The system for determining the grain-size distribution of bulk material particles is described by the example of estimating the dimensions of crushed pieces of iron ore. The stages of the algorithm for processing of iron-ore particle images are illustrated. The results of full-scale testing of the system at the mining and processing plant are analyzed.
Using the video capturing sensor (VCS), the original image of the crushed ore particles lying on the conveyor belt is captured (Fig. 2).
The algorithm for determining the dimensions of bulk particles is as follows. At the first stage, the original image F (i, j) is processed by a median filter. The aperture of the median filter W (i, j) is a square matrix of size 3x3. Traditionally, the filter response is calculated as:
MED[W (i, j)] MED[{ f (i k, j l)}; k,l 1,1]
The nature of the impulse distortions on the resulting video images is in the form of "pepper" with an area of not more than 3 pixels. Given this fact, a sliding hybrid median filtering is used, which is as follows.
W (i, j) {clm(i, j 1), clm(i, j), clm(i, j 1)}, where clm(i, j) { f (i k, j), k 1,1} Then the response of the sliding hybrid median filter will be calculated according to the formula When shifting W (i, j) by one position, we get
HMED[W (i, j)] MED[MED[clm(i, j 1)], MED[clm(i, j)], MED[clm(i, j 1)]]
HMED[W (i, j 1)] MED[MED[clm(i, j)], MED[clm(i, j 1)], MED[clm(i, j 2)]]
It can be seen from (
G( j, k ) As a result of these transformations, a contour image of the visible layer of crushed ore pieces F c (i, j) is obtained (Figure 3), which is the aggregate of M disjoint regions
M F c (i, j) Sk . k 1
It is noted in [9] that one of the main features that allow us to classify an isolated region Sk as a particle of crushed ore is a sign of roundness calculated by formula
L (dik avg(dik ))2 Rk i1
L where dik (xik C k )2 ( yik C k )2 is a distance from each boundary point (xik , yik ),i 1, L of the region Sk to x y its center (Cxk ,Cyk ) (mean(xmk ), mean( ymk )),{xmk , ymk} Sk ,
L dik and avg (dik ) i1
L
The segmentation procedure is implemented as follows. For each Sk , k 1, M Rk is calculated. If Rk <THRESHOLD,
it is classified as an image of crushed ore particles. If Rk > = THRESHOLD, then the selected fragment of the image is
subdivided into the subdomains by the "watershed" algorithm [11,12]. In Fig. 4 the results of this algorithm is shown.
(
At the next stage, for each selected area, its main axis is calculated, the length of which E2a is determined by the
formula
(
In Fig. 5 lines corresponding to the maximum width of the ore particle region are drawn in white. These axes will characterize the dimensions of bulk material pieces. 3
Industrial tests of the GRANICS complex were carried out, in particular, at the Erdenet Mining Corporation, Mongolia. According to the program of industrial tests, the central research laboratory of the plant (CRL) sampled the ore after fine crushing from the conveyor No. 18 in order to determine the condition and nominal grain-size, with parallel fixation of the same parameters of the crushed ore according to the GRANICS output. The grain-size distribution of the ore is shown in Tables 1 and 2.
To perform a comparative analysis of the obtained data, the results of determining the nominal grain-size of the crushed ore and the content of the "+15 mm" grain-size class are presented in Table 3.
System GRANICS and CRL have different principles of evaluation of the same technological process. However, the above results show that the relative error of the GRANICS complex in determining of the nominal grain-size of the crushed ore was only 3.01% compared to the CRL, and in the evaluation of the controlled class "+ 15mm" the error is 5.48%.
Thus, the performed industrial tests illustrate that the complex accurately determines the grain-size distribution of the crushed particles.