variability in working modes, dynamic behaviors observed in sequential data, and the inadequate utilization of energy consumption information have resulted in poor universality and diminished prediction accuracy of conventional deep learning-based quality prediction approaches. To this end, an exergy-related soft sensor modeling framework is designed in this paper. Specifically, an integrated correlation coefficient is proposed to select exergy-related variables. In order to fully use the historical and future dynamic information of exergy-related variables, bidirectional minimal gated unit (BiMGU) is combined with broad learning system (BLS) for feature extraction and quality prediction. After that, the expansion ability of the proposed method is further discussed. Finally, the proposed soft sensor framework is applied to a real hot rolling process (HRP). The experimental results indicate that the proposed framework has higher prediction performance than the other deep learning methods. Compared to traiditional BiMGU, the mean absolute error (MAE) and the root mean squared error (RMSE) of the proposed method are reduced by 0.0105 and 0.0114, respectively.