Study On The Method Of Determining End Point Of Converter Steelmaking Based On Feature Extraction Of Flame Image Blowing Information

Iron and steel industry plays a significant role in the national economy, Basic Oxygen Furnace (BOF) as the major method is widely used in the world. Accurate determination of the BOF endpoint has a great significance for improving steel quality, reducing smelting costs and environmental pollution, which is also a long-standing problem. In recent years, BOF endpoint determination based image processing has been rapidly developed due to its low costs and anti-interference characteristics. Chemical composition and temperature changes in the bath will be reflected in the flame color, texture and flame dynamic changes in the furnace mouth. The flame features extracted during the blowing process is the essential part of the endpoint determination based on image process. The existed algorithms are mainly extract static characteristic in the gray space, which, in general, underused the abundant color texture features and ignored the dynamic deformation characteristics of flame. Dynamic deformation features and color texture features are extracted from flame images and applied them to the endpoint determination of the converter. The research contents:(1) A method is proposed to build the flame color texture model use co-occurrence matrix in HSI color space. Entropy, Angular Second Moment and Inverse Difference Moment are extracted based the model to represent the flame color texture. In this procedure, non-uniform quantization is operated to HSI color space to ensure the calculation speed. Gaussian normalization is used to normalize the feature vector. Compared with the present gray-level co-occurrence matrix and gray differential statics, the proposed algorithm shows obvious advantages.(2) To extract the dynamic change features of the flame, a method for describing and representing dynamic deformation of flame boundary is proposed. Firstly, framework is used to locate the center of the flame which is in line with human visual perception. Secondly, a Polar is established for flame boundary. Finally, the extraction method of dynamic deformation amplitude spectrum is defined based on the Polar, and the features extracted from the spectrum are applied to identify the blowing endpoint. For the effectiveness of the description and representation, pretreatments are done to the flame boundary. Compared with the present differential chain code, circular or edge invariant line-moments, the proposed method exhibits good advantages.(3) A model fused multi-flame characteristics is established to determine the blowing endpoint. Flame color mean feature is extracted and together with color texture features, dynamic deformation features as input of generalized regression neural network. Software system is designed and callback functions are written to realize the system functions.In this paper, the flame color texture and border dynamic deformation characteristics are extracted, which is an important complementary to existing methods, and are applied to determinate the endpoint of BOF. Compared with the existing algorithms, the effectiveness of this method is demonstrated. A new attempt is made to the problem of determining BOF endpoint by flame image processing, which exhibits a high practical value.