The Research And Application Of Pulverized C Oal Injection Detetction System In Blast Furnace Tuyere Based On Video

The blast furnace is the main place for iron making and modern steel production.As demand of steel continues to grow,the original way of relying solely on coke as a fuel is no longer applicable.In the 1980s,pulverized coal injection(PCI)technology was widely used in blast furnace production.The technology can reduce the smelting coke ratio for smelting,which achieves high production and low consumption.Due to the continuous increase in the amount of pulverized coal injection in blast furnaces,strict requirements have been placed on the stability and safety of coal injection in blast furnaces.The accident can be avoided in time through the monitoring of the coal injection status in the blast furnace tuyere,which is of great significance for the stable operation and precise control of the blast furnace.In recent years,the blast furnace tuyere camera has been applied in steel mills,the raceway images are disturbed by severe noise during transmission and processing.Moreover,the visual impression of images varies between different raceways.The above characteristics not only bring difficulties to the detection of pulverized coal injection,but also requires the detection method should be robust and convenient to fine-tune for different raceway images.To address the problems mentioned above,this thesis proposes a monitoring method in pulverized coal injection based on image processing and computer vision technology,and then builds an intelligent monitoring system.The method can detect the condition and uniformity of the PCI in real time.An image preprocessing method combining image filtering and grayscale transformation can remove noise and improve the quality of the image.Adaptive threshold segmentation combining the segmentation method based on convolutional neural network are used to extract coal group region.Moreover,three coal group feature extraction methods are proposed to characterize the coal injection information by analyzing the characteristics of coal group region,which can effectively monitor the condition and uniformity of PCI.Firstly,the characteristics information of the raceway image collected in blast furnace is analyzed.An adaptive preprocessing algorithm combining image denoising and gray-scale transformation is proposed.For the problem of severe image noise,the noise type of raceway image is judged by analyzing the spectral information and histogram information.An improved image denoising algorithm combining frequency domain filtering and spatial domain filtering is investigated.The quality of the raceway image is seriously degraded and exists gray level imbalance,because the tuyere camera has been in dusty and high temperature environment.Image gray-scale transformation algorithm combined with improved adaptive grayscale stretching and power transformation is applied to improve image quality.Experiments on the raceway image collected by steel mills show that the proposed preprocessing algorithm not only can effectively preserve the edge to remove noise,but also improve the contrast of each region in the image,which lays a good foundation for the subsequent processing.Secondly,this thesis proposes an intelligent pulverized coal extraction method.The binarization algorithm combining the global threshold and the local threshold is used to accurately obtain the lance and pulverized coal connection region.Considering that the amount of pulverized coal injection is large,the pulverized coal and the coal gun are"adhesive" in the binarization process.One-dimensional Hough ellipse is proposed to determine the location of the tuyere region,which can be used to separate the pulverized coal from and furnace wall.There is a fact that the lance region is close to the pulverized coal region,which causes the traditional binarization algorithm cannot separate the regions.In addition,the lance needs to be dismantled and repaired during the blast furnace blow down,and the secondary installation may make the position of the lance change,which menas the single fixed background template method is not applicable.In this thesis,an improved full convolutional neural network(FCN)is proposed for lance region segmentation.The pulverized coal region can be obtained by filtering out the lance detected by FCN in the binarized image.Experiments on a large number of raceway images show that the algorithm can accurately segment the pulverized coal.Finally,the area method,weighted area method and ellipse fitting method are proposed to calculate the pulverized coal region characteristic information,whcih can reflect the coal injection information.The tkinter in the Python GUI is used to build a blast furnace tuyere state intelligent monitoring system,and the tuyere videos collected from industrial site to verify the algorithm.The experiments show that the proposed method can effectively monitor the working condition of PCI.In the existing tuyere raceway video dataset,the accuracy rate is over 95%.At the same time,it can detect the uniformity of PCI in the blast furnace,and the error is only 5%,which means the algorithm has practical application value.