The Research Of Sinter FeO Real-time Content Detection System Based On Image Processing And Fuzzy C-means Clustering

With the fast development of the economy, there is no doubt that the need for steel grows, which adds to the awkward situation of the shortage of iron ore in our country. The natural, high-quality and high-grade iron ore is a non-renewable resource and the amount of it will become less and less year by year after being exploited. The iron ore with low quality cannot be put into the stoves immediately to produce iron but it should be converted into agglomerate first of all. Meanwhile, with the development of the stoves of large-scale and the technique of smelt, higher requirement to the refined material are put forward. Nowadays, most iron ore used in stoves to produce iron are agglomerate. Therefore, the production of agglomerate becomes more and more important.There are many ways today to get the chemical components, including the judgment of firers by eyes, chemical examination and so on. The firers control the process of production by observing the images at the end of the machine directly and analyzing all the parameters of agglomerate. Although the result of chemical examination is accurate, the whole process from the sintering of mixed materials to the cooling cost2to3hours. Artificial interpretation not only tires the firers easily but also is not so intelligent. Hysteretic quality makes the chemical examination out of control of us. All of them have apparent defects.In recent years, people have done a large scale of researches to the forecast of the agglomerate’s function and the optimizing control of the process. Besides, it appears an agglomerate forecast model in the market through traditional modeling, the defects of which are that the parameters are too many and the time costs too much. What’s more, the accuracy cannot meet the requirement of the process of commercial run and it cannot realize the timely and the on-line deduction need. Therefore, it’s necessary to exploit a system to test the amount of FeO in agglomerate on line and automatically. This article tells about how to combine fuzzy C-means clustering and BP to achieve the forecast of the amount of FeO according to the characters when FeO reacts. To begin with, cluster the cross-sectional images based on a certain feature by FCM. Next, divide the result of the clustering into three classes according to the effective characters, input them into BP to get trained and then gain the neural network. Then, use BP after training to forecast the non-classified agglomerate. By this way we can adjust the technological parameters during the process in time and realize the optimizing control of the sintering process. Finally, analyze and discuss the model training and the result. The experiment shows that model-forecasting the amount of FeO and the result tested from the laboratory can match perfectly.