Analysis of a Coal Preparation Plant. Part 1. Changes in Water Quality

1 Analysis of a coal preparation plant. Part

2 1. Changes in water quality, coal seam,

3 and plant performance

5 Ghislain Bournival 1, Masataka Yoshida 1,3, Nicholas Cox 2, Noel Lambert 2, Seher Ata 1*

7 1 School of Minerals and Energy Resources Engineering, University of New South Wales, Sydney, 8 NSW, 2052, Australia 9 2 Clean Process Technologies Pty Ltd, 700 Standen Drive, Lower Belford, NSW, 2335, Australia 10 3 Mitsubishi Development Pty Ltd, Level 16, 480 Queen Street, Brisbane, QLD, 4000, Australia

11 * To whom correspondence should be addressed

12 Email: [email protected]

16 Abstract 17 Coal preparation plants are under increasing pressure to reduce their consumption of fresh water 18 leading to the use of recycle water. Recycled water generally contains a large quantity of dissolved 19 inorganic electrolytes, which affect coal flotation. This paper reports the conductivity and pH 20 measurements of water in tailings stream of an industrial scale flotation cell covering a period of 21 approximately two years. The different coal seams processed at the site were also compared and the 22 influence the water quality on the overall yield of coal assessed. This study presents the large 23 variation in inorganic content in recycled water observed for a wash plant and determine its possible 24 effect on coal preparation. The maximum daily temperature was found be an important factor 25 controlling the amount of inorganic electrolytes in the water and, very significantly, the water 26 management system implemented. The overall plant performance was not significantly affected by the 27 use of water containing high concentrations of inorganic electrolytes. The effect of the quantity of 28 ions in the water on the flotation process is explored in Part 2.

30 Keywords: coal, coal preparation, flotation, water quality, water recycling, inorganic electrolytes

31 1. Introduction 32 Mining operations have been known to impact the quality of natural water streams [1]. The quality of 33 the water from coal mining could have an adverse effect on aquatic life, on the deterioration of 34 concrete and metal structures, and result in increased costs of water treatment. These environmental 35 effects may be problematic considering the shortage of water supply around coal mines [2]. To limit 36 the environmental impact and reduce their fresh water usage, coal preparation plants have developed 37 water usage strategies. The most common strategy is to recycle process water. However, other sources 38 of water with high salt content such as bore water and sea water may also be used [3]. Wang and Peng 39 [4] and Bournival, Muin, Lambert and Ata [5] have reported compositional analyses of water samples 40 from Australian coal preparation plants. They have found a wide range of water conductivities 41 reporting values from 0.24 to 12.86 mS/cm. The pH was mostly in the alkaline range but acidic water 42 with a pH as low as 2.6 has been noted. The main elements found in the water are presented in Table 43 1. It is surprising to find that iron is a relatively low impurity in the process water while pyritic iron is 44 a major impurity in coals. In all cases sodium was to the major cation in process water in coal 45 preparation plants.

46 Table 1. Characteristics of process water from some Australian coal preparation plants.

Bournival, Muin, Lambert and Wang and Peng [4] Ata [5] Parameter Bournival, Zhang and Ata [6] * Minimum Maximum Minimum Maximum pH 7.1 9.2 2.6 8.8 Conductivity, mS/cm 0.24 12.86 3.80 10.95 Na, mg/L 385 3100 586 1212 K, mg/L 3.4 54 7.65 33.8 Ca, mg/L 6 365 4.11 470 Mg, mg/L 3 180 2.9 458 Fe, mg/L -- -- 0 30.8 Si, mg/L -- -- 2.32 49.4 Cl, mg/L 333 2360 423 1011

SO4, mg/L 57 4800 177 5084 47 * Both references combine a total of 7 different sites of which one site was analysed at two different times for a total of 8 48 analyses. The data in the table represent a small sample of the 125 coal mining sites in Australia listed in 2015 [7] and a 49 small sample in time.

50 The chemical composition of the water used in minerals and coal processing plants is determined by 51 the sources in which the water is derived, properties of the feed going through the plant (e.g. soluble 52 minerals like potash), processing routes applied in upgrading the ore and retention time of the 53 recycled water in the system. As a result of changes in chemical composition, mineral processing 54 operations requiring chemically demanding processes such as froth flotation and leaching may be 55 affected [8, 9]. The constituents present in the water may affect, through adverse or synergistic 56 interaction, the effectiveness of chemical reagents used for beneficiating minerals or coals e.g. [5, 10- 57 15].

58 Another important factor concerning water quality in the flotation of coal is pH. The pH of the water 59 in a coal preparation plant may vary greatly [5, 16] and has been shown to affect the combustible 60 recovery of coal through a number of mechanisms. The pH of the pulp influences the adsorption of 61 ionic collectors [17] as well as directly altering the surface property of the coal. The wettability of 62 coal varies with surface composition [18-20], which can be controlled by changing the surface groups 63 through the pH [21]. The pH of the slurry affects the charge of the coal particles with the value of the 64 isoelectric point being dependent on the impurities in the coal [22-24]. A change in the charge of the 65 coal has been shown to modify the interaction forces between a particle and a bubble, which affects 66 flotation efficiency [25-27] and the particle-particle interaction (i.e. rheological properties) [28, 29]. It 67 has also been found that specific clay slimes may coat the coal particles and prevent their attachment 68 to air bubbles [10, 30-32]. This phenomenon is regulated by the differences in charge between the 69 coal particles and the clays, which is controlled by the pH of the pulp. As well, the quantity of ions 70 dissolved in the slurry from the coal is strongly dependent on the pH of the slurry. Thus a sudden 71 change in the pH from acidic, with a high content of dissolved metal ions, to alkaline led to the 72 precipitation of inorganic electrolytes, which decreased the combustible recovery of coal [33, 34]. 73 Therefore in assessing the quality of the water the pH ought to be included along with any measure of 74 ionic strength (e.g. concentration of salts, conductivity).

75 Most studies investigating the impact of water chemical composition on coal combustible recovery 76 used either artificially prepared solutions in a laboratory environment or water collected from a 77 preparation plant at a specific time. The chemical composition of such samples is very well 78 characterized but does not capture changes in the day to day water quality and how those changes 79 affect the operation of the wash plant. This paper presents the quality (i.e. conductivity, pH) of the 80 process water measured for the equivalent of two years over a three year period in a coal preparation 81 plant in the Hunter Valley in NSW, Australia. However, these measurements were analyzed post hoc. 82 As such they were not obtained through a rigorous experimental campaign, which normally 83 accompanies planned experiments. Despite this shortcoming, it is believed that the analysis presented 84 here is relevant due to the period of time it represents and because it is unbiased since the data was not 85 produced to fit a particular experimental protocol or aimed at getting a specific outcome. In addition, 86 it represents the largest such dataset published to the best of the authors’ knowledge. The first part of 87 this study presents a survey of a typical coal preparation plant with an emphasis on the variations in 88 water quality. The effect of water quality on flotation is presented in the second part of this study [35]. 89 In addition to developing correlations between the different measured factors the current work 90 reinforces the importance of instrumentation and controls in mineral processing plants as discussed by

91 Shean and Cilliers [36] as well as demonstrating the presence of strong and uncontrolled variations in 92 mineral and coal processes [37].

93 2. Experimental Procedure 94 2.1 Coal preparation plant 95 The coal handling and preparation plant from which the data was obtained was located in NSW, 96 Australia. The plant processes mostly coal of thermal quality as well as some semi soft coking coal. 97 As such the mineralogy changes considerably for the different seams washed. A total of 5 different 98 seams were washed in the period of time considered, including several splits for each of the seams.

99 A schematic representation of the washing circuit is presented in Figure 1. The run-of-mine is crushed 100 to 50 mm. The oversize of the screened product (+ 1 mm) is washed by a series of two dense medium 101 cyclones (DMC). The cleaned coal is de-watered in a vibrating basket centrifuge. The finer fraction is 102 de-slimed using a hydrocyclone. The hydrocyclone oversize is cleaned using a spiral. The cleaned 103 product from the spiral is sized using a sieve bend (S/B) of size 150 μm. The oversize undergoes 104 washing in a reflux classifier while the undersize is washed by flotation in a Jameson cell. The 105 flotation circuit consists of a single flotation cell with 16 downcomers. It is to be noted that there may 106 be a large variation in the yield of coal from the flotation circuit due to the numerous seams treated. 107 However, the ash produced upon combustion of the flotation coal product is about 10 % of the coal 108 weight.

Vibrating basket ROM centrifuge Cleaned coal

DMC DMC +1 mm

Thickener d50≈100 μm

S/B≈150 μm Reject

Jameson cell

OSPDM

Clean coal

Screen-bowl centrifuge 109

110 Figure 1. Coal preparation plant process flowsheet. The instrument used to measure the conductivity 111 of the water as well as the average particle density is located at the tailings of the Jameson cell.

112 Most coal preparation plants must put in place strategies in order to preserve their water supply [2]. 113 Such strategies may lead to variability in the quality of the water supply. This particular site used 114 clarified water from the thickener, which was recycled back into the circuit as well as raw water from 115 tailings dams. Approximately 90 % of the water was recycled. The raw water was made of runoff 116 water and water from the pits both collected and pumped into the dams. The site has three major dams 117 in addition to a couple of smaller dams. More details on the particulars of the water management 118 system is discussed in the following section.

119 2.2 Water 120 The CHPP washes coal for a few operations in the region and has an agreement with one of the 121 surrounding operations for sourcing water. The authors did not have access to the record of the water 122 management systems, e.g. source of the water supply to the CHPP supplied at any one time and the 123 volumetric flow rate. However, the company’s water management documentation* provides general 124 information on the systems put in place. Figure 2 is a schematic representation of the water 125 management system of the mine and the preparation plant. Water dam A is the major dam in the 126 system and takes water from the underground workings and other sources such as stockpile, run-off 127 water, etc. Some of the water is periodically piped to Water dam B, which is occasionally used to 128 supply Water dam C. Most of the supply of water for the CHPP is from water dam C although it must 129 be remembered that the quality of the discharged water from Water dams A and B affects Water dam 130 C.

Tailings Water dam C Others Water dam A 45 ML 400 ML

Discharge Water dam B CHPP 200 ML

Potable supply

Surface Underground workings Discharge workings 131

132 Figure 2. Schematic of the main water circuit. The solid lines are regular flows and the dashed lines 133 are occasional flows. The black lines are mine water while the blue lines are discharge water.

134 A few factors are worth to mention concerning the water. From 2013, the company allowed some of 135 the underground workings to be retained in the mine. The exact volume is unknown but it did affect 136 the water balance. It should also be noted that the groundwater make became a more important 137 contributor to the volume of water held in the main dam from 2010. As suggested by Figure 3, which 138 shows the conductivity of surface and ground water, a larger contribution of ground water to the 139 stored water caused an increase in the salinity of the dam water as reported by the company.

* Reference withheld for confidentiality

a b

140

141 Figure 3. Water conductivity measurements from (a) surface workings and (b) underground workings 142 for reported locations.

143 Concerning the ground water of the underground workings, it was stated that a rapid change in the 144 salinity of the water might indicate that a source of surface water or near-surface groundwater may 145 have infiltrated the mine workings. Changes in water management strategy also affected the quality of 146 the water. As previously mentioned, water has been stored in the underground workings since 2013. It 147 has caused the conductivity of the water to decrease from about 4 mS/cm to approximately 2 mS/cm. 148 The decline was caused by the use of potable water (conductivity less than 0.5 mS/cm) for dust 149 suppression in the underground workings, which lowered the overall salinity of the water removed 150 from the underground while mine water was stored underground. These water strategies may 151 influence the process water (see Figure 2) as will be shown later.

152 2.3 Data acquisition 153 For the purpose of assessing the quality of the water, conductivity measurements were used since they 154 correlate well with the concentration of dissolved salts and inorganic materials. Neglecting the effect 155 of individual ionic species, a higher concentration of dissolved inorganic electrolytes results in a 156 higher conductivity. In the present study the conductivity was measured in an Online Slurry Particle 157 Density Meter (OSPDM) (CleanProTech, Australia) [38]. The OSPDM uses the conductivity of the 158 water in order to correct the calculation of the average particle density in the slurry. The OSPDM was 159 installed at the tailings stream of the Jameson cell, as indicated in Figure 1. Other parameters were 160 measured and are presented in Table 2. Table 2 presents a summary of the information that is relevant 161 to Part 1.

162 Table 2. Measured parameters from instruments on a coal preparation and handling plant. Some of 163 the measured parameters were used in the calculation of other parameters as indicated in the table.

Factor Units Additional information 5 different seams coded A, B, C, D, and E. The different splits Coal seam -- were paired for each seam. Conductivity measured on the tailings stream of the flotation cell reported at 20 °C. It was measured by the OSPDM using an Conductivity mS/cm inductive conductivity sensor with an accuracy of 0.5% of the reading plus 0.5 μS/cm. Date at which the measurement was taken (measurements Date dd/mm/yyyy averaged from measurements taken at two minute intervals) Average density of the particles in the tailings stream of the Particle density t/m3 flotation cell as measured by the OSPDM.

pH value of the pulp in the thickener measured using a Thermo Thickener pH -- Scientific Alpha pH 500 controller/transmitter with a standard double-junction pH electrode. Yield % Overall yield of the plant. 164

165 2.4 Data processing

166 2.4.1 Data cleaning 167 The plant was not continuously in operation such that only a portion of the data was retained for 168 analysis. While the instruments were continuously recording there are indicators which could be used 169 to indicate the washery was not in operation. Markers employed were: an overall yield of 0 % (or low 170 yields around 0 % yield values, which may indicate a start-up or shut down), a conductivity less than 171 or equal to 1 mS/cm, and an average particle density less than 1.2 t/m3 or greater than 3 t/m3 since the 172 density of clean coal is approximately 1.4 t/m3 while the density of a coal reject is approximately 2.4 173 t/m3. When such values were encountered a listwise deletion method was employed (i.e. all data 174 pertaining to that particular time entry was deleted). Although such a method may present drawbacks 175 in some situations [39] it was believed to be appropriate to get data of the plant at steady-state for the 176 whole plant.

177 2.4.2 Data analysis 178 The data was received as .csv readable files containing measurements every two minutes. The files 179 were compiled into Excel and a VBA macro was written to clean the data, which was then averaged 180 on a daily basis. The statistical analysis of the data was performed using the open source software R 181 run in RStudio v1.1.383 (RStudio, Inc.). Since the datasets are separated by approximately one year in 182 time, the data was not always paired for statistical analysis. It should also be noted that the data did 183 not allow any mass balancing to be performed. Thus the analysis was conducted on the raw data. The 184 data spanned a period of three years with most of the recordings taking place for the year 2014 and the 185 year 2016. Although each set includes measurements in 2013 and 2017, respectively, the two datasets 186 will be referred to as 2014 and 2016 for simplicity.

187 3. Results and discussion 188 Part 1 of this study is reporting the variations in the quality of the water, the coal seams, and the effect 189 of the water quality on the overall plant operation. The variations in water quality are presented in 190 section 3.1, followed by the analysis of the coal seams (section 3.2) and the performance of the 191 preparation plant (section 3.3).

192 3.1 Assessment of water quality

193 3.1.1 Conductivity data 194 The variation in water conductivity is shown in Figure 4 as a function of time. The conductivity 195 follows a cycle with both sets of data showing lower values around the month of July. Seasonal 196 effects causing variations in the properties of the water (e.g. anti-coalescing properties) have been 197 noted by others [40]. As such, the data is consistent with the fact that the concentration of ions varies 198 significantly in coal process water between different sites and within a site [4, 5]. Seasonal effect will 199 be further explored in the next section. It is noted that the conductivity of the water was measured in 200 the tailings stream of the flotation cell and is reported at 20 °C, as mentioned in Table 2. Although 201 water additions throughout the circuit will alter the conductivity of the water, a change in the

202 measured conductivity is assumed to occur in the entire circuit over time as water is recycled 203 (although it is expected to be different at different locations at any particular point in time).

204 Figure 5 shows the histograms for the two different datasets and the combined data. The conductivity 205 for 2014 and 2016 are bimodal with the modes being further apart for 2016. The lower mode in 2016 206 corresponds to the conductivity values below 8 mS/cm in February of 2016 and 2017 (cf. Figure 4b). 207 The vertical dashed lines in Figure 5 represent the mean values. The mean conductivity increased 208 from 5.0 mS/cm in 2014 to 10.8 mS/cm in 2016. It is believed the difference in conductivity is caused 209 by the decrease in the salinity of the water pumped out of the underground workings to the main dam 210 as discussed in section 2.2. The lower values in conductivity in 2016 could be attributed to the use of 211 water from another source, noting a two week shutdown between the higher conductivity values at the 212 end of 2016 and low conductivity values at the beginning of 2017. As a result of the significantly 213 different distributions between the two sets of data, the combined distribution also presents different 214 modes, with a mean value of 7.6 mS/cm. It is noted that the conductivity of the process water is 215 higher than that of the measured conductivity in various locations (see Figure 3). However, the 216 conductivity of the dam itself was not recorded in the secondary data. Since it is clearly different, it 217 may be hypothesised that the sources of water had a higher salinity or the recycling of the water 218 caused the accumulation of dissolved ions over time.

a b

219 220 Figure 4. Measured conductivity of water with time in (a) 2014 and in (b) 2016 with the vertical axis 221 tick marks on the first day of the month.

222

223

224 Figure 5. (a) Histograms of conductivity measurements in 2014, 2016, and combined. The vertical 225 dashed lines represent the mean values.

226 3.1.2 Factors affecting the water conductivity 227 Differences in water sources can be used to explain the difference in mean conductivity between the 228 two datasets. However, gradual variations with a dataset may be better accounted by seasonal 229 variations. Data for daily rainfall and maximum temperature were recovered from the Australian 230 Bureau of Meteorology [41]. The results for the rainfall are presented in Figure 6. Although some of 231 the raw water was coming from runoff water Figure 6 shows no correlation between the rainfall and 232 the conductivity of the water with a large spread of conductivity values when no precipitation was 233 recorded. It shows the low contribution of precipitation to the total water balance. It is also noted that 234 the monthly average pan evaporation exceeds the monthly average precipitation in this region.

235

236 Figure 6. Conductivity as a function of rainfall for the period of (solid red circle) 2014 and (solid 237 green square) 2016.

238 On the other hand, the maximum temperature significantly affected the conductivity of the water as 239 shown by the linear regressions in Figure 7. It should be noted that full lines were used to represent 240 the regressions of the different dataset, i.e. 2014 and 2016. The dashed lines are linear regressions for 241 conductivity values greater than 8 mS/cm (in green), which represent most of the dataset from 2016 242 and for conductivity values less than 8 mS/cm (in red). The latter set of regressions is believed to 243 reflect more closely the effect of temperature for the different sources of water, noting that the low 244 conductivity values in 2016 did not display a gradual change in the conductivity (pointing to a 245 different source of water).

246

247 Figure 7. Conductivity as a function of maximum temperature for (solid red circle) 2014 and (solid 248 green square) 2016. Solid lines represent linear regressions with the solid lines fitting each dataset and 249 the dashed lines fitting (green) the data greater than 8 mS/cm and (red) the data smaller than 8 mS/cm.

250 Figure 8 presents the residuals from the different fittings. The 2016 dataset is poorly fitted by the 251 linear regression due to the presence of outliers, which are believed to be from a different source of 252 water. The fittings of the data grouped according to the conductivity of the water (i.e. dashed lines in 253 Figure 7) produce good residual plots. The locally weighted smoothing (LOESS) lines in the residual 254 plots show that most of the residuals are approximately randomly located around 0, except for Figure 255 8b where the presence of outliers affected the fitting as previously discussed.

256 Table 3 shows the estimates of the coefficients used to fit a first order model in the form y = β0 + β1x 257 (y is the independent variable and x the dependent variable). Also included are the standard errors on 258 the estimates and the p-values, which is the probability that the coefficients will take on a value at

259 least as extreme as the estimated values when the null hypothesis (i.e. H0: βi = 0) is true.[39, 42] In 260 order words, it is the smallest level of significance that would lead to the rejection of the null 261 hypothesis while considering the data being tested. The coefficient of determination (R2) presents the

262 proportion of the variability in the response that can be explained by the predictors β0 and β1. The 263 analysis confirms that the grouping of the conductivity values by the source of the water provides a 264 better fit. In such fittings the p-values indicate that the effect of the daily maximum temperature is 265 significant. Thus the positive correlation between the temperature and the conductivity for a given 266 source of water suggests evaporation of the water from the dams (and/or thickener), which would 267 result in concentrating the inorganic electrolytes. In this context an increase in the maximum

268 temperature of 1 °C produced an increase, with some scatter, in the order of 0.13 – 0.14 mS/cm (Table 269 3) in water conductivity at this particular coal preparation plant. It should be noted that this value 270 includes any instrument error associated with the measurement of the conductivity at different 271 temperatures. The effect of the time of the sampling is consistent with the measurements of Bournival, 272 Muin, Lambert and Ata [5] who found an increase in the conductivity of water samples collected in 273 November and December (from the plant from which this data was obtained) measured at the same 274 temperature. The subsequent decrease in conductivity may be attributed to the increased use of 275 potable water after the summer season when the water levels in the dams is likely lower. Evaporation 276 can significantly affect the water re-use strategy. As an example the covering of open site water 277 storages helped BHP as part of a water use reduction cost operation at Olympic Dam, Australia [43]. 278 Similarly, such strategy may be used to prevent seasonal changes in water conductivity.

a b

c d

279

280 Figure 8. Residuals from the fittings of the regressions from Figure 7 by index. Residuals for the 281 fitting of the (a) 2014 dataset, (b) 2016 dataset, (c) conductivity values less than 8 mS/cm. and (d) 282 conductivity values greater than 8 mS/cm. The fitted line in the residual plots are locally weighted 283 smoothing with its standard error.

284 Table 3. Summary of the linear regression coefficients in the form y = β0 + β1x for the effect of the 285 temperature on the conductivity of the process water. The individual datasets were fitted with solid 286 lines and the conductivity values were grouped between less than and greater than 8 mS/cm.

Coefficient Estimate Standard p-value R2 error -9 β0, 2014 solid line (residual - ●) 1.781 0.284 2.82 × 10 0.445

-16 β1, 2014 solid line (residual - ●) 0.130 0.011 < 2 × 10 -9 β0, κ < 8 mS/cm dashed line (residual - ○) 1.682 0.272 3.45 × 10 -16 0.480 β1, κ < 8 mS/cm dashed line (residual - ○) 0.137 0.010 < 2 × 10 β solid line (residual - ■) 10.522 0.811 < 2 × 10-16 0, 2016 0.001 β1, 2016 solid line (residual - ■) 0.011 0.030 0.713 -16 β0, κ > 8 mS/cm dashed line (residual - □) 7.893 0.480 < 2 × 10 -12 0.338 β1, κ > 8 mS/cm dashed line (residual - □) 0.140 0.018 4.31 × 10 287

288 3.1.3 Changes in water pH 289 The pH is an important parameter in any coal preparation plant as previously discussed. Figure 9 290 shows the pH for different conductivity values assuming that changes in pH in the overall process are 291 reflected in the thickener where the pH was measured. Most of the pH is in the range of 8 to 9 (and 292 less importantly 7.5 to 8) as shown by the marginal histogram for the paired data. These values are 293 consistent with those reported by Wang and Peng [4] although acidic values are also found similar to 294 those noted by Bournival, Muin, Lambert and Ata [5]. There is no variation in the pH caused by a 295 change in the conductivity. In addition, there was no correlation between the pH and the rainfall and 296 maximum daily temperature (not shown). Since the pH is relatively constant, it is not expected to 297 influence the flotation of coal through the precipitation of clays as previously discussed.

298

299 Figure 9. pH of the water (measured in the thickener) as a function of the conductivity for (solid red 300 circle) 2014 and (solid green square) 2016 with marginal histogram for the paired datasets.

301

302 3.2 Coal seams 303 With the aim to get an overall appreciation of the operation of the coal preparation plant and the effect 304 of water quality, the different coal seams were analyzed using the available data. Many researchers, 305 particularly Napier-Munn [37], have recognised the impact of variations in the quality of the feed in 306 evaluating changes in the mineral (and coal) processes. The quality of coal seams vary widely in 307 different locations [44-46]. One of the determining factors influencing the yield of the plant is the 308 washability of the coal being processed. The yearly average yield of a plant varies significantly. 309 Figure 10 shows the histogram of the overall yield of the plant under investigation for 2014 and 2016. 310 The mean yields in 2014 and 2016 were 71.8 % and 59.5 %, respectively. Since the distributions are 311 not normally distributed, as determined by the Shapiro-Wilk normality test, the Kolmogorov-Smirnov 312 test was used as a non-parametric test to compare the two distributions [47] and were found to be 313 significantly different.

314

315

316 Figure 10. (a) Histograms of the overall yield for 2014, 2016, and combined. The dashed lines 317 represent the mean values.

318 It is evident from the box plots of the yields obtained from various seams, presented in Figure 11, that 319 large variations in the overall yield may be associated to the seams of coal themselves. The washing 320 of seam at different times, cf. dataset 2014 and 2016 for seams A, B, and C, also produced different 321 yields. It should be noted that the results presented in Figure 11 paired the various splits for each 322 seam. The variation found between each coal seam was statistically analyzed.

323

324 Figure 11. Overall yield for the different coal seams washed. Seams washed in 2014 are in red (left) 325 and those treated in 2016 are in green (right). Seams D and E were washed in 2016 only.

326 The Kruskal-Wallis test is a non-parametric test used to compare the population means (μ) of k

327 independent samples [48]. The hypotheses tested are: H0: μ1=μ2=…=μk; HA: μi≠μj. The alternative 328 hypothesis applies for any pair of i and j. The p-values for the Kruskal-Wallis test when testing the 329 means in the yield of the different coal seams were 1.375×10-4 in 2014 and 2.201×10-9 in 2016. The 330 relatively low p-values are essentially associated to the smallest level of significance that would lead 331 to the rejection of the null hypothesis while considering the data being tested [42]. A Tukey HSD 332 (Honest Significant Difference) test was conducted as a post hoc test on the datasets to compare all 333 the pairwise combinations and build a confidence interval on the difference in the mean values for 334 each pair. Figure 12 shows the 99 % confidence interval on the difference of the means for (a) 2014 335 and (b) 2016. The Tuckey HSD difference of means plots statistically confirm what is visually 336 apparent from the box plots in Figure 11. In 2014 it was found that washing seam A resulted in a 337 higher yield than seam B since the mean difference with a 99% confidence does not contain the zero 338 difference in mean level. In 2016, the plant processed coal from 5 different seams where seams A, C, 339 and E seem to be of a higher quality for that year such that the pairwise comparison of a high yield 340 seam (A, C, or E) with a low yield seam (B or D) produced a statistically significant difference. The 341 fact that the Tuckey HSD test outcomes on seams C and B were different in 2014 and 2016 may be an

342 indication of heterogeneities of different splits within the same seam or operational variations.

a b

343

344 Figure 12. Graphical display of pairwise comparison of differences in means from Tukey’s HSD test 345 for the yield of different coal seams with a 99 % confidence level for (a) 2014 and (b) 2016. The bars 346 in red include a difference in mean level of zero, which imply no difference in mean yield between the 347 pairs.

348 The differences in the average yearly yield in 2014 and 2016 (see Figure 10) is partly due to the use of 349 different seams. In 2014, high yield seams (A, C) were used 63 % of the operating time while that 350 proportion dropped to 46 % in 2016 for seams A, C, and E.

351 Models have been developed to optimize the economic output of coal plants by determining the 352 overall yield of the plant and the quality, which should be produced [49]. In the following analysis the 353 average monthly price for thermal coal was used for all calculations. The optimization of a plant 354 should be performed in light of a constant incremental ash and maximum yield to meet defined 355 specifications [50, 51]. However, the specifications for each seam (or blend) and the price of coal may 356 have dictated the choice of a particular seam over another. The incremental ash used for the specific 357 gravity setpoint is unknown but a relation of the yield with the price of coal was evaluated, which can 358 be seen in Figure 13. It shows that the yield was fairly constant despite changes in coal price. The 359 production time weighted mean coal price was statistically similar at US$ 72.22 and US$ 69.60 in 360 2014 and 2016, respectively. As such there was no economic incentive to change the yield of coal, 361 which were statistically different. However, the total throughput may have changed.

362

363 Figure 13. Production yield of coal as a function of the average price of coal for (solid red circle) 364 2014 and (solid green square) 2016. The trend lines represent regression lines for (dashed red line) 365 2014, (dashed green line) 2016, and (solid black line) overall trend.

366 Despite the similarities in yield with coal price, particular coal seams may have been preferred with 367 respect to the coal price. Figure 14 shows that specific coal seams were processed depending on the 368 price of coal. The most significant example of processing of particular coal seams at high coal prices 369 can be seen in 2016 for seam D and E (Figure 14b), and to a smaller extend seam B. The mean price 370 difference for the washing of seams D and E was in general $30 to $40 higher than when other seams 371 were washed. It is interesting to note that the overall yields for those seams were lower than for the 372 other seams (Figure 11), which may indicate coals of lower washability and taking advantage of a 373 higher coal price to process lower quality coal.

374

a b

a’ b’

375

376 Figure 14. Box plot of the price of coal when processing various seams of coals in (a) 2014 and (b) 377 2016. A pairwise comparison of differences in mean prices using Tukey’s HSD for the different coal 378 seams with a 99 % confidence level is presented for (a’) 2014 and (b’) 2016.

379

380 3.3 Overall yield of coal

381 3.3.1 Effect of water conductivity on the overall yield 382 Wet cleaning processes in coal washing is now dominating dry methods.[52] The particular site under 383 investigation operates dense medium cyclones (DMC), spirals, a reflux classifier, and a flotation cell. 384 The use of water and recycled water might affect other separation processes. An attempt was made to 385 correlate the yield of the coal preparation plant with the quality of the water.

386 The overall yield of the coal was tested against the pH of the water and the conductivity of the water. 387 These interactions can be seen in Figure 15. The solid black line in Figure 15a indicates the general 388 effect of dissolved ions on the overall yield of coal. The water conductivity had a substantial effect on 389 the overall yield, but the model did not account for much of the variance in the data, i.e. 390 approximately 8% of the change in yield can be explained by the conductivity of the water through a 391 simple linear model. However, a look at the trends of the individual coal seam reveals that seam B 392 (dashed red line) showed a significant decrease in the mean yield with increases in the water 393 conductivity. It is worth to note that different splits of seam B were washed in 2014 and 2016 (cf. 394 Figure 11), which explains the difference in overall yield since in this case time is also associated with 395 a difference in water conductivity (cf. Figure 5).

396 Although the pH may affect flotation as previously discussed, it is not believed that phenomena such 397 as the dissolution of minerals or precipitation of clays on the surface of larger lumps will affect 398 density separation. Figure 15b shows the effect of pH on the overall yield (solid black line). There 399 were only a few instances in which the pH was low and seam B (dashed red line) and to a lesser 400 extent seam C (dashed orange line) were the main reasons of the upward trend.

a b

401 402 Figure 15. Overall yield as a function of (a) the conductivity of the process water and (b) the pH of 403 the process water for the dataset (solid circle) 2014 and (solid square) 2016. The colors are for the 404 different coal seams with (blue) seam A, (red) seam B, (orange) seam C, (purple) seam D, and (green) 405 seam E. The dashed lines represent the overall trends with the filling being the standard error for each 406 coal seam while the solid black line is the overall trend.

407 3.2.2 Effect of the flotation circuit on the overall yield 408 As previously mentioned (Section 2.1) the OSPDM was installed on the tailings stream of the 409 flotation cell. A low average particle density may indicate that a large amount of coal was reporting to 410 the tailings and not contributing to the overall yield of the plant. In Figure 16 the yield is compared 411 with the particle density of the flotation tailings stream in order to determine a possible correlation 412 between the two. A LOESS was applied to the data to indicate any trend. As a result of the LOESS, a 413 linear regression and a quadratic model were fitted to the paired dataset. The fittings were evaluated 414 based on the Bayesian Information Criterion (BIC) also called the Schwarz Information Criterion 415 (SIC). The BIC is known to provide the correct order of parameters in relatively large samples [53]. It 416 follows [54]:

퐵퐼퐶 = −2 ∙ log ℓ(휃̂) + 푘 ∙ log 푛 1 417 where the term 휃̂ is the maximum likelihood estimate of the model parameters and ℓ(휃̂) is log- 418 likelihood. The second term is the penalty on the log-likelihood and is dependent on n the sample size 419 and k the number of parameters in the model. A slight improvement was obtained by using the 420 quadratic model, which is the model displayed on Figure 16.

421 It was believed that a higher average particle density (i.e. low loss of clean coal to the flotation 422 tailings or entrainment of coal rejects) would result in an increase in the overall yield. The data tend to 423 draw a different picture in which a lot of coal may be lost (i.e. low average particle density) but a 424 significant amount of good coal is also recovered as indicated by the high yield. So, there could have 425 been an opportunity to further increase the yield. However, any definite conclusion is difficult to 426 establish due to the scatter of the data and the type of data on which interpretation has to be made (i.e. 427 the flotation yield was not measured). The particle density of the flotation tailings indicates a

428 somewhat limited effect of flotation on the overall yield. The throughput to the flotation circuit 429 relative to the total throughput was unknown in this particular case although up to 40 % of the raw 430 feed may be processed by flotation [55] but this figure could be as low as 10 % in Australia [56]. 431 Though the economic benefit of flotation has been recognised, e.g. [57], the small throughput may not 432 be as significant on the overall yield and any trend may be insignificant.

433

434 Figure 16. Overall yield as a function of the average particle density in the flotation tailings stream. 435 The dashed blue line represent the LOESS fitting and its shaded area is the standard error while the 436 solid black line is the quadratic fitting.

437 4. Conclusions 438 Secondary data on water quality, expressed as water conductivity and pH, was analyzed to determine 439 the extent of the variation in quality for the equivalent of two years spanning three years. An attempt 440 was also made, using the limited number of parameters available, to correlate the overall performance 441 of the coal preparation plant with the water quality and the seam of coal washed. The effect of water 442 quality on the flotation operation is discussed in the second part of the study.

443 The analysis of the plant data shows that the concentration of inorganic electrolytes is high in a site, 444 which uses recycled water. It is important to realise the very large fluctuations in water quality taking 445 place. Some of the fluctuations are attributed to changes in the water management strategy and others 446 to seasonal variations. Regarding the latter, the conductivity of the water was strongly correlated to 447 the maximum daily temperature. The evaporation of the water could have led to the concentrating of 448 the inorganic electrolytes while increases in the use of potable water at the end of the summer season

449 can reduce the conductivity of the process water. The pH of the water was more consistent except in a 450 few instances in which water from a dam known to contain acidic water may have been used.

451 The coal preparation plant washes a variety of coals with seams of various quality. The mean yield of 452 coal produced was clearly dependent on the seams of coal treated. It should be mentioned that the 453 additional value brought by flotation could not be strongly determined due to the lack of data relevant 454 for such calculations and the scatter in the data. In addition the yield of coal was relatively constant, 455 within a given year, despite changes in the price of the commodity although some seams were clearly 456 preferred when the price of coal was higher.

457 Acknowledgements 458 The authors are indebted to the preparation plant for sharing the data and allowing its publication. We 459 also thank ACARP for its support for the project C26013.

460 References 461 [1] J.D. Powell, Origin and influence of coal mine drainage on streams of the United States, 462 Environmental Geology and Water Sciences, 11 (1988) 141-152. 463 [2] M.S. Klima, B.J. Arnold, P.J. Bethell, Challenges in fine coal processing, dewatering, and disposal 464 (Electronic version), in, Society for Mining, Metallurgy, and Exploration, Inc. (SME), 2012, pp. 389. 465 [3] S. Alexander, J. Quinn, J.E. van der Spuy, J.A. Finch, Correlation of graphite flotation and gas 466 holdup in saline solutions, in: J. Drelich (Ed.) International symposium on water in mineral 467 processing, SME, Seattle, 2012, pp. 41-49. 468 [4] B. Wang, Y. Peng, The effect of saline water on mineral flotation - A critical review, Minerals 469 Engineering, 66-68 (2014) 13-24. 470 [5] G. Bournival, S.R. Muin, N. Lambert, S. Ata, Characterisation of frother properties in coal 471 preparation process water, Minerals Engineering, 110 (2017) 47-56. 472 [6] G. Bournival, F. Zhang, S. Ata, Effect of flotation water chemistry on coal chemistry, fluidity, and 473 coke quality (stage 1), ACARP Report (Project C25011), 2018. 474 [7] Geoscience Australia, Operating mines dataset, in, 2015. 475 [8] S.R. Rao, R. Espinosa-Gomez, J.A. Finch, R. Biss, Effects of water chemistry on the flotation of 476 pyrochore and silicate minerals, Minerals Engineering, 1 (1988) 189-202. 477 [9] S.R. Rao, J.A. Finch, A review of water re-use in flotation, Minerals Engineering, 2 (1989) 65-85. 478 [10] S. Kelebek, U. Demir, O. Sahbaz, A. Ucar, M. Cinar, C. Karaguzel, B. Oteyaka, The effects of 479 dodecylamine, kerosene and pH on batch flotation of Turkey's Tuncbilek coal, International Journal 480 of Mineral Processing, 88 (2008) 65-71. 481 [11] P.K. Naik, P.S.R. Reddy, V.N. Misra, Optimization of coal flotation using statistical technique, 482 Fuel Processing Technology, 85 (2004) 1473-1485. 483 [12] L. Wang, Modeling of bubble coalescence in saline water in the presence of flotation frothers, 484 International Journal of Mineral Processing, 134 (2015) 41-49. 485 [13] B. Michaux, M. Rudolph, M.A. Reuter, Challenges in predicting the role of water chemistry in 486 flotation through simulation with an emphasis on the influence of electrolytes, Minerals Engineering, 487 125 (2018) 252-264. 488 [14] Y. Xia, R. Zhang, Y. Xing, X. Gui, Improving the adsorption of oily collector on the surface of 489 low-rank coal during flotation using a cationic surfactant: An experimental and molecular dynamics 490 simulation study, Fuel, 235 (2019) 687-695. 491 [15] W. Liu, C.J. Moran, S. Vink, A review of the effect of water quality on flotation, Minerals 492 Engineering, 53 (2013) 91-100. 493 [16] Z. Liu, D. Yuan, Z. Shen, Effect of coal mine waters of variable pH on springwater quality: A 494 case study, Environmental Geology and Water Sciences, 17 (1991) 219-225. 495 [17] D.P. Ferrel, C.P. Huang, The removal of fine coal particles from water by flotation, Chemical 496 Engineering Communications, 35 (1985) 351-371.

497 [18] B.J. Arnold, F.F. Aplan, The hydrophobicity of coal macerals, Fuel, 68 (1989) 651-658. 498 [19] D.W. Fuerstenau, M.C. Williams, K.S. Narayanan, J.L. Diao, R.H. Urbina, Assessing the 499 wettability and degree of oxidation of coal by film flotation, Energy & Fuels, 2 (1988) 237-241. 500 [20] T. Chaturvedi, J.M. Schembre, A.R. Kovscek, Spontaneous imbibition and wettability 501 characteristics of Powder River Basin coal, International Journal of Coal Geology, 77 (2009) 34-42. 502 [21] D.W. Fuerstenau, J.M. Rosenbaum, J.S. Laskowski, Effect of surface functional groups on the 503 flotation of coal, Colloids and Surfaces, 8 (1983) 153-174. 504 [22] D.W. Fuerstenau, J.M. Rosenbaum, Y.S. You, Electrokinetic behavior of coal, Energy & Fuels, 2 505 (1988) 241-245. 506 [23] J.A.L. Campbell, S.-C. Sun, An electrokinetic study of bituminous coal froth flotation and 507 flocculation, in: Coal Research Board Special Research Report No SR-74, Pennsylvania State 508 University (Department of Mineral Preparation Engineering), 1969, pp. 262. 509 [24] P.A. Harvey, A.V. Nguyen, G.M. Evans, Influence of electrical double-layer interaction on coal 510 flotation, Journal of Colloid and Interface Science, 250 (2002) 337-343. 511 [25] R.-H. Yoon, The role of hydrodynamic and surface forces in bubble-particle interaction, 512 International Journal of Mineral Processing, 58 (2000) 129-143. 513 [26] R.Q. Honaker, M.K. Mohanty, J.C. Crelling, Coal maceral separation using column flotation, 514 Minerals Engineering, 9 (1996) 449-464. 515 [27] J. Piñeres, J. Barraza, Energy barrier of aggregates coal particle-bubble through the extended 516 DLVO theory, International Journal of Mineral Processing, 100 (2011) 14-20. 517 [28] S.K. Mishra, P.K. Senapati, D. Panda, Rheological behavior of coal-water slurry, Energy 518 Sources, 24 (2002) 159-167. 519 [29] M. Pawlik, J.S. Laskowski, F. Melo, Effect of coal surface wettability on aggregation of fine coal 520 particles, Coal Preparation, 24 (2004) 233-248. 521 [30] Z. Xu, J. Liu, J.W. Choung, Z. Zhou, Electrokinetic study of clay interactions with coal in 522 flotation, International Journal of Mineral Processing, 68 (2003) 183-196. 523 [31] B.J. Arnold, F.F. Aplan, The effect of clay slimes on coal flotation. Part I: The nature of the clay, 524 International Journal of Mineral Processing, 17 (1986) 225-242. 525 [32] B.J. Arnold, F.F. Aplan, The effect of clay slimes on coal flotation. Part II: The role of water 526 quality, International Journal of Mineral Processing, 17 (1986) 243-260. 527 [33] D. Liu, P. Somasundaran, T.V. Vasudevan, C.C. Harris, Role of pH and dissolved mineral 528 species in Pittsburgh No. 8 coal flotation system - I. Floatability of coal, International Journal of 529 Mineral Processing, 41 (1994) 201-214. 530 [34] M.S. Celik, P. Somasundaran, Effect of pretreatment on flotation and electrokinetic properties of 531 coal, Colloids and Surfaces, 1 (1980) 121-124. 532 [35] G. Bournival, M. Yoshida, N. Cox, N. Lambert, S. Ata, Analysis of a coal preparation plant. Part 533 2. Effect of water quality on flotation performance, Fuel Processing Technology, Submitted (2019). 534 [36] B.J. Shean, J.J. Cilliers, A review of froth flotation control, International Journal of Mineral 535 Processing, 100 (2011) 57-71. 536 [37] T.J. Napier-Munn, Detecting performance improvements in trials with time-varying mineral 537 processes - Three case studies, Minerals Engineering, 8 (1995) 843-858. 538 [38] N. Lambert, N. Cox, R. Nicholson, Online slurry particle density meter, in: V. Litvinenko (Ed.) 539 XVIII International Coal Preparation Congress, Springer, Saint-Petersburg, Russia, 2016, pp. 445- 540 450. 541 [39] H.E.A. Tinsley, S.D. Brown, Handbook of applied multivariate statistics and mathematical 542 modeling, in, Academic Press, Sydney, 2000, pp. 796. 543 [40] J.E. Nesset, J.R. Hernandez-Aguilar, C.A. Acuña, C.O. Gomez, J.A. Finch, Some gas dispersion 544 characteristics of mechanical flotation machines, Minerals Engineering, 19 (2006) 807-815. 545 [41] Commonwealth of Australia, Climate data online, in: Bureau of Meteorology (Ed.), 2018. 546 [42] D.S. Montgomery, G.C. Runger, Applied statistics and probability for engineers, John Wiley & 547 Sons, Inc., New York, 2003. 548 [43] ICMM, Water management in mining: a selection og case studies, in, International Council on 549 Mining and Metals, London, UK, 2012, pp. 30.

550 [44] C.R. Ward, Z. Li, L.W. Gurba, Variations in the maceral chemistry with rank advance in the 551 German Creek and Moranbah Coal Measures of the Bowen Basin, Australia, using electron 552 microprobe techniques, International Journal of Coal Geology, 63 (2005) 117-129. 553 [45] C.R. Ward, Z. Li, L.W. Gurba, Variations in elemental composition of macerals with vitrinite 554 reflectance and organic sulphur in the Greta Coal Measures, New South Wales, Australia, 555 International Journal of Coal Geology, 69 (2007) 205-219. 556 [46] R.A. Durie, The characteristics of Australian coals and their implications in coal liquefaction, in: 557 D.D. Whitehurst (Ed.) Coal liquefaction fundamentals, American Chemical Society, Washington, 558 D.C., 1980, pp. 53-73. 559 [47] P.N. Tattar, S. Ramaiah, B.G. Manjunath, A course in statistics with R, John Wiley & Sons, Ltd, 560 2016. 561 [48] J. Verzani, Using R for introductory statistics, Chapman & Hall (CRC Press), New York, 2005. 562 [49] S. Mohanta, S. Chakraborty, B.C. Meikap, Prediction of economic operating conditions for 563 Indian coal preparation plants, Fuel Processing Technology, 92 (2011) 1696-1700. 564 [50] G.H. Luttrell, M.J. Mankosa, Strategies for the instrumentation and control of solid-solid 565 separation processes, in: A.L. Mular, D.N. KHalbe, D.J. Barratt (Eds.) Mineral processing plant 566 design, practice, and control, Society of Mining, Metallurgy, and Exploration, Vancouver, 2002, pp. 567 2152-2163. 568 [51] G.H. Luttrell, P.J. Bethell, R.Q. Honaker, Designing and operating fine coal processing circuits 569 to meet market specifications, International Journal of Coal Preparation and Utilization, 34 (2014) 570 172-183. 571 [52] B.G. Miller, Clean coal engineering technology, Butterworth-Heinemann (Elsevier), New York, 572 2011. 573 [53] R.H. Shumway, D.S. Stoffer, Time series analysis and its applications: with R examples, 3rd ed., 574 Springer, New York, 2011. 575 [54] J.D. Cryer, K.-S. Chan, Time series analysis - with applications in R, 2nd ed., Springer, New 576 York, 2008. 577 [55] J.W. Leonard, Coal preparation, in, Society for Mining, Metallurgy, and Exploration, Inc., 578 Baltimore, 1991, pp. 1184. 579 [56] M.R. Riazi, R. Gupta, Coal production and processing technology, in, CRC Press (Taylor & 580 Francis Group), Boca Raton (FL), 2016, pp. 535. 581 [57] D. Osborne, J. Euston, Value of the Jameson cell to the Australian economy 1990 - 2014, in, 582 Manford Coal Technology Consultants, 2015, pp. 40.

583

Graphical abstract

Tailings Water dam

Change in water quality due to water CHPP strategy (?) Performance (?)