특성 군집화를 활용한 전로 공정 종료 시점의 용강의 온도와 산소 예측 기법

Abstract

Temperature and components are important factors in the steelmaking process. The internal quality of the steel product is determined from liquid steel to solid as slab, because the physical properties of the steel product change depending on the temperature and components at the time of solidification. Basic oxygen furnace (BOF) is a process of removing impurities by blowing oxygen into hot metal and controlling temperature. In the steelmaking process, 90% of the control for temperature and components are performed by converter process (BOF). After converter process, temperature and components are finely controlled in secondary refining processes such as Bubbling Station, Ladle Furnace. Generally, temperature and oxygen are measured at the end point of the converter process for process analysis. Measurement is used by a one-time probe, and cost occur. Unfortunately, even if we create a model that predicts the data at the end point, measurements are required for model verification and re-training. This study proposes a method that does not require measuring the temperature and oxygen at the end point. The prediction target was set as the measurement data of the next process after converter process, and the data were used from the converter process to the measurement of the next process. In addition, a model that can be understood as domain knowledge was implemented to be used in actual industrial sites. Also, group information is generated using unsupervised learning methods to improve prediction accuracy.