Computer vision detection of mold breakout in slab continuous casting using an optimized neural network

Abstract

A visual breakout prediction method for mold monitoring is proposed based on temperatures measured by thermocouples in the mold combined with the use of computer vision technology. This mold temperature rate thermography allows the characteristics of abnormal temperature regions to be captured and extracted, including the rate of temperature change with time, the geometry, and propagation velocity. On the basis of these characteristics, a back-propagation (BP) neural network model is constructed to detect mold breakout. The weight and threshold values of the model are optimized using the Levenberg–Marquardt (LM) algorithm and a genetic algorithm (GA) through an iterative process of sample training and testing. The results show that the GA–LM–BP neural network model is better than both the traditional BP and the LM–BP models. This breakout prediction model has a higher accuracy rate (83.3 %) and a lower false-alarm rate (0.05 %). The GA–LM–BP model has also been compared with an actual BOPS used in continuous casting production. Meanwhile, it provides a way of detecting abnormalities visually for continuous casting process. The results of this work also provide a positive example of the application of intelligent monitoring and visualization methods to continuous casting.