Note: Descriptions are shown in the official language in which they were submitted.
CA 2720080 2017-05-16
1 -
MEASUREMENT OF BULK DENSITY OF THE PAYLOAD
IN A DRAGLINE BUCKET
[0001] This invention relates to the measurement of bulk density of the
payload in a dragline
bucket and has been devised particularly though not solely for assessing the
dig and blast
performance of overburden removal in an open-cut mine.
[0002] Large electric draglines are typically used in open-cut mining to
remove overburden
after blasting operations and to shape the configuration of the open-cut pit.
[0003] The requirement for any dragline is to move the largest amount of
material per unit
time, typically measured in tonnes per hour. High productivity achieved at the
cost of high or
undesirable loads on the dragline will generate increased maintenance costs
and downtime so
it is therefore important not to overload the bucket of a dragline in order to
increase
productivity. Research has indicated that the bulk density of blasted material
in a dragline pit
can vary greatly depending on blast performance, particularly in throw blasts.
This variation
has a significant effect on bucket size required to achieve desired or optimal
payload (thus
rated suspended load) as well as the digability of the material.
[0004] It is desirable to provide accurate estimates of the bulk density in
order to provide the
benefits of reliable assessment of dig and blast performance, improved bucket
size selection to
achieve consistent suspended load targets, and decreased production costs by
reduced
dragline damage and improved productivity through reduced probability of
bucket overloads.
[0005] Although work has been done in the past in determining material density
in other
open pit mining situations such as in excavator buckets or in haul trucks, it
is extremely difficult
to provide real-time density determination in a dragline bucket due to the
difficulty in
determining the accurate bucket pose estimation, payload extraction, and
filtering. The bucket
is attached to free moving ropes and thus the dynamics of the bucket at any
point in time are
unknown.
[0006) The bulk density of the payload material in the bucket of a dragline
is typically
determined by measuring payload weight and dividing that weight by payload
volume.
Determination of payload weight is reasonably well known and able to be
determined from
proprietary products which measure rope lengths and motor currents to
determine the load on
the dragline hoist ropes at any point in time and hence enable calculation of
the weight of the
CA 2720080 2017-05-16
- 2 -
payload in the dragline bucket. The main objective of the present invention is
to provide a
method of accurately determining the volume of the payload in the bucket
during the carry
phase of the dragline dig and dump cycle in order to allow real-time
calculation of the bulk
density of the material in the dragline bucket.
[0007] Accordingly, the present invention provides a method of measuring
the bulk density
of the payload in a dragline bucket during dragline operation, comprising the
steps of scanning
a loaded dragline bucket during an operating cycle of the dragline to provide
mathematical
data relating to the volume of the loaded bucket, calculating the volume
enclosed by the
surfaces of the payload and the known base and side surfaces of the bucket
from the
mathematical data to give the payload volume, and dividing the payload volume
into the
payload weight to give the payload bulk density.
[0008] In particular embodiments, the process of scanning the loaded bucket
during an
operating cycle of the dragline occurs during the carry phase of the cycle,
between the lifting
phase and the dumping phase.
[0009] In particular embodiments, the process of scanning the loaded bucket
is performed
by moving the bucket during the carry phase through the beam of a suitable
scanner.
[0010] In particular embodiments, the suitable scanner comprises a laser
scanner.
[0011] In other embodiments, the suitable scanner comprises a radar
scanner.
[0012] In particular embodiments, the mathematical data is analysed to
screen out data
points originating from surfaces other than those of the bucket and the
payload.
[0013] In particular embodiments, the process of calculating the volume
enclosed by the
surface of the bucket and the known base and side surfaces of the bucket
includes analysing
the collected data to rebuild the bucket structure by estimating the bucket
motion between
scans to allow for bucket sway.
[0014] In particular embodiments, the process of calculating the volume of
the loaded
bucket includes the steps of collecting hoist and drag rope length data and
using that data to
determine the bucket displacement between each scan.
CA 2720080 2017-05-16
a
- 3 -
[0015] Alternatively, the step of determining the volume of the loaded
bucket includes
measuring the displacement between each scan as a function of bucket velocity
as it passes
through the scanner beam and using the displacement to rescale bucket points
in a direction
orthogonal to the scanner beam.
[0016] In other embodiments, the processes of calculating the volume
enclosed by the
surface of the payload and the known base and sides of the bucket include
determining the
pose of the loaded bucket in order to provide reference surface data and
enable known
features of the bucket to be deducted from the volume calculation.
[0017] In particular embodiments, the payload volume is determined using an
elevation map
representation.
[0018] Notwithstanding any other forms that may fall within its scope, one
preferred
embodiment of the invention will now be desdibed with reference to the
accompanying
drawings in which:
[0019] Figure 1 is a laser beam visualisation of the bucket passing through
the beam;
[0020] Figure 2 is a plot of a scan at the bucket;
[0021] Figure 3 is a plot of bucket sway over various scans;
[0022] Figure 4 is a plot of the x co-ordinates before and after sway
correction;
[0023] Figure 5 is a three dimensional plot of ICP model fitted to
reconstructed data points;
[0024] Figure 6 shows the results of scans taken in dragline bin;
[0025] Figure 7 is a perspective view of a dragline in use, showing the
location of hardware
used in the invention;
,
[0026] Figure 8 is a simulation of the bucket moving through the scan
plane;
[0027] Figure 9 is an elevation showing the sensors assembled on the Pan
Tilt Unit;
[0028] Figure 10 is a sensor assembly schematic;
CA 2720080 2017-05-16
. A
- 4 -
[0029] Figure 11 is a control room networking schematic;
[0030] Figure 12 shows a screen shot of the bucket visualiser, alongside a
video capture of
the corresponding bucket; and
[0031] Figure 13 is an elevation of a dragline showing ground truth static
scanning. In the
preferred form of the invention, a laser scanner 1 (Figure 9) is mounted at 2
on the boom 3 of a
large electric dragline 4 (Figure 7) in order to scan a loaded dragline bucket
5 as it passes
through the beam of the laser scanner during the carry phase in the cycle of
dragline
operation, between the lifting phase and the dumping phase.
[0032] Accurate scanning of the loaded bucket poses a number of problems,
exacerbated
by the fact that the bucket is suspended from the dragline boom by hoist ropes
6 which allow
degrees of movement of the loaded bucket during the carry phase, and also
because the
bucket and the scanner pass over varying terrain 7 during the carry phase as
the dragline
house rotates about its base.
[0033] These constraints and problems require a very difficult analysis as
is set out and
explained in the following section.
Bucket Detection
[0034] The foremost step to measuring the in-bucket payload volume is to
firstly scan and
identify the bucket. Bucket detection is critical to isolate the relevant data
from background
noise such as points from the terrain. Each point is assigned to a specific
class; terrain, bucket
or noise. Since we are only interested in the terrain and the bucket (namely
the payload) other
items such as hoist and drag ropes are discarded as noise.
[0035] This can be seen in Figure 1 where the beam from the laser at 8 casts a
shadow 9
on the ground 10, revealing the outline of the bucket at 11.
[0036] After major clusters are identified, the number of major clusters in
each scan is used
to determine the presence of the bucket. For dxample:-
[0037] The invention uses point clustering techniques to identify similar
points within the
scan to improve the performance of data classification and overcome. typical
thresholding
CA 2720080 2017-05-16
- 5
issues (see figure 2). The individual clusters are classified by their overall
shape an
dimensions as expected for bucket and terrain.
Bucket Reconstruction
[0038] Due to the typical swing motion of the dragline, the bucket can
exhibit extensive out-
of-plane motion referred to as bucket sway. This kind of motion produces an
artefact
resembling a wavy shaped bucket that is caused by the lengthy duration (of
approximately 2
seconds) for the bucket to pass through the beam. As a consequence the bucket
data from
each scan line is to some extent shifted with respect to the previous scan
line. This step
analyses the collected bucket data to rebuild the bucket structure by
estimated the bucket
motion between scans.
Sway Correction
[0039] The amount of bucket sway was measured by the translation of bucket
points
between scans. This was critical to determine the required transformation used
to recover the
actual bucket shape. The bucket points of each scan are translated by their
mean x
coordinate, which centre points about x=0 as shown in Figure 4..
[0040] Irregulatirties in the payload profile can cause significant changes
in the mean x
coordinate between scans as seen in Figure 3 by the outlying points. These
outlyers inflict a
rapid and incorrect change in the estimated motion of the bucket. The
estimated motion of the
bucket is smoothed by applying a either a polynomial or sine wave fit to the
mean x
coordinates of each scan.
Rescaling
[0041] The displacement between each scan is a function of the bucket velocity
as it passes
through the laser beam This displacement is used to rescale the bucket points
in the direction
orthogonal to the laser beam. Assuming the bucket passes through the beam at a
constant
velocity the displacement of the bucket between scans can be deduced as
follows:
I cos 5
Ay
n ¨1
[0042] Where / is the known length of the bucket, 6 is the carry angle of
the bucket
(determined by the rigging), and n is the number of scans taken of the bucket.
CA 2720080 2017-05-16
- 6 -
[0043] An interface to the hoist and drag rqpe lengths supplied through the
onboard DCS
monitor allowed for a more direct approach to evaluating the translation
between scans.
Between each scan the lengths of the extended hoist and drag ropes is used to
estimate the
Cartesian position of the bucket relative to the machine. This method is
capable of measuring
any change in velocity of the bucket during the scanning process. However, up-
to-date rope
length offsets are required to make this approach feasible. Generally, these,
are entered into
the dragline monitor software after each rope cropping, however this wasn't
made available on
the PLC interface. In practice, the bucket position as measured from the laser
was used to
estimate the offsets.
Pose Estimation
[0044] The pose of the bucket is required in later steps to filter out
known features of the
bucket in addition to providing a reference surface to calculate the volume of
the payload. Two
methods for determining the pose of the bucket have been investigated on the
scaled system,
with the second trialled on a full scale dragline.
[0045] The first method involves placing fovr reflectors at known locations
on the bucket
which are segmented from rest of the scan based on the intensity of the
returns. Laser retro-
reflector tape was used as it provides a high intensity reading and allows for
intensity based
segmentation. Often there are multiple returns per reflector and the localised
mean of these
returns are used to define the location of these reflectors. These points are
matched to the
reflector locations in the bucket frame and by using a Levenberg-Marquardt
numerical solver,
the pose of the bucket is computed. Problems with this method are that some
reflectors are
often occluded and for the full scale application would not be able to
withstand the harsh
environment. This option is commercially unfeasible any future dragline
buckets with this
system installed would need reflectors welded across their arch and rim. In
addition to this the
small amount of reference points used resulted in a large transformation error
that forms a
basis for the volume calculation error exceeding 10%. The second method of
using ICP was
chosen in an attempt to overcome the drawbacks of using reflectors. ICP better
fits a model
point set to the entire bucket point cloud as shown in Figure 5.
Payload Extraction
[0046] Payload points need to be segmented from other points on the bucket
such as the
bucket arch, spreader bar and jewellery while taking into account noisy
outliers. This process
is summarised by firstly removing known features of the bucket such as the
arch of the bucket.
CA 2720080 2017-05-16
- 7 -
Next, the algorithm is used to filter out noise and identify clusters
representing the payload.
Finally points are added to the payload in regions occluded by the sensor to
ensure full
coverage of the bucket surface.
Bucket Feature Filtering
[0047] Using the bucket pose information, particular features such as the
arch and rim of
the bucket can be removed as they are not part of the payload. These known
features are
stored in the form of a cylinder represented by two points (at the centre of
each circular face)
and a radius. The points are transformed into the sensor frame using the
bucket pose and any
data point enclosed by a feature cylinder is removed.
Cluster Density Segmentation
[0048] The payload points are characterised by a large regions of high
density within the
point cloud due to the surface being rather orthogonal to the ray produced by
the sensor. The
previous step of removing known features such as the arch and rim of the
bucket would also
reduce the point density in these regions. This leaves the payload points as
the largest high
density point clusters in the remaining sample set.
,
Addition of Occluded Points
[0049] Due to the effects of shadowing caused by the arch and spreader bar,
the outer
boundary of the bucket's payload is often occluded. This effectively reduces
the area covered
by the visible payload points and thus reduces the total sensed volume of the
points. By using
the pose estimation of the bucket we can assume that the payload forms a
continuous smooth
surface up to the inner edges of the bucket. Points near the transformed
bucket teeth are
added to the payload. This ensures that the payload volume is continuous from
the sensed
payload to the teeth after interpolation. Similarly points positioned on the
inner rear surface of
the bucket are added as this region is often occluded by the spreader bar.
Volume Measurement
[0050] With the payload points subdivided from the bucket, and the bucket's
pose known,
the payload volume can be measured. A relatively straightforward method for
representing a
surface is the height grid (or elevation map) representation. The grid is
constructed by dividing
up the area (in x, y plane) covered by the point cloud into uniformly sized
cells. The height
CA 2720080 2017-05-16
- 8 -
value at each cell is equal to the average z value of all points with x, y
values bounded by the
cell.
[0051] Shadows from the arch and spreader bar may cause some cells to contain
no points
and thus leave the height undefined. To overcome this, an interpolation method
is used to fill in
the missing data between the known cells, forming a convex shape in the x, y
projection of the
grid.
[0052] To calculate the volume of the payload, a reference surface of the
inside of an empty
bucket is required. This is found by transforming a pre-computed height grid
generated from a
bucket CAD model by the pose estimated from ICP. This height grid now
represents the
bottom surface of the payload against the base of the bucket.
[0053] The volume between the payload height grid and the reference height
grid is
computed by summing the height differences over the aligned cells multiplied
by the cell area.
Results of Pilot Studies
[0054] The simulated payload material has relatively uniform granule size
to ensure minimal
compaction when transferring from the measuring apparatus to the bucket. The
material is
measured independently before each scan and after with any discrepancy
averaged to ensure
an accurate ground truth measurement.
[0055] The velocity of each run and the material is slightly varied to test
the robustness of
the algorithms. When the bucket is swaying the variance of the volumes
slightly increases as
seen in Figure 6, but maintains minimal bias.
Static Runs Dynamic Runs All Runs
RMS Error 4.1% 5.6% 4.9%
Mean Error 0.8% 0.3% 0.5%
Standard Deviation 4.1% 5.8% 4.9%
Table 1: Result summary with error expressed as a percentage of bucket
capacity.
,
CA 2720080 2017-05-16
- 9 -
System Design
Overview
[0056] Figure 7 gives a brief overview of th'e system hardware. The system
is comprised of
two primary elements, the first being a number of sensors mounted on the boom
3, and the
second being equipment housed in the dragline's control room.
Scanner
[0057] The primary sensor of the system is the scanner, used to generate a
'point cloud"
image of the payload in the dragline bucket, as well as the surrounding
terrain. Two options
considered were a 94 GHZ FMCW radar and the Sick LD-MRS Laser, a commercial
off the
shelf scanning laser.
[0058] Table 2 compares key performance criteria of both sensors, including
the radar with
a proposed upgrade.
[0059] The LD-MRS Laser scanner was the sensor chosen for this project, as
simulations
showed it to be able to scan the dragline bucket payload with greater
accuracy. A description
of both sensors, as well as further details of the criteria applied for
selection between them is
given below.
Radar Upgraded Laser
Radar
Max Range (m) 300 300 200'
Scanning Frequency (Hz) 4.5 10 12.5
Angular Resolution (deg) 0.81 0.37 0.25
Range Accuracy (m) 0.5 0.5 0.1
Beam Width (deg) 1.49 0.89 0.8/0.082
Table 2: Scanner Performance Comparison (Notes: 1. The actual maximum range of
the laser is dependent
on the reflectivity of the surface being scanned. 2. Vertical/Horizontal beam
width).
Radar
[0060] The primary advantage of the 94GHz frequency modulated continuous-wave
(FMCW) radar is its immunity to adverse environmental conditions. As the
millimetre wave
,
CA 2720080 2017-05-16
- 10 -
beam can penetrate dust and water particles, it can produce images even under
zero visibility
conditions. However, critical weaknesses of the radar are its poor angular
resolution, range
accuracy, and scan speed.
Laser
= .
[0061] The Sick LD-MRS, formerly a product of IBEO is a scanning laser
designed for use
in the automotive industry. The laser offers far greater range accuracy and
angular resolution
than the radar. Additionally the LDMRS is capable of simultaneously producing
four scanning
planes and recording up to three echoes per transmitted pulse. These features
provide some
immunity to environmental conditions such as dust and rain, but not to the
same degree as the
radar.
Scanner Selection
[0062] The main criterion for the selection of a 3D Scanner was the ability
to accurately
calculate the volume of the dragline bucket payload. To compare the
performance of both
sensors accordingly, a simulation of the volume calculation was performed. The
simulation
approximates the expected sampled surface of a typical bucket moving through
the scan plane
at a rate of 3m/s (nominal velocity of bucket during swing cycle), using the
performance
statistics of Table 2. The simulation scenario is illustrated in Figure 8. The
results of this
simulation are shown in Table 3.
Upgraded'
; Radar Radar ,LD-MRS*
:dx (across width) 0.71 :0.32 10.22 1m
'dy (down length) :0.67 0.30 ;0.24
' Points across width 7 15 121 points/line
'Lines down length 9 21 ,26 ;lines '1
.Total points on bucket 163 :315 1546 !points
Simulated Volume Error# 110.18 i8.84 3.55 11
Table 3: Simulation Results
[0063] The simulation clearly shows the Sick LD-MRS Laser as the only sensor
expected to
produce a volume accurate to within the acceptable level of 5%. Furthermore,
as the upgraded
radar did not produce a markedly better results, it can be surmised that the
radar's poor range
accuracy was the limiting factor. Some caveats of the simulation that should
be noted are:
CA 2720080 2017-05-16
11 -
= The simulator does not include effect of: Shadowing, Bucket Dynamics or
segmenting
payload from bucket features, and
= The dragline bucket payload was approximated by a simple box.
Navigation Sensor
[0064] The navigation sensor chosen for tliis project was the Xsens MTi-G.
This sensor
combines a GPS receiver and Inertial Measurement Unit, allowing it to operate
with an
intermittent GPS signal. The sensor provides 6 degree-of-freedom position and
orientation,
with a position accuracy of 2.5m. While this accuracy does not compare
favourably to the real
time kinematic (RTK) navigation systems, it was deemed sufficient, as this
project's primary
objective of in- bucket volume measurement did not require highly accurate
navigation
sensing.
Sensor Assembly
[0065] The Sick LD-MRS Laser, along with an off the shelf IP camera 12, were
mounted to
a Directed Perception PTU-D100 Pan Tilt Unit (PTU) 13. Although the scanner
was held
stationary for the purpose of scanning the dragline bucket, the PTU allowed
fine tuning of the
scan plane angle on the fly, as well as the scanning of the terrain. In
anticipation of possible
dust issues, the mounted laser 1 was enclosed in a protective enclosure with a
compressed air
purge line 14. The laser 1, camera 12 and PTU 13 are shown fully assembled in
Figure 9, as
mounted at 2 on the dragline boom 3 (Figure 7).
[0066] The PTU, and all boom mounted sensors were connected to an Ethernet
network.
As the PTU and Navigation sensor only provided a serial interface, a converter
was necessary.
This network of sensors was then connected to the previously mentioned fibre
link. The serial
converter and the network switch were housed in fibre converter were housed in
a separate
boom mounted enclosure, along with the navigation sensor. A schematic outline
of the boom
mounted hardware is shown in Figure 10.
Supporting Hardware
[0067] In the dragline control room, a network of supporting devices for
purpose of
computation, telemetry and interfacing with the dragline PLC were connected to
the other end
CA 2720080 2017-05-16
- 12 -
of the fibre link. A schematic overview of this control room network is
provided in Figure 11,
with further details of the individual devices given in the sections below.
PLC Interface
[0068] The prediction of bucket movement during scanning, as well as the
calculation of
bulk density, necessitated the collection of data from the dragline's control
system. The
required data included rope lengths, rope'veldcities, and payload weights. For
this purpose,
Drives and Controls Services (DCS), the maintainer of the control system,
installed and
configured a Prosoft MVI56-MNETC PLC Module. This module would transmit the
required
data via the Modbus TCP/IP Protocol.
3G Cellular Router
[0069] To facilitate telemetry, a 3G cellular router was connected to the
control room
network. This router was connected to a high gain antenna mounted on the roof
of the dragline
operator's cab. The 3G router, used in conjunction with a dynamic DNS service
would allow
connection to the system over the internet.
Embedded PC
[0070] An embedded PC was connected to the control room network. The computer
would
allow secure remote access and run the project software.
System Software
[0071] The initial software for the scaled pilot trials was implemented in
MATLAB for
computing volumes from data collected in the 1:20 scaled dragline facility.
The algorithms were
later ported to C++ in time for the full scale installation to ensure real-
time performance.
Software drivers for the full scale hardware were extensively bench tested
before
commissioning. The real-time performance requirement of the software is that
the volume
computation will be completed before the third party dragline monitor reports
the payload
weight of the last bucket.
[0072] Figure 12 shows an illustration of how the payload mesh looks after
segmentation
and filtering. The raw 3d point cloud is overlaid to show that the spreader
bar is excluded from
CA 2720080 2017-05-16
- 13 -
payload mesh generation. Typically, 800 to 1000 bucket points are collected
across the bucket
(including rigging) of which generally 600 or more points are used to generate
the payload
mesh. The visualizer is capable of showing the irregularities of the payload
surface with rocks
17 extruding over the top of the payload and occasional gaps 18 in the payload
when the
material doesn't always fall to the rear of the bucket.
Ground Truth Data Collection Method
[0073] Figure 13 illustrates the collection of ground truth data. After the
bucket was pulled
out of each dig the operator rested the bucket on the pad in a position
directly under the laser
position 20 on the boom 21. The laser then takes four high-resolution sweeps
of the bucket
that produce 6000 to 8000 samples across the bucket surface. This is
approximately ten times
the number of samples collected dynamically as the bucket moves to the dump
zone.
Accuracy of Static Bucket Sweep
[0074] A static sweep of an empty bucket was carried out during the production
time as a
reference. Each individual sweep on the empty bucket were analysed with a
volume of -0.3
cubic metres or 0.56% of the rated bucket capacity. This is considered to be
an acceptable
measurement error for the measurement of the ground truth data set.
[0075] Similarly the loaded bucket sweep scans were individually analysed
and the median
sweep volume is used for each bucket.
,
,
