Research On Oxygen Content Prediction Of Molten Steel In LF

The oxygen in molten steel mainly comes from the oxygen blowing process, and it mainly exists in oxide.The deoxidation in refining process aims at two sides. One is to avoid the difficulty in tapping and pouring of molten steel, which is caused by the fierce reaction of oxygen and carbon, and the other is to avoid the harm done to the steel product caused by the oxide in molten steel, which separate out from the molten steel as the temperature falls. The oxygen content is a reflection of the clearness of steel product, and high quality steel making always require low oxygen content. It is an important premise to predict accurately the oxygen content of molten steel in LF for organizing production, improving steel quality and controlling the oxygen content.The deoxidation in LF is a complex process, and we often use w[T.O] as the standard of oxygen. In this thesis, the theory of deoxidation is explained from metallurgy perspective, and the modeling methods are analyzed in three part, the static equation, dynamic equation and AI modeling. The static equation described the relationship between w[Oe] and [Als] in balanced state. The dynamic equation use the speed of nonrestrictive to take place of the speed of whole process, which contains the increasing part and decreasing part. The mechanism modeling belongs to empirical formula, but the factors affected oxygen content can be inferred from mechanism analyses, which have a nonlinear relationship with the oxygen content. So the neural network is chosen as AI modeling methods, and RBF is most suitable because of its good performance, which can be decried as best fitting and fast training. Considering the methodology of 300t LF in bao-steel, the input of the network are defined as follows:initial oxygen content, aluminum p.t.w, CaO p.t.w, synthetic slag p.t.w, slag's thickness, argon blow p.m, alloy p.t.w, and the out put is the oxygen content.The network was trained by 300 sampled data, which came from 400 sampled data collating from actual sampled data. The model was examined by the other 100 data.The result testified that the model is feasible, and it can be used as a instruction of producing.