The invention relates to a metallurgical industry, which comprises classifying the vibrating signal into without dreg, mixed dreg and all dreg period as the training data, vectorization the signal frequency amplitude, forming model base, inputting observation data into each model by the vibrating signal in the pouring process, selecting the max. output probability working status as the identification result, with the system composed of treatment control unit module, power supply management unit module, electronic appliance control unit module, alarming unit module and front end control unit module. It can better solve the random issue of steel melt vibrating signal, with low cost, convenient seal, dismantle, transportation, maintenance, and easy mastery of the operator, with higher accuracy and stability.