Real-Time Prediction of Molten Pool Velocity Distribution in Converter Steelmaking Based on Deep Learning

As the main equipment of converter steel production, the oxygen lance plays an important role in stirring, slagging, decarburizing, and heating the molten pool. The velocity distribution of the melt pool is based primarily on a computational model of fluid mechanics. This model has a long calculation time and requires more computing resources. Given the complex working conditions on site, it is not possible to assess the timing. The deep learning algorithm is selected to replace the numerical calculation model, which has acceptable accuracy and shorter calculation time. The oxygen lance process parameters are mapped to the hidden code of the convolution autoencoder via a multilayer perceptron, and the input oxygen lance parameters are used to predict the velocity distribution of the molten pool on the millisecond time scale. The frequency distribution of the velocity error value is counted, the prediction accuracy is 93.12, and the calculation time is only 0.12 seconds. The real-time calculation of the dead zone area and impact depth of the melt pool is helpful for the field personnel to meaningfully adjust the parameters of the oxygen lance.