End-point Composition Prediction Of Converter Steelmaking Based On Image And Spectrum Fusion

As the country's economy turns from rapid to stable,the demand for steel has also shifted from high quantity to high quality.Currently,BOF steelmaking is the main steelmaking method in our country.The temperature and impurity content of molten steel during tapping are the most important indicators to measure the quality of steel.Therefore,it is of great significance to study the real-time control of the steelmaking endpoint.This paper is based on the application background of BOF steelmaking.By analyzing the flame image and spectrum of the furnace mouth,a BOF steelmaking endpoint composition prediction system is designed and established.This system also has the functions of smelting process judgment,furnace mouth obstruction alarm,and carbon content prediction.The effective flame area in the image is extracted by multi-color space fusion segmentation with Otsu threshold segmentation method.The color texture feature of the flame image is extracted by Multi-directional dual-channel color co-occurrence matrix.The pixel value distribution of the flame image is described by Local binary patterns conversion.The norm property of the flame image matrix is calculated with the help of mathematical analysis.A spectral characteristic wavelength selection method based on window competitive adaptive reweighted sampling combined with iterative successive projection algorithm was proposed.Combining window rough selection with iterative selection,the characteristic wavelength of the flame spectrum is extracted.The flame image features and spectral features are fused.The characteristic data is calculated by principal component analysis,and the original features are replaced by principal components.The degree of correlation between each feature and the end point carbon content is measured by calculating the maximum information coefficient.According to the results,invalid features are screened out.Finally,the BOF steelmaking endpoint composition prediction system is established using support vector machines.The system can realize the monitoring and judgment of the smelting process,the timely alarm of the shielding on the furnace mouth,and the real-time prediction of the carbon content of the molten steel.The final system structure is determined by comparing the effects of different feature inputs on the carbon content prediction results.System parameters are optimized through particle swarm optimization and chicken swarm optimization.System operation interface is designed,and the smooth operation of the system is realized.