The Use of a Colour Parameter in a Machine Vision System, SmartFroth, to Evaluate Copper Flotation Performance at Rio Tinto’s Kennecott Copper Concentrator S H Morar1, G Forbes1, G S Heinrich1, D J Bradshaw1, D King2, B J I Adair2 and L Esdaile3 ABSTRACT Traditionally, operators on mineral flotation plants have relied on visual inspection of the froth surface when making adjustments to process set-points. Machine vision systems are being developed to relate froth surface descriptors, which can be consistently measured in real time, to flotation performance. By investigating the impact of various selected operating variables on both froth surface characteristics and metallurgical performance it is possible to develop a management strategy for the control and optimisation of the process. This paper presents the results of testwork conducted at Kennecott Copper mine which investigates the relationship between the froth colour and the concentrate grade and discusses the factors which affect the froth colour. This work shows that the main factors that influence the froth colour in this system are molybdenum, copper and iron concentrate grade, pulp and concentrate per cent solids and pulp iron grade. It goes on to develop models to predict the concentrate grade based on colour information and velocity and stability information. It also shows that the most accurate prediction is made with combined colour and stability information.
INTRODUCTION Froth flotation has traditionally been controlled manually by operators, based on their experience and judgement. Operators typically look at the surface of the froth and make judgements based on their experience to obtain qualitative indicators of the performance of a flotation circuit. Many factors, such as the bubble size, shape, colour and motion of the froth, are observed and monitored by operators. Experienced operators are known to be able to predict various performance characteristics of a flotation cell to a high level of accuracy based on these observations. Research has been undertaken to develop machine vision techniques that reliably and quantitatively relate froth surface descriptors with metallurgical factors to predict and control performance characteristics of the flotation process based on the froth appearance (de Jager et al, 2003, 2004; Sweet et al, 2000). This investigation evaluates the relationship between the colour of the froth surface and the concentrate grade. Previous work, performed by Heinrich (2003), where analyses were performed in the laboratory and on industrial flotation cells showed that the froth colour showed a statistically significant relationship to concentrate copper grade in a chalcopyrite/pyrite system. This paper addresses some of the recommendations of the previous work, and evaluates other factors that influence the concentrate colour. Variables such as the velocity and stability 1.
Mineral Processing Research Unit, Department of Chemical Engineering, University of Cape Town, Private Bag, Rondebosch 7701, South Africa.
2.
Kennecott Utah Copper Corporation, 8400 West 10200 South, Bingham Canyon UT 84006, USA.
3.
Rio Tinto Technology, 1 Research Avenue, Bundoora Vic 3083.
Centenary of Flotation Symposium
were measured during a set of testwork where collector type was varied. These variables, together with colour, were used to model the concentrate grade.
Colour measurement The machine vision system consists of a video camera viewing the top of a flotation cell, which sends a video signal to a computer at a remote location for processing. The objective of this investigation was to use this system to measure the colour of the froth. The accurate measurement of colour in the flotation environment is a difficult problem. There are two primary external variables that change the observed colour of the froth surface. These factors are the daily and seasonal variation in ambient light and long-term degradation in signal due to the lighting and camera setup. Previous work (Heinrich, 2003) established the need for isolation of the system from ambient light interference and colour calibration. A system comprising a hood to shield out ambient light and a calibration object, which is constantly in the view of the camera, has been developed to provide an accurate robust froth colour measurement (de Jager et al, 2004). It was also established that the colour information should be analysed in a colour space where the luminance component is decoupled from the colour information. The CIE Lab colour space, as shown in Figure 1, is used (Heinrich, 2003) because not only does it decouple the luminance from the colour components, but it was designed to associate similar looking colours closer to each other within the colour space, and thus detects colour in a similar way to the human eye. Previous authors (Bonfazi et al, 2001) have looked at luminance independent colour reference systems, but these have only extended to HSI (Hue, Saturation and Intensity) and HSV (Hue, Saturation and value).
Brisbane, QLD, 6 - 9 June 2005
FIG 1 - CIE Lab colour space.
1
S H MORAR et al
The Lab L parameter represents the luminance component within the colour space, extending from white to black. The Lab a parameter represents an axis spanning redness to greenness and the Lab b parameter represents an axis spanning yellowness to blueness.
Otherwise: L* = 903.3Y / Yn and: a = 500(f(X / Xn) - f(Y / Yn))
EXPERIMENTAL METHODOLOGY
b = 200(f(Y / Yn) - f(Z / Zn))
An experimental campaign was performed at the Kennecott Utah Copper Concentrator, Bingham Canyon, Utah, where four collectors were tested. The Kennecott ore consists mainly of chalcopyrite and pyrite, with a copper head grade of 0.5 to one per cent and four per cent pyrite. The aim of this campaign was to optimise the chalcopyrite recovery relative to the pyrite recovery through the use of selective collectors. The program consisted of step tests, where the collector type was changed. The step tests were performed, after an hour of conditioning time at the new condition, by taking three concentrate samples at ten minute intervals. Images of the froth surface were recorded before, during and after the tests for off-line processing. The colour analysis that was performed on the video images, using SmartFroth, included normalisation against a calibration patch, which was present within the image throughout the test, and a transformation to the CIE Lab colour space. The normalisation to the patch was performed in RGB space by scaling the froth RGB colour components relative to the colour patch. The normalised RGB values were converted into XYZ values using: [X Y Z] = [R G B][M] where: M
is the transformation matrix defined by the PAL specification
The X and Z XYZ colour components are then transformed to produce a constant luminance (Y) component. These XYZ values along with the tristimulus values (XnYnZn) of the PAL white point are calculated and these are then used to determine the Lab colour parameters using the following standard XYZ to Lab conversion: For Y / Yn > 0.008856: L = 116(Y / Yn)1 / 3 – 16
where for f(t), where for t > 0.008856: f(t) = t1 / 3 Otherwise: f(t) = 7.787t + 16 / 116 This procedure is described in more detail by de Jager et al (2004). Iterative multivariate linear regressions were used to model the effect of independent variables on a parameter, such as Lab a, or the concentrate grade. This was performed using multivariate linear regressions where a potentially significant variable was added to the regression. The R2 value, determined from the correlation coefficient, was compared between regressions with and without this additional variable. The potentially significant variable was only added to the multivariate linear regression if it increased the R2 value by greater than five per cent, and showed a significance of greater than 95 per cent. The results from the iterative multivariate regression reported include the model coefficients, the significance of the variable in the regression, the variable’s independent R2 value against the modelled parameter, the multivariate regression R2 value on addition of the variable to the multivariate regression and the incremental effect on the overall multivariate regression R2 value. The model prediction is performed using a linear combination of the regression coefficients, which are reported in Tables 1 to 5.
RESULTS AND DISCUSSION Effect of operating conditions on the concentrate Four different collectors were used in this investigation. Each collector induced a different response to the floatability of each mineral component in the system. This response caused different minerals compositions to report to the froth which affected the froth colour.
TABLE 1 Regression statistics for the modelling of the Lab a colour parameter. Independent R2
Multivariate R2
Incremental R2
100.00
0.34
0.34
0.34
99.99
0.17
0.52
0.18
99.92
0.14
0.60
0.08
Coefficients
Significance
2.14
15.99
Concentrate Mo
-57.6
Concentrate solids
0.422
Pulp solids
1.41
Intercept
TABLE 2 Regression statistics for the modelling of the Lab b colour parameter.
Intercept
Coefficients
Significance
-3.20
88.79
Independent R2
Multivariate R2
Incremental R2
Pulp Fe
1.21
98.28
0.32
0.21
0.21
Concentrate Fe
0.618
100.00
0.26
0.43
0.22
Concentrate Cu
-0.276
99.94
0.16
0.52
0.09
Pulp solids
-0.238
99.98
0.12
0.61
0.09
2
Brisbane, QLD, 6 - 9 June 2005
Centenary of Flotation Symposium
THE USE OF A COLOUR PARAMETER IN A MACHINE VISION SYSTEM, SMARTFROTH
TABLE 3 Regression statistics for the modelling of the concentrate copper grade using colour information only. Independent R2
Multivariate R2
Incremental R2
99.99
0.16
0.19
0.19
97.67
0.04
0.25
0.06
Coefficients
Significance
Intercept
17.1
100.00
Lab b
1.09
Lab a
0.0898
TABLE 4 Regression statistics for the modelling of the concentrate copper grade using the velocity and stability information only. Independent R2
Multivariate R2
Incremental R2
100.00
0.31
0.45
0.45
99.96
0.05
0.55
0.09
Coefficients
Significance
Intercept
-59.5
100.00
Correlation peak
101
Velocity
-1.32
TABLE 5 Regression statistics for the modelling of the concentrate copper grade using the combined colour, velocity and stability information. Independent R2
Multivariate R2
Incremental R2
100.00
0.31
0.45
0.45
99.87
0.16
0.52
0.06
-1.45
99.99
0.05
0.60
0.08
0.0750
98.80
0.04
0.64
0.04
Coefficients
Significance
Intercept
-48.1
100.00
Correlation peak
85.0
Lab b
0.638
Velocity Lab a
FIG 2 - The mean and 95 per cent confidence limits of the concentrate’s copper grade for the tests using different collectors.
From Figure 2 it can be seen that both the lowest grades and the lowest variation in grade over the tests was observed with collector 3, whilst the use of collectors 1 and 4 led to very similar responses with the highest grades.
Factors that influence froth colour The iterative multivariate regression performed on the system variables showed that in addition to grade, there were other main factors that affected froth colour. The results of this regression are shown in Tables 1 and 2 and Figures 2 and 3 where factors such as the per cent solids in the pulp and concentrate, and the amount of iron in the pulp showed significant relationships to the colour parameters.
Centenary of Flotation Symposium
FIG 3 - The predicted Lab ‘a’ value obtained using the concentrate Mo grade, concentrate and pulp per cent solids versus the measured Lab a value.
The results reported in Table 1 and Table 2 show that the strong influences on colour for the Lab a parameter are concentrate molybdenum grade, concentrate per cent solids and pulp per cent solids. The strong influences on the Lab b colour parameter are pulp and concentrate iron grade, concentrate copper grade and pulp per cent solids. It can be seen in Tables 1 and 2 that concentrate solids and copper grade only affect the colour of the froth by less than ten per cent, whereas the more significant contributions to changes in the froth colour are related to the iron and molybdenum concentrations in both the pulp and the concentrate.
Brisbane, QLD, 6 - 9 June 2005
3
S H MORAR et al
Figures 2 and 3 show that the colour variation is not fully described by the parameters in the model. It is likely that there are other factors that could account for up to 40 per cent of the uncharacterised variation in froth colour.
Modelling of concentrate grade The modelling of concentrate grade can be performed using machine vision data, using the multivariate linear regression models shown in Table 3, 4 and 5.
Colour information only Figure 4 shows that colour information only is not sufficient to predict copper grade accurately. The regression statistics in Table 3 suggest that there is a relationship that explains 25 per cent of the variation between the colour and the grade; however, it is evident that the copper grade is not the only factor that affects colour. FIG 5 - The predicted versus measured colour grade where only the colour information has been used.
FIG 4 - The predicted Lab ‘b’ value obtained using the pulp and concentrate Fe grade, concentrate Cu grade and pulp per cent solids versus the measured Lab b value.
These results confirm the findings from the previous work performed by Heinrich (2003), which suggested that the colour information only could be used to predict grade if the model is continuously updated. It was proposed that an on-stream analyser is used to analyse the concentrate grade. The results from this model periodically update the colour to grade model as operating and ore conditions change.
Velocity and stability information There is potential to use stability and velocity information (Hatfield and Bradshaw, 2003) to model the concentrate grade, as shown by the regression statistics in Table 4 and in Figures 5 and 6. This model predicts the concentrate grade to within 4.5 per cent Cu with 90 per cent confidence. This relationship has been verified on data obtained from Northparkes Mine, NSW, Australia, and is currently under further investigation. The input parameters in the model shown in Table 4 are ‘velocity’ and ‘correlation peak’. The ‘correlation peak’ information is related to the stability of the froth. Here, it can be noted that as the ‘correlation peak’ rises (the froth gets more stable), the modelled concentrate grade increases. If the velocity of the froth overflowing the launder increases, the modelled concentrate grade decreases.
4
FIG 6 - The predicted and measured copper grade for each sample point where only the stability and velocity information has been used.
An increase in froth velocity results in a decrease in the concentrate grade. This is due to the increase of entrained gangue particles at high froth velocity. Any change in entrained material will not be detected by the froth surface colour, as it is transported within the plateau borders which occur at the intersection of the bubble lamella. Typically, froth images are dominated by bubble lamella on the froth surface. It thus follows that the major colour components are influenced by the material that reports to the froth due to true flotation, and not entrainment. A more appropriate investigation between colour and grade would include the decoupling of the entrained component in the concentrate by use of top of froth samples. The mechanistic significance of the stability observations will be discussed in a later publication.
Combined information The colour, stability and velocity information was combined using iterative multivariate linear regression (Table 5) and used to model the colour grade. This model shows the most accurate
Brisbane, QLD, 6 - 9 June 2005
Centenary of Flotation Symposium
THE USE OF A COLOUR PARAMETER IN A MACHINE VISION SYSTEM, SMARTFROTH
FIG 7 - The predicted versus measured colour grade where only the stability and velocity information has been used.
FIG 9 - The predicted versus measured colour grade where the colour, stability and velocity information has been used.
The findings suggest that experienced flotation operators intuitively predict the concentrate grade using more than just the colour information, incorporating information such as the froth stability and motion. This work has shown that colour information alone is not enough to accurately predict concentrate grade and the inclusion of other measurements such as velocity and stability information is necessary. Top of froth grade and froth entrainment measurements are recommended to further investigate the colour, velocity and stability relationship to grade.
ACKNOWLEDGEMENTS Thanks go to the staff at the Kennecott Utah Copper Concentrator for all their help and assistance. This work was supported by Rio Tinto. The authors would also like to acknowledge Anglo Platinum and Rio Tinto for their support in the research and development of SmartFroth. FIG 8 - The predicted and measured copper grade for each sample point where the colour, stability and velocity information has been used.
prediction (Figures 7 and 8), with a 4.0 per cent Cu error with 90 per cent confidence, since it combines the information provided by the froth colour and the information provided from the froth stability and velocity. It can be seen from Table 5 that the colour information explained ten per cent of the relationship whilst the stability and velocity explained 45 per cent and eight per cent respectively. There is an overlap between the colour, velocity and stability information, since the colour only relationship explained 25 per cent of the variation, compared to the ten per cent explained here.
CONCLUSIONS Machine vision data can be used to predict concentrate grade. The froth colour provides a limited amount of concentrate grade information; however the best model was obtained using a combination of the colour, velocity and stability measurements. The colour of the froth in a chalcopyrite/pyrite system has been found to be a function of molybdenum, copper and iron concentrate grade, pulp and concentrate per cent solids and pulp iron grade.
Centenary of Flotation Symposium
REFERENCES Bonfazi, G, Serranti, S, Volpe, F and Zuco, R, 2001. Characterisation of flotation colour and structure by machine vision, Computers and Geosciences, 27:1111-1117. de Jager, G, Francis, J J, Hatfield, D P, Oostenborp, B G, Bradshaw, D J, Morar, S H, Nicholls, F C, Forbes, G R, Heinrich, G S and Markham, H W, 2004. A method and a control system for extracting valuable minerals from mined ore, ‘SmartFroth’, Adams & Adams Patent Attorneys, Pretoria, A&A Ref: v16402 (1-16), Provisional patent. de Jager, G, Hatfield, D P, Bradshaw, D J, Francis, J J and Morar, S H, 2004. A method and a control system for extracting valuable minerals from mined ore, ‘SmartFroth’, Adams & Adams Patent Attorneys, Pretoria, A&A Ref: v16148 (1-9), Provisional patent. de Jager, G, Hatfield, D P, Bradshaw, D J, Francis, J J and Rapacz, B M, 2003. The extraction of valuable minerals from mined ore, ‘SmartFroth’, Adams & Adams Patent Attorneys, Pretoria, A&A Ref: v15597 (1-11), Provisional patent. Hatfield, D P and Bradshaw, D J, 2003. The relationship between concentrate yield and descriptors from a machine vision system in a platinum flotation application, in Proceedings XXII International Mineral Processing Congress. Heinrich, G S, 2003. An investigation into the use of froth colour as sensor for metallurgical grade in a copper system, MSc thesis, University of Cape Town. Sweet, C G, Bradshaw, D J, Cilliers, J J L, Wright, B A, de Jager, G and Francis, J J, 2000. The extraction of valuable minerals from mined ore, ‘SmartFroth’, Adams & Adams Patent Attorneys, Pretoria, A&A Ref: 142745 (1-13).
Brisbane, QLD, 6 - 9 June 2005
5
6
Brisbane, QLD, 6 - 9 June 2005
Centenary of Flotation Symposium
