Study On Breakout Prediction System Based On Artificial Neural Networks In Thin Slab Continuous Casting Process

With the development of the technology for continuous casting, high efficientcontinuous casting technology has become the main direction of the research incontinuous casting field. It is highly valued by big steel enterprises, engineeringcompanies and equipment manufacturers. High efficient continuous casting technology isa high utilization casting system technology with high rate aiming at high quality and nodefects of casting billet production. High speed of casting is the core part of the efficiencycasting technology. However, with the increasing of the casting speed, the risk of breakoutis getting higher. Then, the breakout becomes the key obstacle for the drawing speedincrease in the continuous casting process. Employing the real-time and effective breakoutprediction system for the recognition of the breakout characteristics is the main means toprevent the breakout accident during casting.At present, the research on the breakout prediction system of our country is still inearly stages. System research on the formation mechanism and the process of breakout islack. Occasionally, certain false alarms and missed alarms issued by the existing breakoutprediction systems used in the process of continuous casting lead to the stagnation ofproduction and equipment damage. Thus, the breakout prediction system in the process ofsteel continuous casting was discussed on the subject in this paper, with the CSP slabcasting machine in Handan Steel plant as the research object. Series of research works onthis subject were carried out in this paper as follows.The mechanism and the main forms of breakout were analyzed in the paper; themechanism of the sticking type breakout has been analyzed and discussed emphaticallycombining the behaviors of the liquid steel meniscus. Further, related preventing stepswere provided according to the inductive factors.The wavelet-based de-noising method was introduced into the breakout predictionsystem in the continuous casting process to conduct the noise reduction. Temperature datacollected by thermocouples was multi-scale decomposed. The Birge-Massart thresholdwas applied to the wavelet coefficients, the temperature signal was reconstructed and the temperature date was de-noised. The trend of the temperature changing can be well shownby the de-noised data.According to the defects of local optimal solution and slow convergence rate in thetraining process of the BP neural network, the BP neural network was optimized with theparticle swarm optimization algorithm, genetic algorithm and LM algorithm. Then, theGA-LM-BP neural network breakout prediction model and the PSO-LM-BP breakoutprediction model had been built. And they were trained and tested using the on-sitemeasured date. The test result indicates that the identification accuracy of the GA-LM-BPneural network breakout prediction model for breakout prediction was the highest.Finally, the thin slab continuous casting visual breakout prediction system wasaccomplished with the help of programming software Microsoft Visual C++6.0andANSYS secondary development function. At the same time, the system passed the testsmoothly. The results show that the system can make accurate judgment timely when thebreakout occurs. The dynamic and real-time display of the temperature curve of data andthe mold copper plate hot-side contour in the interface of the system is helpful foroperators to make auxiliary judgments.