Sinter Quality Prediction with Multimodal Information Fusion

The accurate real-time estimation of the quality index of iron ore sintering is essential for maintaining the stability of the production process. To address the dynamic challenges and time delays inherent in the sintering process, a hybrid approach combining deep residual network (ResNet) and long short-term memory network (LSTM) for predicting the concentration of ferrous oxide (FeO) in sintering processes is developed. The proposed ResNet-LSTM model integrates the strengths of conventional models and is capable of processing multiple-modality data, including both image and time series. The efficacy of the proposed model is evaluated through experimental analysis using real-world data collected from the iron ore sintering process. The results indicate that the ResNet-LSTM model outperforms competing models, achieving the lowest prediction error rates.