Engineering & Mining Journal

JAN 2013

Engineering and Mining Journal - Whether the market is copper, gold, nickel, iron ore, lead/zinc, PGM, diamonds or other commodities, E&MJ takes the lead in projecting trends, following development and reporting on the most efficient operating pr

Issue link: https://emj.epubxp.com/i/102812

Contents of this Issue

Navigation

Page 48 of 99

BAUXITE (a) (b) Figure 5—Correlation between the sample data used in the calibration (a) and in the verification (b) of the moisture analyzer. Step 1: Pre-Calibration (Initial Model Building)—The pre-calibration is the initial determination of values for the parameters ({a0, ci} in Model 3). This is best done with the moisture analyzer at a laboratory workbench. Using a set of material samples with known moisture contents mci, achieved by the addition of known masses of pure water to the dry masses of the ore samples, each sample was put inside a non-metallic (fiberglass) box to form a layer of ore with known height (level) hi. The box was then placed between the microwave antennas of the analyzer to sense the microwave attenuation αi and phase-shift φi corresponding to the moisture content mci, level hi, and density ρ of the ore sample. For this step, it's not necessary to take the ore samples from stop belt sampling. One can use ore collected from other easily accessible locations, like the plant stockyard, provided that the collected material has similar properties (particle size distribution and density) to the material carried by the conveyor on which the analyzer would be installed. This initial model building is difficult due to the need for so many ore sample creations, for better representativeness within the intended measuring range for the analyzer (0–18%w). This task used seven reference moisture values (0, 7, 9, 11, 13, 15, and 17%w), five reference levels (within 60–390 mm), two sample box positions (front and rear), and five sample measurement collection, resulting a total of 350 sample measurements of microwave attenuation and phase-shift values. It took seven days, with 12 hours/day of work, to complete the tasks. Figure 3 shows some pre-calibration data for the chosen set of reference moisture values. www.e-mj.com The graphs show the relationship between the quantities αL/M and φL/M versus the moisture content mc, for model (2). After sample data collection in laboratory, the set of experimental data values {mci, ρ, hi, αi, φi} was statistically analyzed to detect and eliminate possible outliers. The remaining data set was then used in a multivariate regression to determine values of the model parameters {α0, ci} that best fit the experimental data. The model was then loaded into the analyzer electronics by the manufacturer, through Ethernet communication, from a laptop running a proprietary configuration software. Step 2: On-Site Calibration (Model Refinement)—Having concluded the precalibration step, the moisture analyzer was installed on conveyor TC-123-02. An overview of the installation is shown in Figure 4. Because of some differences between actual field conditions and the laboratory conditions of the pre-calibration step, an onsite calibration for model refinement is necessary. This refinement is simply an adjustment on the model parameter values {α0, ci} determined in the pre-calibration step. While the analyzer is running, a set of composite samples is taken from conveyor belt stops, and their corresponding moisture values measured by the analyzer are recorded. The samples are sent to laboratory to be analyzed through the LoD method. A measurement check is performed based on the deviations between the analyzer measurements and the laboratory results. If necessary, the model parameter values {α0, ci} are adjusted to better match the laboratory results. The greater the number of samples collected from the conveyor, the more consistent the calibration would be. Ideally, a minimum of 30 samples should be collected. In this work, however, due to restrictions to stop the conveyor belt, a set of only 14 composite samples of wet ore, each one involving three cuts, were collected. The sampling uncertainty (σsampling) is given by the root mean squared error (RMSE) of all the cut sample values of all the composite samples, and resulted as σsampling = 0.391%w. The criterion to consider a cut sample value as an outlier was a cut sample deviation greater than two standard deviations. One of the samples was regarded as an outlier and removed from the sample data set. The remaining 13 valid sample results are shown in Figure 5(a). For each of those valid composite samples, the error εi was computed between the moisture value from the analyzer and its corresponding result from laboratory Equation 5. Equation 5a. Equation 5b. JANUARY 2013 • E&MJ; 47

Articles in this issue

Links on this page

Archives of this issue

view archives of Engineering & Mining Journal - JAN 2013