- Geowissenschaftliche Sammlungen
The mineralogy of iron ore carriers is one of the key tools for understanding the reduction behavior. By analyzing images from polished sections it is possible to get parameters like mode and porosity in a short period of time compared to point counting. For the evaluation of iron ore carriers different limiting factors are identified and algorithms are developed to predict the quality of the charge material.
Principle image processing of iron ore carriers
It is necessary to keep the settings for image acquisition constant. The number of images for processing depends on the grain size and the uniformity of the texture and ranges between 200 and 1000 images per grain fraction. The positions of the image acquisition raster’s are randomly distributed and the size of the raster depends on the grain size and texture of the sample too.
In order to respond to the characteristics of the different types of iron ore carriers their evaluation algorithms are realized with the measurement and automation programming language LabVIEW developed by National Instruments. First, before the actual processing routine, the quality control is performed. The quality control comprises of an edge detection program. Every image is executed by a differentiation filter. This filter produces continuous contours by highlighting each pixel where an intensity variation occurs between itself and its neighbors. The mean value of the highlighted pixels for every image is calculated and diagrammed by a boxplot. This display option gives a first impression of the overall image quality. Every image below a defined drop out value is removed. With this algorithm it is possible to eject blurred images and mostly black images (e.g. out of grain/sample boundary). Next step removes color fringes between dark and bright areas by correction of the color layers. Image noise is minimized by a suitable filter routine like Gaussian filter. The evaluation routine starts with phase identification by thresholding. Each phase gets unique intervals for RGB (red, green and blue) and HSL (hue, saturation and luminescence). Depending on the texture of the ore and effects like interior reflections it is necessary to correct the phase identification with morphology tools like erode/dilate or removing small particles.
Hematite, magnetite and limonite are identified on considering lump ores. For pellets hematite and magnetite are combined to Fe oxide. In addition glass and pores are identified. The identified area to image area is accumulated image by image. Referring to the basic equation from stereology V
A based on the principle of Cavalieri the percentage of volume is calculated for each phase.
The quality evaluation of lump ore is based on a simulation of the reduction process by a concentric phase front movement. It is a distance contouring by encoding a pixel value of a particle as a function of the location of that pixel in relation to the distance to the border of the particle. As a result every particle is subdivided into concentric shells from border to core. The algorithm is called Danielsson distance map by Erik Danielsson.
Depending on the reducibility of each phase an assured number of shells is excluded from one calculation step to the next, starting at the border of the phases. The remaining area is measured. This simulates the progress of the reduction front in a certain period of time which is distinctive for every mineral phase. For each of these calculation steps the removed area is displayed as diagram (amount of steps/ cum. removed area). The results correlate with the reduction progress (time/ reduction grade) as measured with standard reduction tests for the same samples performed at the Department of Metallurgy.
The pellet evaluation is based on a pore size distribution model. The pore area is measured and the diameter of a coextensive circle is calculated. The result is displayed as a cumulative distribution diagram (equivalent diameter/ relative frequency). Compared to the reduction curves from standard reduction tests the performance of pellet samples is assumed to be positively influenced by a high amount of large pores.