Short Course on Comminution

& Semi-autogenous Grinding

Session 8 Functional Performance

Slide 1
Functional Performance
• Conceived by Rob McIvor as a way to
measure different efficiencies in grinding.
• Separates grinding efficiency into liberation
and classification.
• Determines efficiency for each step separately.
• Metcom does whole courses on Functional
Performance (CMP, etc.) – recommended!

Slide 2

We will be talking about some circuit benchmarking methods later, things like the
Work Index efficiency of a circuit. These benchmarks tell you if a circuit is operating
efficiently, but don’t help much in identifying where to look inside the circuit to
improve efficiency.
So we will introduce Functional Performance, a framework for separating different
types of efficiency that are mixed up inside an operating ball milling circuit. It gives
metallurgists and operators a more insightful understanding of the mechanisms at
work inside a grinding circuit and permits a more rapid identification of the causes of
poor grinding efficiency.
Metcom operates both live and online courses on grinding efficiency and Functional
Performance; they are highly recommended.
https://www.metcomtech.com
Types of efficiency in grinding
• Electromechanical efficiency (motors)
• Pump efficiency
• Product “recoverability” in flotation
– overgrinding, undergrinding
• Circuit size reduction efficiency
Functional
• Mill grinding efficiency Performance

• Classification system efficiency

Slide 3

Optimizing a grinding circuit involves checking various types efficiency and then
focusing on improving the low efficiency areas.
Electromechanical and pump efficiency can (mostly) only be improved by changing
equipment – installing more efficient motors or optimized pump impellers
Recoverability efficiency comes from the flotation circuit and won’t be covered in this
course.
The remaining three types of efficiency can be assessed using Functional Performance
by separating the different types of process efficiency that get jumbled together in a
grinding circuit.
Intention
• Identify and measure specific grinding
efficiencies.
– classification system efficiency
– product production rate
– ore grindability
• Use these different efficiencies to drive circuit
optimization.

Slide 4

Functional Performance considers three types of efficiency:

1. classification system efficiency, affected by the circulating load and hydrocyclone
operation. (Not to be confused with classification efficiency which is how well your
hydrocyclones are working.)
2. mill grinding efficiency, affected by the mill operating conditions
3. ore grindability, a disturbance that operators can’t control but that can affect the
other two efficiencies.
Objectives
• To be able to calculate each of these
efficiencies for a ball mill circuit.
• To combine them to obtain overall circuit
efficiency.
• To judge if these are “good or bad”.
• Optimizing strategies for particular situations.

Slide 5

The calculations needed for Functional Performance are pretty simple, and the
necessary data is already collected is most mill surveys:
●
the power draw of the grinding mill, corrected to the mill shell;
●
a circuit mass balance by size fraction gives both the solids throughput of a mill
and the particle size distribution of the mill feed and product; and
●
a Bond ball mill work index test on a sample of ore.
The efficiency values that come out of the analysis need some experience to interpret,
but AGD suggests a SABC circuit ball mill functioning well should observe:
●
classification system efficiency (CSE) 70% or better ;
●
grinding efficiency 80% or better.
Observed efficiency less than these benchmarks invites future optimization.
Definitions
• Coarse are particles that need to be ground
(too big).
• Fines are particles that are the correct size
(need to exit the grinding circuit).
• The circuit target P80 defines the split between
coarse and fine.

Slide 6

These are pretty simple definitions. “Coarse” are particles that need grinding and
“Fines” are particles that don’t.
The only potential confusion here is that “fines” can also refer to particles so small
that they flow with the water in a dewatering system – these tend to be –45 µm sizes.
These “ultrafines” still fit the Functional Performance definition, and they also tend to
follow the water phase in the hydrocyclone underflow.
Observation – circulating load
• Higher circulating load generally gives better
energy efficiency in grinding (goes back to the
time of Taggart, early 1900’s)
• Caused by lower retention times in the mill
– the moment a coarse particle breaks into fine
particles, you want those particles to exit the mill
and be classified.

Slide 7

It has long been known that milling circuits operating at higher circulating loads
operate at higher energy efficiency. Metcom’s training materials show a chart from
Gaudin’s 1939 book “Principles of Mineral Dressing” that demonstrate the mill
efficiency (-105 µm material produced) increases with circulating load, going
somewhat asymptotic beyond 500%. Some large SABC plants can’t operate at such
high circulating loads, largely due to the capital costs associated with pumping and
classification.
AGD has interpreted a small number of surveys for very large copper mines (in BC &
Chile) and finds the maximum classification efficiency of 71%–74% happens in ball
mills operating between 550% and 650% circulating load. Circuits operating below
200% circulating load rarely approach 70% classification efficiency.
Circulating load
• Higher circulating load means less retention
time in the mill
– less opportunity for “fine” particle to overgrind.
– more rapid chance for fine particles to classify.
• Removal of fines means more grinding energy
is focused on the coarse.
– Big benefit – the mill is grinding coarse particles!

Slide 8

Two concepts here, the first is the simplest: only grind coarse material. This means
the mill should contain more coarse material and less fine material.
The second concept is a bit more nuanced: the moment that a coarse particle breaks
into several smaller “fine” particles, those particles should exit the mill as quickly as
possible and report for classification. A higher “superficial velocity” of particles
down the mill (ergo, a lower retention time) will more rapidly flush the finished fine
product out of the mill, hopefully before those fine particle absorb more grinding
energy and over-grind.
So a higher circulating load means less over-grinding of fine particles and more
opportunities for classification.
Classification system efficiency
• McIvor defines CSE (classification system
efficiency) as the fraction of coarse particles in
the mill charge.
– more fine particles means poor efficiency
– more coarse particles means good efficiency
• Easy to measure, just sample the mill feed and
product for % retained at the P80 size.
– efficiency = average(%retained in feed & product)
Slide 9

Another really simple definition: the classification system efficiency is the fraction of
the particles inside the mill that are coarser than your desired product size (212 µm in
our case).
Measure the PSD of the mill feed and the mill product, read off the %retained of this
desired size in both streams, and assume that the mill contents will be the average of
the feed & product.
So the average of the %+212 µm in the mill feed and the %+212 µm in the mill
product is your circuit’s classification system efficiency (CSE).

Be aware that “classification efficiency” (without the “system”) is a different

efficiency. Classification efficiency describes how well your classifier (the
hydrocyclone) is classifying. Function Performance only describes CSE.
Classification system efficiency
• The less fines in the mill feed and product
means higher efficiency
– less opportunity for overgrinding
• Classification dictates %retained in mill feed
– fines carried with interstitial water
• Run hydrocyclone underflow as dense as
possible.

Slide 10

The classification system efficiency is affected by a couple of factors:

●
circulating load (which we already discussed), and
●
hydrocyclone performance (classification efficiency without the ‘system’).
The hydrocyclone performance determines how much fine product is mis-classified
into the underflow, thereby returning to the mill feed. The geometry of hydrocyclones
will affect the split size and separating efficiency of near-size material; tweaking the
geometry and operating pressure of hydrocyclones can optimize the amount of near-
size misclassification.
Ultrafines are carried along with the water in the hydrocyclone underflow.
Minimizing the %solids in the hydrocyclone underflow is the best way to control this
sort of fine particle in the mill feed.
Production rate of fines
• t/h of fines = %passing P80 × t/h
• Mill production of new fines, as t/h, is
(%pass P80 in disch – %pass P80 in feed) × t/h

• Allows you to assess if your mill circuit can

keep up with grinding requirement
– if you need to make 1000 t/h of fines for flotation,
but are only making 800 t/h, then problem!

Slide 11

A ball mill of a particular geometry and with a particular set of grinding conditions is
only capable of making a certain amount of fines. If an operator pushes the mill
throughput beyond the rate of fines production, then the circuit will build up coarse
material until the coarse starts appearing in the hydrocyclone overflow.
Knowing the production rate of fines in your ball mills, when combined with the
quantity of fines in the SAG mill discharge, will tell you the maximum amount of
throughput the circuit can handle before the flotation feed begins to get too coarse.
Functional Performance calculates this production rate, which is good to know.
If you want higher throughput, there are ways to nudge the fines production rate
higher. The first step in solving the problem (of limited fines production) is knowing
that you have a problem.
Production rate of fines
• Another way to calculate:
– new fines = mill kW × CSE × grinding rate
• Grinding rate is the energy going into making
new fines
– McIvor defines it as kg of new fines per kWh.
– possible to compare this to a lab ball mill Wi test.

Slide 12

So how does an operator assess if the fines production rate of the grinding circuit is
good or bad? Is there scope to improve the grinding efficiency, or does the feed rate
need to be lowered or the P80 target relaxed?
Functional Performance has a way to look a bit deeper into the production rate of fines
by working out the grinding specific energy consumed to make those fines and then
comparing that specific energy to a standard laboratory test. If the grinding circuit
specific energy consumption per amount of fines is similar to the laboratory, then
there isn’t much scope to improve the grinding efficiency. The operator is stuck with
“Plan B”, either reducing throughput or increasing the P 80 target. If the lab obtains
significantly better specific energy per unit of fines production, then the operator has
some hope of improving the operation of the grinding circuit.
Example survey
• Ball mill in copper porphyry SABC-A circuit
– BM feed 22.4% –212 µm (77.6% retained)
– BM product 51.4% –212 µm (48.6% retained)
– CSE = avg(77.6,48.6) = 63%
• Dry solids feed rate to ball mill = 1,723 t/h
• New fines = 1,723 × (0.514-0.224) = 500 t/h
• Mill power draw at mill shell = 12,708 kW

Slide 13

The ball mill feed (hydrocyclone underflow) and BM product samples were screened
and the data passed through a MolyCop Tools mass balance calculation. Then the
balanced particle size distributions are queried to find the amount of “product size”
212 µm in each stream.
The classification system efficiency is a measure of how much product size material is
in the mill and at risk of over-grinding. Normally want this efficiency to be greater
than 70%.
The mass balance calculation also predicts the ball mill was being fed 1,723 t/h of dry
solids (the hydrocyclone underflow solids flow). Doing the math predicts that BM is
creating 500 t/h of new -212 µm “product”.
Example survey
• BM new fines = 500 t/h; 12,708 kW;
63% classification system efficiency
– grinding rate, new fines per kW consumed
= 500 t/h ÷ 12,708 kW ÷ 0.63 c.eff = 62 kg/kWh
• Is this good or bad? Compare to Bond BM
test done during survey, 1.26 g/rev.
– use Levin B=1.45×10-5 kWh/rev, 1.26 g/rev = 87 kg/kWh
• BM grinding efficiency is 62/87 = 72%.
Slide 14

The grinding rate is a combination of the mill’s grinding efficiency and the ore
hardness. Using a standard laboratory test, the Bond ball mill work index
determination, gives a consistent measurement of the ore hardness, so the mill’s
grinding efficiency can be determined.
McIvor & Metcom treat the work index differently: they simultaneously multiply &
divide by the g/rev; AGD thinks this method using the Levin B parameter is more
useful and easier to understand.
The Levin B parameter is measured when doing a ball mill work index determination.
The parameter is the energy consumed per revolution of the laboratory mill, in units of
kWh/rev. More on the B parameter later.
Calculating the grinding efficiency with the B parameter will give different results
that using Metcom’s approach, so the 72% efficiency measured above is not directly
comparable to their database. You should build their own database of grinding
efficiency as you conduct surveys, then can interrogate the database to find the best
mill operating conditions.
Levin B parameter
• Measured during a
Bond BM test
(method published
by Levin, 1989)
– kWh consumed per
lab BM revolution
(1.45×10-5)
Source: Josefin & Doll, 2018

Slide 15

Levin was originally looking for a way to measure the properties of regrind circuits
using a Bond-type laboratory ball mill. Bond’s model with its fixed –0.5 exponent
usually doesn’t work for regrinding, so Levin needed a way to measure the exponent
experimentally for each ore.
On the way to achieving his goal, Levin deduced the amount of energy delivered to
the ore in each revolution of the lab mill, a value he labelled “B” and he tabulated for
a series of different ores that he experimented with. The equation is:
4.9×10−3×G 0.18
B= 0.23
P100 (100−U )
where, G is the grams per revolution in the test, g/rev
P100 is the closing screen size of the test, µm
U is the percent of –P100 undersize in the –6# feed
What affects ball mill
efficiency?

Slide 16

The first chart showing the effect of media size on breakage rates – it is from a book
chapter by McIvor (1998) who credits it to Lo and Herbst, 1986. It shows how the
ball size affects the rock breakage inside a ball mill – too large a ball and the mill is
inefficient is fine sizes, too small a ball and the mill can’t break the coarsest size.
Mixing different ball sizes can help a bit, see Metcom’s CMP 2018 paper for some
experimental results of mixing ball sizes.
The second chart is from another paper by McIvor (SME 2005) that shows the effect
on grinding efficiency (as kWh/(g/rev)) as a function of the mill percent solids (%vol).
Higher %solids gives better grinding efficiency. This can been explained by the
higher viscosity pulp adhering to the balls and then getting pinched and broken as the
ball charge moves. A more fluid pulp (lower %solids) means the particles can move
out of the way (avoiding breakage) when two balls come together.
Mass balancing
• Some survey
samples are better
than others.
• Perform matrix-type
mass balancing of all
sizes and streams.

Slide 17

The accuracy of calculations like Functional performance will improve when the
survey data is “cleaned up” using a size-by-size mass balance. The plot above shows
the example survey around ball mill hydrocyclones that was fitted using the
‘CycloBal’ calculation in MolyCop Tools. The “weight” factor assigned to the
cyclone overflow and circuit feed samples was higher than the weight assigned to the
cyclone feed (feed sample is much harder to get and didn’t represent the whole time
interval).
The balance calculation made negligible adjustments to the PSD of the hydrocyclone
overflow and underflow (blue & black), but a significant adjustment was made to the
feed (red).
Case studies
• McIvor et al, CMP 2017

Slide 18

The case study Metcom provided to CMP 2017 is for an unnamed plant operating a
small ball mill circuit. A target product size of 106 µm is used in the Functional
performance calculations.
●
Classification system efficiency = 65.6% (430% circulating load)
●
Amount of new 106 µm production = 126.6 t/h × (0.762–0.286) = 60 t/h
●
Mill grinding rate is 60.3 t/h / (955 kW × 0.656) = 96.2 kg/kWh
●
Ball mill test gives 2.07 g/rev, assume Levin B = 160×10 -7 kWh/rev
●
Laboratory grinding rate: 2.07 g/rev ÷ 160×10 -7 kWh/rev=129.4 kg/kWh
●
Grinding efficiency = 96.2 ÷ 129.4 = 74%
Note the diagram above is from Metcom’s new “Streamline” software which has
functional performance metrics built into it.
Large SABC-B plant
• A large Chilean SAG plant operates SABC-B
with crushed pebbles sent to ball mills.
– Classification sys efficiency 71%, Circ. load 580%
– BM specific energy consumption 11.5 kWh/t
– plant grinding rate 75 kg/kWh
– laboratory grinding rate 140 kg/kWh
– BM grinding efficiency 54%

Slide 19

AGD did an analysis of a survey done by others at a large Chilean copper mine that
operates an SABC-B circuit. We’ve seen earlier that ball mills aren’t great at grinding
really coarse material unless they have large balls, but large balls aren’t much good
for making a 106 µm product.
The survey wasn’t intended for Functional Performance (it was geared towards JK
SimMet), so some of the parameters require educated guesswork, especially the Bond
work index lab test (2.24 g/rev, Levin B: 160×10 -7). The circuit wasn’t very stable
either, and the mass balance needed quite a bit of forcing.
In spite of these warnings, it is reasonable to ask “is there a grinding efficiency
penalty to milling crushed pebbles?” The answer appears to be “yes”, the ball mill
circuit grinding efficiency here is a lot less than we’ve seen in the other examples.
Selbaïe, grinding media test
• Grinding media A • Grinding media B
– class eff: 71% – class eff: 72%
– ore BWi: 11.8 – ore BWi: 12.7
metric, 2.31 g/rev metric, 1.98 g/rev
– OWi: 11.7 (Wi – OWi: 11.3 (Wi
efficiency: 101%) efficiency: 112%)
– Grinding eff: 38.8 – Grinding eff: 46.5
(kg/kWh)/(g/rev) (kg/kWh)/(g/rev)

Slide 20

Functional Performance is useful for running experimental tests because it allows the
ore hardness variation to be removed from the analysis. This example from McIvor
(2006) shows a couple of metrics that don’t depend on ore hardness, the ratio of the
operating work index to the laboratory ball mill work index and the grinding
efficiency using Metcom’s preferred metric, (kg/kWh)/(g/rev).
Both metrics tell the same story, that grinding media B (which is obviously from
MolyCop ☺) works better than media A. Because the ore was harder during the
survey for media B, the throughput and specific energy of the survey did not reveal
the higher efficiency. Only by untangling the ore hardness using Functional
Performance could the real benefit of media type B be observed.

Operating work index (OWi or WiO) is calculating a work index from a plant survey
where E, P80 and F80 are known:
OWi = (Esurvey/10) (P80-½ - F80-½)-1
Hydrocyclone models
• Not really Functional Performance, but this is
a reasonable place to discuss hydrocyclones.
• Several models exist:
– Arturburn (Krebs), simplest
– Lynch & Rao
– Plitt and CIMM (MolyCop Tools)
– Nageswararao (JK SimMet)

Slide 21

The typical list of parameters that models consider are:

●
hydrocyclone body diameter, vortex finder diameter & apex diameter,
●
length of the hydrocyclone body, cone angle,
●
%solids of feed, feed rate, feed pressure,
●
and some include terms like liquid viscosity.
Some of these parameters are related to hydrocyclone geometry and can’t change
rapidly as part of a control strategy. The parameters that can rapidly change tend to
all be interrelated, making it hard to tweak (for example) the feed rate without also
affecting the pressure and the %solids.
Something is missing from many of these models: the density (or mass) of the
particles being classified. Dense-medium hydrocyclones separate dense material (like
rocks) from non-dense material (like coal); these density effects aren’t commonly
built into the models above.
Hydrocyclone Corrected d50
• Very fine particles are
not classified; they
“follow the water” and
bypass to the u/f
• Modelling
hydrocyclones requires
removing these fines
from the equations

Slide 22 Krebs Engineering, Arterburn model

Most hydrocyclone models work on the basis of the 50% passing size, usually referred
to as d50, which describes what you see in a real plant. But hydrocyclone models
generally start with a hypothetical “corrected” size distribution having the form:
●
d50c = f (geometry, feed, fudge factors)
where the models is always working with an idealized situation of fines not bypassing
the classification effects and ending up in the underflow. The real d 50 is then re-
created by working out the water recovery to overflow and underflow, then using the
same recovery to allocate the fines to overflow and underflow.
Procedure is to simulate the separation of the coarser material using the d 50c equation,
then simulate the separation of the fine material using the water recovery, and finally
adding the two together to get the “uncorrected” actual PSD and recovery predictions.
MolyCop Tools example

Slide 23 Parameters (fudge factors) appear here

The MolyCop Tools package has some excellent tool for dealing with hydrocyclone
surveys and predicting the parameters for modelling hydrocyclones.
The “CycloBal” spreadsheet takes survey information and both balances it (to give
you an actual mass balance) and then will predict the classifier parameters (if you
enter the hydrocyclone details in the appropriate place).
MolyCop Tools example,
simulate

Slide 24 Parameters (fudge factors) go here

The classifier parameters can now be copied to a different MolyCop Tools spreadsheet
where different hydrocyclone conditions can be simulated. The model will predict
effect of changes to the feed size distribution, the hydrocyclone geometry or the feed
conditions.
The calculation output is a prediction of the solids and water split around the
hydrocyclone and particle size distributions for the underflow and overflow streams.
The calculation probably requires iteration until the desired target overflow (product)
size is obtained.
References
• Josefin, Y. & Doll, A.G. (2018) Correction of Bond Ball Mill Work Index Test for
Closing Mesh Size, Proceedings of Procemin·GEOMET 2018, Santiago, Chile.
• Levin, J. (1989) Observation on the Bond standard grindability test, and a proposal
for a standard grindability test for fine materials, J. S. Afr. Inst. Min. Metall., Vol
80, №1, Jan 1989, 13–21.
• McIvor, R.E. (1998) The Effect of Media Sizing on Ball Milling Efficiency,
http://www.onemine.org/document/abstract.cfm?docid=2033
• McIvor, R.E. (2005) Industrial validation of the functional performance equation –
a breakthrough tool for improving plant grinding performance, Proceedings of the
SME Annual Meeting, Salt Lake City, USA, Preprint 05-31.

Slide 25
References
• McIvor, R.E. (2006) Ball Mill Circuit Models for Improving Plant Performance,
Advances in Comminution, SME, Ed. S.K. Kawatra.
• McIvor, R.E., Bartholomew, K.M., Arafat, O.M., & Finch, J.A. (2017) Use of
Functional Performance Models to Increase Plant Grinding Efficiency, Proceedings
of the 49th Annual Canadian Mineral Processors Operators Conference
• Metcom Technologies (2013) Workshop on Practical Tools for Maximizing Plant
Grinding Efficiency, short course presented at 2013 CMP Annual Meeting, Ottawa,
Canada.

Slide 26