Molten Steel Level Detection by Temperature Gradients With a Neural Network

Molten steel level is difficult to measure as a result of the high-temperature melt and the covering flux. For the measurement, in our previous work, a novel principle by using the temperature gradient was proposed, and a refractory sensor was inserted into the metallurgical container to sense the temperature gradients of the flux and the molten steel. However, variations of temperature gradient distributions are large when the fluctuation speed of the molten steel level is fast, causing difficulties in the detection of the flux-steel interface. For this issue, a neural network is introduced to learn the features of the temperature gradients around the flux-steel interface, and the convolution of the neural network is developed to detect the flux-steel interface from the temperature gradient distribution. The numerical data obtained from the heat transfer model is used to train and test the detection method. The detection method gives a good performance with the numerical data. But for actual on-site applications, noise in the temperature gradient distributions affects the reliability and accuracy of the detection results. To improve the reliability of the detection method in practice, lifting of the sensor at the wave crest of the flux level is adopted to ensure the large temperature gradients around the flux-steel interface. The statistics show that the detection errors of the flux-steel interface are within ±5 mm with a confidence of 98.3%.