Carbon (C) and phosphorus (P) prediction in steel shop (BOF) with artificial neural networks

In today's technology, models are commonly developed and applied to predict and control the end point of any processes, due to obtaining sustainable product quality. The power of model and it's usage will be more precise and accurate in case when the system is explained in detail and defined completely. Nevertheless, during the steelmaking process there are several plants that modelling becomes challenging. Blast Furnace (BF) and Converter process are the most difficult processes that can be modelled due to liquid, metal and gas reactions and a large number of input variables that can influence reaching the end point. In this study, Feed Forward Back Propagation Multi-Layer Neural Network in Matlab with training function Trainscg is proposed for prediction of the phosphorus and carbon at blowing end (TSO phase) and inblow (TSC phase ,%80 of blowing time) in BOF Converter. In order to ensuring and validating the randomness of input data, a single BOF plants data is collected. The data set is filtered with a strict limitation method according to the plant specific metallurgical interactions, bottom stirring effect and 10 Fold Cross Validation used for clustering in order to have homogenous learning process. The simulated results hit rate %92 within the error range ±0,01 for end-point carbon and %89.4 within the error range ±0,025 for end-point phosphorus are observed. For inblow the simulated results hit rate %83 within the error range ±0,02 for phosphorus and %93 within the error range ±0,15 are observed. The results showed that the output could be used in software to calculate P and C during the end of blowing and inblow without interrupting the blowing process like TSO or TSC measurement.