Real-time particle size analysis in wet closed-circuit milling

37 Powder Technology, 12 (1975) 37-50 @ Elsevier Sequoia S-A., Lausanne - Printed in The Netherlands

Real-Time

A.L.

Particle

HINDE

Chamber (Received

Size Analysis

in Wet Closed-Circuit Milling

and P.J.D. LLOYD

of Mines Research December

Organisation,

P.O. Box 809,

Johannesburg

2000

(Republic

of South Africa)

12,1974)

SUMMARY An attempt is made to summarise the state of the art of real-time particle size analysis as applied to the mill-classifier process in wet grinding circuits. Problems associated with the extraction of representative samples from the overflows of industrial classifiers are discussed. A critical analysis is presented of the use of industrial hydrocyclones and small classifying devices as real-time size analysers. Recent developments of rapid response automatic sieving devices are described. An assessment is given of the performance of a commercial onstream size analyser and sampling system.

1. INTRODUCTION Recent reviews [1,2] have given details of available instrumentation for rapid particle size assessment in the mineral processing industry, but little has been said about the specific problems of product sizing in closed wet grinding circuits. This paper endeavours to remedy this deficiency_ Although each individual milling circuit has its own special difficulties and characteristics, there are nevertheless many common problems worthy of discussion. This review attempts to answer a few of the questions which frequently arise in any discussion on realtime particle size analysis, such as: (a) Why are accurate real-time size analysers needed? (b) Is it easy to obtain a representative sample? (c) Can an installed industrial hydrocyclone be used to infer product size distributions? (d) Can small inexpensive classifying devices

be used for accurate assessment of statistical size distribution parameters? (e) How good are readiIy available, b-ut possibly expensive, commercial size-analysis systems? (f) Can conventional batch-sizing techniques be automated and speeded-up? In this review, an attempt is made to answer these questions by comparing, on a statistical basis, the accuracy of the sampling and measurement sub-systems employed in the various potential size analysers. The accuracy of the conventional grab-sample and sieve analysis is taken as the starting point of the comparison, and the accuracy of the factors which determine the performance of the sampling and measurement sub-systems is calculated to indicate the ways in which real-time systems may achieve an equivalent accuracy.

2. THE NEED FOR ACCURAT’:

REAL-TIME

SIZE

ANALYSERS

Milling circuits can be very unstable and difficult to control to within narrow operating limits. Unwanted fluctuations in particle size, pulp densities and volume flow rates can lead to the inefficient use of grinding capacity and to poor extraction of the val~ab!e mineral. In some instances working profits can be extremely sensitive to fineness of grind. For example, on one gold mine studied [ 31, a drop in daily average grind of only 1% minus 75 pm (200 mesh, Tyler) could cause a drop in monthly revenue of about $4800 (at a gold price of $150 per fine oz.). Unfortunately, sizing data from daily or shift averages of bulked samples are of little

36

value in any attempt to stabilise changes in

The slope of the line, called the distribution

product size distributions, because the rate of change may be as high as *lo% minus 75 I.rm per minute over a period of a minute or two. It is, therefore, not surprising that there is a demand for rapid response instrumentation for particle size analysis. But just how precise should such instrumentation be? This is not a simple question to answer, but one thing is obvious; and that is that the precision of any alternative particle size analysis system should be at least of the same order as that of the existing means of size analysis, i.e. the cutting of samples from a classifier overflow and subsequent analysis by sieving. This implies an accuracy of about 22% (at 68% confidence level) by weight passing the chosen mesh sizes. As will become apparent, this accuracy is difficult to achieve.

modulus, is a measure of the spread of sizes in the product. The maximum size given by the straight line is called the size modulus of

3. SOME PROPERTIES OF THE SIZE DISTRIBUTIONS OF PRODUCT FROM CLOSED MILLING CIRCUITS

Before discussing any instrumentation, it is important to know a little of the properties of the size distribution of product of industrial wet classifiers in closed grinding circuits. These size distributions are commonly expressed in cumulative form as the weight per cent passing through a number of chosen mesh sizes, Fig. 1. It is convenient to plot such distributions on log-log graph paper since they can be crudely approximated by a straight line in most cases.

Size

the distribution. The precise shape of the actual distribution curve is determined by a number of parameters such as size distribution and mineralogical properties of fresh ore feed; conditions within grinding mills; circulating loads; classifier design and operating parameters.

In spite of the large number of controlling factors, the whole size distribution can very often be defined precisely by the measurement of the per cent by weight passing a single mesh size. This property follows from the fact that if a large family of size distributions, all originating from any one milling circuit, are plotted as in Fig. 1, then at no point will these curves intersect. It follows that if the percentage passing, say, 100 mesh is plotted as a function of the percentage passing 200 mesh, then the result should be a smooth curve. Some results of such an analysis for a gold ore milling circuit are shown in Fig. 2. These data, collected over a six-month period, show some scatter which reflects changes in the performance of mills, hydrocyclones, pumps and non-equilibrium conditions of the circuit arising from inadequate control and poor functioning of equipment. The percentage passing 200 mesh could be used to predict the percentage passing 100 mesh to within +2% 100 mesh with a confidence of about 92% (79 data points). Hairily, therefore, the classifier itself may

modulus

Fig. 1. Cumulative size distributions of prodcst from a closed grinding circuit with hydrocyclone

classification.

39

Fig. 2. Relationship between the minus 200 mesh and minus 100 mesh fractions of product from a closed grinding circuit.

be used as a size analyser if the operating and design parameters of the classifier can be related to a single mesh size. This possibility is discussed in greater detail in Section 5. However, it must be remembered that the basis for the use of a classifier in this fashion, and in fact the basis for many potential size analyser systems, is the above property of the resultant size distribution, and it does not seem impossible that a classifier may be developed for which the product size distribution does not have this very convenient property. One possible reason for attempting to use the operating parameters of a classifier to determine a particle size in the product stream is that sampling problems may be avoided. Accordingly the nature of the sampling problem is considered next.

4. SOME SAMPLING AND SLURRY HANDLING TECHNIQUES

The process streams from industrial wet classifiers can vary in volume flow rate, solids concentration and particle size distribution. Moreover, such variations can be significant over periods considerably shorter than a minute. Any sampling technique should be capable of coping with these variations without affecting

the representativeness of the extracted sampie.

For those size analysers requiring a batch sample, the extraction of a representative sample is, in principle, relatively easy. Simple automatic devices are availabie where a sampling slot is traversed intermittently acrcss a free falling slurry. Unfortunately, the extraction of a representative sample for continuous realtime size analysers is far from an easy task. One cannot improvise with conventional sequential cutting devices since such samplers introduce pulsating flow conditions which cannot be matched directly to the steady flow requirements of continuous size analysers. It is possible to average out these pulsations in a large well-agitated surge tank, but the use of such a tank would also average out the very changes in size distribution one is trying to measure in real-time. Two novel cutting devices have, however, been described which allow continuous samples to be extracted from cyclone overflows without the introduction of unwanted pulsations in flow rate. Cross 143 used a slotted pipe mounted vertically in the overflow compartment next to the vortex finder of the hydrocyclone (Fig. 3b). This cutter samples a revolving umbrella of pulp. Only one grading comparison has been reported [4]; the percentage of plus 200 mesh given by the sampler was 29.3 and the corrc-

sponding conventional cut from the cyclone overflow launder was 29.5. Osborne [5] has described a sampler which consists of a narrow siot continuously rotated on an axis parallel to the slurry flew (Fig. 3a). This device has been tested over a wide range of size distributions (90 - 10% passing 150 mesh) and solids concentrations (20 - 55% by weight). An accuracy of &O.S6% 150 mesh and 1.12% solids. (SC% confidence levels I has been reported. Even though it is possible to get a representative sample free of unwanted fluctuations, one is still faced with the problem that the sample flow may vary a.ccording to the flow rate of the origin;J process stream. Since most continuous size analysers require a small constant volume flow rate, further subdivision of the sample is often necessary. One solution to the problem has been described by Osborne (51. The varying sample stream is fed to a relatively small, well-agitated tank of about seven litres capacity with an overflow to reject some of the sample (Fig. 3~). Although some

Fig. 3. Sampling devices: (a) rotating slot of Osborne [S], (b) slotted pipe of Cross 141, (c) ssmpling tank of Osborne [ 51.

size segregation can take. place in the tank, it is believed that it is possible to find a zone inside the tank which is representative of the incoming sample stream. By inserting the intake of a constant displacement pump in this zone it is possible to withdraw a representative sample at a controlled flow rate. No data are available to show how this sampler copes with large variations in the feed rate or large changes in feed size and solids concentration. Autometrics [23 have indicated in the design of their system that if the sampling requirements are large (60 - 90 litres/min) it is possible to by-pass the preliminary stages of sample reduction and sample directly from the process stream. Implementation of this direct sampling technique is shown in Fig. 4(a). The purpose of the gaps beneath the weirs is to prevent sanding in the region of the sample intake screen. The setting of some of the weir gaps is adjustable to s&ow for some control over

Fig. 4. More sampling devices: (a) Autometrics

design for direct sampling of process stream, (b) Autometrics design for extracting small calibration samples.

41

TABLE Sampling

1 accuracy

____-___--

Size fraction

Wm ) --~-__ + 44 + i4 +104 +I47 +aos .___~

44 74 -10’7 -147 -208 .___

Average wt.% from cyclone (26 results>

Average wt.% from size analyser (26 results) -

53.5 14.5 14.1 10.0 7.1 1.5

51.2 15.1 15.5 10.6 6.5 0.9

although the differences in the above averages are small they are unlikely to be caused by random fluctuations. Another way of looking for systematic errors is to find the best linear fit to the data shown in Fig. 5. If this is done, the best straight line through the data can be represented by the equation Fig. 5. Results obtained by direct sampling of process stream as described in Fig. 4(a).

y = (0.669

+ (21.8 * 5.8)

where _v is the percentage minus 200 mesh material

the location of the zone from which a representative sample may be taken within the sampling tank. A check was made on the reliability of this sampling technique for an installed system. Conventional hand cuts were taken directly from the cyclone overflow. A. few seconds later (to allow for residence time between hydrocyclone and size analyser), a sample was withdrawn from the sensor zone of the analyser. Comparison of the percentages of 200 mesh material in the corresponding pairs of samples are displayed in Fig. 5. About 62% of the data points were within 22% minus 200 mesh (26 data points) of the ideal sampling line. Much of the scatter could be attributed to real char.ges in the size distribution during the finite period of sampling. In spite of the difficulties of measuring a “noisy” process stream it is possible to investigate systematic errors by taking the average weight percentage of the different size fractions from the cyclone and from the size analyser. These mean percentages are shown in Table 1. From the table it follows that the sample to the size analyser are slightly depleted of the very coarse and very fine size fractions. A statistical analysis of the data indicates that

f 0.086)x

in the

size

analyser

sensor

samples

and x is the percentage minus 200 mesh material in the production hydrocyclone overflow. It follows that sampling is adequate so long as the process stream contains about 65% minus 200 mesh. However, for a coarse process stream containing, say, 50% minus 200 mesh the sample to the size analyser will contain about 55% minus mesh. In some instances it is necessary to subdivide the sample stream even further. For example, a small sample must be withdrawn from the sensor zone of the Autometrics system for calibration purposes. Subdivision of small sample streams can be accomplished by syphoning from a vertical fast-flowing stream as in Fig. 4(b). Sources of random error in this technique would appear to be very small as judged by the good correlation between the ihdicated reading of the analyser and the results of careful sieve analyses on samples taken from the sensor zone. One solution to the sampling problem is to employ the whole process stream in determining the size distribution. Attempts have been made to utilise the production hydrocyclone as an on-line size analyser. The concept is particularly attractive since not only does one eliminate sampling problems, but nothing

more sophisticated than a gamma density gauge, a magnetic flow meter and a pressure transducer are required to estimate size distributions.

5. INFERENCE FROM

OF PARTICLE

MEASUREMENTS

SIZE DISTRIBUTIONS

ON INSTALLED

HYDRO-

CYCLONES

Work related to the application of industrial hydrocyclones as on-line particle size analysers was pioneered by Lynch et ai. [6] _ In their approach to the problem, they proposed that the per cent by weight less than scme chosen mesh size in the cyclone overflow can be related directly to the dsoc parameter of the cyclone, provided the size distribution of the feed to the cyclone does not change appreciably The dsoc parameter is that size of the classified particle which, when the classifier performance has been corrected for the water split, reports 50% to the underflow and 50% to the overflow. The parameter can be &lculated from bath the operating parameters of the cyclone such as the volumetric flow rates, the pulp densities and the operating pressure, and the design parameters of the cyclone such as the spigot and vortex finder diameters, the cyclone diameter, the cone angle, etc. The calculation requires empirical equations relating dsoc to these parameters by multiple regression techniques. Typical results of this approach are shown in Fig. 6, for silica classified in a 20-inch cyclone operated in a test circuit with feed of constant size [ 63 _ Analysis of these results indicates that the percentage passing 200 mesh can be predicted to within 22% at approximately 50% confidence. In closed production circuits, however, there may be marked changes in the size distributions of the cyclone feed caused by changes in fresh ore tonnage or hardness. In order to compensate for the effect of such changes, Draper and Lynch [7] suggest that the cyclone performance can be adequately linked to the fraction of -200 mesh material in the product by an equation of the form logle(% al -

-200

mesh) =

=2 loglO(dssc) -

as T

(1)

where T is the tonnage of fresh feed to the circuit.

ZOO Cdculoted

r~lucot

.+,

Icorrected,

250

,rr,crans!

,Fig. 6. Relationships given by Lynch et al. [S] for inference of product size distribution from measured hydrocyclone parameters.

Lees [8,9] has refined the regression equations and has also pointed out that a far more complex model of the cyclone t.han was originally proposed would be necessary to cope with those industrial hyclrocyclones which operate at very high solids conterns, or with coarse feed, or where a high percentage of feed water reports to the underflow_ However, successful application of this sizing technique has been reported by Fewmgs IlO] whereby product size distributions can be controlled to within 1 - 1.5% minus 200 mesh with 95% confidence. Assessment of this accuracy was evaluated from product sampled every 20 seconds and bulked over i5-minute time intervals. Thus results are the average of 45 samples and consequently it is difficult to assess the standard deviation of a single measurement. If deviations can be attributed to random changes, it would appear that this error would be about * (3 - 5%) 200 mesh at a 68% confidence level. This result is certainly comparable to those attainable with expensive on-line size analysers w-hen one considers the overall accuracy of instrumentation and sampling. It is interesting to speculate on the limits of the use of hydrocyclones as size analysers

45

arising from their sensitivity to small changes in feed pulp density. This effect is made clear if one examines one of the regression equations given by Lees [ll] : log, d5,,c = 0.047 PSF + 0.14

VF + 0.015

0.47 SPIG

P + 3.33

(2)

where PSF = per cent solids by weight in cyclone feed, SPIG = spigot diameter in inches, VF = vortex finder diameter in inches, P = cyclone feed pressure in lb/in2, and dsoc = cyclone parameter in microns. If an error of t G(PSF) OCCUTSin the estimate of pulp feed density, then the resultant error in dsoc is Wc150,)

=

d 5oc(

0.047)

6 (PSF)

Cyclone

Fig. 7. Real-time

relationships

to speed up and automate

the sieving process

to allow regular up-dating of the cyclone model.

As an example, consider a nominal value of d,, of 100 microns and let s(PSF) be *l% solids by weight. Then 6(ds0,) = 4.7 microns. If one now refers to Fig. 6, it follows that an error in pulp density measurement of 2 1% solids could lead to an unwanted product size error of about +2.6% minus 200 mesh. In practice, the degree of sensitivity of particle size to feed solids concentrations can vary considerably from plant to plant. For example, consider Fig. 7, where the recorder outputs from cyclone feed and product density monitors are shown together with the particle size assessment on a corresponding time scale. These plots were obtained for a gold plant operating with two stages of classification involving two 610-mm

Cyclone

hydrocyclones in series. A 1% solids by weight change in feed concentration to the first cyclone produced a 1 - l%% minus 200 mesh change in the final product size distribution_ The specified accuracy of gamma density gauges is about ?O.OOl specific gravity units (about +O.l% solids by weight for pulp density of about 40 - 50% solids specific gravity of 2.7). However, such instruments are susceptible to drift and must be standardised regularly; standardisation must sometimes be done daily. Consequently it is important to keep a regular check on the performance of the cyclone as a size analyser of the cyclone overflow product. There would appear to be considerable incentive

feed

overflow

between

This is discussed

6. CLASSIFICATION

in Section

DEVICES

8.

FOR

RAPID

It is clear from the above discussion that while it is possible to use the classifier installed in the circuit to estimate the particle size distribution of the product with sufficient accuracy, the application of this technique requires very thorough analysis of the circuit and repeated checking of the resultant empirical equations relating particle size to operating parameters. An alternative approach has been to accept the inherent difficulties of sampling, and to

density

densi

cyclone

overflow

SIZE

ASSESSMENT

and feed densities and particle size distribution.

install smaller, more precise classifiers alongside the production classifiers. Although some attempts have been made to use small calibrated hydrocyclones [12] and elutriation columns [ 2 J as continuous size analysers, only the device developed by HollandBatt [13] has been subjected to a rigorous investigation [ 51. The device takes the form of a helical tube of rectangular cross-section. Sample pulp densities and volume flow rates (about 10.5 litres/min) in the feed to the device are held constant. Size distribution parameters can be determined by beta-ray attenuation measurements of the densities, at different locations, of the size-segregated slurry emerging from the classifying helix. A possible limitation to the accuracy of the device is its sensitivity to changes in feed pulp density. An estimate of this sensitivity can be deduced from the calibration nomogram shown in Fig. S. As an example, consider a feed density of l(?% solids by weight and a particle size distribution corresponding to 30% plus 150 mesh. if the density reaches 15% solids by weight, then the read-out from the instrumen .t will indicate that the size distribution conesponds to 40% plus 150 mesh. In other words, a 1% error in the per cent solids estimate of the feed pulp density could lead to an error of about 2% plus 150 mesh. This sensitivity to

Fig. 8. Calibration nomogram for helical classification device described by Holland-Batt (133.

pulp density appears also to be a limiting factor for all similar means of estimating particle size, such as smali calibrated cyclones aud elutriation columns. A possible solution to the density dilemma has been presented by Kelsall and Res;;arick [14] in their design of the Cyclosen%or. This device has the disadvantage of being a batch size analyser. It relies on the assumption that the whole size distribution can be precisely defined by a singIe position on the curve, as discussed earlier. The basic principles behind the operation of this device can be understood by referring to Fig. 9. An extremeiy dilute sample of the milled ore is introd-aced at constant volume flow rate, to a coarse separator in the form of a tangentially fed cylindrical screen. The coarse fraction is allowed to settle and the finer fraction is further separated with an efficient hydrocyclone into a fine size fraction and a very fine fraction. The very fine fraction is discarded and the fine fraction allowed to settle. The ratio of the times taken to fill the coarse and the fine fraction collection vessels to the indicated levels can be related directly to the particle size distribution.

Fig. 9. The Cyclosensor

of Kelsall and Restarick

[ 141.

The device has a sensitivity whereby a change cf +1.8% passing 100 mesh can yield a 7% change in the ratio of the settling times. The reproducibility of the device is such that for the same feed rate of the same solids the ratio of the times remains constant to better than l%, and an increase in solids feed rate of 30% has no effect on the ratio. In this discussion of classifying devices, emphasis has been given to the effects of pulp density. However, it must be remembered that the performance of such devices also depends upon particle shapes, solids densities and liquid viscosities. Consequently many indefensible assumptions must be made in any attempt to correlate results of this type of size analysis with statistical size distribution parameters determined by other methods, such as sieve analysis. Moreover, additional errors can arise from the extraction of a small representative sample from a high tonnage process stream, as discussed earlier. It would appear, therefore, as if it would be essential in employing such devices to have a continual checking procedure built in to the operating cycle. This possibility is discussed in Section 8.

T. THE

AUTOMETRICS

SYSTEhl

&=c

The various means of estimating the particle size of a classifier product stream outlined above have relied upon some form of hydraulic classification in a gravitational or centrifugal force field to determine the particle size. There is, however, a unique analyser, which has found wide application in determining particle size in mill pulps, which relies instead on acoustical forces to determine the particle size. This is the Autometrics “PSM System loo”, whicl. %as been quite widely reported in the literature, although little appears to have been reported about the physical principles by which it performs size analysis. A feasibility study of Flammer 1153 into the use of ultrasonics in the measurements of suspended sediment size distributions and concentration gives a quantitative insight into the physical principles of the system. If a pulsed beam of plane uitrasonic waves is transmitted through a sample slurry stream, the intensity of the wave in the suspension can be given approximately by I = I, exp(-2ax)

where & = absorption coefficient, Ia = initial intensity of wave, and _Y= distance travelled by wave. The absorption coefficient is determined by two primary mechanisms: viscous and scattering losses. Viscous losses are associated wit.11 the relative movement of liquid and solid. The particles vibrate in response to the ultrasonic wave, but with a phase lag and different amplitude. Extremely small particles t.end to move in phase with the fluid, and losses are very small. As size increases, the particles tend to lag more and more behind the movement of the fluid and t.he loss per particle increases, but at the same time the solid-liquid interface area, and therefore the loss, per unit mass decrease. These opposing factors result in an absorption maximum [ 161. The second loss mechanism is the scattering of energy due to the absorption of a small amount of energy from the directed beam by each particle, and its subsequent radiation away from the point of absorption. For particles of a discrete size and for X 3 Z&r, it is possible to express the relationship between (Yand the properties of the solids and liquid by the equation

(3)

[

+z(Y-1)2

2+(y+ T)or 1 (4) s

where r = particle radius, h = wavelength of ultra.soiGc wave, c = volume concentration of solids, K = angular wave number of ultrasonic wave in water, y = ratio of densities of particles and water, w = angular frequency of ultrasonic wave, s = (9/4@)[1 + (l/fir)], B = (w/2?J)i’2, u = kinematic viscosity of water, and T = (l/2) + (g/4@). This equation is valid up to about 20% solids, above which the dependence on 2 becomes more complex. The first term in eqn. (4) represents the attenuation due to scattering loss and the second that due to viscous loss. For a given frequency, at very small particle sizes the viscous loss is predominant but as the size increases it becomes insignificant and the qcat.+c:isg loss becomes important. For the case where X 4 27ir, eqn. (4) is no longer valid and must be replaced by the expression (5)

16

Fig. 10. Relationships ultrasonic attenuation of discrete sizes.

given by Flammer [ 151 for the of slurriescontainingparticles

LO

In this range (known as the diffraction range), each particle can be considered as casting a simple shadow so that (Yis proportional to n? per particle. In his thesis Flammer [ 153 indicated that the scattering and diffraction terms in eqns. (4) and (5) were equivalent to similar expressions for the scattering of light by small particles. By analogy, he proposed that in the transition regime where X = 25rr the value of (Yis independent of particle size. Figure 10 represents Flammer’s interpre+tation of eqns. (4) and (5). When the particles correspond to a wide distribution of sizes the integrated forms of eqns. (4) and (5) are very complex, although the qualitative features displayed in Fig. 10 probably still apply. By choosing an appropriate frequency it is possible to find a regime where quite large changes in size distribution will not affect the attenuation of an ultrasonic wave. In such an instance the value of (Yis determined by the solids concentration only. By choosing a different frequency it is possible to find a regime where even a small change in size distribution will result in a major change of the absorption coefficient. Since all particles contribute to the magnitude of the absorption coefficient, it is imperative that the measured size distribution should be uniquely defined by a single point on the size distribution curve. If this were not the case, the results of attenuation measurements at only two frequencies

’

Fig. 11. Results obtained analyser.

with the Autometrics

size

would lead to ambiguous interpretations of the size distribution. However, commercial experience confirms the feature of size distributions to which attention was drawn earlier. In the Autometrics system only two pairs of receiving and detecting transducers are necessary to proouce adequate accuracy. Figure 11 shows some results for a gold mine. The readout of the size analyser is shown plotted as a function of the results of size analyses on small samples extracted from the sensor zone of the analyser. About 94% of the readings agreed with the sieve analyses to within *2% 200 mesh (78 data points). A major problem area in the early development of the Autometrics system was that traces of air could lead to substantial attenuation losses. The air must be stabiliied or removed to allow accuratemeasurement of particle size. The problem was solved by removing the air with a device which utilises a combination of centrifugal force and reduced pressure. This need to remove air has increased the cost of the overall system significantly, and made it an expensive instrument when compared with other instrumentation often -G&Mled in grinding circuits. Nevertheless, at present it

appears to be perfectly competitive with other approaches when its inherent reliability and long-term stability as an accurate size analyser are taken into account.

8. AUTOMATIC

SIEVING

MACHINES

FOR

PLANT

USE

Sieve analysis techniques have been accepted generally as the absolute standard for assessing the size distributions of product from miiling circuits. Until about two years ago very little attempt had been made to automate end speed up sieving analysis. Recently a prototype automatic wet sieving device was described [ll] which would determine a single point on the size distribution curve within a few minutes witho-ut any need to dry samples. The principle of operation of a modified design of the prototype can be understood by referring to Fig. 12. The sieving vessel (about one litre capacity) can be topped

Fig.

12.

Wet

sieving

device

for rapid

size analysis.

up with water or slurry to a specified level with the aid of solenoid valves activated by conductivity probes. The vessel is first filled with slurry containing the particles to be sized, and topped up with water to a precise level to allow an accurate determination of the mass of solids added. This weight. (WI ) can be calculated easily by applying Archimedes’ principle. The fine fraction is next removed from the vessel through the discharge valve. Screening is hastened by propeller agitation and with ultrasonics applied to maintain the sieve mesh completely free of pegged material. The weight of the residue (Wz) is determined by further application of Archimedes’ principle, and the fraction coarser than the screen is then given directly by W2 /II’, . The ability to remove pegged material from the screen by the use of ultrasonics is of considerable importance. By this means, the free area of the sieve can be held constant throughout the sieving process, and the ideal rate of sieving can be approached. It is of some interest to note that the rate of sieving under these conditions is found experimentally to be independent of the mass of solids to be sieved. This is shown in Fig. 13, in which the rate of sieving is shown to be independent of the

Fig.

13.

Sieving

kinetics

of a wet-sieving

device.

48

solids concentration in the slurry fed over the range approximately 40 - 160 g/l. A simple theory of the sieving process has been developed to account for these results. If wj(t) is the weight of particles retained by the screen at any one time, which lie in the size range xi t0 Xi+17 and if Pi is the rate of passage of unit mass of particles in this same size range through a sieve of aperture size (I, and if first-order kinetics on a mass basis are assumed, then

dW9 --

dt

=

dWi -=

-Wi

ctt

P2

-wz

Pi

dWn -=-W,, dt

P,,

Each of these may be integrated, and the result summed over all sizes to give an equation for IV(t), the total mass through the screen at time t:

an order of magnitude higher than that normaUy recommended for conventional dry test sieving. Comparisons between conventional sieving analysis using a “Ro-Tap” sieve shaker indicated that. the accuracy of a laboratory manually operated device was 22% 200 mesh at the 93% confidence level (31 data points). The data corresponded to samples containing between 0 and 100% 200 mesh and for distribution moduli varying from 0.58 to 0.69. Pulp densities were between 10 and 30% solids by weight. In a recent paper, Schijnert et al. [ 181 describe a fully automatic sieving machine which can determine several points on the size distribution curve. In this machine, samples are wet sieved, using a technique involving a pulsating water column and the application of ultrasonics. Screen charges are dried and weighed automatically. Although development work on these auiomatic sieving devices is still in progress, it is hoped that their performance in plant operation will be reported in the near future.

11

IV(t)

=

C

i= 1

wi(0)e-Pi'

(‘7)

or in integral form Smax

Iv(t)

= H’(0)

J

(8)

0 where x,,, is the maximum size of particle in the feed and F(X) is the cumulative weight size distribution of the feed. -4 suitable form for P(x) has been given previously by Rendell 1173 : m

(9) with k and m being constants close to 1 and 2 in value, respectively; LIbeing the sieve aperture, in the same units as x, and b being a constant determined by the aperture shape, etc. Use of this function in eqn. (8) leads to a decrease in W(t) with time very much as shown in Fig. 13. It will, of course, be observed that in eqn. (8), the ratio W(t)/W(O) is independent of W(O) as required by the results of Fig. 13. It is interesting to note that the capacity of the rapid wet sieving device expressed as screen charge mass per unit screen area is more than

CONCLUSION

A few of the difficulties of real-time size analysis applied to mill-classifiers processes have been discussed. Most size analysers reported in the literature are certainly more reliable than the “sixth sense” of a good mill operator, but few systems can yield an overall accuracy which is as reliable as the simple routine sieve analyses on hand-cut samples. However, commercially available sizing systems are rapidly becoming commonplace in many wet grinding circuits. Where such systems have been used for product size control their initial expense has been amply rewarded by a fast pay-back period. Even with such success, however, it must be remembered that at best the analyser systems can yield results for a single point on the size distribution curve with an accuracy no better than about 22% of the 200 mesh size fraction. Table 2 summarises the estimated accuracy of the various analyser systems. Unfortunately, no instrument is yet available for plant use which can determine independently and accurately several points on the size distribution curve. For conventional closed grinding circuits this presents few prob-

TABLE

2 ___

----.Analyser system or sub-system

Accuracy

and confidence

levelsa

Remarks

___~

Sampling

Measuring

(i) Grab sample plus sieve

Not

2(66)

(ii) Rotating slotted sampler

1(6S)

Not

(iii)

2(62)

Not applicable

Systematic errors depending solids and flow rates

Suction near weir sampler

applicable

-__

_~___

.-__-I__--

Accuracy based on random errors. Systematic errors cannot be ascertained since system forms absolute standard. System not real time Details

appIicabIe

of sampler

given in ref. [i-J ]

on size,

(iv) Hydrocyclone

Not

applicable

2(50)

Applicable to cyclone with fined feed size distribution

(v) Hydrocyclone

Not

applicable

3 - 5(68)

For plant in closed circuit. kcuracy good over range 40 - 85,‘; minus 200 mesh

(vi) Helix

3(-)

2(-)

Estimated

for a plant environment

Not applicable for grab sample

I(-_)

Not

tested

in a plant environment

Not applicable for grab sample

2(93)

Not

tested

in a plant environment

(vii)

(viii)

(ix) ~-

tube

Cyclosensor

Wet sieving

Autometrics

aAccuracy

as % 2200

Same as (iii) ___-__ -~-____ mesh. confidence

2(94) ____

level (in brackets)

lems, since the whole curve can be defined by a single point. However, this limitation may well prove to be a disadvantage, especially if one considers likely developments in the design of classifiers and comminution machinery in the not too distant future. It has long been recognised that the performance of a conventional production hydrocyclone is far from ideal. Poor classification efficiencies lead to unnecessarily high circulating loads and poor grinding capacities. Although little data have been reported, it is known that efficiencies can be improved and product size distributions modified significantly by injection of elutriation water near the hydrocyclone apex or by passing the underflow through an ultrasonic screen separator. Conventional grinding mills cannot be controlled easily to give selective breakage of minerals in their natural particle sizes. However, this can be done, although to a limited extent, by adopting unconventional milling techniques such as high-

_-

Fully __

xell-de-

tested in plant environment ---._ _.__ _~. __. --

as R.

speed centrifugal milling or the Snyder process. Although the above comments related to likely developments are very speculative, it does appear inevitable that radical changes in comminution techniques will take place and that. rapid-response size analysers for measuring the xvhole size distribution with high precision will be needed as standard research tools. ACKNOWLEDGEMENTS

The authors would like to express their thanks for the help and co-operation of the staff of the following organisations: Blyvooruitzicht Gold Mining Company, Limited; Aut.omeirics Company; Mount Isa Mines, Limited; Julius Kruttschnitt Research Centre and Commonwealth Scientific and Indust.rial Research Organisation Minerals Research Laboratories. This paper was read at the Engineering Foundation Conference on Particulate Matter Systems, Henniker, N-H., August, 1974.

50 REFERENCES 1 R. Davies, Rapid response instrumentation for particle size analysis. Parts I, II and III, Amer. Lab., (Dec.) (1973) 17-23; (Jan.) (1974) 73-S6; (Feb.) (197-i) 47-55. 2 A.L. Hinde, A review of real-time partick size analysers, J.S. Afr. Inst. Mining RIet., 73 (1973) 258-268. 3 hl.1. Brittan and E.J.J. van Vuuren, Computer analysis, modelling and optimisation of gold recovery plants of the Anglo-American Group, J. S. Afr. Inst. hlining Met., 73 (1973) 211 - 222. 4 H.E. Cross, Automatic mill control system. Parts I and II, Mining Congr. J., (July) (1967) 62-67; (Aug.) (1967) 72-76. 5 B.F. Osborne, A complete system for on-stream particle size analysis, CIhl Bull.. 65 (19i2) 97-107. 6 A.J. Lynch, TX. Rao and N.J. Whiten, Technical note on on-stream sizing analyss in closed grinding circuits, Proc. Australasian Insr. Mining hlet., 223 (1967) 71-‘73. i N. Draper and A.J. Lynch, Operation and control of mineral grinding circuits, Trans. Mech. Chem. Eng., (Nov.) (1968) 207-217. 6 M.J. Lees, A further study of the hydrocyclone, BSc. Thesis, Univ. of Queensland, 1968. 9 h1.J. Lees, Experimental and computer studies of the dynamic behaviour of an industrial grinding circuit, Ph.D. Thesis, Univ. of Queensland, 1973. 10 J.H. Fewings, Control of a grinding circuit using a digital computer, Proc. Symp. on Automatic

11

12

18

15

I6

17

iS

Control Systems in Mineral Processing Plants, Brisbane, May 1971, Australasian Inst. Mining Met., pp. 333-357. A.L. Hinde, A real-time size analyser for plant use, IFAC Symp. on Automatic Control in Mining, Mineral and Metal Processing, Sydney, Aug. 13-17, 1973, Inst. Engrs., Australia, pp. 45-47. R.E.J. Putman, Optimising grinding-mill loading by particle size analysis, Mining Congr. J., (Sept.) (1973) 68-74. A.B. Holland-Batt, Further developments of the Royal School of Mines on-stream size analyser, Trans. Inst. hlining Met., 77 (1968) C185-Cl90. D.F. Kelsall and C.J. Restarick, The Cyclosensor a simple device for on-stream size sensing, Proc. Symp. on Automatic Control Systems in Mineral Processing Plants, Brisbane, May 1971, Australasian Inst. Mining Met., pp. 361-366. G.H. Flammer, The use of ultrasonics in the measurement of suspended sediment size distribution and concentration, Ph.D. Thesis, Univ. of hlinnesota, 1958. R.J. Urick, The absorption of sound in suspensions of irregular particles, J. Acoust. Sot. Amer., 20 (1948) 283-289. M. Rendell, Separation of particles by sieving and screening, Ph.D. Thesis, University College, London, 1964. K. Schiinert, W. Schwenk and K. Steier, A fully automatic device for screen analyses, Z. _4ufbereit. Verfahrenstech., 7 (1974) 368-372.