Research On Thin Slab Continuous Casting Visual Breakout Prediction System

Breakout is one of the most catastrophic accidents in continuous casting production, which not only endangers the personal safety and causes device damages, but also seriously affects the quality of casting slab that may further cause more substantial economic losses. Therefore, it is of great theoretical and economic significance to study the formation mechanism of the breakout and further develop an effective breakout prediction system to reduce or avoid the breakout.This paper explores the formation mechanism of sticker breakout first, and then discusses the influencing factors and further proposes the preventive measures.An improved BP neural network model is presented by modifying the learning algorithm of the traditional BP neural network based on the Levenberg-Marquardt algorithm, and is further applied to the breakout prediction in the thin slab continuous casting. Time series network is established to check whether the temperature changes of single thermocouple on time sequence conform to the temperature characteristics when sticker happens, in the meanwhile, in terms of the breakout propagation mechanism, the space network is established to further tell whether the transfer of the sticker temperature mode in single thermocouple has happened.Based on WINDOWS platform, the thin slab continuous casting visual breakout prediction system is accomplished with the help of programming software Microsoft Visual C + + 6.0 and ANSYS secondary development function. System function includes: data preprocessing, breakout prediction neural network module, the temperature curve of real-time data and dynamic display and the mold copper plate hotside contour real-time display, alarm information processing and so on.The breakout samples and normal samples are collected in steel plant for test and training of the neural network prediction model. Then the thin slab continuous casting visual breakout prediction system offline simulation is realized in reference to the historical temperature data of a steel plant and relatively good results are achieved.