CM3820.2009.Sampling.

doc Last Updated: January 14, 2009 4:42 PM

Sampling
References:
Taggart, Handbook of Mineral Dressing, Wiley, 1945, Section 19
Allen, Particle Size Measurement, Chapman and Hall, 1981
Khan, Thesis, Bradford University, 1968
Introduction:
Sampling procedures cover the practice of selecting representative quantities of test material in
the field, to evaluate bulk materials. Examples of the test materials are bulk granular solids,
slurries, sludges, grains, and solid fuels. It is necessary to be able to sample bulk materials
during shipment and during processing operations. Taggart (1945) defined sampling as:
The operation of removing a part convenient in size for testing, from a whole which is of
much greater bulk, in such a way that the proportion and distribution of the quality to be
tested (e.g. specific gravity, metal content, recoverability) are the same in both the whole and
the part removed (SAMPLE).
Thus with respect to the samples, it is very important that the sample be representative. In
reality, the condition that the sample be completely representative of the whole as regards all
aspects except bulk, is never fulfilled when heterogeneous materials are sampled. For instance,
you can never get a perfectly representative sample of water from Lake Superior. You may
even ask what good it is or what purpose such a sample would serve. It is important, before
collecting a sample, to decide what the purpose of the sample would be, and then plan the
sampling procedure accordingly so that it will be sufficiently representative for your purposes.
Very often, when the material changes with time or with exposure to air, it is required that the
sample also be fresh. Samples which are known to degrade, oxidize, or change on contact with
air or water should be sealed in a tank, kept in an inert, atmosphere, and/or refrigerated or frozen.
However, it should be remembered that all we are doing is slowing down the degradation process
Methods of Sampling from a Conveyor Belt:
Several methods for sampling material from a conveyor belt are available, and the method
selected depends upon the accuracy desired, manpower available, and the cost. However, the
following basic principles should always be observed:
1. The cutter must cut the entire stream
2. The speed of the cutter must be constant
3. The cutter must have an opening width that is at least three times larger than the largest
particle to be sampled, and should be wide enough to prevent bridging (at least 1 cm for dry
solids, and at least 0.5 cm for slurry streams)
4. The cutter opening must have parallel edges.
5. The flowing stream to be sampled must be in free fall.
6. The sample must pass quickly through the sampler to avoid blockage.
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

There are three basic methods for collecting samples:
Linear Traversing Cut
The cutter moves across the stream in a straight-line path. Usually the path is perpendicular to
the direction of the flow, but can be in the direction of the flow or against the direction of the
flow. If adjusted properly, this is the most accurate type of sampler as it follows all six of the
principles given above, but it is also the most expensive and hardest to maintain.
Rotational Traversing Cut
The cutter is mounted on a rotating axle and moves in an arc through the stream. These cutters
are mechanically simpler and cheaper than linear traversing cut samplers, but they violate
principle #2 (constant speed) and so the sample is not as high-quality.
Stationary Cut
The cutter remains stationary and cuts a particular part of the stream at all times. These cutters
are very cheap, and are used very often. However, they violate both principle #1 (cut entire
stream) and principle #2 (constant speed) As a result, the samples they collect are inherently
less representative and less reliable than either of the other two cutter types.
A typical cross-chute sampler, which is a linear traversing cut type, is shown in Figure 1.

Figure 1. A cross-chute sampler for dry bulk solids. The primary sample is diverted from
the main product flow onto a conveyor or into a collection bin.
Main
Product
Flow
Primary Sample
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Methods for Sampling Material in a Laboratory
In a laboratory before performing a size analysis, chemical analysis, or any other determination
of material properties, it is necessary to take a representative sample from the lot being analyzed.
A very detailed discussion of procedures for laboratory sampling is given by Allen (1981). For
most metallurgical or environmental operations, the methods discussed in the following sections
are adequate.
Grab Sampling
This is the simplest quickest, and most flexible method, as it can be carried out on small
quantities using spatulas, or on large quantities using shovels, and can divide the material into
however many samples are desired. The material is first homogenized by thorough mixing on a
rolling mat, as shown in Figure 2. The mat should be a smooth, flexible sheet that the sample
will not stick to, such as glazed paper, hard vulcanized rubber, or smooth vinyl. The material is
then divided into samples by randomly grabbing small amounts from the homogenized pile on
the cloth, as shown in Figure 3. This method uses the least equipment, but also is the most
prone to human biases and has a higher variance between samples than other methods.

Figure 2: Mixing a sample on a rolling mat. Mix by first drawing corner A so that the
sample rolls towards C, then drawing corner B to corner D, then drawing corner C to
corner A, then corner D to corner B, then repeat.
A B
C D
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM


Coning and Quartering
This method is well-suited for large lots of material, and can be conveniently done using shovels,
or even front-end loaders for very large samples. First, the material is mixed and shoveled into
a uniform conical pile, as shown in Figure 4a. The pile must be made so that the natural
segregation in the cone is radially symmetrical. The cone is then spread from the center to form
a flattened disk of material, as shown in Figure 4b. This disk is then divided into quarters using
perpendicular boards, as shown in Figure 5. One pair of opposite quarters is removed, and the
other pair is used as the sample. If the sample is too large, then it can be coned and quartered
again until the desired sample size is obtained. This procedure is also prone to human error.

Containers
1
2
3
4
5
n
Figure 3: Dividing the mixed sample into subsamples, by incremental division with a
spatula.
Figure 4: Side view of sample (a) in conical heap, and (b) after flattening
(a)
(b)
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM


Sample Splitters
These are often called riffle or chute splitters, and a typical splitter is shown in Figure 6. These
consist of a series of chutes that run in alternating directions, so that when material is poured into
the top of the splitter, it flows through the chutes and is randomly divided into two equal-sized
fractions. One of the fractions can then be split again, and the procedure can be repeated until a
sample of the desired size is obtained. In order to work properly, these splitters must be fed
using a special pan that is exactly the same width as the top of the chutes, otherwise the amount
of material entering the two end chutes will be different and the sample will be incorrect. Also,
if a material is repeatedly split into smaller fractions using a riffle, the errors from each stage of
splitting will be added together, resulting in increasing variance between samples.
Figure 5: Divide the sample into quarters. Use boards to delineate them so that the
sample can be divided precisely.
Board
Board
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM


Rotary Riffle.
The rotary, or spinning, riffle is the best method to use for dividing material into representative
samples, because it produces the lowest variance between samples, and can produce a large
number of samples in a single operation. In these riffles, the material to be sampled is fed from
a feed hopper to a feeder (usually a vibratory feeder, although a screw feeder or small conveyor
would work as well). The feeder drops the material at a uniform rate into a series of bins on a
rotating table, as shown in Figure 7. The turntable speed is set so that each sample container
will pass under the end of the feeder numerous times before the feed hopper is emptied, but
slowly enough that the bin edges do not strike the falling particles hard enough to bounce them
into a different container or throw them out of the machine entirely (generally about 10 to 25
rpm).
Figure 6. Chute-type sample splitter. The material flows through the alternating-direction
chutes, and is split in halves.
Feed to be Split
Half of feed diverted this direction
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM


Comparison of Laboratory Sampling Devices
A comparison of the relative standard deviations of samples made by the above methods is given
in Table 1. It can be clearly seen that rotary riffling is by far the best method of sample division,
and approaches the standard deviation that would be expected from an ideal sample divider
where division of material into samples is perfectly random. Both coning & quartering and
grab sampling perform relatively poorly, indicating that they should only be used when there are
no other practical methods that will work with a given material. Chute-type sample splitters
give intermediate performance, indicating that they are suitable for routine, non-critical work.
Table 1: Standard deviations of samples produced from a 60/40% mixture of fine and
coarse sand (Khan, 1968)
Sampling Method Standard Deviation of Samples, %
Cone & Quarter 6.81
Grab Sampling 5.14
Chute-Type Sample Splitter 1.01
Rotary Riffle 0.125
Random variation for a Theoretically Perfect Sampler 0.076

Figure 7. Rotary riffle splitter. Material is split into multiple samples in a single operation. In
the example shown, the material is split into 12 identical fractions, which allows easy division
of the bulk material into halves, thirds, quarters, sixths, or twelfths.
Feed
Hopper
Vibratory
Feeder
Sample
Containers
Rotating Turntable
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Sampling Calculations Using Gys Method
References:
Gy, P. M. (1979), Sampling of Particulate Materials: Theory and Practice, Elsevier
Wills, B. A. (1992), Mineral Processing Technology, 5th edition, Pergamon Press, pg. 79
Use:
This method is a general-purpose calculation to determine the minimum size of sample needed to
ensure that it will be representative of the whole lot, within specified limits.
Before using, approximate estimates of the following will be needed:
The content of the species of interest in the lot (assay)
The general shape of the particles
The densities of the various species and phases present
The particle size distribution
The degree of liberation, and the grain size
Basic Equation:
2
3
max
s
Cd
M W
WM
=

Where: M= Minimum sample weight needed, grams
W= Weight of the entire lot being sampled, grams
C= Sampling constant for the material being sampled, g/cm
3
d
max
= Dimensions of the largest pieces in the sample, cm
s= value of the standard deviation that will be needed to give the
desired level of assurance (assay units, such as % wt, ounces
per ton, etc.)
When W is much larger than M, this equation can be simplified to:
2
3
max
s
Cd
M =
To calculate the desired value for s, first select the desired certainty range (for example, you
might want a copper assay to be accurate to within +/- 0.1% copper), then divide this value by
the mean probable assay value (so for a 5% copper sample, you get 0.1/5 = 0.02). Then, select
the number of standard deviations that will be needed to give the desired certainty. For a
probability of 95%, this is 1.96 standard deviations, so 1.96s=0.02, and so s=0.0102. If the
desired certainty is 99%, then 2.576 standard deviations are needed, and so for this case s =
0.02/2.576 = 0.0078

Calculating C is more complicated, as follows:
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

fglm C =
where: f= Shape factor
g= Size distribution factor
l= Liberation factor
m= Composition factor
The shape factor, f, is taken to be equal to 0.5 for most common natural materials, but in cases
where the material of interest is highly flattened and elongated, the shape factor is set at 0.2.
The size distribution factor, g, is determined by estimating the 95% passing size (d) and the 5%
passing size (d), and using the following guidelines:
If d/d > 4, then g=0.25;
If 2<d/d<4, then g=0.5;
If d/d < 2, then g=0.75;
If d/d = 1 (monosized particles), then g=1.
The liberation factor, l, is a measure of the degree of dispersion of the valuable material through
the bulk, and of the homogeneity of the material. It is calculated from the expression:
d
L
l =
where: L= the size where the values are essentially completely liberated
(grain size), cm.
The composition factor, m, is calculated from the formula:
( ) ( ) at r a
a
a
m +

= 1
1

where: r= specific gravity of the valuable component
t= specific gravity of the remainder of the material
a= fractional average assay of the valuable substance (for example, if an
iron-bearing slag is being analyzed, and the iron is present as Fe
2
O
3
, use the
fraction of the material that is Fe
2
O
3
, not the fraction that is iron, and it
would be expressed as a decimal fraction, not as a percent)
Calculation with Incomplete Information
For a truly unknown sample, most of the parameters that go into calculating C will be unknown,
and may even be impossible to guess with any accuracy. In this case, make the following
conservative assumptions:
f=0.5 (normal blocky particles);
g=0.75 (narrow size distribution. Use g=1 if the sample is obviously monosized);
l = 1 (grains are as large as the particles)
The value of m will still need to be calculated, based on your best estimate of the assay of the
sample and the densities of the components of interest. This initial sample will be larger than
necessary, but can be analyzed and used to refine the calculations for collection of any
subsequent samples.
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Example Calculation:
Bulk Material Parameters:
Material of Interest: CuFeS
2
in a silica matrix, 1.5% Cu (4.3318% CuFeS
2
); Top Size= 1.5 cm;
CuFeS
2
grain size= 0.01 cm
Desired Sampling Accuracy: +/- 0.02% Cu, certainty of 0.99 (2.576 standard deviations)
CuFeS
2
specific gravity= 4.2; Overall specific gravity: 2.8; Broad size distribution
Calculations:
(Calculations were done using a spreadsheet, and were carried out to more digits than are shown
here. Carrying out the calculations with the values as shown will give slightly different results
due to roundoff variations.)
f = 0.5 (regular shape)
g = 0.25 (d/d > 4 for a broad size distribution)
081650 . 0
5 . 1
01 . 0
= = =
d
L
l
a = 0.043318, r = 4.2, and t = 2.8, and so:
( ) ( ) ( ) 4177 . 91 8 . 2 043318 . 0 2 . 4 43318 . 0 1
43318 . 0
043318 . 0 1
= +

= m
( )( )( )( ) 933028 . 0 4177 . 91 081650 . 0 25 . 0 5 . 0 = = = fglm C
01333 . 0
5 . 1
02 . 0
576 . 2 = = s , so s = 0.0051758
( )
2
3
2
3
0051758 . 0
5 . 1 933028 . 0
= =
s
Cd
M =117,539 grams of sample needed for 99% certainty that the assay
will be within +/- 0.02% Cu of the true assay.
Since the original data was only carried to 2 significant digits, this should be rounded to 120,000
grams of sample, or 120 kg.

CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Sampling Calculations using Gaudins Method
Reference:
A. M. Gaudin, Principles of Mineral Dressing, McGraw-Hill, 1939
Use:
This method is limited to materials where the species of interest is a small fraction (a few percent
or less) of the total volume. It does not allow for selecting the degree of certainty desired in the
results, and it is not as generally useful or as precise as Gys method. It would probably be
most useful for calculating the size of a preliminary sample where minimal information about the
material characteristics is available.
Basic Equations:
2
45 . 0
y
x
n =
3
6
a
n
s

=
n
n
S

=
where: x = volume fraction occupied by the species of interest
y = variation in volume fraction that would correspond to the
maximum allowable error
n = number of particles that will be needed in the final sample
n= estimated number of particles per gram

s
= density of the bulk sample, in grams/cm
3
a = top size of the material being sampled, in cm
S = total sample weight needed, in grams.
Example Calculation:
Bulk Material Parameters:
Same material as for the previous example:
CuFeS
2
, 1.5% Cu (4.3318% CuFeS
2
); Top Size= 1.5 cm;
Desired Sampling Accuracy: +/- 0.02% Cu (+/- 0.0578% CuFeS
2
)
CuFeS
2
specific gravity= 4.2; Overall specific gravity: 2.8
a = 1.5 cm

s
= 2.8 gm/cm
3
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM


(Calculations were done using a spreadsheet, and were carried out to more digits than are shown
here. Carrying out the calculations with the values as shown will give slightly different results
due to roundoff variations.)
028879 . 0
8 . 2
100
2 . 4
3318 . 4
=

= x
00038505 . 0
8 . 2
100
2 . 4
0578 . 0
=

= y
87650
00038533 . 0
028879 . 0
45 . 0
2
=

= n
( )( )
63492 . 0
5 . 1 8 . 2
6
3
= = n
S = (87650)/(0.63492) = 138,050 grams
(compared to 117,599 grams using Gys method)

Again, this should be rounded to 2 significant digits, and so 140,000 grams, or 140 kg, of sample
is required.
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Comparison of Gaudins sampling equations and Gys
sampling equation

Gys equation is a general-purpose sampling equation,

2
3
max
s
Cd
M = , where M = minimum sample mass needed, d
max
= maximum particle size present,
and s
2
= allowable sampling variance. C is a sampling constant, which is calculated for any
given material based on particle shape, size distribution, degree of liberation, density, and
mineralogy.


Gaudins equation is specifically intended for use in calculating sample sizes for precious
metals, and makes a number of simplifying assumptions based on the fact that precious metals
make up a minute fraction of the mass of the ore (typically only a few grams per metric ton).

The equation is normally written as two parts: a calculation of n, the number of particles
needed in the sample, and of n, the expected number of particles per gram:

=
2
45 . 0
y
x
n , where x = ore grade expressed as volume fraction, and y = allowable sampling
error expressed as volume fraction.
3
6
a
n
s

= , where
s
= bulk ore density, and a = maximum particle size present.
The mass of the sample needed, S, is then ( )
2
3
3
2
)( 075 . 0
6
45 . 0
y
a
x
a
y
x
n
n
S
s
s

=

Comparison: We can put Gaudins equation into the same terms as Gys equation by using the
same symbols for equivalent values in each: a = d, S = M. The value of y (a volume fraction)
can be converted into (a weight fraction) by multiplying it by a constant, k, that depends on the
relative densities of the ore and the valuable minerals, and so y = /k. If we make these
substitutions, then Gaudins equation becomes ( )
2
3
2
)( 075 . 0

d
k x M
s
=
If we now take (0.075)(x
s
k
2
) = C, we see that Gaudins equation reduces to same form as Gys
equation. Gaudins equation is therefore simply a special case of Gys equation.
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Sampling Calculations Using Hassialis Method for
Increment Sampling
Reference:
Taggart, Handbook of Mineral Dressing, Wiley, 1945, Section 19
Use:
This method is intended for designing a mechanical sampling system that will account for
normal variations in the characteristics of the process stream being sampled. The sample cutter
is used to collect a series of test increments over a particular size interval, and then each
increment is assayed separately. The variations in the assays between increments are then used
to calculate the number of increments that the sample cutter needs to collect over that span of
time to give the desired level of certainty in the results.
Basic Equation:
2

=
z x
h
n
s

where n = number of increments that will be required to reach the
desired level of certainty
h = number of standard deviations that correspond to the desired
level of certainty (for example, 95% certainty corresponds to
1.96 standard deviations, see Table 2)

s
= experimental standard deviation of the assays of the test
increments collected by the mechanical sampler
x = experimental mean of the assays of the test increments
collected by the mechanical sampler
z = allowable sampling error, expressed as a fraction of the assay
value (for example, if the final sample is 15% Fe and z = .04,
then the allowable variation in analysis of the final sample
will be (0.04)(15) = 0.60% Fe. If the final sample is 30%Fe,
then the allowable variation would be (0.04)(30) = 1.2%Fe).
CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Example Calculation
(Calculations were done using a spreadsheet, and were carried out to more digits than are shown
here. Carrying out the calculations with the values as shown will give slightly different results
due to roundoff variations.)
The following 21 test increment assays were obtained from increments collected at 15 minute
intervals, and took a total of 5 hours to collect. Values are weight % Fe.
10.6 10.9 10.2 10.8 9.4 9.6 10.0 10.1 9.9 9.8 9.5
9.2 9.1 9.7 8.8 10.2 10.4 10.2 9.8 9.2 9.3

Calculate the number of increments that must be collected over a 5 hour interval so that the
probability of the assay of the final composite sample being accurate to within 2% of the true
assay will be 99%
Standard Deviation:
s
= 0.57234
Mean: x = 9.8428 % Iron
Allowable variation: z = 0.02 (that is, the analysis of the sample will be accurate to within 0.02
of the mean, and so the analysis will be 9.8428 +/- 0.1968%Fe)
Number of standard deviations needed: h = 2.576

Table 2: Values of h corresponding to various levels
of probability that a particular sample of a lot will
have an assay value within the desired error limits for
the analysis.
Probability h Probability h
0.90 1.645 0.97 2.170
0.91 1.705 0.98 2.326
0.92 1.750 0.99 2.576
0.93 1.812 0.999 3.291
0.94 1.881 0.9999 3.890
0.95 1.960 0.99999 4.417
0.96 2.054 0.999999 4.892

CM3820.2009.Sampling.doc Last Updated: January 14, 2009 4:42 PM

Number of increments needed to make a composite sample of the desired degree of accuracy:
( )( )
( )( )
09128 . 56
02 . 0 8428 . 9
57234 . 0 576 . 2
2 2
=

=
z x
h
n
s

Since it is not possible to collect a fraction of an increment, this should be rounded up to 57
increments per composite sample, and these increments should be collected over a 5-hour
interval.