Study On The BOF-Steelmaking Endpoint Temperature Control Based On The Convert Mouth Flame Multi-spectral Analysis

Basic oxygen furnace is a pressing important way in the metallurgical industry for years.It is the most important method to make steel by Basic Oxygen Furnace(BOF)in the world.There are 86%steel production from convert steelmaking in our country.The end-point control of BOF consists of temperature control and element control.Steelmaking progress should be controlled precisely and automatically to achieve double-hit element and temperature.Thus we can improve the quality of steel,improve labor efficiency,reduce energy consumption.However,due to the instability of the additional raw materials and the complex chemical reactions,the endpoint of the carbon content and steel temperature is still a pressing problem.Aiming at this problem,based on the spectral characteristics of the flame radiation information,the steel temperature non-contact measurement is studied.Analyzing 345nm～1045nm spectral data from the furnace mouth,furnace flame atomic emission spectra overlap in a continuous or "black body" radiation,which are clear in the visible radiation.In this dissertation,duochrome method,three color method(greybody model,Hottel-Broughton model)and multi-wavelength analysis method are introduced to measure the temperature of BOF flame.The effective emissivity of the gas-particle stream is retrieved by the Radiative Transfer Equation(RTE).The BOF flame project temperature is measured by CCD camera,and then is encoded by pseudo color.The BOF flame image are preprocessed,including gray enhanceme-nt processing,denoise processing,OTSU threshold segmentation processing et al.The flame RGB color images are transformed to other color space,gray mean value of the flame image ROI is calculated;Gray level co-occurrence matrix and Gray difference matrix are counted,which are used to calculate Eneragy,Entropy,Moment of inertia,Relativity,Deficit moment and average value et al.Intensity of characteristic wavelengths、intensity which can reflect the whole wavelengths、smelting time、flame image gray value、texture parameters are used to design Sigmoid-Wavelet neural networks to predict the steel temperature.Sigmoid-Wavelet neural networks had been used to predict the end-point steel temperature successfully.Each neuron in the hidden layer of a feed-forward network is a combination of the sigmoid activation function and morlet wavelet activation function.The output of the hidden neuron is the product of the output from these two activation functions.Five types of recurrent networks are introduced.All of these network models to predict end-point steel temperature are implemented successfully.More than 1000 sets of data presented here in the paper are collected during the actual converter blowing process.The method and technology in this dissertation have certain reference value to achieve the double-hit temperature and carbon content control,experimental results indicate that the constructed endpoint prediction model can meet the demand of the steelmaking field very well.This dissertation also offered the theoretical basis for improving BOF steelmaking automatic control technology level,it has a referential role to the same domestic industry.