Molten Steel Level Identification Based on Temperature Field Distribution Sensed by a Refractory Bar

Abstract

Molten steel level is difficult to measure as a result of high-temperature medium and the covering flux. The crux of molten steel level measurement is to distinguish between the molten steel and the covering flux. The characteristic of the steelmaking process is that a strong stratification of the temperature gradients is formed between the flux and the molten steel. Thus, sequential clustering (SC) by using the temperature gradients is introduced to identify the flux–steel interface. But considering that temperature gradients are sensitive to noise in the temperature field distributions, piecewise linear regression (PLR) by using temperature field distribution is also proposed to look for the flux–steel interface. The two approaches, SC and PLR, are investigated and compared to each other mathematically without loss of generality. It is found that the two approaches cannot predict the same flux–steel interface and the differences of the prediction results decrease with the increase of the slope changes of the temperature distribution curve. Consequently, the two approaches are applied to the temperature field distributions and gradients of the refractory bar obtained from numerical analyses. The findings demonstrate that, overall, SC predicts a result with a smaller error compared with PLR. However, SC may fail with a high noise level while PLR still behaves robustly. Then, thermal images obtained from actual on-site applications are used to validate the two approaches. Finally, the two approaches are both adopted and the prediction results by them are fused for practical applications.