Abnormal Condition Identification for the Electro-fused Magnesia Smelting Process Based on Condition-relevant Information

Abstract To improve the accuracy of feature representation and abnormal condition identification, a new abnormal condition identification method, named integrating multiple binary neural networks based on condition-relevant information (CRI-MBNN), is presented for the electro-fused magnesia smelting...
Ausführliche Beschreibung

Abstract To improve the accuracy of feature representation and abnormal condition identification, a new abnormal condition identification method, named integrating multiple binary neural networks based on condition-relevant information (CRI-MBNN), is presented for the electro-fused magnesia smelting process in this study. Firstly, the features related to each specific abnormal condition, which is named condition-relevant information (CRI), are analyzed and extracted from the multi-source heterogeneous information with the help of limited and consensus domain knowledge. Then, the CRI is fused at the feature-level to provide a comprehensive representation of each abnormal condition. Furthermore, for each abnormal condition, a binary neural network (BNN) is established based on the fused feature. They are further integrated according to the frequency of each condition in the actual production process to form the final abnormal condition identification network, i.e., CRI-MBNN. Finally, the effectiveness and feasibility of the proposed CRI-MBNN are verified by the electro-fused magnesia smelting process. Ausführliche Beschreibung