Mathematical And Physical Modeling Of LF Refining Process

LF refining,as one of the standardized processes in secondary refining,is an important equipment for precise control of steel composition and temperature.Based on the need to optimize the internal flow characteristics of the ladle when a new ladle furnace is built,a 1:3 water model was constructed,and the effects of 13 types of purging plugs arrangements and 56 bottom blowing conditions on the refining effect were tested.Based on the water model experimental results,several groups of well-performing bottom-blowing conditions were screened out,and based on experimental data analysis and fitting,the empirical formulas for calculating the homogeneous mixing time and slag eye area for the equal flow bottom-blowing mode of double purging plugs were proposed.In order to further compare several groups of similarly behaving conditions in the water model,a numerical model of the water model scale was built,and after the calculations reached steady state,User Defined Function was used to classify and count the grid velocities in the water phase region,thus quantifying the percentage of dead zones.Combining the results of the water model and the numerical model,the following conclusions are proposed: if both single and double purging plugs underblowing are considered,the double purging plugs 2/3R,180°,3.435 NL/min underblowing condition should be selected for industrial applications,and if only double purging plugs underblowing is considered,the double permeable brick 1/3R,120°,3.435 NL/min underblowing condition should be selected for industrial applications.After the design requirements were realized,the refractory wear principle at the slag line of ladel furnace was analyzed,and based on the wear principle,an prediction model of corrosion loss was built using BP neural network and convolutional neural network,respectively.The performance of the prediction models using different network types and parameters in the test set was measured using the mean squared deviation,and the top-performing prediction models were selected.The final prediction model used performed well on the test set,with 95% prediction error less than 10 mm.