Bitte wählen Sie ihr Lieferland und ihre Kundengruppe
For the first time, a real time data-driven model on the basis of modern high-efficiency algorithms is used to predict the conditions at the BOF endpoint. The accuracy in predicting the temperature and other target variables is not only equal but even better than using metallurgical models. The prediction model operates in a stable and reliable manner, enhances the understanding of the process by exposing process relations between input and target variables, and thus supports the metallurgical model. However, the application of modern data mining and machine learning to modern metallurgical processes, such as BOF is at the beginning. There are still many possibilities to tune the system, in the area of data selection, data construction, conditioning of data and strategies in the use of algorithms. Steelwork processes are characterized by the fact that target variables usually depends on each other. It is for example difficult to meet a low phosphorous content in the melt in combination with high tapping temperatures. For these cases it is aimed to increase the simultaneous hitting rate of several targets. In the next development stage, a multi-objective control tool will be developed that corrects the running BOF process as early as possible, so that several target values are optimized. These optimization calculations have to be done in real-time to gain the earliest time for corrective measures. On the basis of the online prediction modeling, we will develop this new technology. Beyond there is the intention to develop and implement a full automated model generation. The quality of the input data will be monitored continuously and will be included into the selection of the input variables. Steelwork processes are not fixed for all times, but changed with the modification of the equipment, with the wear of components, drifts of measurement gadgets and other things. Additionally steelwork environment is rough and it can't be guaranteed, that the complete measurement infrastructure works reliable at all times. Models have to deal with the circumstances of changing equipment and data supply. In these cases new adapted models are necessary, to have a real, optimal and multi-purpose applicable automation. Because of the demand of versatile models, the model creation has to be autonomous.