Sei sulla pagina 1di 10

On-Line Mill Charge Analysis for Throughput and

Liner Life Optimization


Peter Blanz, Ville Suontaka and Ari Rantala
Outotec, Finland

ABSTRACT
In many cases, grinding mill power draw, bearing pressure and other typically available
measurements are not ideal variables for optimal grinding process control. To address the need for a
more reliable state indicator and robust control, Outotec has developed a system for analyzing and
controlling a grinding mill’s volumetric charge. The analysis is based on a sensor that attaches
directly to a liner bolt, combining both strain gauges and accelerometers to accurately determine the
position and volume of the mill’s charge. The system also provides other vital information related to
the mill’s operation, such as media trajectory estimates and overall vibration levels. To maximize
reliability while minimizing the time needed for maintenance and re-mounting of the equipment
during relining operations, the system features continuous inductive charging and a quick mounting
mechanism that utilizes super-strong magnets and latches. The embedded multivariable mill
controller option provides new opportunities for reacting faster to common process disturbances,
such as ore hardness and size changes, by adjusting the mill’s ore feed, water feed and rotational
speed. The combination of a reliable mill state measurement and advanced control helps the whole
grinding circuit to achieve maximized throughput, increased availability and less liner damage, all
of which can easily be translated to significant economic gains. These benefits are demonstrated by
industrial case examples.

1
INTRODUCTION
The mill charge volume has a significant impact on grinding efficiency. For efficient grinding,
autogenous (AG) and semi-autogenous (SAG) mills need to be operated with the correct charge to
maximize their throughput while avoiding overloading. While accurately describing the breakage
dynamics in a AG/SAG mill is very complex, we can consider two mechanisms of grinding that are
the main contributors to ore breakage: impact grinding and abrasion grinding. If the charge volume
is too high, impact grinding is hindered. On the other hand, operating with too low a charge volume
leads to hindered abrasion breakage and puts the liners at risk. Therefore, there is an operating point
(in terms of volumetric charge) that will provide the best performance. Figure 1 demonstrates this
using estimated charge positions and outer grinding media trajectories.

Figure 1 Grinding efficiency at different volumetric charge values.

Since there has traditionally been no way to directly measure the volume of the charge, it has become
very popular to use the mill’s bearing pressure as a state indicator in process control. However, there
are several problems in this method. Firstly, the bearing pressure doesn’t always correlate perfectly
with the mill’s mass, since there are other factors (e.g. ambient temperature) that can have a huge
effect on the pressure measured from the lubrication system. And even if the mass was known with
absolute certainty, the mill can have significantly different charge volumes at the same mass. The
mill’s mass changes when the liners wear, the ore’s specific gravity changes, the ball charge (in SAG
mills) changes or when the slurry density changes. With all these uncertainties, we must conclude
that bearing pressure is not an ideal state indicator for process control, and an accurate volumetric
charge measurement system is needed to better control the grinding process. Over the years, the
industry has seen several on-shell (e.g. Tano, Pålsson & Persson, 2003 and Campbell et al., 2001) and
off-shell (e.g. Pontt, 2004 and Pax, 2001) methods that have attempted to address this need.

2
METHODOLOGY
MillSense works by identifying the position of the charge inside the mill, which is accomplished by
continuously measuring the vibration and strain of a liner bolt. The following sections describe both
the analysis equipment and the mathematical principle.

System Components
The MillSense system consists of an on-bolt sensor unit, an on-shell transmitter, a radio receiver, a
connection cabinet, an inductive charging station and an analysis computer. Figure 2 shows all the
field equipment and their locations on the shell and near the mill.

Figure 2 MillSense field equipment.

On-Shell Transmitter
The on-shell transmitter gathers data from the on-bolt sensors (up to 3) via a digital bus and transmits
this data to the analysis system. The transmitter is attached to the mill’s shell using a base plate that
uses super strong neodymium magnets as a secure mounting mechanism, with a layer of double-
sided tape on the bottom of each magnet to provide friction. The combination of magnets and tapes
provides approximately 7200 N of force in the normal direction and 4000 N in the tangential direction.
The amount of force is massive in comparison to the transmitter weight, making it impossible for the
transmitter to become detached from the mill. The magnet-based mounting mechanism is ideal,
because it is suitable for all mills and doesn’t hinder the work of the relining crew.

3
On-Bolt Sensor
The on-bolt sensor features a circular strain gauge, designed to accurately measure the liner bolt’s
tension. The sensor also has a three-dimensional accelerometer that serves two purposes; reliably
determining the mill revolution’s start and end times and measuring the bolt’s vibration. This is the
only component that needs to be handled during mill relining, and the analysis software has been
designed in a way that allows the sensor to be reattached in any orientation. These features ensure
that the system stays functional after the relining process.

Inductive Charging Station


The on-shell components need power to function, but it is not convenient to use unreliable
pendulums or batteries in an instrument that needs to remain reliable in all conditions. That is why
we chose to use an inductive charging station to power the on-shell electronics. This has the added
benefit of providing us with a means of shutting down or resetting the on-shell system components,
even when the mill is running. For quick testing purposes, the on-shell components can also use
battery power.

Receiver
The radio receiver collects data packets from the transmitter and sends them to the analysis computer
by TCP/IP.

Analysis Computer
The analysis computer gathers the data packets into continuous signals, determines where a
revolution starts and ends and finally calculates the charge position from the revolution’s profiles.
The computer can – based on the customer’s preference – be an industrial PC that is placed directly
in the cabinet, a separate server computer or even a virtual computer.

Charge Volume Measurement


MillSense can detect the mill’s toe angle either from the revolution’s bolt strain or bolt vibration
profile. When considering the strain profile, the toe can be identified by finding the point where the
signal changes most rapidly. When the vibration signal is used, the toe can be found at or near the
maximum. Figure 3 illustrates the vibration and strain profiles for a single mill revolution.

4
Figure 3 Strain (i.e. force) and vibration profiles of a single mill revolution.

In Figure 3, the vertical lines mark the locations of the detected toe (first half of the graph) and
shoulder (second half of the graph) angles for both signals. In steel-lined mills, the vibration signal
usually produces a more consistent toe angle measurement, while the strain signal tends to be
extremely clear in rubber-lined mills. Once the toe angle is known, the mill charge is calculated using
the Julius Kruttschnitt model (Napier-Munn et al., 1999) for charge shape:
𝝅
𝝓𝒕𝒐𝒆 = 𝟐. 𝟓𝟑𝟎𝟕(𝟏. 𝟐𝟕𝟗𝟔 − 𝑱)(𝟏 − 𝒆−𝟏𝟗.𝟒𝟐(𝒏𝒄−𝒏𝒑) ) + (1)
𝟐

𝒏𝒄 = 𝟎. 𝟑𝟓(𝟑. 𝟑𝟔𝟒 − 𝑱) (2)

where
𝜙𝑡𝑜𝑒 = 𝑐ℎ𝑎𝑟𝑔𝑒 𝑡𝑜𝑒 𝑎𝑛𝑔𝑙𝑒 (𝑟𝑎𝑑)
𝐽 = 𝑣𝑜𝑙𝑢𝑚𝑒𝑡𝑟𝑖𝑐 𝑐ℎ𝑎𝑟𝑔𝑒(0 … 1)
𝑛𝑐 = 𝑐𝑜𝑚𝑝𝑙𝑒𝑡𝑒 𝑐𝑒𝑛𝑡𝑟𝑖𝑓𝑢𝑔𝑖𝑛𝑔 𝑠𝑝𝑒𝑒𝑑(𝑝𝑟𝑜𝑝𝑜𝑟𝑡𝑖𝑜𝑛 𝑜𝑓 𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 𝑠𝑝𝑒𝑒𝑑)
𝑛𝑝 = 𝑚𝑖𝑙𝑙 𝑠𝑝𝑒𝑒𝑑 (𝑝𝑟𝑜𝑝𝑜𝑟𝑡𝑖𝑜𝑛 𝑜𝑓 𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 𝑠𝑝𝑒𝑒𝑑)

The volumetric charge (J) cannot be directly solved from the equations, but a moderate number of
iteration steps or a lookup table yield good results in practical applications. While the shoulder angle
of the charge is not needed for the calculation of volumetric charge, it can still provide valuable
insight on the mill’s dynamics to the operators and metallurgists. The shoulder angle can be detected
most reliably from the strain profile.

Additional Measurements
While the volumetric charge measurement is the system’s primary output, additional state indicators
are included to provide more insight for the operators and metallurgists. The accelerometers are also
able to provide a measurement of the mill’s overall vibration level that can help in, for example,

5
identifying damaged equipment. If the profiles of the lifters are known, outer media trajectories can
be calculated using Powell’s trajectory model (Powell, 1991).

Charge-Based Mill Control


To get the full benefit out of the accurate mill charge measurement, the primary mill’s fresh ore feed
should be controlled based on this measurement. While this type of controller can also be
implemented with the tools typically available in modern DCS systems, an advanced process control
(APC) platform provides better tools (e.g. advanced PID controllers) for this application. The
standard ore feed control in Outotec’s APC platform (ACT) is implemented with an advanced PID
block that supports dead-time compensation. Dead-time compensation is critical, since some
processes can have a very long conveyor belt between the controlled feeder and the mill. If the
standard implementation is not suitable for some process, the platform also offers an extensive APC
toolkit, including rule-based logic, fuzzy logic and model predictive control.

RESULTS AND DISCUSSION


In this section, we present a set of validation measurements and a small study on the relationship
between volumetric charge and mill throughput.

Comparison with Laser Scanner Measurements


In 2016, MillSense was installed in a process that features a large steel-lined SAG mill. The plant
personnel regularly use a laser scanner during mill stoppages to get (along with crucial liner
information) accurate measurements of the mill’s charge, which made it possible for us to compare
the 10-minute averages of MillSense and bearing pressure measurements before each shutdown to
the laser-scanned charge volumes. Figure 4 shows the bearing pressure’s correlation with the laser-
scanned volume.
50.00
45.00
40.00 y = 0.011x - 37.404
Scanned charge (%)

35.00 R² = 0.3594
30.00
25.00
20.00
15.00
10.00
5.00
0.00
4500.00 4700.00 4900.00 5100.00 5300.00 5500.00 5700.00 5900.00 6100.00 6300.00 6500.00
Discharge pressure (kPa)

Figure 4 Pressure vs. scanned charge.

The pressure seems to have a weak correlation with the actual charge volume. This supports the
assertions made in the beginning of the article. Figure 5 shows the MillSense charge measurement’s
(as calculated by the JK model) correlation with the scanned volume.

6
50.00
45.00
40.00 y = 1.0848x - 13.71
Scanned charge (%)
R² = 0.6016
35.00
30.00
25.00
20.00
15.00
10.00
5.00
0.00
25.00 30.00 35.00 40.00 45.00 50.00
Charge, JK model (%)

Figure 5 JK-modelled charge vs. scanned charge.

The correlation is clearly better than the pressure’s, but the JK charge model is based on a big set of
data collected from numerous mills (Napier-Munn et al., 1999), and therefore the model isn’t ideal
for any individual mill. To demonstrate this, let’s examine the raw toe angle measurement, as shown
in Figure 6.
50.00
45.00
40.00
Scanned charge (%)

35.00 y = -0.6503x + 168.42


30.00 R² = 0.7049
25.00
20.00
15.00
10.00
5.00
0.00
200.00 205.00 210.00 215.00 220.00 225.00 230.00 235.00 240.00
Toe angle(°)

Figure 6 Toe angle vs. scanned charge.

As we can see, the direct correlation with toe angle is better than any of the previous measurements.
Since the toe angle is not only affected by the charge volume but also by the mill speed, we should
make a small correction to each toe angle measurement. With the correction, the equation for charge
becomes:
𝑱 = 𝒌𝒕𝒐𝒆 ⋅ (𝝓𝒕𝒐𝒆 + 𝒌𝒄𝒐𝒓𝒓 ⋅ (𝟎. 𝟕𝟓 − 𝒏𝒑 )) + 𝒄𝒕𝒐𝒆 (3)

where
𝑘𝑡𝑜𝑒 = 𝑡𝑜𝑒 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡, 𝑐𝑡𝑜𝑒 = 𝑡𝑜𝑒 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡, 𝑘𝑐𝑜𝑟𝑟 = 𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑖𝑜𝑛 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡

The correction coefficient can be fitted by iteratively searching for a value that maximizes the
correlation between the calculated charge and the scanned charge. In each iteration, the toe coefficient

7
and the toe constant can be found with linear least-squares fitting. The results calculated with the
fitted charge model are shown in Figure 7.
50.00
45.00
40.00
Scanned charge (%)

35.00 y=x
30.00 R² = 0.7156
25.00
20.00
15.00
10.00
5.00
0.00
10.00 15.00 20.00 25.00 30.00 35.00 40.00
Charge, OT model (%)

Figure 7 OT-modelled charge vs. scanned charge.

This method of calculating charge appears to be the most accurate one, but it should be noted that a
model cannot be reliably validated with the same data set that was used for fitting the model.
Therefore, more scan results would be needed to verify the method.

Performance During a Grind-Out


To verify MillSense’s ability to track the charge volume in a wide range of operating conditions, we
have visualized the results from a full SAG mill grind-out.
40 300

35
280
30
260
Toe angle (°)

25
Charge (%)

20 240

15
220
10
200
5

0 180
8:30:00 9:00:00 9:30:00 10:00:00 10:30:00 11:00:00 11:30:00

Charge Toe angle

Figure 8 MillSense measurement during a grind-out.

As we can see from Figure 8, MillSense successfully tracks the charge all the way from 35 to 5 %.

8
Potential Throughput Improvement
To study the relationship between volumetric charge and mill throughput, we took data from a two-
month period, removed the data from abnormal process conditions, clustered the remaining data
based on the volumetric charge and compared the average throughput values of the clusters. To get
a large sample count for each cluster while still being able to show differences between the clusters,
a cluster width of one percent (volumetric charge) was selected. Clusters with less than 100 samples
(i.e. very low and very high mill charges) were discarded from the analysis. Furthermore, the data
sets of ‘hard’ and ‘soft’ ore (as determined by the power consumption per tonne) were considered
separately. The results are illustrated in Figure 9.
100

98
Cluster average throughput (% of max)

96

94

92

90

88

86

84

82
27 29 31 33 35 37 39 41
Cluster average charge (%)

Both ore types Soft ore Hard ore

Figure 9 Clustered throughput data with different ore types.

These results indicate that there are significant performance variations between different charge
levels and ore types. The next logical step int this process is to build a charge-based controller to
replace the existing pressure-based one. To prove the charge controller’s superiority, we suggest an
extensive period of tests, during which the plant personnell alternate between the two control
methods in a randomized way.

CONCLUSION
We have presented a method for volumetric charge measurements that is suitable for all known SAG
and AG mills. The instrument has been proven with steel, rubber and composite liners, in seven mills
globally, with many more installations coming. For the reasons outlined in this article, we assert that
volumetric charge measurement is the best state indicator for efficient grinding control and will gain
much more popularity during the coming years.

9
NOMENCLATURE
𝜙𝑡𝑜𝑒 charge toe angle (rad)
𝐽 volumetric charge (0…1)
𝑛𝑐 complete centrifuging speed (proportion of critical speed)
𝑛𝑝 mill speed (proportion of critical speed)

𝑘𝑡𝑜𝑒 toe coefficient


𝑐𝑡𝑜𝑒 toe constant
𝑘𝑐𝑜𝑟𝑟 correction coefficient

REFERENCES
Pontt, J. (2004) ‘MONSAG: A New Monitoring System for Measuring the Load Filling of a SAG Mill’,
Minerals Engineering, vol. 17, pp. 1143 – 1148.
Powell, M.S. (1991) ‘The Effect of Liner Design on the Motion of the Outer Grinding Media in a Rotary
Mill’, International Journal of Mineral Processing, vol. 31, pp. 163 – 193.
Napier-Munn, T.J., Morrell, S., Morrison, R.D., Kojovic, T. (1999) Mineral Comminution Circuits: Their
Operation and Optimisation, Julius Kruttschnitt Mineral Research Centre, Queensland.
Pax, R. A. (2001) ‘Non Contact Acoustic Measurement of In-Mill Variables of SAG Mills’, Proceedings
of the International Conference on Autogenous and Semiautogenous Grinding Technology Vol. II, University
of British Columbia, Dept. of Mining and Mineral Process Engineering, Vancouver, pp. 386 – 393.
Tano, K.T., Pålsson, B.I., Persson, S.V. (2003) ‘Continuous Monitoring of a Tumbling Mill’,
Proceedings: XXII International Mineral Processing Congress, South African Institute of Mining and
Metallurgy, Cape Town, pp. 1514 – 1523.
Campbell, J., Spencer, S., Sutherland, D., Rowlands, T., Weller, K., Cleary, P., Hinde, A. (2001) ‘SAG
Mill Monitoring Using Surface Vibrations’, Proceedings of the International Conference on Autogenous
and Semiautogenous Grinding Technology Vol. II, University of British Columbia, Dept. of Mining and
Mineral Process Engineering, Vancouver, pp. 373 – 385.

10

Potrebbero piacerti anche