An effective strategy for pellet size monitoring via fiducial marker-assisted image reconstruction: Enabling high-precision granulation process control

Abstract

Abstract

An online automatic detection model for determining the particle size of green pellets incorporates multiple image processing technologies, realising the online automatic detection of the particle size of green pellets. This approach solves the problem that the current detection of the particle size of green pellets is primarily dependent on manual screening and sampling, which is low-efficiency and has significant lags in detection. During the production process, the green pellets on the pellet discharge conveyor belt are mostly in a stacked state. To address the issue that green pellets are clumped and challenging to segment in images, this method combines bilateral filtering, histogram equalisation and greyscale morphological operations to reduce image noise. The green pellet images are marked via the maximum between-class variance method, morphological operations and distance transformation. The marker-based image reconstruction and watershed (MIW) algorithm was experimentally compared with four other algorithms. It can successfully segment the overlapping green pellet particles in both sparse and stacked images. After watershed segmentation is performed on the reconstructed marked images, the accuracy of green pellet segmentation exceeds 94.1%, and the accuracy of detection results 93.7%. An on-site detection experiment was conducted on the grate kiln of a certain steel plant. The discrepancy between the proportion of green pellets detected by the system and that obtained through manual screening is within 3%. The online detection of the pellet particle size on the basis of the reconstruction of marked images provides a robust basis for optimising decision making in the pelletising process.