The accurate prediction of mold sticking breakout is an important prerequisite to ensure the stable and smooth production of the continuous casting process. When sticking breakout occurs, the sticking region expands vertically along the casting direction and horizontally along the strand width direction, forming a V-shaped area on the strand surface. This paper uses computer vision technology to visualize the temperature of mold copper plates, extract the geometric and movement characteristics of the sticking region from time and space perspectives, and construct feature vectors to characterize the V-shaped sticking breakout region. We train and test the auxiliary classifier WGAN-GP (ACWGAN-GP) model on true and false sticking feature vector samples, developing a breakout prediction method based on computer vision and a generative adversarial network. The test results show that the model can distinguish between true sticking breakout and false sticking breakout in terms of mold copper plate temperature, providing a new approach for monitoring abnormalities in the continuous casting process.

对粘模爆头的准确预测是保证连铸工艺稳定、顺利生产的重要前提。当发生脱粘时，粘着区域沿浇注方向垂直扩展，沿铸坯宽度方向水平扩展，在铸坯表面形成V形区域。本文利用计算机视觉技术对模具铜板的温度进行可视化，从时间和空间的角度提取粘着区域的几何和运动特征，并构建特征向量来表征V形粘着突破区域。我们在真假粘贴特征向量样本上训练和测试辅助分类器 WGAN-GP（ACWGAN-GP）模型，开发基于计算机视觉和生成对抗网络的突破预测方法。试验结果表明，该模型可以根据结晶器铜板温度区分真粘漏和假粘漏，为连铸过程中的异常监测提供了新的途径。