Model For Breakout Diagnosis And Prediction In Continuous Casting Mold With High Casting Speed

Without question, the breakout caused mainly by the sticking of solidifying shell is the most catastrophic accident associated with continuous casting process. With adverse impact on performance, product quality and equipment life, it badly affects the efficiency of continuous casting process. At present, by use of data measured and collected with thermocouple technique, most of breakout prediction systems apply models based on logical decision and neural network to predict the breakout. With higher accuracy and performance, neural network-based prediction models become leading direction in the research recently. On basis of the technology innovative project "R&D of Mold Technology Integration for Continuous Casting Mold with High Casting Speed" in a domestic steel work, breakout and shell sticking prediction especially the critical prediction model was investigated in this paper. The followings are the main research work carried out and the correspondent results obtained:(1) During analyzing the overall flow of breakout prediction in continuous casting process, a effective pre-processing method was proposed. This method could completely distinguish the steady temperature pattern from the pattern of breakout, and fully reflect the characteristic of temperature-wave pattern.(2) Closeness-based improved FCM and Conditional FCM algorithms were proposed after introducing a self-defined closeness which belongs to fuzzy pattern recognition into the corresponding fuzzy clustering algorithms. These advanced algorithms greatly optimized fuzzy clustering effect and improved accuracy of the forecasting system.(3) By introducing the closeness-based improved FCM and Conditional FCM algorithms into two steps of network training to initially calculate and ulteriorly optimize values of clustering centers respectively, the model based on advanced FRBF neural network was proposed in this paper for breakout prediction. The improved model optimized learning algorithm of FRBF network. It also showed its advantage in both capability of patern recognition and predicting performance.(4) The simulation testing results showed that the occurrence of false alarms focused on the process of cast starting and increasing of casting speed after tundish quick-change. It is possible that the temperature behavior in mould of these process was similar with process of sticking breakout.(5) With history data acquired in a domestic steel work, the FRBF and corresponding advanced neural network were applied to the breakout prediction in continuous casting process. The simulation results showed that the prediction rates of the advanced model for two typical temperature patterns of sticking breakout were 94.9% and 98.3% respectively, and both of the quote rates were 100%, that indicates the model is more effective in predicting possible leakages of liquid steel.