Multisource Heterogeneous Data Fusion-Based Process Monitoring of the Reheating Furnace in Steel Production

Abstract

The reheating furnace is the key piece of equipment in the hot rolling process of steel production. In order to fully exploit all of the data recorded from the production process representing different information, this paper designs a process monitoring algorithm with multisource information fusion by integrating multiple information to comprehensively monitor the operating state of the reheating furnace. Multisource information fusion combines process variable data of the reheating furnace and heating process data of the slab. To overcome the challenge of fusion of heterogeneous data due to different sampling patterns, univariate time series and multivariate time series data are fused by a transformer. In the fusion scheme, univariate time series data are represented by bidirectional gated recurrent unit for one-dimensional temporal representation, multivariate time series data are represented by temporal convolutional network for two-dimensional temporal representation, and multivariate time series data are represented by eigenvalue decomposition for correlation representation between variables. To evaluate the performance of the proposed method, computational experiments based on actual data are carried out. In univariate and multivariate time series representations, the highest predictions are obtained for bidirectional gated recurrent unit and temporal convolutional network by comparison with different regression algorithms, respectively. By comparing with fusing different fusion objects and different fusion schemes, the proposed algorithm achieves the highest accuracy (91.33%), precision (91.46%), and recall (92.59%), proving the effectiveness of the fusion approach. The process monitoring performance is compared with multivariate statistical process monitoring algorithms, which achieve the highest accuracy (95%), precision (93.45%), and recall (97.08%).