A method of and system for monitoring a mining operation

Abstract

A system for monitoring a mining operation includes a first sensor arrangement which is positioned at a first location and which is configured to scan / sense part of a portionof ore at the first locati

Description

Conventionally, integration of ore sensors within mining operations is reliant on offsite calibration work. This preparatory work is typically executed in an offsite third-party laboratory using a limited sample of reference material to create a reference library of static data models. These static calibration models, based on the grade of a limited number of reference materials, are exported digitally to the sensor and are then either left in place long term or updated occasionally in conjunction with the sensor provider using assays executed by site personnel. These calibrated models are then typically adjusted from time to time according to conventional site requirements, usually targeting a limited number of interpreted outputs (such as elemental (Cu, Fe, etc.), mineralogical (Serpentinite, Kaolinite, etc.) and / or physical (moisture, density, etc.) which can be reported in a conventional short form CSV sheet. This calibration process however is manual and offline. As a result, sensor output can drift during intervals of weeks or months between calibration operations. It should be noted that mining sensor calibration is usually dependent on elemental or mineralogical assay. Mining sensors are not typically calibrated for lithology despite the critical nature of lithological information within most mining and fleet management systems The Inventors wish to address at least some of the issues mentioned above. SUMMARY OF THE INVENTION In accordance with a first aspect of the invention there is provided a method of monitoring a mining operation, wherein the method includes: characterising a portion of ore (hereinafter referred to as the “ore portion”) at a first location by operating a first sensor arrangement to scan / sense part of the ore portion and utilising data obtained from the first sensor arrangement to perform the characterisation; transporting at least a sub-portion of the ore portion (hereinafter referred to as the “ore sub-portion”) to a second location which is different from the first location; characterising the ore sub-portion at the second location by operating a second sensor arrangement to scan / sense part of the ore subportion and utilising data from the second sensor arrangement to perform the characterisation; and comparing the characterisation of the ore portion at the first location with the characterisation of the ore sub-portion at the second location. The first sensor arrangement may include at least one sensor. The first sensor arrangement may include a plurality of sensors. The first sensor arrangement may include a sensor fusion arrangement. The first sensor fusion arrangement may therefore integrate a plurality of sensors or sensor technologies. The second sensor arrangement may include at least one sensor. The second sensor arrangement may include a plurality of sensors. The second sensor arrangement may include a sensor fusion arrangement. The second sensor fusion arrangement may therefore integrate a plurality of sensors or sensor technologies. The method of monitoring a mining operation may more specifically be to monitor sensor performance of a sensor arrangement used within a mining operation. The comparison step may be performed by a processor. The method may include tracking the ore sub-portion as it is transported from the first location to the second location in order to determine when the second sensor arrangement is scanning / sensing part of the ore sub-portion at the second location. The transportation step may include loading the ore sub-portion onto a vehicle (e.g. a truck / tipper) and driving the vehicle towards the second location, wherein the ore sub-portion may be tracked during transport by tracking the vehicle. The method may include tagging the ore portion based on the characterisation of the ore portion at the first location. The ore portion may be tagged in terms of lithology, mineralogy, texture, physical properties, colour, and / or elemental compositions. The step of characterising the ore portion at the first location may include utilising data obtained from the first sensor arrangement within a machinelearning model (hereinafter referred to as the “first machine-learning model”) which has been trained on historical sensing data, in order to perform the characterisation. The step of characterising the ore portion at the first location may include determining, using a processor, a lithology of the ore portion by utilising data obtained from the first sensor arrangement within the first machine-learning model. The step of characterising the ore sub-portion at the second location may include utilising data obtained from the second sensor arrangement within a machine-learning model (hereinafter referred to as the “second machine learning model”) which has been trained on historical sensing data, in order to perform the characterisation. The step of characterising the ore sub-portion at the second location may include determining, using a processor, a lithology of the ore sub-portion by utilising data obtained from the second sensor arrangement within the second machine-learning model. The step of characterizing the ore portion at the first location may include determining, using a processor, a mineralogy, texture, physical properties, colour, and / or elemental compositions of the ore portion scanned / sensed by the first sensor arrangement, by utilising data obtained from the first sensor arrangement within the first machine-learning model. The step of characterizing the ore sub-portion at the second location may include determining, using a processor, a mineralogy, texture, physical properties, colour, and / or elemental compositions of the ore portion scanned / sensed by the second sensor arrangement, by utilising data obtained from the second sensor arrangement within the second machinelearning model. The data obtained from the first sensor arrangement and used within the machine-learning model may be raw, full frequency, broadband data. The method may include implementing a principal component analysis (PCA) on the raw, full frequency, broadband data and using an output of the PCA analysis as an input to the first machine-learning model. The data obtained from the second sensor arrangement and used within the second machine-learning model may be raw, full frequency, broadband data. The first location may be at an area of interest within an open pit mine. The method may include transporting the ore sub-portion from the first location to the second location using a vehicle. The second location may be at a location before the sub-portion of the ore portion has been processed through a crusher. The second location may be at a location where the sub-portion of the ore portion is sensed / scanned while the sub-portion is still carried / loaded on the vehicle (e.g. carried on a tipping bed of a truck). The second location may be at a location where the vehicle discharges the ore from the vehicle (e.g. during tipping when the vehicle is a tipping truck). The sensing by the second sensor arrangement may be performed while the sub-portion is being discharged (e.g. during tipping).The second location may be at a particle sorter. The second location may be at a gantry system where the ore sub-portion is scanned / sensed by the second sensor arrangement while it is located on the vehicle. The second location may be on a conveyor belt after the ore sub-portion has been crushed / processed by a crusher. In accordance with a second aspect of the invention there is provided a system for monitoring a mining operation, wherein the system includes: a first sensor arrangement which is positioned at a first location and which is configured to scan / sense part of a portion of ore (hereinafter referred to as the “ore portion”) at the first location; a second sensor arrangement which is positioned at a second location which is different from the first location and which is configured to scan / sense part of at least a sub-portion of the ore portion (hereinafter referred to as the “ore sub-portion”) at the second location; and at least one processor which is configured to perform a first characterisation by characterising the ore portion using data obtained from the first sensor arrangement, perform a second characterisation by characterising the ore sub-portion using data obtained from the second sensor arrangement, and compare the first characterisation with the second characterisation. The first sensor arrangement may include at least one sensor. The first sensor arrangement may include a plurality of sensors. The first sensor arrangement may include a sensor fusion arrangement. The first sensor fusion arrangement may therefore integrate a plurality of sensors or sensor technologies. The second sensor arrangement may include at least one sensor. The second sensor arrangement may include a plurality of sensors. The second sensor arrangement may include a sensor fusion arrangement. The second sensor fusion arrangement may therefore integrate a plurality of sensors or sensor technologies. The system for monitoring a mining operation may more specifically be to monitor sensor performance of a sensor arrangement used within a mining operation. The at least one processor may be configured to determine when the second sensor arrangement scans / senses part of the ore sub-portion, by utilising tracking data which relates to a time-based location of the ore sub-portion. The at least one processor may be configured to tag the ore portion based on the characterisation of the ore portion at the first location. The at least one processor may be configured to tag the ore portion in terms of lithology, mineralogy, texture, physical properties, colour, and / or elemental compositions. The first characterisation may be done by utilising data obtained from the first sensor arrangement within a machine-learning model (hereinafter referred to as the “first machine-learning model”) which has been trained on historical sensing data, in order to perform the characterisation. The second characterisation may be done by utilising data obtained from the second sensor arrangement within a machine-learning model (hereinafter referred to as the “second machine-learning model”) which has been trained on historical sensing data, in order to perform the characterisation. The second location may be at a location before the sub-portion of the ore portion has been processed through a crusher. The second location may be at a location where the sub-portion of the ore portion is sensed / scanned while the sub-portion is still carried / loaded on the vehicle (e.g. carried on a tipping bed of a truck). The second location may be at a location where the vehicle discharges the ore from the vehicle (e.g. during tipping when the vehicle is a tipping truck). The sensing by the second sensor arrangement may be performed while the sub-portion is being discharged (e.g. during tipping). The second location may be at a particle sorter. The second location may be at a gantry system where the ore sub-portion is scanned / sensed by the second sensor arrangement while it is located on the vehicle. The second location may be on a conveyor belt after the ore sub-portion has been crushed / processed by a crusher. In accordance with a third aspect of the invention there is provided a method of monitoring a mining operation wherein the method includes: characterising a portion of ore (hereinafter referred to as the “ore portion”) at a first location by utilising data obtained from a first sensor arrangement which was used to scan / sense part of the ore portion at the first location; characterising the ore portion, or a sub-portion thereof, at a second location, which is different from the first location, by utilising data obtained from a second sensor arrangement which was used to scan / sense part of the ore portion, or the sub-portion thereof, at the second location; and comparing the characterisation of the ore portion at the first location with the characterisation of the ore portion, or the sub-portion thereof, at the second location. The first sensor arrangement may include at least one sensor. The first sensor arrangement may include a plurality of sensors. The first sensor arrangement may include a sensor fusion arrangement. The first sensor fusion arrangement may therefore integrate a plurality of sensors or sensor technologies. The second sensor arrangement may include at least one sensor. The second sensor arrangement may include a plurality of sensors. The second sensor arrangement may include a sensor fusion arrangement. The second sensor fusion arrangement may therefore integrate a plurality of sensors or sensor technologies. The comparison step may be performed by a processor. The step of characterising the ore portion at the first location may include utilising data obtained from the first sensor arrangement within a machinelearning model (hereinafter referred to as the “first machine-learning model”) which has been trained on historical sensing data, in order to perform the characterisation. The step of characterising the ore portion at the first location may include determining, using a processor, a lithology of the ore portion by utilising data obtained from the first sensor arrangement within the first machine-learning model. The step of characterising the ore portion, of the sub-portion thereof, at the second location may include utilising data obtained from the second sensor arrangement within a machine-learning model (hereinafter referred to as the “second machine-learning model”) which has been trained on historical sensing data, in order to perform the characterisation. The step of characterising the ore portion, of the sub-portion thereof, at the second location may include determining, using a processor, a lithology of the ore sub-portion by utilising data obtained from the second sensor arrangement within the second machine-learning model. The step of characterizing the ore portion at the first location may include determining, using a processor, a mineralogy, texture, physical properties, colour, and / or elemental compositions of the ore portion scanned / sensed by the first sensor arrangement, by utilising data obtained from the first sensor arrangement within the first machine-learning model. The step of characterizing the ore portion, of the sub-portion thereof, at the second location may include determining, using a processor, a mineralogy, texture, physical properties, colour, and / or elemental compositions of the ore portion scanned / sensed by the second sensor arrangement, by utilising data obtained from the second sensor arrangement within the second machinelearning model. The data obtained from the first sensor arrangement and used within the machine-learning model may be raw, full frequency, broadband data. The method may include implementing a principal component analysis (PCA) on the raw, full frequency, broadband data and using an output of the PCA analysis as an input to the first machine-learning model. The data obtained from the second sensor arrangement and used within the second machine-learning model may be raw, full frequency, broadband data. The first location may be at an area of interest within an open pit mine. The second location may be at a location before the sub-portion of the ore portion has been processed through a crusher. The second location may be at a location where the sub-portion of the ore portion is sensed / scanned while the sub-portion is still carried / loaded on the vehicle (e.g. carried on a tipping bed of a truck). The second location may be at a location where the vehicle discharges the ore from the vehicle (e.g. during tipping when the vehicle is a tipping truck). The sensing by the second sensor arrangement may be performed while the sub-portion is being discharged (e.g. during tipping). The second location may be at a particle sorter. The second location may be at a gantry system where the ore sub-portion is scanned / sensed by the second sensor arrangement while it is located on the vehicle. The second location may be on a conveyor belt after the ore sub-portion has been crushed / processed by a crusher. The method may include any one or more of the features / steps mentioned in respect of the first aspect of the invention. BRIEF DESCRIPTION OF THE DRAWINGS The invention will now be described, by way of example, with reference to the accompanying diagrammatic drawings. In the drawings: Figure 1 shows a schematic layout of the system in accordance with the invention; Figure 2 shows a schematic layout of part of a mining process flow; Figure 3 shows a photo of different ore samples placed on a conveyor belt; Figure 4 shows a step-by-step graphical illustration of how data from an XRF sensor, which was used to scan different ore samples conveyed on a conveyor belt, was obtained and processed and then used to provide classification predictions for the different ore samples in terms of lithology; Figure 5 shows a step-by-step graphical illustration of how data from a hyperspectral sensor, which was used to scan different ore samples conveyed on a conveyor belt, was obtained and processed and then used to provide classification predictions for the different ore samples in terms of lithology; Figure 6 shows a step-by-step graphical illustration of how data from a LIBS sensor, which was used to scan different ore samples conveyed on a conveyor belt, was obtained and processed and then used to provide classification predictions for the different ore samples in terms of lithology; Figure 7 shows a table which summarizes the performance results of a classification / characterisation system of the present invention when different sensors are used over a range of loading and treatment conditions; Figure 8 shows a confusion matrix which was obtained from an output of an experiment (see Experiment 2) where a PGNAA sensor was used; Figure 9A shows a graph which illustrates the true lithology under test in Experiment 2 where a PGNAA sensor was used; Figure 9B shows a graph which illustrates to a predicted under test in Experiment 2 where the PGNAA sensor was used; Figure 10 shows a table which sets out the different conditions under which Experiment 2 was performed in respect of the PGNAA sensor. Figure 11A shows a graph which illustrates sensor performance of various sensors which were used in Experiment 2; Figure 11B shows a graph which illustrates how an XRF sensor ore classification gets affected by moisture and loading conditions as per Experiment 2; Figure 12 shows a table which sets out the different conditions under which Experiment 2 was performed in respect of the XRF sensor. Figure 13 shows a schematic flow diagram of how data from cross-belts sensors 12 are processed and how a decision-making system of a mine can be updated and calibrated over time; Figure 14 shows a graph which illustrates a result of lithology identification testing at a mining site (see Experiment 3); Figure 15 shows a process flow which was used when visual images were used for lithology training and identification; Figure 16 shows a graph which shows the experimental results, in the form of F1 scores, of a system in accordance with the invention, which was trained by visual images; DESCRIPTION OF PREFERRED EMBODIMENTS Figure 1 provides an overview of a system 10 or process in accordance with the invention which is implemented at a mine, while Figure 2 provides a view of a typical mining process flow. One of the aims of the present invention is to compare a characterisation of ore sensed by a first sensor arrangement in one place with a characterisation of at least a portion of that ore sensed by a second sensor arrangement in another place. This may be done despite changes in presentation that may have occurred between the two locations. The exact locations of the two places may differ / vary. For example, the one location (also referred to as the “first location”) may be inside the pit (e.g. a sensor arrangement may scan an area of interest in an open-pit mine), while the other location may be located at any position downstream therefrom. The other location (also referred to as the “second location”) may for example be: • at a location where ore is sensed / scanned while still being carried / loaded on a vehicle (e.g. obtained from the first location). More specifically, at a gantry system where a sensor arrangement is used to scan the ore contained in the trays / tipping beds of trucks (see PCT Patent Application No. PCT / IB2024 / 051991 (which is incorporated herein by reference) for a more detailed description of the gantry system) (i.e. pre-crusher), • at a location where the vehicle discharges the ore from the vehicle (e.g. during tipping when the vehicle is a tipping truck), • at a post-crusher position, such as on a conveyor belt (post crusher) where a cross-belt sensor arrangement senses the ore, • at an autosampler and robolab upstream or downstream of a crusher, • near / at a free flowing ore stream at a transfer point (where ore falls off a conveyor belt), or • at a particle sorting device which incorporates a sensor arrangement. Sensor arrangements are often employed within a mine pit during excavation to gather data concerning the ore being mined and guide downstream actions, such as haul truck fleet management. These sensor arrangements, referred to as in-pit sensor arrangements 22, facilitate in-pit measurements. The in-pit sensor arrangement 22 includes at least one sensor. In one example, the in-pit sensor arrangement 22 may include a plurality of sensors. In a more specific example, the in-pit sensor arrangement 22 may be / include a fusion sensor arrangement. Following the in-pit scanning, the ore may be transported to a crusher 42. Alternatively, the ore may be directed to a waste pile, stockpile or leaching heap. Post-crusher, the ore is conveyed via a belt through a sensor arrangement 12, which conduct further sensing functions on the ore during transport. These sensor arrangements 12 are known as cross-belt sensor arrangements 12 or on-belt sensor arrangements 12. The cross-belt sensor arrangement 12 include at least one sensor. In one example, the cross-belt sensor arrangements 12 may include a plurality of sensors. In a more specific example, the cross-belt sensor arrangements 12 may be / include a fusion sensor arrangement. Conventionally, the output from these sensor arrangements 12 12, 22 typically includes grade and / or mineralogy data (refer to blocks 20 and 30 in Figure 1). In other words, the ore can be characterised in terms of grade and / or minerology. Examples of sensors included in such sensor arrangements 12, 22 include, but are not limited to, Prompt Gamma Neutron Activation Analysis (PGNAA), Pulsed Fast Thermal Neutron Analysis (PFTNA), X-ray Fluorescence (XRF), Laser-Induced Breakdown Spectroscopy (LIBS), Red, Green, Blue imaging (RGB), Visible Light Detection and Ranging (Visible LIDAR), Near Infra Red (NIR) and Hyperspectral Sensing. As mentioned, these sensor arrangements 12, 22 may be used conventionally to determine grade and mineralogy. With the present invention, data obtained from these sensor arrangements 12,22 can be used to determine / predict lithology. In order to provide a system which is capable of doing so, the sensor arrangements 12, 22 are adapted / altered in order allow the raw full spectrum of data (also referred to as raw multispectral data) acquired during measurements to be obtained (see blocks 14 and 24 in Figure 1) to be used for further processing. The process of developing such a system (also referred to as a “classification system” or “characterisation system”) included the following steps: • Ore samples of specific lithologies (e.g. specific ore types) were obtained. These samples may have been gathered at different mining sites. In one example, the samples may be Copper (Cu - high grade copper), CGT (Conglomerate ore from an iron ore mine). BIF (Banded Iron Formation from an iron ore mine), Shale (shale from an iron ore mine), hematite (e.g. high grade hematite from an iron ore mine), and an empty tray (an empty tray, of the type loaded with ore to hold the other samples). • These samples were then placed on a conveyor belt (e.g. within a testing environment or on site), with each sample being tagged with its specific location / position along the belt. A sensor was then used to scan the samples as the belt moves the samples past the sensor. The raw, full frequency, broadband data of the sensor (mentioned earlier) is then obtained at a rate of approximately once per second, making it possible to evaluate the broadband data for every point along the loaded belt (i.e. for each sample) with a high spatial resolution {e.g. many spectra per meter). • A Principal Component Analysis is then used (by using one or more processors) in order to identify characteristics (characteristic feature groups) from the raw spectral data with respect to the sample types. • An output of the PCA for each sample, together with its associated tag (identifying the specific sample lithology, e.g. CGT or BIF) are then used (by using one or more processors) to train and develop a machine learning model / algorithm which predicts ore lithology based on raw spectra data obtained from the relevant sensor / sensor type. The above was done for various different sensor types in order to effectively develop a machine learning model / algorithm 70 for each type of sensor. In other words, the machine learning model / algorithm 70 for one sensor type will be different from that of another sensor type. The sensor types used were an XRF sensor, a LIBS sensor, a hyperspectral (Fourier Transform Infrared Spectroscopy) sensor. Experiment 1 - Testing accuracy of machine learning model / algorithm In order to test the efficiency / accuracy of the machine learning model / algorithm 70, the following experiment was conducted for different sensor types: • The following samples were obtained: o Copper(Cu - high grade copper), o CGT (Conglomerate ore from an iron ore mine), o BIF (Banded Iron Formation from an iron ore mine), o Shale (shale from an iron ore mine), o hematite (high grade hematite from an iron ore mine), and o an empty tray (an empty tray, of the type loaded with ore to hold the other samples). • These samples were then placed on a conveyor belt (see Figure 3), with each sample being tagged with its specific location / position along the belt. The sensor (i.e. the sensor which was used for the development of the machine learning model / algorithm 70) was then used to scan the samples as the belt moves the samples past the sensor. The raw full spectrum data of the sensor (mentioned earlier) was then obtained for every point along the loaded belt (i.e. for each sample). • A Principal Component Analysis was then used in order to identify characteristics (characteristic feature groups) from the raw full spectrum data. • An output of the PCA was then used as an input to the machine learning model / algorithm 70. The experimental steps and results for the XRF sensor are illustrated visually in Figure 4. In Figure 4: • reference numeral 50 refers to the different ore samples (i.e. Copper (Cu), CGT, BIF, Hematite, as well as an empty tray) which represent different lithologies (also referred to as lithology type); • reference numeral 52 refers to a graphical illustration of the so-called “ground truth” which indicates the correct tags for each location along the belt (the Y axis indicates the location on the belt and the X axis refers to the sample type / material). In essence, a perfect classifier would give an output identical to the tags shown in this graph; • reference numeral 54 refers to a graphical illustration of the raw XRF spectra data obtained from the XRF sensor (the Y axis indicates the location on the belt and the X axis refers to photon energy (eV) (i.e. electron volt energy level value)); • reference numeral 56 refers to a graphical illustration of the so-called “transformed data” which is after a PCA has been performed on the raw XRF spectra data (the Y axis indicates the location on the belt and the X axis refers to specific principal components); and • reference numeral 58 refers to a graphical illustration of the predictions (also referred to as the characterisation / classification) put out by the machine learning model / algorithm 70 which utilised the transformed data (the Y axis indicates the location on the belt and the X axis refers to the predicted lithology (more specifically the predicted material type). It was found that the machine learning model / algorithm 70 developed for the XRF sensor was able to characterise / classify the sample, in terms of lithology, in about 84% of circumstances, with primary ambiguity between BIF and shale ores. The same experiment set out above was also implemented for a hyperspectral sensor and a LIBS sensor. With the hyperspectral sensor, the machine learning model / algorithm 70 (developed for the hyperspectral sensor) was able to characterize / classify the sample, in terms of lithology, in about 88% of circumstances. In this regard, reference is made to Figure 5 where the experimental steps and results for the hyperspectral sensor are illustrated visually (in a similar manner to Figure 4). With the LIBS sensor, the machine learning model / algorithm 70 (developed for the LIBS sensor) was able to characterize / classify the sample, in terms of lithology, in about 95% of circumstances. In this regard, reference is made to Figure 6 where the experimental steps and results for the LIBS sensor are illustrated visually (in a similar manner to Figures 4 and 5). Experiment 2 Since loading and moisture can have an effect on the accuracy of the machine learning models / algorithms (also referred to as the “classification model” or “spectral classification model”), the same experiment set out above was conducted over a range of loading and treatment conditions for the following sensors: Near Infra Red (SFA) sensor, an XRF sensor a PFTNA sensor, a PGNAA sensor and a LIBS sensor. The effect of mass loading on each sensor’s performance is summarised in the table shown in Figure 7. For the sake of completeness, reference is made to Figure 8 which shows a confusion matrix which was obtained from an output of the experiment when the PGNAA sensor was used, which indicates a high uniqueness for lithology (e.g. ore type). More specifically, output of the spectral classification model is lithology (e.g. an ore type) of the material being sensed. This output of the classification system (or characterisation system) is compared to the original ground truth label of the ore. In this case, it was shown that PGNAA technology can be used to identify the different lithologies (e.g. ore types) with a high degree of certainty. The graph shown in Figure 9A indicates the true lithology under test, and can be compared to the prediction (using the PGNAA sensor) shown in Figure 9B under the conditions set out in the table shown in Figure 10. Figure 11A illustrates sensor performance of various sensors which were used in Experiment 2. Figure 11B illustrates how an XFR sensor ore classification gets affected by moisture and loading conditions as per Experiment 2. Reference is now made to Figure 13 which illustrates the way in which data from cross-belts sensors 12 (e.g. XRF, NIR, PGNAA and PFTNA) are processed and how a decision-making system of a mine can be updated and calibrated over time. First, ore arrives at the cross-belt sensor 12 on an operational belt which has already been installed at an operational mine. Data is acquired from the sensor 12. More specifically the data may be acquired / captured at regular time intervals (e.g. at 1 second intervals). Each time data is captured, the said data can be referred to as a block of data. Each block of data also has an associated time stamp. The data is processed in two ways. First, the data is processed using built-in algorithms, (e.g. provided by the manufacturer of the sensor) in order to provide a conventional output, which is typically grade (Chemical grade and / or mineralogical grade) (see block 60). This conventional output (also referred to as conventional vendor data) is, in practice, sent to an industrial control network 64 (e.g. using a MODBUS OPC protocol). The data from the cross-belt sensor 12 is also saved in raw form with the full spectrum data. The raw full spectrum data (i.e. the full frequency spectrum data) is sent to a processing system 66 (also referred to as a Raw Spectrum Reading system). The processing system 66 may be implemented on a server (e.g. a local high-powered Operational Technology (OT) server). Since each block of data has a time stamp, the raw full spectrum data for each block can be correlated with lithology data obtained either (i) through physical sampling (also referred to as belt sampling) (see block 68) using conventional sampling techniques, or (ii) by estimating the ore type using data obtained from a fleet management system of the mine, in order to develop a machine learning model / algorithm (also referred to as a lithology identification model / algorithm - see reference numeral 70) for the particular sensor 12, so that it can be used to determine / predict the lithology of ore. The lithology data (hereinafter referred to as the “verified lithology data”) obtained through the physical sampling, or by using the data obtained from a fleet management system, therefore act as “Ground Truth”. Physical sampling refers to where ore is physically taken from the belt and analysed physically to check lithology (mineralogical and elemental composition may also be checked) Since the raw full spectrum data is time stamped, each data block of raw full spectrum data can be correlated (using one or more processors) to a specific part / portion of the verified lithology data which indicates the determined / verified lithology of a particular portion of ore which was sensed / scanned by the sensor 12 at the time stamp associated with the particular block of data. The part of the verified lithology data which is correlated to a specific block of raw full spectrum data can then act as its associated ’’tag”. A Principal Component Analysis (PCA) is then used (by using one or more processors) in order to identify characteristics (characteristic feature groups) from the raw full spectrum data for each block of data. An output of the PCA for each block of raw full spectrum data, together with its associated tag (identifying the specific verified lithology, e.g. obtained through physical sampling or by using the data obtained from a fleet management system) is then used (e.g. by using one or more processors) to train and develop the machine learning model / algorithm which predicts ore lithology based on raw full spectrum data obtained from the particular sensor 12. It should be appreciated that although lithology is used in this example, the same methodology can also be used with various other ore properties / characteristics such as mineralogy, texture, physical properties, colour, and / or elemental compositions. In other words, the present invention is not limited to the specific application of comparing ore lithology at one location (sensed by a first sensing arrangement) with ore lithology at another re location (sensed by a second sensing arrangement), but the same can also be done by comparing mineralogy, texture, physical properties, colour, elemental compositions, etc. From the above, it will be apparent that the process is similar to the process described above with reference to Figures 3-6. However, instead of using the specific positions of the ore on the conveyor belt, time stamps are used. It should be noted that the fleet management system lithology data quality is variable and depends on operational factors including GPS (global positioning system) availability in-pit and pre-blast drill sample quality. This correlation process and the development of the machine learning model / algorithm can be implemented from time to time in order to help improve the accuracy of the machine learning model / algorithm. The initial development of the machine learning model / algorithm can be done through the process described above with reference to Figure 13 (i.e. on site) or it can be implemented in a test environment (i.e. as described above with reference to Figures 3-6). The data obtained from the sensors can be transported to a cloud storage for offline analysis (e.g. using a Simulated Integrated Model (SIM)). After the machine learning model / algorithm 70 has been developed and implemented in practice, an output of the machine learning model / algorithm which indicates the lithology identification (ID) (see block 72) for the particular block of data (i.e. the machine learning model / algorithm in this example characterises / classifies in terms of lithology) is sent back to the industrial control network 64 where it is tagged in a manner that allows the lithology ID to be merged with the conventional output which was previously captured from the sensor 12. This output can then be analysed according to a value based ore control model, which takes into account all of the measurement interpretations making it possible to estimate the market value of the particular packet of ore under observation. This final value-based output can then be stored within the cloud 76 and used to make mining decisions. Quality control can also be executed on the basis of observations concerning signal statistics including signal-to-noise ratio, persistent / misleading peaks or dips, signal stability, noise stability, and / or amplitude drift. It should be noted that the process described above with reference to Figure 13 to develop a machine learning model / algorithm 70 for an on-belt sensor 12 (one example of the position of the on-belt sensor 12 is shown in Figure 2 - see reference sign 44) and then update it from time to time, can also be implemented for in-pit sensors (see Figure 2, reference sign 40) or any other sensors within a mining environment. When the training is done at a particular mining site, it is important to understand that a fleet management system (FMS) of a mine plays a key role in managing mine material flows. Geological data gathered from drilling / sampling and visual inspection is embedded in the fleet management system to create material type “polygons”. The polygons are available to the excavator fleet, which uses this information to ensure that material is extracted in a sensible manner consistent with the mine plan. Haulage trucks are then tagged according to the material labels for each polygon to enable fleet management optimised for material properties. Experiment 3 - Training Performed at a Mining Site Raw, full-frequency spectra were acquired using cross-belt sensors 12 already deployed at a site (e.g. see Figure 2, reference sign 40). Following a period of several weeks, material type tags stored in a fleet management historian (i.e. on a database) were then associated with the spectra according to time logs by implementing the methodology described earlier. In other words, the sensors were altered to enable delivery of raw full spectrum data in parallel to the conventional calibrated output sent to the control system. This raw data was then tagged according to ore classification data obtained from a block model of a mine, and passed through a machine learning model to develop the result shown in Figure 14 (the model performed well as indicated by the favourable ratio of True to False results). In other words, the machine learning algorithm for the sensor on site was developed using a previous data set, and then deployed on a follow-on set (to characterise / classify the set). The block model of a mine is a synthetic representation of the entire resource and is an essential tool for mine planning. It can be updated iteratively through the employment of typical geology methods (e.g. drilling, sampling, etc.) Using this approach, the sensor spectra could be correlated with lithology (e.g. ore / material type), making it possible to identify the lithology automatically (i.e characterise / classify the ore in terms of lithology) using output from the cross-belt sensor(s) 12 already installed on site (e.g. at tier one mining operations) with no disruption to operations Alternative Visual Technique In the examples and experiments described above, the raw frequency spectrum data from specific mining sensors was used to develop a machine learning algorithm which could identify ore lithology automatically. In this alternative embodiment, the same process described above is implemented in order to develop a machine learning algorithm / model, except that image data captured from a camera(s) is used instead of the raw full spectrum data of a mining sensor. In one experiment, each of 7 lithology classes were presented to the system, leading to a limited training set of 307 images. The processing chain used is shown in Figure 15. Following multiple training iterations the system accuracy (i.e. of the machine learning algorithm) was measured using an “F1 Score”, which is a conventional measurement for classifier quality. The multiple training iterations refer to initial training, then retraining on updated labels (i.e. training was done twice on initial, and secondary sets of pictures). In this case, an average F1 score across all classes of 99% was obtained. For real-world deployments, this capability can be embedded on an edge camera device without reliance on cloud communications, rendering it possible to develop a field-deployable tool for ore lithology classification without the need for high data bandwidths. Potential uses for this technique include underground ore type identification at-face or on-belt to support processing stream selection, quality assurance for ex-plant product in bulk operations where %fines and pellet quality must be monitored, and flotation operations where operators may rely on visual cues for indicators of process quality. Figure 16 shows the visual system’s training results using 7 different classes of ore from a particular mining site with an F1 Score comparison before and after retraining. Referring now to Figure 1. In practice, machine learning algorithms / models are developed for both an in-pit sensor arrangement 22 (see block 22) and on-belt / cross-belt sensor arrangement 12 (see block 12) in the manner as described earlier in the specification. In other words, the raw full multispectral data for both the in-pit sensor arrangement and cross-belt sensor arrangement 12 are obtained (see blocks 14, 24) and then analysed using its associated machine learning algorithm / model (see blocks 16 and 26 - “ore fingerprinting” refers to the implementation of the machine learning algorithm / model). Each machine learning algorithm / model then performs a characterisation / classification function by determining the lithology of the ore which was sensed / scanned by the particular sensor arrangement. As mentioned, conventional output data of these sensor arrangements (e.g. grade and mineralogy) can also be obtained (see blocks 20 and 30). Due to the conditions under which the in-pit sensor arrangements operate, their performance may degrade over time or they may become faulty. In practice, early detection of sensor performance degradation / sensor faults is very difficult to detect. Sensing output tends to be impacted by bias and background noise, the combination of which drives the signal-to-noise ratio. Signal-to-noise ratio is a key determining factor in the usability of data; if noise overwhelms the signal, decision quality will be reduced leading to destruction of value. With respect to standoff sensing, signal quality can be reduced by both sensor-driven elements (reduction in laser power, deterioration of electronics integrity over time) or environmental effects (increased dust or droplets in the airborne signal pathway, increased absorption or scattering of energising light caused by water or mud accumulation on rock surfaces, high vibration levels causing defocusing or electronic cross-talk, mud or dust accumulation on receivers or spectrometers). Further, system outputs can become ambiguous as the sensor is applied to lithologies outside of the original training envelope. Therefore, standoff sensor performance and application value is anticipated to deteriorate over sensor lifetime. In order to address this issue, the same ore / ore portion which is sensed by an in-pit sensor arrangement is later identified when it is sensed by the on-belt sensor arrangement 12 which is installed downstream after the crusher 42. More specifically, part of the ore portion is sensed by an in-pit sensor arrangement. The ore portion, or a sub-portion of the ore portion (not all the ore is necessarily transported to the crusher 42), is then transported and processed, and part thereof is then sensed by the on-belt sensor arrangement 12. In doing so, the lithology determined by the in-pit sensor arrangement and its associated machine learning algorithm / model for a particular portion of ore can be compared (e.g. using a processor) to the lithology determined by the on-belt sensor arrangement 12 and its associated machine learning algorithm / model for the same portion of ore (see block 32). If the determined lithologies for the same portion of ore are identical, then no action is required. However, if the determined lithologies for the same portion of ore are not identical, then it may indicate that the performance of the in-pit sensor arrangement may be degrading or that the in-pit sensor arrangement is faulty and require replacement. More specifically, ore will be scanned in the pit “as it lays”. A mining environment map (“polygons”) is then updated so that high resolution polygons can be created, each with a lithology tag. The ore is then mined conventionally, and each bucket load is tagged according to a local material polygon for the mined area. Trucks are loaded and the lithology tags for each truck follows the labels assigned to the buckets based on polygon labels. Ore is then delivered to the crusher 42. Output from the crusher 42 then passes under the cross-belt sensor arrangement 12. The output from the cross-belt sensor arrangement 12 is then analysed for lithology. The lithology identified post-crusher is then compared to the lithology which had originally been identified in the pit. A more detailed explanation of the technical steps for this method / process is outlined below: a) An area of interest within an open pit mine is evaluated according to lithology using a sensor arrangement (as described above). b) Data from sensor arrangement is collected on a geospatial basis and delivered to a mine management system. c) Excavation is executed within the area of interest. d) As ore is excavated from the area of interest, the ore is tagged digitally according to the locally inferred lithological properties (i.e. using the data obtained from the sensor arrangement) on a bucket-by-bucket basis. e) Trucks are loaded with ore which has been tagged digitally according to the locally inferred lithological ore properties. f) Truck destinations are determined according to the lithological properties of loaded ore according to guidelines for the particular mine at the particular time of handling. Destinations may include: crushers, stockpiles, waste piles, and heap leach piles. g) Ore which is delivered to the crusher is analysed using a cross-belt sensor arrangement. The cross-belt sensor arrangement is calibrated to deliver lithology observations as a function of time. The timestamp of a given ore parcel is associated with the time of arrival for a tagged truck using a time calculation which may incorporate delays incurred by (a) crushing time and (b) transfer through chutes and on conveyor belts to the analyser. h) The analyser output is assembled in a time series showing lithology measurements for each truck over time. i) Retrospective analyses are built using appropriate time constants (hourly, shift-by-shift, daily, monthly). In these analyses, lithology output observed from the cross-belt analyser for each tagged truck is compared with lithology output anticipated based the in-pit sensor arrangement. Analysis may be done using a multi-class correlation technique, wherein a single lithology assessment is an observation within a confusion matrix. Higher confidence output will indicate that the system is functioning according to design, while lower confidence output may be used to indicate ambiguities requiring an intervention. A multi-class correlation technique is a statistical approach used to find relationships between multiple classes or categories. In this context, a class refers to a specific type of lithology. The confusion matrix is used to visualize the performance of an algorithm. Each row of the matrix represents the instances of a predicted class, while each column represents the instances of an actual class. It is called a confusion matrix because it shows how often the algorithm confuses two classes. In the case of the sensor arrangements, each lithology assessment is treated as an observation that is placed within the confusion matrix. The output of the analysis is a measure of confidence, which indicates how well the system is performing. A higher confidence output suggests that the system is functioning as designed and can reliably distinguish between different lithologies. On the other hand, a lower confidence output may indicate that there are ambiguities or uncertainties in the sensor readings, which could require further investigation or intervention. It will be appreciated that crushed ore is physically different than uncrushed ore. Therefore, the output gathered from an in-pit sensor arrangement using uncrushed ore may be different the output from a post-crusher sensor arrangement using the same ore after crushing. However, repeat calibration (see reference sign 33 in Figure 1) can be executed to overcome this challenge. Further, the calibration of cross-belt sensor arrangement and inpit sensor arrangement can be executed using physical sampling as described earlier in the patent specification, i.e. taking material off the belt or out of the pit and sending for chemical / mineralogical assay. It should be appreciated that by comparing the characterisation at the first location using the first sensor arrangement with the characterisation at the second location using the second arrangement, the system / methodology effectively implements an effective quality control function which can indicate when there may be sensor performance degradation.

Claims

1. A method of monitoring a mining operation wherein the method includes:characterising a portion of ore (hereinafter referred to as the “ore portion”) at a first location by operating a first sensor arrangement to scan / sense part of the ore portion and utilising data obtained from the first sensor arrangement to perform the characterisation;transporting at least a sub-portion of the ore portion (hereinafter referred to as the “ore sub-portion”) to a second location which is different from the first location;characterising the ore sub-portion at the second location by operating a second sensor arrangement to scan / sense part of the ore subportion and utilising data from the second sensor arrangement to perform the characterisation; andcomparing the characterisation of the ore portion at the first location with the characterisation of the ore sub-portion at the second location.

2. The method of claim 1, wherein the comparison step is performed by a processor.

3. The method of claim 2, which includes tracking the ore sub-portion as it is transported from the first location to the second location in order to determine when the second sensor arrangement is scanning / sensing part of the ore sub-portion at the second location.

4. The method of claim 3, wherein the transportation step includes loading the ore sub-portion onto a vehicle and driving the vehicle towards the second location, wherein the ore sub-portion is tracked during transport by tracking the vehicle.

5. The method of claim 1, which includes tagging the ore portion based on the characterisation of the ore portion at the first location.

6. The method of claim 5, wherein the ore portion is tagged in terms of lithology, mineralogy, texture, physical properties, colour, or elemental compositions.

7. The method of claim 1, wherein the step of characterising the ore portion at the first location includes utilising data obtained from the first sensor arrangement within a machine-learning model (hereinafter referred to as the “first machine-learning model”) which has been trained on historical sensing data, in order to perform the characterisation.

8. The method of claim 7, wherein the step of characterising the ore portion at the first location includes determining, using a processor, a lithology of the ore portion by utilising data obtained from the first sensor arrangement within the first machine-learning model.

9. The method of claim 7, wherein the step of characterising the ore subportion at the second location includes utilising data obtained from the second sensor arrangement within a machine-learning model (hereinafter referred to as the “second machine-learning model”) which has been trained on historical sensing data, in order to perform the characterisation.

10. The method of claim 9, wherein the step of characterising the ore subportion at the second location includes determining, using a processor, a lithology of the ore sub-portion by utilising data obtained from the second sensor arrangement within the second machinelearning model.

11. The method of claim 9, wherein the step of characterizing the ore portion at the first location includes determining, using a processor, a mineralogy, texture, physical properties, colour, or elementalcompositions of the ore portion scanned / sensed by the first sensor arrangement, by utilising data obtained from the first sensor arrangement within the first machine-learning model.

12. The method of claim 11, wherein the step of characterizing the ore sub-portion at the second location includes determining, using a processor, a mineralogy, texture, physical properties, colour, or elemental compositions of the ore portion scanned / sensed by the second sensor arrangement, by utilising data obtained from the second sensor arrangement within the second machine-learning model.

13. The method of claim 10, wherein the data obtained from the first sensor arrangement and used within the machine-learning model is raw, full frequency, broadband data.

14. The method of claim 13, which includes implementing a principal component analysis (PCA) on the raw, full frequency, broadband data and using an output of the PCA analysis as an input to the first machine-learning model.

15. The method of claim 14, wherein the data obtained from the second sensor arrangement and used within the second machine-learning model is raw, full frequency, broadband data.

16. The method of claim 1, wherein the first location is at an area of interest within an open pit mine and wherein the method includes transporting the ore sub-portion from the first location to the second location using a vehicle.

17. The method of claim 16, wherein the second location is at a gantry system where the ore sub-portion is scanned / sensed by the second sensor arrangement while it is located on the vehicle.-SI-18. The method of claim 16, wherein the second location is on a conveyor belt after the ore sub-portion has been crushed / processed by a crusher.

19. A system for monitoring a mining operation, wherein the system includes:a first sensor arrangement which is positioned at a first location and which is configured to scan / sense part of a portion of ore (hereinafter referred to as the “ore portion”) at the first location;a second sensor arrangement which is positioned at a second location which is different from the first location and which is configured to scan / sense part of at least a sub-portion of the ore portion (hereinafter referred to as the “ore sub-portion”) at the second location; andat least one processor which is configured toperform a first characterisation by characterising the ore portion using data obtained from the first sensor arrangement,perform a second characterisation by characterising the ore sub-portion using data obtained from the second sensor arrangement, andcompare the first characterisation with the second characterisation.

20. The system of claim 19, wherein the at least one processor is configured to determine when the second sensor arrangement scans / senses part of the ore sub-portion, by utilising tracking data which relates to a time-based location of the ore sub-portion.

21. The system of claim 19, wherein the at least one processor is configured to tag the ore portion based on the characterisation of the ore portion at the first location.

22. The system of claim 21, wherein the at least one processor is configured to tag the ore portion in terms of lithology, mineralogy, texture, physical properties, colour, or elemental compositions.

23. The system of claim 19, wherein the first characterisation is done by utilising data obtained from the first sensor arrangement within a machine-learning model (hereinafter referred to as the “first machinelearning model”) which has been trained on historical sensing data, in order to perform the characterisation.

24. A method of monitoring a mining operation, wherein the method includes:characterising a portion of ore (hereinafter referred to as the “ore portion”) at a first location by utilising data obtained from a first sensor arrangement which was used to scan / sense part of the ore portion at the first location;characterising the ore portion, or a sub-portion thereof, at a second location, which is different from the first location, by utilising data obtained from a second sensor arrangement which was used to scan / sense part of the ore portion, or the sub-portion thereof, at the second location; and comparing the characterisation of the ore portion at the first location with the characterisation of the ore portion, or the sub-portion thereof, at the second location.33

Images