Identification of Partial Molten in Electrofusion Magnesium Furnace Based on Bayesian Multi-Scale Convolutional Neural Networks

The advancement of industrial automation and intelligent manufacturing technologies has made the identification and control of operational conditions in electrofusion magnesium furnace critical for enhancing production efficiency and product quality. Traditional methods for identifying partial molten anomalies suffer from issues such as high subjectivity, limited adaptability, and insufficient real-time performance. In this study, we propose a Bayesian multi-scale convolutional neural network (MSCNN)-based method to identify partial molten anomalies in electrofusion magnesium furnace. This method analyzes key variables, such as infrared video and electric current, and integrates the strengths of Bayesian theory and convolutional neural networks. The multi-scale convolutional layer captures feature information at different scales, while the Bayesian inference layer enables probabilistic reasoning. Experimental results demonstrate that the MSCNN model surpasses traditional methods in accuracy, robustness, false positive rate, and leakage rate, highlighting its potential for real-world production applications. This study offers a novel approach for identifying partial molten anomalies in electrofusion magnesium furnace and serves as a reference for condition monitoring and fault diagnosis in other industrial equipment.