Copyright © (FAC Automation in Mining, Mineral a nd Me tal Processing, Helsinki , Finla nd , 1983
A COMPUTER SYSTEM FOR PHOTOMETRIC MINERAL SORTING I. Maenpaa, P. Malinen and R. Soderstrom Dy Part ek Ab , Pargas, Finland
Abstract. The sorting of minerals is often done manually based on visual information. The human limitations, such as boredom, fatigue, lack of motivation, etc. can influence the sorting process and make worse the result. In most cases the sorting result, and moreover the recovery of the process, can be improved by an automatic sorting system. This sets, however, very high requirements for the system performance. The limestone quarries of Partek have been provided with an automatic, microprocessor based sorting system. The system has been designed to be very easily adapted into various surroundings and requirements. The experiences have so far been very positive. The system contains several computing units connected to a common bus. Due to parallel processing and special hardware modules the system is very high speed and functions in real time . The visual information from a semiconductor camera is ana1yzed by a preprocessing unit, which extracts different feature values for recognition. These values are fed to classification units for final classification. The whole operation is controlled by a master processor, which also communicates with an operator. Keywords. Mining; ore sorting; quality control; multiprocessing systems; pattern recognition. two of them combine with flotation for further cleaning.
BACKGROUND The need for separating useful minerals from gangue minerals is as old as the mining industry itself. The oldest method - hand sorting by using different colours or other visible features as criteria - has in most cases been replaced by more cost-effective methods, based on other properties than colour. These other properties - density, magnetic susceptibility, chemical surface reactions etc. - have been easier to adapt for industrial processes than colour.
The first commercial machines applied to opti c al sorting we r e studied in the early 60's. These machines, developed from food sorting machines, had rather small capacities and we re found to be too inefficient to justify the costs. In 1977 a test with Finnish limestone was performed in Canada on a photometric sorting machine, called Model 16 Sorter, developed by RTZ Ore Sorters. The tests showed encouraging results and one machine was bought and installed for use on the fraction 45-85 mm at the Lappeenr a nta limestone quarry of Oy Partek Ab. This was a fraction that was too small to be effectively hand picked. By installing the Model 16 the limestone recover y could be improved.
High quality industrial products such as burnt lime, Portland cement and white fillers cannot normally be manufactured directly from Finnish run of mine lime stone. The impurities consist of rather wide dark silicate veins, often pegmatites or amphibolites, crossing the deposit randomly. Apart from hand pic k ing flotation can be used as a cleaning process, but this is in most cases too expensive. Also, due to the fineness, the product from the flotation process can only be used for certain other processes.
Some years before purchasing the Model 16 Sorter, an intensive programme of research and development on optical sorting had been started by Oy Partek Ab. The work was concentrated on detection and processing of data using the latest techniques available. The development work was carried out in cooperation with the Technical Research Centre of Finland. The results of this work were tested by using the mechanical feeding system of the Model 16 in combination with the
Because of the reasons mentioned some 75 % of the Finnish limestone mined has to pass through some kind of optical sorting. Hal f of the 12 quarries or mines use optical sorting,
499
I. Maenpaa, P. Malinen and R. Soderstrom
500
development made by Oy Partek Ab. The results were encouraging and the licence for manufacture for mineral sorting purposes was later sold to RTZ Ore Sorters, who applied it on their Model 22 Sorter. PRELIMINARY TESTS WITH THE FEED MATERIAL From the very beginning of the development project the requirement was to use a simple, reliable and economical visual sensor in the sorting system. The two possible sensor types were a laser scanning system and a solidstate camera. According to the requirements above the obvious choice was the solid-state camera. The improvements that had been made to the solid-state cameras after their first introduction in the early 70's made the decision easier. The modern cameras are very good as to the eveness from element to element and as to the low frequency of black elements. Although we were convinced about the performance of the solid-state cameras generally, we were not sure about their applicability to this particular problem of rock sorting. It was for this reason that Oy Partek Ab ordered a test from the Technical Research Centre of Finland in order to clarify the above mentioned problem. The rock samples were delivered to the research centre from the Lappeenranta mine of Oy Partek Ab.
which the reflectance factors differ with the coefficient 2 to 3 within the whole measured area. Only the red pegmatite makes a difference. It becomes darker with wave lengths under 0.55 pm but is almost white with wave lengths 1.1 pm. Hence the wave length band choice is critical only when measuring red pegmatite. In Fig. 2 the sensitivity curve of RETICON is shown as a function of wave length. It can be very easily seen that red pegmatite is also sortable from the other rocks by RETICON. Its reflectance differs from that of the others in the area where the RETICON has its maximum sensitivity. RETICON Measurements with Individual Sample Rocks The following reflectance features of the sample rocks were measured: mean value maximum value minimum value peak to peak value mean deviation The measurements were all carried out, rock being in the following different states: dry wet dry wet
and and and and
dusty dusty clean clean
The rock types to be sorted were as follows: 1.
white rocks limestone wollastonite
2.
red pegmatite
3.
dark rocks amphibolite felsite
The applicability tests were the following: 1.
measurement of the reflectance of different rocks as a function of light wave length
2.
sortability of sample rocks using solidstate camera
The reflectance spectrums were measured with Perkin-Elmer Recording spectrophotometer. The camera type used in the actual sortability tests was a RETICON solid-state line scan camera. The illumination was constructed of normal incandecent lamps. Spectrum Measurements The reflectance of the sample rocks was measured between wave lengths from 0.4 pm to 2 . 2 pm. The spectrums are shown in Fig. 1. From this figure it can be seen that the rocks are clearly separated into two groups between
As an example there are shown in Fig. 3 the measurement results of the mean value. From Fig. 3 it is easy to see that by using this feature there will be no problem in separating white, red and black rocks from each other . However, separation of lime rock and wollastonite is impossible by this feature. Fig. 3 also shows that washing of the rock significantly improves the sortability. Conclusions Derived from the Preliminary Tests 1.
The solid-state camera is very suitable for the rock sorting purpose
2.
The main feature to be used in the further development is the mean value of the reflectance. In addition, in some cases it is also reasonable to use the maximum, the minimum, and the mean deviation values of the reflectance. FEATURES OF THE SYSTEM
Before the designing phase ,the following requirements were stated for the system performance: universal; the system has to be easily adapted into different sorting tasks high speed and capacity
A Computer System for Photometric Mineral Sortin g flexible; measurement of different physical values easy to use reliability Partek produces a great variety of different products which need inspection and sorting. For this reason the system was designed to be as universal as possible. On the other hand the applicability to various cases sets new demands for the system's flexibility and easiness to use. The system must be able to measure different kinds of physical features and must have the ability to be rapidly changed for new environments. For example, rock quality varies according to the place where rocks are quarried. Furthermore, the high tonnage throughput in quarries requires that the system is high speed and has a high capacity. Finally, the operation of the system is expected to be very reliable. The solution of the above requirements was reached by the following characteristics : universal architecture, modularity parallel processing architecture of a preprocessor learning ability multiprocessor system, solid-state sensing The system architecture is hierarchical so, that different tasks are executed hierarchically by different units. The modular structure has been preserved, the expansion and reduction of the system is achieved simply by increasing/decreasing the number of modules in the system. Due to parallel processing the system is high speed. The preprocessing unit is also of modular construction. The signal processing is carried out by hardware modules in real time. The physical features to be measured can be changed by plugging in new modules. The system has two modes, a learning mode and a sorting mode. During the learning mode typical samples of each class are fed through the machine and the system stores the necessary classification parameters. In the sorting mode these parameters are used in classification. It is possible to feed these parame ters manually - or to adjust them during sorting mode. Because of the multiprocessor system the reliabilit y of the system is very good. This is due to the fact that if some unit fails, another unit takes its place and the operation can continue. Sensing takes place by a solid state line scan camera without any moving parts, this in itself increases the reliability as compared with the systems using mechanical scanning systems. Figure 4 is a sketch of the operation of the mechanics for the sorting of limestone. The rocks are fed to a 80 cm wide conveyor belt, which has a speed of 4 m/so The rocks should be stable and have no speed component in
501
relation to the belt, neither should not overlap. When leaving the belt the rocks pass through the field of view of a solid state camera. The visual images from the camera are fed to the system electronics and each rock is classified to be accepted or rejected. Rejected rocks are deflected by the air nozzles, which get the information of the size and position of a rock to be deflected. A more detailed scheme of the feeding system, manufactured by Ore Sorters Ltd. is shown in Fig . 5. STRUCTURE OF THE SYSTEM General Structure Sorting can be divided into three subtasks: measurements classification separation Hence a sorter consists of different subsystems performing these three individual tasks. This division enables "naturally" defined subsystems with well defined internal communication. The structure of the system is given in Fig. 6 . The solid-state camera continuously scans perpendicular to the conveyor belt movement with an interval of 1 ms. The camera signals are preprocessed in the line analyzer by hardware modules, which extract specific features of each object in front of the camera. This is necessary due to the requirement o f the real-time function of the system within the given speed requirements. During the same sweep several particles may be scanned by the camera. For each rock particle its specific features are extracted and given to a classification unit , which is dedicated to this one particular object . The unit accumulates information of an object from subsequent sweeps and does the analysis of the object. The analytical result is given to the master processor, which controls the rejection system (air nozzles) . Sweep Analyzer The block diagram of the sweep analyzer is described in Fig. 7. This subsystem performs two different tasks: detection of the objects calculation of the specific features of the object The video signal delivered by the camera to the anal yzer is shown in Fig. 8. In this case there are 3 rocks in the field of view of the camera. During the sweep a rock is detected and an "object" signal is generated. The special feat ures of each rock are extracted on-line
I. Mae npaa, P. Malinen and R. Soderstrom
502
during the sweep and fed forward to the classification units. The features are calculated on the basis of the video signal amplitude and the present system generates the features mentioned in section 2: mean value of the video maximum value of the video minimum value of the video It can of the is one 1n the change change
be seen in Fig. 7 that the structure sweep analyser is very modular. There plug-in circuit board for each feature analyser rack, therefore, in order to a feature, all that is required is to the plug-in units.
The positions of the rocks are also transformed to the classification units. The position of a rock is defined by the coordinates of the leading and trailing edges of the "object" signal during every sweep. The objects on the field of view of the camera have in reality three dimensions, but the analyzed visual image is two-dimensional. Due to illumination, and to the forms of the objects, the system may see shadows around the object, which may disturb the visual reception. This disturbance is eliminated as indicated in Fig. 9. Around the objects an area is formed from which the video information is not used. The size of this area can be changed.
waiting recursion classification In the waiting state the unit is ready to accept new classification tasks, which it will be given by the master processor. When a new rock is for the first time observed in front of the camera the classification task of this rock is dedicated to the first classification unit in the waiting queue. This classification unit goes then into the recursion state. It collects during every sweep the extracted features of its "own" rock and connects the features from different sweeps to each other. During this state the hardware of the unit catches the features just of its own rock. When the rock has passed the camera, the classification unit registers this and passes to the classification state. In this state it calculates the class of the rock. From this calculation the result is either accept or reject, which together with the coordinates and the size of the rock is transferred to the master processor and the classification unit goes again into the waiting queue. Master Processor The master processor contains a single board computer with 4 k bytes RAM and 4 k bytes ROM memories. The main tasks of the master processor are
Classification Unit The system can contain up to 255 classification units. The necessary number depends on how many rocks may occur in a line across the conveyor belt. In the present systems for rock sorting 16 units are used. Each unit consists of a one-chip microprocessor with associated memory and hardware for object detection and bus communication. The main tasks of a classification unit are a)
in sorting mode: to collect the feature information of the rock dedicated to it by the master processor, at the end of the rock, according to the collected feature information, carry out the classification and send the result (blast or no blast) to the master processor,
b)
in learning mode: the same as in sorting mode at the end of the rock send the collected data directly to the master processor for prototype vector calculation (i.e. the reference value of each class in question).
During the sorting mode the unit may be in one of the following states:
supervision of the function of the classification units control of the reject mechanics communication with the operators During the learning mode the master processor calculates the sorting parameters, these are given to the classification units in the initialization of the classification. In the classification mode the master processor follows the communication on the data bus and when some object is not caught automatically by a classification unit, the master processor gives it to the first unit in the waiting queue. In the classification mode the master processor controls the reject mechanism in accordance with the information given by the classification units. This mechanism consists of 40 air nozzles pendendicular to the line of stone movement. Depending on the size and location of the objects to be rejected the processor calculates, which nozzles should be active and for how long time they should be working. The operator can communicate with the system using three different panels : main panel local adjustment panel local teaching panel
A Computer System for Photometric Mineral Sorting The main panel is the normal operation station, all necessary communication with the system can be done via this panel (e.g. start and stop of conveyors, initialization of the system, parameter setting, etc.). The adjustment panel can be located near to the collection conveyors for the sorted material. If the operator is not satisfied with the sorting result, he can change the sorting criteria by this panel. The changes can, however, be done in small steps only. The teaching panel is located near to the conveyor belt and it can be used to start and stop the learning mode of the system and to control the different states in the learning mode. In addition to the above listed functions the master processor has a serial interface for connection with a higher level process control system. It also has software to control this interface and feed certain statistical data from sorting process to the higher control system. The main tasks of the software are as follows: to collect distribution of the feed material as a function of desired feature (normally, the mean value of reflectance is used) to calculate the size distribution of the feed material through the sorter to calculate the total feed rate, and the rates of each product in tons/hour. All the above mentioned information can be sent to the higher lever control system, and the sorter can also receive commands from there. The information can also be shown on the VDU screen of the main control panel, if required. At the moment (January 1983) the software for adaptive control of sorting limits is being subjected to further development. It will be finished during spring 1983. EXPERIENCES At the Lappeenranta quarry Oy Partek Ab has been using one Model 16 Sorter since 1979 and two Model 22 Sorters since 1981. Furthermore, three Model 22 machines have been used at the Pargas quarry since March 1982. The reasons for making the fairly high investment involved with these machines, and the impressive infrastructure needed, are purely economical: since the mechanical sorting can go down to a rock size of 20 mm (even to 10 mm, if needed), the recovery of high quality limestone is improved, and thus the lifetime of a quarry may rise, AM-Q*
503
the raising labour costs are a good motive for rationalization, the recruiting problems of hand pickers are already being experienced - this kind of work hardly motivates someone to his best. The quality of the product achieved by photometric sorting never exceeds the quality achieved by skilled and responsible hand pickers. This is due to the fact that the machine has a certain - although low - percentage of misblasts, mainly resulting from unstable rocks when presented and blasted. Dirty rocks and rock with one half consisting of waste also represent a problem, which in many cases can be avoided by a human eye. On the other hand, the human eye, both that of a picker and that of the supervisor, becomes tired. The Lappeenranta installation was originally designed to operate without washing of the rocks. Later on a wetting of the rocks was installed, however, without any greater success. The Pargas installation has a complete washing system and if the results are good, Lappeenranta will have it as well. Since the temperature of the rocks will be below freezing point in winter time - in the Pargas case even the machine hall - the washing cannot take place around the year, which will affect the sharpness of the sort. The stability of the rocks on the main belt is very important. This has to be checked by controlling the speed of accelerator roll (compensation for wear is important), stabilizer and main belt. The narrower the fraction, the better the rocks will stabilize. The stability is to be seen by eye, but in special cases it can be controlled by means of a slow motion camera. Regular preventive maintenance inspections are very important. Generally, the quality of well-instructed highly skilled maintenance men cannot be overemphasized. The condition of blast valves, lamps, camera, accelerator, etc. has to be regularly checked. The maximum rock size is defined as 140 mm. The best way to prevent oversized material to be fed into the machine is to use a square mesh screen deck. A grizzly might allow too big rocks to enter the machine. If the quality of the sort is very critical, there are two ways to control it. One is to install facilities for controlling of the sort by hand pickers, the other is to have a second machine to do the same job. In the first case the most economic way usually is to have two picking persons to do the checking. The capacity of the machine is defined as a rule of thumb: "The capacity in tons/hour is the same as the higher number of millimetres in the fraction to be sorted." E.g. with the fraction 45-85 mm the maximum tonnage should
504
I. Ma enpaa , P. Mal i nen and R. Soder s trom
be 85 tons/hour. In our case the upper limit has not been reached - we have 'landed' at about 75 tons/hour. E.g. with a higher density of the rock (now 2.7), or a raise of the lower limit from the present 45 mm, the capacity will increase. The highest cost is that of the compressed air. The amount needed is dependant on the amount of rock blasted. Variations in air pressure will affect the sort. In our case, investment costs were cut, as we at the same time switched over from pneumatic to hydraulic drilling, which gave us free compressor capacity. The maintenance costs can be rather high if there are much unexpected damage wear
parts. The labour costs depend on the amount of machines. One operator can handle 3-5 machines depending on the quality requirements of the sort. Generally spoken, presorting of ores is a mean to prevent barren rock to enter the following benefication stages. Hence, presorting is a way to lower the overall costs for mineral processing. It might be a feasable solution for ores that by conventional technique are considered uneconomical to process. However, to establish its profitability, this technique has to be tested in full scale.
A Computer System for Photometric Mineral Sorting
505
10 0
-
""
100
80
8
wo1lastoni te iroestone
?
--=-
~ --....::::::::-
do1anite
---
-:-----
red ,...,.."",.I-'il-" _ ____ wollastonite .1ditrk)
60
............
-
~
400
450
500
550
600
650
700
800
100e, 1200 1400 1600
--.J
1.0 w o.S
(f)
z
0
0..
(f)
w 0.6
Cl::
W
> 04 i=
--1
0.2
200 WAVELENGTH (nm) FIGURE 2.
1800 2000
WAVELENGTH (nm)
FIGURE 1.
Cl::
40 30 ·
10
(RETICoo)
w
I
20
amP1iboli te
i
360
50
2200
506
I. Ma enpaa, P. Ma linen and R. Sod erstrorn
~
8 7
lime stone wollastonite red pegrnatite amphil:xlli te
DRY AND DUSTY
6
5 4 3 2
20
40
60
80
100
120
140
160
180
200
MEAN VALUE OF REFLECrANCE
(relative value)
10 9 8 7 6 5 4 3 2
WE!' AND DUSTY
1
20
40
60
80
100
120
140
160
180
200
MEAN VALUE OF REFLECrANCE
10 9
8 7
6 5 WE!' AND CLEAN
4
3 2
20
40
60
80
100
120
140
160
180
200
MEAN VALUE OF REFLECTANCE
10 9
8 7 6
DRY ,AND CLEAN
5 4 3 2 20
40
60
80
100
120
FIGURE 3. Measuranents of reflectance
with RETICCN
140
160
180
200
MEAN VALUE OF REF'LI:r1'ANCE
507
A Compu ter System fo r Pho t ome t ric Mi nera l Sor t i ng
Q
ca mera
I I
feeder
~~) ~ ~'i-(~--~--=--=--:=.----,-..C) b-; ~ -:; '" " " . ""
con veyor
00 0
belt
a ir nozzles \
"
\
OGa
Fig. 4. Application of th e sorter for sorting of lime-stone
LINE (AN AHER
o
O
Q
0
Gel 0
oVO O\JG
Fig. 6. The structure of the sorter
, Qc.
11
..--,
accept
re ject
Ln
o
CO
RTZ
ORE
SORTERS
~
OUAnz HAI.DGEN
J\ETICON CAM(M
LI&H1S
O~TlCS
5£Co..c.RY FEEDER
1
HOUS'H(,
SUo( PlATE -,
AIR CURTAIN FAN
C[ED€R
Jo£X:ELfRATOR ROLLER
7
\
~
/""
AS5£>
~ (\l
/'
::I
'"Cl
ill: ill:
'"
~ ....
1-'-
::I
(\l
BLAST "'A.t-iI'Olb
::I
A~l£,...eL'1
ill ::I
A.
'>
:::
0:
A. (\l
Ii
en
~
~
Ii
0:
"'\I N E" LT S.
S TA8ILI SE R eor'V(yER ASSE"'SLY
P~CESSOR
CAClN£T
Le:.CI";IjF\J B~ACr. ..r ;:> FIGURE 5. Mechanics of the Sorter M 22
!3
509
A Computer System for Photometric Mineral Sorting
EDGES r-
~
CCNI'ROL UNIT
CAMERA INI'ERFACE
J-
-
r
to classification units (8-bit bus)
MEAN VALVE
MAXIMLM VALVE
MINIMUM
CAMERA
~
VALVE
VARIANCE "---
>-
FEATURE "X"
,L..-
1
FIGURE 7. Block diagram of sweep analyzer
VIDEO
OBJECT
FIGURE 8. Video and object signals from camera
______~~--------1
==:~~:===~ n- 1 ===~~~~~-=-=~=n-2
- - - - -.....~~---- ,.,
Fig. 9. Elimination of shadows