A real-time prediction method of carbon content in converter steelmaking based on DDMCN flame image feature extraction

As an important operation at the end of converter steelmaking, the key to the end-point control is the accurate and real-time prediction of carbon content. And the oxidation rate of carbon content in the molten pool can be reflected in the variation of the flame texture at the furnace mouth. Therefore, the extraction of accurate characteristics of flame texture is the key to predict end-point carbon content. However, the difficulty of flame texture feature description lies in its multi-directional and multi-scale irregular characteristics. This paper proposes a derivative nonlinear mapping direction weighted multilayer complex network color texture descriptor, which conforms to the multi-scale and multi-directional characteristics of flame irregular texture. Firstly, the fire flame image under the HSI space is mapped to the phase space to enhance spatial location-related information. Then, based on the complex network, a weighting formula of the derivative relationship that reflects the continuous changes between the vertices of different scales is given. And the multi-scale irregular direction weighted color texture complex network of the furnace mouth flame image is constructed by combining the direction information. Finally, the direction weighting degree feature of the vertex is calculated to quantify the connection mode of the complex network topology, and the color texture feature of the flame is constructed. And the end carbon content is predicted by the KNN regression model. The results show that the algorithm meets the real-time requirements of the actual converter steelmaking process.