Localization of electrorefining and electrowinning cells in an infrared image for short circuits monitoring
Pandey, Sijan (2022)
Pandey, Sijan
2022
Bachelor's Programme in Science and Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. Only for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2022-05-06
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202205104622
https://urn.fi/URN:NBN:fi:tuni-202205104622
Tiivistelmä
Electrorefining and electrowinning processes, which are based on electrolysis, are carried out to produce pure metal on an industrial scale. In these processes, short circuit occurs when a cathode touches an anode. Shortcircuits are detrimental to production efficiency. Hence, quick detection and removal of short circuit is better for production. Short circuit in an electrode is characterized by an increase in temperature of the electrode. One of the ways to detect the short circuits is to manually scan around the production tankhouse for hot electrodes with an infrared scanner. However, since the production tankhouse consists of hundreds of electrodes and has a toxic environment, it is a challenging task to accomplish manually. Metso Outotec has a short circuit monitoring system already in use that automates the monitoring of short circuits. It uses infrared images of production cells taken by an infrared camera mounted on a crane that traverses above the production area in a tankhouse. Temperature of the electrodes is analyzed using the captured infrared image to monitor short circuits. For the temperature analysis, it is essential to locate the regions in the images where the temperature analysis needs to be conducted. The current system has an algorithm to locate the region of interest. Its performance has been unreliable and hence needs replacement. This thesis proposes an algorithm to replace the existing algorithm.
The proposed algorithm locates the region of interest in the image and prepares a bounding box around it. In addition, it visualizes the electrodes, where temperature analysis is done, within the proposed bounding box. The algorithm seeks locations of dark horizontal bars visible within a production cell in the infrared image. It is accomplished by subjecting the infrared image to a series of image processing methods that highlight these bars and calculating the coordinates of the bars from the output. This information is combined with a user-provided configuration to prepare a bounding box that highlights the region of interest. Then, image strips corresponding to each electrode within the bounding box are extracted. For each strip, a bounding box is created. Temperature data embedded in the pixel intensities of these strips are analyzed to monitor short circuits.
The proposed algorithm was evaluated on 298 randomly chosen test images. A test was setup to examine how close the algorithm’s outputs were to the ground truth. Those images where the outputs were sufficiently close to the true values were considered passed test cases. The algorithm passed the test on 96.6% of the total test images. For performance comparison, the average output of the proposed algorithm was considered a baseline. The baseline was also tested on 298 randomly chosen test images. It passed the test on 38.2% of the total test images.
The proposed algorithm locates the region of interest in the image and prepares a bounding box around it. In addition, it visualizes the electrodes, where temperature analysis is done, within the proposed bounding box. The algorithm seeks locations of dark horizontal bars visible within a production cell in the infrared image. It is accomplished by subjecting the infrared image to a series of image processing methods that highlight these bars and calculating the coordinates of the bars from the output. This information is combined with a user-provided configuration to prepare a bounding box that highlights the region of interest. Then, image strips corresponding to each electrode within the bounding box are extracted. For each strip, a bounding box is created. Temperature data embedded in the pixel intensities of these strips are analyzed to monitor short circuits.
The proposed algorithm was evaluated on 298 randomly chosen test images. A test was setup to examine how close the algorithm’s outputs were to the ground truth. Those images where the outputs were sufficiently close to the true values were considered passed test cases. The algorithm passed the test on 96.6% of the total test images. For performance comparison, the average output of the proposed algorithm was considered a baseline. The baseline was also tested on 298 randomly chosen test images. It passed the test on 38.2% of the total test images.