Abstract: A method of recognizing abnormal condition levels during heating and melting process of fused magnesium furnace utilizing multi-source information fusion is proposed. Current， image and sound signal features are extracted to be serialized and normalized under the premise of analyzing abnormal condition degrees during heating and melting process. Different features are selected according to the characteristics of various anomalies. The models for recognizing mild semi-melting and severe overheating conditions based on support vector machine， for moderate and severe semi-melting conditions based on rule reasoning， and for mild and moderate overheating conditions based on decision tree support vector machine， are established， respectively. The simulation results show that the proposed method can effectively recognize the abnormal condition levels of semi-melting and overheating conditions.

Key words: fused magnesium furnace, abnormal level recognition, rule-based reasoning, support vector machine(SVM), decision-tree SVM