Computer Vision for Sensed Images Approach in Extremely Harsh Environments: Blast Furnace Chute Wear Characterization

• Published in: IEEE sensors journal Vol. 21; no. 10; pp. 11969 - 11976
• Main Authors: Lay-Ekuakille, Aime, Ugwiri, Moise Avoci, Okitadiowo, John Djungha, Chiffi, Cosimo, Pietrosanto, Antonio
• Format: Journal Article
• Language: English
• Published: New York IEEE 15.05.2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; backprojection; Blast furnace chemistry; Blast furnace practice; Blast furnaces; Computer vision; extremely harsh environments; High temperature effects; image processing; Machine learning; machine vision; monadic processing; Neural networks; nondestructive tests; Radon; Sensor phenomena and characterization; Sensors; sinograms; Steel; Temperature measurement; Two dimensional displays; Virtual sensors; Wear; Wear characterization and measurement
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2021.3063264

Measurements and characterization for extremely harsh environments require accurate approach especially by means of image-based computer vision. Because of harsh conditions, such as high temperature, pollution, turbulences, radioactive exposure, high energy, direct measurements through conventional sensors are not easy even with recent sensing technologies. Live and/or shortest time-delayed sensing, by means of imaging, can come to help to overcome the aforementioned constraints. The paper outilines the use of sensed images for characterizing the effects of high temperatures, at the inlet of a blast furnace, during the discharge of materials using a chute. This latter is subject to wear due to chemico-physical reactions at around 350-450 °C. Given the specific application related to the harsh environment, two algorithms are comparatively proposed and updated for the purposes of the paper; they are based both on computer vision, namely monadic technique and conventional neural network. For the first technique, virtual sensors have been introduced within the image thanks to sinogram and backprojection subtechniques. The results highlight the effects of the environment on the layers of anti-wear compounds applied on the chute, then they permit to understand the chute life-cycle. Quantitative percentage of material detection has been included as well as specific metrics for machine learning expression.