Abstract: Accurate identification and prediction for the sticking breakout is of great significance for the control of continuous casting process. Sticking breakout is the most dangerous accident in continuous casting. If the breakout cannot be predicted and disposed in time and accurately, the resulting missing and false alarms will seriously damage the caster equipment and affect the quality of the slabs, which causes huge economic losses. The hierarchical clustering and dynamic time warping (DTW) are combined to capture and identify the temperature timing characteristics during sticking breakout, and then, a new sticking breakout prediction method based on machine learning is developed. Compared with the in-service breakout prediction system, the proposed method can reduce the number of false alarms by about 60% with a 100% correct true alarm rate. The method can greatly improve the prediction accuracy and avoid the casting speed reduction or shutdown caused by false alarms, which is of positive significance for smooth stability and slab quality. The sticking breakout prediction method based on clustering shows excellent application potential and provides a new approach for abnormal monitoring in continuous casting.

Key words: timing characteristics, hierarchical clustering, dynamic time warping, breakout prediction