An industrial image processing-based approach for estimation of iron ore green pellet size distribution

Abstract Green pellets are spherical objects mainly made from crushed iron and water in rotating pelletizing disk. Pellets play an important role in the process of direct reduction steel manufacturing. For high quality steel production, pellets should be of an appropriate size. Oversized pellets cannot be properly cooked in furnace and undersized pellets may pass through metal grid of the furnace, which affects the performance of the process. In most of the steel manufacturing companies, operators use sequential sampling and traditional methods to measure size of pellets, however, limited number of tested pellets, low speed of the quality control and human errors are the most common drawbacks of this method. A solution to these shortcomings is continuous monitoring and accurate measurement through a fully-automatic approach. In this paper, a method using image processing technique is proposed and verified for estimating size distribution of iron-ore green pellets. For segmentation of single and multiple pellets in images captured from live video, we have utilized morphological methods, watershed algorithm, and linear searching. Then, a Support Vector Machine (SVM) is employed for classification of segmented pellets. The proposed method has already been implemented in Mobarakeh Steel Complex, where the method was tested on about 1000 sample pellets. Results show that the accuracy of method is 95.1% in detection of single pellet elements.