• Data mining for mine-mill ore grade reconciliation at Erdenet Mining Corporation

      Sarantsatsral, Narmandakh (2016-11)
      This project investigates the relationship between the mined ore and the produced copper at the Erdenet Mining Corporation (EMC) surface copper mine in Mongolia. Four and half years of data (from 2011-2015) was obtained from the open pit mine and mineral processing plant of EMC. The mine-mill data was collected on a shift basis. The data was examined carefully using process knowledge and exploratory data analysis techniques to detect and eliminate errors. Ultimately, two years of data (2013-2014) was selected for further analysis. As is common in all mines, the material flow between the mine and mill is complicated by numerous stockpiles. The copper grade going into a stockpile may not be directly related to the copper grade exiting a stockpile. Therefore, data mining techniques applied to detect the relationship between mined ore and milled copper had to overcome the complications introduced by the presence of stockpiles. Multiple data sets were created by aggregating the original dataset by different periods. For example, in one case, data was aggregated by three shifts, to convert the data from shift-basis to daily-basis. Aggregation is an ideal way to absorb variations in material flow (tonnage and grade) between mine and mill. Data was aggregated by 3, 4, 5, 6, 7, 8, 9, 10, 11 and 12 shifts ("aggregated widths" or AW). Correlation analysis was then conducted on each version of the data to determine if a relationship existed between mine data and mill data. Correlation was computed for various period lengths, but not exceeding 28 days. Therefore, in a given year, several correlation plots were produced at each aggregated width. The number of times the correlation coefficient exceeded 0.8 in a year was measured. Results showed that correlation improved with aggregation width. The highest correlations occurred at AW of 7 or 8. This suggests that the stockpiles aggregate material for 2-3 days. Correlation analysis also included examining a time shift ("lag") between mine data and mill data. This is useful to detect whether material takes a certain amount of time before it is processed and produced as copper. However, results indicated that once the data was aggregated, a time lag greater than 0 only worsened correlation.