| Continuous casting is the key point of modern iron and steel production. Along with the rapid development of high continuous casting speed technology in recent years, the heat load of molds is increasing more obviously due to the heavy passing steel. Many surface defects of slabs and breakout accidents still often occur, which put forward higher requirement to monitoring technologies in production process. Therefore, developing fresh online process detection methods and technical equipment, which complies with the visual and intelligent trend of modern industry, are very helpful to improve the monitoring and control of mold process further. It would be of benefit to development of high efficient continuous casting technology.This work is based on the computer vision and artificial intelligence methods, and focuses on development of new breakout prediction method and the visualization system for mold process. The main contents are as follows:Firstly, the influence of casting parameters on sticker breakouts, formation and propagation of sticker are investigated based on the data of 4190 casting heats and 64 sticker breakouts, which are from an arc continuous caster of wide and heavy plate in a domestic steel plant. The main effective parameters on sticker breakout are analyzed in the view of slab dimension, casting speed, mold level fluctuation and heat flux, et al. Meanwhile, temperature patterns of sticker breakout are studied according to the thermocouple measured temperature, especially concentrating on the propagation features such as sticker moving velocity and angle. These results provide a reference for the later research of visual prediction method of breakout.Secondly, basing on computer vision technology, the researches of mold process visualization method and thermal imaging technology are conducted. Abnormal temperature regions of breakout and longitudinal crack are labelled and extracted, using some image processing algorithms:frame difference, threshold segmentation,8-connected region labelling, boundary tracing, et al. The common features of mold sticker are collected in terms of temperature velocity, geometry, position and movement of abnormal temperature regions, and the comparison with the false stickers are conducted. The results show that these features above occurred simultaneously with breakouts formation can be the main criteria to distinguish the true and false sticker breakouts.Basing on above researches, visualization and intelligent prediction methods of sticker breakout are studied further more. Breakout prediction model based on BP neural network is built up, which takes temperature velocity, geometry and moving features of sticker hot areas as the input parameters. Then, BP neural network model is trained and improved by LM (Levenberg-Marquardt) algorithm and GA(Genetic Algorithm). The traced and collected 64 sticker breakouts and 200 false sticker breakouts are used to test and analyze network model of BP, LM-BP and GA-LM-BP respectively. And the model accuracy and efficiency are investigated and compared comprehensively.Finally, according to above developed methods, such as process visualization, computer vision and artificial intelligent algorithm, mold process visualization and breakout prediction expert system is designed and developed based on detected signals, including mold temperature, hydraulic oscillation, caster equipment, casting parameters in continuous casting field. This system integrates multi-functions such as copper plates temperature detection, mold friction detection, oscillation state detection, process visualization and breakout prediction. The system has been run parallelly with a commercial system together at same time to monitor the mold process for an arc caster of slabs in a domestic steel plant for almost one year. The results show that our system does not have the missed alarm and the false alarms are reduced greatly compared with the commercial system. Therefore, the accuracy of breakout prediction is improved remarkably. The system can ensure the production smoothly and has a good application prospect in slab continuous casting. |
PDF Full Text Request