Academia.eduAcademia.edu

Figure 3 – uploaded by javad gholamnejad

See full PDFdownloadDownload figure
Source: Kasmaee et al. (2010) Figure 3. Location of HP stockpile at Choghart mine of Iran (see online versions for colours)

Figure 3 Source: Kasmaee et al. (2010) Figure 3. Location of HP stockpile at Choghart mine of Iran (see online versions for colours)

Related Figures (8)

Figure 1 Structure of MLP neural network n a multilayer Feed-forward Neural Network (FNN), supervised learning algorithm is applied to train a network. Training a neural network to find a good mapping function can be done by adjusting the connection weights between the neurons of a network using a suitable earning algorithm while fixing the network architecture and activation function. Through he process of training, a network learns general properties of the input-output relationship of a system and thus generalises beyond training data points. There are many learning algorithms used in grade estimation papers, like Newton’s method, Conjugate gradient’s or Levenberg—Marquardt (LM) optimisation techniques (Hagan and Menhaj, 1994; Samanta st al., 2006). The LM technique is widely accepted as the most efficient one in the sense of realisation accuracy and speed (Hagan and Menhaj, 1994).
Figure 1 Structure of MLP neural network n a multilayer Feed-forward Neural Network (FNN), supervised learning algorithm is applied to train a network. Training a neural network to find a good mapping function can be done by adjusting the connection weights between the neurons of a network using a suitable earning algorithm while fixing the network architecture and activation function. Through he process of training, a network learns general properties of the input-output relationship of a system and thus generalises beyond training data points. There are many learning algorithms used in grade estimation papers, like Newton’s method, Conjugate gradient’s or Levenberg—Marquardt (LM) optimisation techniques (Hagan and Menhaj, 1994; Samanta st al., 2006). The LM technique is widely accepted as the most efficient one in the sense of realisation accuracy and speed (Hagan and Menhaj, 1994).
Figure 2 Location of Choghart iron ore mine (see online versions for colours)
Figure 2 Location of Choghart iron ore mine (see online versions for colours)
Figure 4 Sample points locations in HP stockpile (see online versions for colours) The samples are not only scanty but also sparse. They are also located on an irregula: grid. The northing, easting and elevation coordinates were used as input variables and the iron attribute was selected as the output variable. Fe grade distribution in this stockpile demonstrates some special features. Characterisation of the spatial distribution of the dat is, therefore, deemed essential for meaningful ore grade modelling. Towards this goal, < preliminary statistical analysis of data set was performed for Fe values. This includes the estimation of basic statistics involving the mean, median, standard deviation, skewness anc kurtosis. Figure 5 presents the frequency distribution and statistics parameters of Fe.
Figure 4 Sample points locations in HP stockpile (see online versions for colours) The samples are not only scanty but also sparse. They are also located on an irregula: grid. The northing, easting and elevation coordinates were used as input variables and the iron attribute was selected as the output variable. Fe grade distribution in this stockpile demonstrates some special features. Characterisation of the spatial distribution of the dat is, therefore, deemed essential for meaningful ore grade modelling. Towards this goal, < preliminary statistical analysis of data set was performed for Fe values. This includes the estimation of basic statistics involving the mean, median, standard deviation, skewness anc kurtosis. Figure 5 presents the frequency distribution and statistics parameters of Fe.
in variable of interest. The process of dumping of material from various parts of open pit to the stockpile is expected to decrease the continuity and increase the nugget effect.
in variable of interest. The process of dumping of material from various parts of open pit to the stockpile is expected to decrease the continuity and increase the nugget effect.
Figure 6 Experimental semi-variogram parameters for Fe
Figure 6 Experimental semi-variogram parameters for Fe
As can be seen, the variogram are interpreted as having a larger nugget effect and shorter ranges of continuity than is suitable for geostatistical modelling. If the data is interpolated using the high nugget effect, the resource model is over-smoothed. To reduce the impact of high nugget effect, which was identified during the prliminary variography analysis, the stockpile was divided into three homogeneous zones. These are made up of three zones, which had been separated by elevation: level 1: z= 1165, level 2: z= 1163, level 3: z= 1158. Once the statistical analysis is completed and the domains are identified, spatial variability of the data values must be analysed again. This is carried out using variogram analysis, which was previously described. Repeating the variography for each zone shows that the nugget effect is reducing (Table 1), but still it is high.
As can be seen, the variogram are interpreted as having a larger nugget effect and shorter ranges of continuity than is suitable for geostatistical modelling. If the data is interpolated using the high nugget effect, the resource model is over-smoothed. To reduce the impact of high nugget effect, which was identified during the prliminary variography analysis, the stockpile was divided into three homogeneous zones. These are made up of three zones, which had been separated by elevation: level 1: z= 1165, level 2: z= 1163, level 3: z= 1158. Once the statistical analysis is completed and the domains are identified, spatial variability of the data values must be analysed again. This is carried out using variogram analysis, which was previously described. Repeating the variography for each zone shows that the nugget effect is reducing (Table 1), but still it is high.
Table 3 Final report of estimation of high-grade stockpile
Table 3 Final report of estimation of high-grade stockpile
Figure 7 The results of testing process for HP Choghart stockpile data (see online versions for colours) The mean squared error and correlation coefficient in this network are 2.36 and 0.80, respectively, for the available data. Estimation of grade and tonnage of ore dumps can convert the stockpile to valuable ore reserve with regard to increasing of metal demand and depletion of in situ ore reserves in this mine. The final report of the estimation in all levels is given in Table 3.
Figure 7 The results of testing process for HP Choghart stockpile data (see online versions for colours) The mean squared error and correlation coefficient in this network are 2.36 and 0.80, respectively, for the available data. Estimation of grade and tonnage of ore dumps can convert the stockpile to valuable ore reserve with regard to increasing of metal demand and depletion of in situ ore reserves in this mine. The final report of the estimation in all levels is given in Table 3.
Related topics:
EngineeringMultilayer PerceptronArtifi Cial Neural Networks
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved