Shadow detection based estimator for particle size distribution of crushed oreThe particle size distribution of the crushed ore plays a significant role in autogenous grinding process. In Pyhäsalmi, the estimation and prediction of the incoming particle size distribution is valuable information for the mine. The measurement is carried out by using a specially designed machine-vision-based system. Introduction Pyhäsalmi Mine Oy, owned by the Canadian Inmet Mining Corporation, operates a copper, zinc and pyrite mine in central Finland. The Pyhäsalmi mine is the oldest operative mine in Finland and one of the deepest in Europe since the new shaft reaching 1410 meters at depth was opened in the summer of 2001. The mine production has steadily increased and is currently about 1.3 million tons per year. The Pyhäsalmi mine is considered to be one of the most cost-effective and best performing underground mine world-wide.
A schematic diagram of the mining process is shown in Figure 1. After blasting the ore is mucked to the ore passes or tipped directly into the jaw crusher. Then it is hoisted to the underground bin beneath the headframe from where it is transported by a conveyer belt to the mill. Before hoisting there is also a smaller bin as a buffer. Time delay between the ore crushing and the ore grinding can be several days, because the bins between the mill and the crushing plant can store over 15 000 tons of ore.
 Measurement & Benefits
In order to get some knowledge of the particle size distribution in advance, a novel image-analysis-based measurement system was developed and installed to the mine (exact position indicated in Figure 1). As shown in Figure 2, a monochrome camera is installed on top of the conveyor belt and shadows are created by a single light source coming from side.
  The main principle behind the particle size measurement is to first measure shadow length distribution from a preprocessed image and then to utilize this information (combined with the mass flow measurement) to get an estimate of the actual particle size distribution (see Figure 3). The image analysis is carried out by using dedicated algorithms running on Visual Basic™ and Matlab™ based programs. 
The calibration model shown in Figure 3 can be chosen arbitrarily. The authors have successfully tested both neural network and Partial Least Squares (PLS) based models. Currently the measurement utilizes the PLS model, which was generated and validated by manually screened ore. In the frequent validation tests the presented analysis method has proven to be very accurate: Correlation coefficients for different size classes are between 0.77 and 0.95 (for the PLS case). The particle size measurement calculates the same three particle size classes which are screened at the mill. Therefore the changes in the size distribution can be observed much earlier, which makes it possible to predict the feed ore particle size distribution of the grinding circuit up to three days in advance.
|
