Deep Vision-Based Intelligent Measuring Method for Two Liquid Levels of Aluminum Electrolysis

The two liquid levels of aluminum electrolysis refer to aluminum liquid level and electrolyte level, which are important control parameters for the aluminum electrolytic production. The traditional measurement of the two liquid levels is time-consuming, labor-intensive and dangerous. In this paper, we propose an intelligent measuring method applied to inspection robots, which overcomes the drawbacks of the traditional measurement. This method uses an RGB-D camera to capture images of the brazing rod to be measured. Then, the coordinates of the full-length endpoints of the two liquid levels are located by the threshold segmentation algorithm based on HSV color space. Due to the difficulty in locating the boundary between the two liquid levels through image processing algorithms, this paper applies deep learning to industrial production. We train a semantic segmentation model to segment the characteristic electrolyte region. Then the edge of the segmentation region is used to locate the coordinates of the boundary, which improves the robustness. Finally, we project the obtained coordinates into three-dimensional space and calculate the Euclidean distance. The tests show that the error between the algorithm-calculated values and the truth values is within 1cm, which meets the industrial requirements. The proposed method has the advantages of safety, stability and accuracy, which promotes the automation of the aluminum electrolysis production.