| Molten steel level is one of the important processing parameters in strip casting,and the variation of height directly affects the quality of products and the stability of production.Therefore,accurate detection of molten steel level is of great significance to strip casting.Compared with the traditional level detection method,the new detection method based on machine vision technology has the characteristics of high detection precision,fast detection speed and small footprint,which can better meet the needs of molten steel level detection for strip casting.Based on the strip casting project of a domestic steel plant,this paper carried out related research on the development and application of molten steel level visual detection system,mainly focused on systematic analysis,research on the construction of visual detection system,image processing process,calibration and calculation method and the problems in its practical application.Relevant research results have been successfully applied on production practice and have achieved good effects.The main results are as follows:(1)Aiming at the problem of Gaussian noise and salt and pepper noise in the image under high temperature environment,combined with the mean filtering method,the conventional median filtering algorithm is improved and optimized.The algorithm mainly sets a reasonable signal trimming amount R according to the actual detection condition,centered on the middle of the field,and do the mean processing for all the pixels within R.As a result,the imagine boundary can reserve more accurately by using new algorithm,which ensured the efficient running of subsequent image processing program.(2)Aiming at the "double boundary" which is frequent appeared in the process of molten pool level detection,a new image segmentation algorithm based on the traditional OTSU threshold segmentation algorithm was proposed.According to the optimal threshold variation law,a piecewise function based on OTSU threshold has been designed.The feature region of images from different level has been exacted more accurately by using new image segmentation algorithm.The detection range is expanded to four times as the single-use OTSU detection range,which meets the actual detection requirements.(3)According to the molten steel level detection scheme,an imaging projection model based on traditional camera calibration method was established to determine the corresponding relationship between two-dimensional object points and three-dimensional object points.By using different width calibration plates to simulate molten steel levels at different heights and simplifying the camera calibration model to reduce system calibration parameters,to achieve the purpose of improving the system detection speed and facilitating the industrial application of the system.(4)Combined with the practical application requirements of the liquid level visual inspection system,three key technical problems of boundary recognition,detection range and coordinated control were elucidated.An integrated processing scheme of "optimizing image processing-expanding the range of quantitation-PID algorithm regulation" was proposed.The molten pool level can be controlled within the design ranges,which meet the requirements of strip casting process.(5)The molten steel level visual inspection system has been put into industrial field application,which meets the actual detection requirements in terms of detection accuracy and speed.The actual measurement results show that the machine vision system used for the detection of molten steel liquid level has accurate detection results and reasonable system design. |
PDF Full Text Request