AU2019335607A1

AU2019335607A1 - Monitoring ore

Info

Publication number: AU2019335607A1
Application number: AU2019335607A
Authority: AU
Inventors: Kristo BOTHA; Adriaan Lodewicus HAASBROEK; Shaun George IRWIN; Simon Jacobus STREICHER
Original assignee: Stone Three Digital Pty Ltd
Current assignee: Stone Three Digital Pty Ltd
Priority date: 2018-09-07
Filing date: 2019-09-06
Publication date: 2021-04-01
Anticipated expiration: 2039-09-06
Also published as: BR112021004248A2; PE20210695A1; AU2019335607B2; CL2021003235A1; CL2021000547A1; WO2020049517A1; CL2021003234A1

Abstract

Systems and methods for estimating the magnitude of an ore load, determining particle size distribution (PSD) of ore, recognizing foreign material in ore images, monitoring ore, and monitoring mineral processing equipment are disclosed. An image of an ore carrying vehicle (1012) is received from an image capturing device (1016). Data relating to the ore load (1011) on the ore carrying vehicle (1012) is sensed with a scanner (1018). An estimated magnitude of the ore load (1011) is calculated by accessing data relating to the ore load, the image data and data relating to the ore carrying vehicle (1012). A machine-learning module is used to identify an ore region in an ore image and to recognize ore particles (1010) or objects within the image. A PSD value of the ore is calculated. An indicator that estimates the condition of mineral processing equipment is generated by using the PSD value and operating parameters.

Description

MONITORING ORE

CROSS-REFERENCE TO RELATED APPLICATION

This application claims priority from South African provisional patent application number 2018/06000 filed on 7 September 2018, which is incorporated by reference herein.

FIELD OF THE INVENTION

This invention relates to mineral processing and monitoring ore.

More particularly, but not exclusively, this invention relates to estimating the magnitude of an ore load on an ore carrying vehicle, analysing ore images, determining particle size distribution (PSD) of ore, monitoring mineral processing equipment, and associated prediction of required maintenance to the mineral processing equipment.

BACKGROUND TO THE INVENTION

Mineral processing is performed to separate commercially valuable minerals from ore. Various types and techniques of mineral processing exist, including: comminution such as particle size reduction by crushing and grinding; sizing whereby a screening or classification method is used to separate particle sizes; concentration such as gravity concentration or froth floatation; and dewatering where liquids and solids are separated.

Ore is typically mined and delivered by large trucks to a tipping area where the trucks discharge their ore loads, sometimes directly into a feeding opening of a crusher or onto a conveying mechanism. Mine carts or other vehicles are also used to transport the mined ore to wherever it is required. A known method of determining the weight of the ore load is to use a scale or weighbridge to determine the combined weight of the vehicle with the ore load and to subtract the empty weight of the vehicle therefrom. Weighbridges are expensive, require substantial maintenance and are sometimes not available at the location where the ore load is discharged from the vehicle. A weighbridge also needs to be continually calibrated which may be time consuming and the accuracy of measurements made by the weighbridge is dependent on these calibrations. Many weighbridges yield incorrect results due to poor calibration if calibrations are even performed at all. Vehicle sizes also vary significantly and some of the vehicles that transport ore are very large. These large trucks may carry more than a hundred tons of ore, sometimes almost five hundred tons of ore which makes it difficult and cumbersome to use a weighbridge to determine the ore load. Even when weighbridges are provided, truck drivers may tend not to use them because of the added time it takes to weigh the truck and its load. Factors such as the type of truck and the amount of fuel on board may also make it difficult to determine the weight of the ore load on the truck. Some trucks have built-in load cells for measuring the ore load. These load cells can lack accuracy and they need to be regularly calibrated which can become very expensive. Moreover, foreign objects or parts of objects such as shovel teeth, drill pipes, tyres, wood and other scrap metal can sometimes become mixed with ore during processing. These foreign objects can cause damage to crushers or other equipment which may result in costly downtime and repairs.

Particle-size distribution (PSD) of ore is a mathematical function that defines the relative amount, typically by mass of particles present in ore according to their respective sizes. A known method of determining the PSD of ore is to take a sample of the ore from a stream of ore in a mineral process. Flowever, the sample must be withdrawn from the stream in such a way that the sample has the same proportions of particle sizes as the stream. Sampling of a heap of ore may be inaccurate, depending on where the sample is taken.

Other known methods for measuring the PSD of ore on conveyor belts and trucks generally involves“watershedding” and other classical computer vision methods where images of the ore are analysed. These techniques require careful tuning of parameters and tend to struggle in the presence of complex lighting, such as shadows and changing light conditions. Furthermore, it is difficult to accurately segment (detect) the very small particles, as well as the large particles, using the same algorithm. The watershedding approach suffers from many inaccuracies and particles are detected which would seem unintuitive to a human observer. Particles are sometimes detected in the background i.e. on empty parts of the conveyor and the presence of water may exacerbate the inaccuracies. A further problem with the watershedding approach is that particles are often erroneously detected in parts of the ore that include finer material (referred to in the art as“fines”) which may bias the PSD calculation because these false detections cause the particle distribution of the ore to be incorrectly weighted.

Many different types of crushers are used to reduce the size of the ore particles, so that the ore may be processed further. Examples of types of crushers used include jaw crushers, gyratory crushers, cone crushers, horizontal shaft impactors, and vertical shaft impactors. Components of crushers are subjected to a substantial amount of wear during operation and replacement of these components is both costly and cumbersome, especially because of the downtime required. Gyratory crushers usually include a central movable mantle and a stationary concave liner (often simply referred to as“the concave”) surrounding the mantle. The mantle is then moved in an eccentric fashion, crushing the ore when a gap between the mantle and the concave is reduced.

Replacing components of a crusher is an expensive routine maintenance task. Components of the crusher that generally need replacement include the mantle and the concave (sometimes, the mantle and concave are collectively referred to as a liner) of the crusher. Getting maximum production out of a crusher liner before replacement may contribute to minimising operational costs. However, due to a lack of visibility on the state or“health” of a crusher liner, it is not uncommon to find that many crusher operators choose to perform liner maintenance on fixed schedules. It may be difficult or impossible to determine the state of “health” of a crusher liner without doing a physical inspection. There are ways of visually inspecting interior parts of a crusher but this requires physical access to the equipment with the following possible drawbacks: downtime; disassembly of the crusher being needed; and risks of physical harm to personnel performing the visual inspection.

The crusher gap is one of the main manipulated variables available for controlling the reduction ratio of a crusher and determines the maximum size of particles that are able to pass through the crusher. Depending on the circumstances, the crusher gap may be defined by a crusher closed side setting (CSS), an open side setting (OSS) or an average gap setting. As the crusher liner wears, regular adjustments to mantle position need to be made to maintain a target crusher gap. Failure to do so may have a negative impact on downstream comminution and beneficiation processes. However, getting an accurate measurement of crusher gap is difficult and usually involves passing an aluminium ball or other physical measuring device through the crusher. This procedure often requires downtime to complete. Ideally, adjustments to control crusher gap should be performed often, but it is not uncommon to find that some operations only opt to perform this task very infrequently or ad-hoc, when there is a suspicion that a significant adjustment might be necessary. An incorrect gap size may result in a product size that is outside of required specifications. This can cause significant downstream inefficiencies, equipment choking or damage, reduced milling throughput etc.

There is accordingly scope to address the aforementioned problems and deficiencies, or at least to provide a useful alternative to the known devices, processes, systems and methods.

The preceding discussion of the background to the invention is intended only to facilitate an understanding of the present invention. It should be appreciated that the discussion is not an acknowledgment or admission that any of the material referred to was part of the common general knowledge in the art as at the priority date of the application.

SUMMARY OF THE INVENTION

In accordance with an aspect of the present disclosure there is provided a computer-implemented method for estimating the magnitude of an ore load on an ore carrying vehicle, the method comprising:

receiving an image including image data of the ore carrying vehicle from an image capturing device;

sensing data relating to the ore load on the ore carrying vehicle with a scanner; and calculating an estimated magnitude of the ore load by accessing the data relating to the ore load, the image data and data relating to the ore carrying vehicle.

Further features provide for the method to include one or more of the steps of: using the data relating to the ore load and the image data to generate a three-dimensional model of the ore load; and calculating the estimated magnitude of the ore load by using the three-dimensional model of the ore load and the data relating to the ore carrying vehicle. Each of these features may be used separately, or they may be used in conjunction with one another.

Still further features provide for the data relating to the ore carrying vehicle to be pre-stored, and to include a three-dimensional model or scan of the ore carrying vehicle when it is empty; for the step of calculating the estimated magnitude of the ore load to include calculating a difference between the three-dimensional model of the ore load and the three-dimensional model of the empty ore carrying vehicle; and for the method to include comparing the data relating to the ore carrying vehicle to the data relating to the ore load on the ore carrying vehicle obtained from the scanner, and based thereon, determining a type of the ore carrying vehicle. Each of these features may be used separately, or they may be used in conjunction with one another.

Yet further features provide for the calculated magnitude of the ore load to be the volume of the ore load, which volume may be used in combination with an estimated density of the ore to calculate a weight of the ore load.

Further features provide for the method to include one or more of the steps of: receiving from the image capturing device, a plurality of images that each includes a plurality of pixels; tracking movement data of the ore carrying vehicle by using the images; combining or superimposing the image data of the plurality of images with the sensed data relating to the ore load to convert pixel coordinates to three-dimensional coordinates or real-world coordinates; and using the movement data together with the real-world coordinates to generate the three-dimensional model of the ore load; and for the three-dimensional model of the ore load to be a point cloud that is representative of the ore load. Each of these features may be used separately, or they may be used in conjunction with one another.

Still further features provide for the step of tracking movement data of the ore carrying vehicle to include: tracking linear or curved or arbitrary movements of the vehicle by tracking one or more data points on the vehicle; determining the velocity of the one or more data points; and using the tracked linear or curved or arbitrary movements of the vehicle and the calculated velocity of the data points to generate the three-dimensional model representative of the ore load. Each of these features may be used separately, or they may be used in conjunction with one another.

A yet further feature provides for the data relating to the ore load sensed by the scanner to include a profile of the ore load, or a plurality of profiles of the ore load.

A further feature provides for the step of generating the point cloud to include reconstructing or stitching the plurality of profiles of the ore load together by using the movement data.

A still further feature provides for the method to include the step of selecting an ore region in the three-dimensional model of the ore load when calculating the estimated magnitude of the ore load.

Yet further features provide for the scanner to be in the form of a three-dimensional scanner, Lidar scanner, laser scanner or any other sensor capable of determining the relative positions of real- world objects; and for the image capturing device to be in the form of a digital camera.

A further feature provides for the method to include the step of detecting the presence of the ore carrying vehicle and automatically initiating data capturing by the image capturing device and/or the scanner when the presence of an ore carrying vehicle is detected.

In accordance with a further aspect of the disclosure there is provided a system for estimating the magnitude of an ore load on an ore carrying vehicle, the system including a memory for storing computer-readable program code and a processor for executing the computer-readable program code, the system comprising:

an image receiving component for receiving image data of the ore carrying vehicle captured by an image capturing device;

a scanner for sensing data relating to the ore load on the ore carrying vehicle; and a magnitude calculation component for calculating the magnitude of the ore load by accessing the image data, the data relating to the ore load and data relating to the ore carrying vehicle.

Further features provide for the system to include a three-dimensional model generating component for generating a three-dimensional model of the ore load by using the data relating to the ore load and the image data; for the three dimensional-model to include a point cloud model or reconstruction of the ore load; and for the three-dimensional model generating component to be configured: to track linear, curved or arbitrary movement data of the ore carrying vehicle by tracking one or more data points on the vehicle; to calculate the velocity of the one or more data points; and to use the tracked linear or curved or arbitrary movement data of the vehicle and the calculated velocity of the data points to generate the three-dimensional model representative of the ore load. Each of these features may be used separately, or they may be used in conjunction with one another.

In accordance with another aspect of the present disclosure there is provided a computer- implemented method comprising:

receiving an image of ore from an image capturing device;

using a machine-learning module that is trained with one or more training images to identify an ore region in the received image and to separate data relating to the ore region from a remainder of the image; and

using the machine-learning module to recognize one or more objects within the image and outputting data relating to the recognized objects.

Further features provide for the method to include one or more of the steps of: using the machine learning module to recognize ore particles within the ore region and outputting data relating to the ore particles; and calculating a particle size distribution value of the ore using the output data relating to the ore particles and the data relating to the ore region.

Still further features provide for the machine-learning module to include a neural network; for the neural network to have a deep learning network architecture; and for the deep learning network architecture to be a fully convolutional deep neural network.

Yet further features provide for the method to include one or more of the steps of: training the machine-learning module with the one or more training images labelled according to known particle size distributions of the training images to recognize ore particles; predicting or estimating a boundary of each ore particle in the ore region with the machine-learning module and/or estimating a centre of mass of each ore particle in the ore region; separating a remainder of the ore region that falls outside the estimated boundaries and classifying this region as a fine ore region; and using the estimated boundaries of the recognized ore particles and the fine ore region to calculate the particle size distribution value of the ore. Each of these features may be used separately, or they may be used in conjunction with one another.

Further features provide for the method to include one or more of the steps of: implementing the method in mineral processing or where comminution of ore is performed by reducing the size of particles of the ore; providing the image capturing device before the comminution or after the comminution; alternatively, providing a plurality of image capturing devices with a first image capturing device provided up-stream, or before the comminution of the ore and a second image capturing device provided down-stream, or after the comminution of the ore; recognizing or identifying particles in images received from the first and/or second image capturing devices and calculating an up-stream particle size distribution value and a down-stream particle size distribution value; and comparing the respective up-stream and down-stream particle size distribution values with one another to yield a result. Each of these features may be used separately, or they may be used in conjunction with one another.

A still further feature provides for the method to include the step of generating a mask to be applied to the received image to separate data relating to the ore region from a remainder or background of the image.

Further features provide for the method to include using the machine-learning module to recognize foreign material in the image and outputting data relating to the foreign material; for the foreign material to include non-ore material; for the data relating to the foreign material to be transmitted to a server computer for further processing; and for the machine-learning module to be configured for recognizing one or more boundaries or edges of the foreign material.

Still further features provide for the step of training the machine-learning module with the one or more training images to include training the machine learning module with images that include foreign material to enable the machine learning module to recognize foreign material and to distinguish between foreign material and ore particles.

Yet further features provide for the method to include one or more of the steps of receiving data relating to one or more operating parameters of mineral processing equipment; and estimating a condition of the mineral processing equipment by using at least the determined particle size distribution value and the one or more operating parameters. The step of estimating the condition of the mineral processing equipment may include determining an indicator that estimates the condition.

A further feature provides for the method to include the step of controlling the equipment based on the estimated condition or based on the determined indicator.

A still further feature provides for the equipment to be a crusher and for the condition to be the wear of a component of the crusher.

A yet further feature provides for the step of controlling the equipment to includes adjusting a gap of the crusher.

Further features provide for the step of controlling the equipment to be performed during operation of the equipment and for the method to include determining particle size distribution of ore in real time during operation of the equipment.

A still further feature provides for the one or more operating parameters to be selected from the list comprising: power draw, throughput and an estimated magnitude of an ore load on an ore carrying vehicle.

Yet further features provide for the method to include the step of determining a rate of change of the estimated condition and extrapolating the rate of change over time to predict a remaining lifetime of the equipment.

The method may include inputting the one or more images into the machine learning module to recognize one or more objects within each image and outputting data relating to the recognized objects.

In accordance with a further aspect of the present disclosure there is provided an ore monitoring system, the system including a memory for storing computer-readable program code and a processor for executing the computer-readable program code, the system comprising:

an image receiving component for receiving an image from an image capturing device; and

a machine-learning module that is trained with one or more training images to identify an ore region in the received image, to separate data relating to the ore region from a remainder of the image, wherein the machine-learning module is configured for recognizing one or more objects within the received image and outputting data relating to the recognized objects. Further features provide for the objects to be ore particles, alternatively for the objects to be foreign material recognized in the image with the machine-learning module.

Further features provide for the machine learning module to further be configured for recognizing ore particles within the ore region and/or for outputting data relating to the ore particles; for the system to include a calculating component for calculating a particle size distribution value of the ore using the data relating to the ore particles and the data relating to the ore region.

The image may include image data of ore in the image.

Further features provide for the system to include an indicator determining component for determining an indicator that estimates the condition of the mineral processing equipment by for example applying multivariate statistics using one or more of the calculated particle size distribution value and one or more operating parameters relating to the mineral processing equipment.

Yet further features provide for the system to be used in mineral processing or where comminution of ore is performed by reducing the size of particles of the ore; for the image capturing device to be provided before the comminution or after the comminution; alternatively for the system to include a plurality of image capturing devices with a first image capturing device provided up stream, or before the comminution of the ore and a second image capturing device provided down-stream, or after the comminution of the ore; for the system to recognize particles in images received from the first and/or second image capturing devices and to calculate an up-stream particle size distribution value and a down-stream particle size distribution value; and for the system to include a comparing component for comparing the respective up-stream and down stream particle size distribution values with one another to yield a result.

In accordance with another aspect of the present disclosure there is provided a computer- implemented method for monitoring mineral processing equipment, the method comprising: conducting a method for determining a particle size distribution value of ore as defined above;

receiving data relating to one or more operating parameters of the mineral processing equipment; and

determining an indicator estimating the condition of the mineral processing equipment by using at least the determined particle size distribution value and the one or more operating parameters.

Further features provide for the equipment to be in the form of a crusher; and for the condition to be the wear of a component of the crusher such as a liner, mantle or concave of the crusher.

Still further features provide for the method to include the step of applying multivariate statistics when determining the indicator; for the method to include the step of controlling the equipment based on the determined indicator; for the step of controlling the equipment to include: adjusting a gap of the crusher, or adjusting a relative position of the liner or mantle according to a predefined target crusher gap; and for the step of controlling the equipment to be performed during operation of the equipment.

Yet further features provide for the method to include determining particle size distribution of ore in real-time during operation of the mineral processing equipment; and for the particle size distribution of the ore to be determined at an up-stream location and/or at a down-stream location of where comminution of the ore is performed by crushing, grinding, milling or another particle size reduction process of the ore.

Further features provide for the one or more operating parameters to include one or more of: power draw, throughput, magnitude of an ore load on an ore carrying vehicle as defined above, ore composition or hardness, closed side setting or reduction ratio of the crusher, whether the crusher is choke fed or not, and data relating to a froth phase of a flotation cell such as data relating to a bubble segmentation image from a digital froth image for estimating a rate of recovery or a beneficiation rate of mineral processing.

Still further features provide for the step of applying multivariate statistics to include using an inferential model; for the method to include the step of determining a rate of change of the indicator and extrapolating the rate of change over time to predict a remaining lifetime of the equipment and/or to determine when maintenance or replacement of the equipment is required.

In accordance with a further aspect of the present disclosure there is provided a system for monitoring mineral processing equipment, the system including a memory for storing computer- readable program code and a processor for executing the computer-readable program code, the system comprising: a system for determining particle size distribution of ore as defined above; and an indicator determining component for determining an indicator that estimates the condition of the mineral processing equipment by applying multivariate statistics using at least the determined particle size distribution and one or more operating parameters relating to the mineral processing equipment.

Each of the aforementioned aspects of the present disclosure may be used separately, or they may be used in conjunction with one another.

Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

In the drawings:

Figure 1 is a high-level block diagram illustrating examples of a system for estimating the magnitude of an ore load on an ore carrying vehicle, a system for determining particle size distribution of ore, and a system for monitoring mineral processing equipment;

Figure 2 is a high-level flow diagram illustrating an example of various stages of mineral processing monitored by the systems and methods disclosed herein;

Figure 3 is a high-level block diagram illustrating a backend server that may form part of the system for estimating the magnitude of the ore load;

Figure 4 is a graph showing a plurality of data points captured by a scanner, the data points being representative of an empty ore carrying vehicle (in this embodiment a truck);

Figure 5 is an image of a top view of the empty truck captured by an image capturing device, the image corresponding to the scan of Figure 4;

Figure 6 is a screenshot of a graph showing a plurality of data points captured by the scanner (in this embodiment a laser scanner) of a full truck; Figure 7 is an image captured by the image capturing device (in this embodiment a camera) of the full truck corresponding to the scan in the screenshot of Figure 6;

Figure 8 is a further image captured by the camera showing a number of data points selected as“key points” by the system for estimating the magnitude of the ore load;

Figure 9 is an image similar to Figure 8, but showing a first group of the key points located on the truck with a second group (shown in Figure 8) ignored or removed by the system, because they are not located on the truck;

Figure 10 is an image illustrating an example of a line of data points captured by the laser scanner;

Figure 1 1 shows selected key points on the truck where lighter key points are closer to the camera and scanner and darker key points are further away from the camera and scanner;

Figure 12 shows another image of the truck with key points toward the right of the image moving relatively faster (illustrated as lighter key points) and key points toward the left of the image moving relatively slower (illustrated as darker key points);

Figure 13 shows an example sequence of images for a substantially curved approach of the truck;

Figure 14 shows an example sequence of images for a substantially straight approach of the truck;

Figures 15-16 show two sequential frames or images captured by the camera with respective key points on the truck selected by the system in each of these images, to determine how the key points move from one frame to the next;

Figure 17 is a graph illustrating an example of calculated transformation parameters; Figure 18 is another graph illustrating a result of the transformation;

Figure 19 shows sequential lines of data points captured by the scanner for the example curved approach of the truck;

Figure 20 shows sequential lines of data points captured by the scanner for the example straight approach of the truck;

Figure 21 shows an example of stitched profiles of the ore load;

Figure 22 shows an example of interpolated and binned data points;

Figure 23 shows an example of a first point cloud of the truck when filled with the ore load;

Figure 24 shows an example of a second point cloud of the truck when the loading region of the truck is substantially empty;

Figures 25-26 show another point cloud of a truck with a relatively large ore load, a corresponding image of the truck with the large ore load and an ore region that may be selected by the system;

Figure 26A shows another embodiment where the system uses pixels that are associated with the ore carrying vehicle as key points;

Figures 27-28 are similar to Figures 25 and 26, but show a point cloud of a relatively smaller load than that in Figures 25 and 26;

Figure 29 shows calculations performed by a magnitude calculation component forming part of the system for estimating the magnitude of the ore load on the ore carrying vehicle;

Figure 30 shows an example image of ore on a conveyor captured by a camera forming part of the system for determining particle size distribution (PSD) of ore;

Figure 31 shows an example of so-called“blobs” representative of ore particles that are detected in the image of Figure 30;

Figure 32 shows blob edges, boundaries or outlines of the ore particles identified by the system for determining PSD of ore;

Figure 33 shows a mask used by the system for determining PSD of ore, to separate an ore region from a remainder of the image;

Figure 34 is an example of ore particles identified or estimated by a machine learning module forming part of the system for determining PSD of ore;

Figure 35 is a high-level block diagram of a backend server that may form part of the system for determining PSD of ore;

Figures 36 to 50 show visual side-by-side comparisons of: ore images (marked Original image” in Figures 36, 39, 42, 45 and 48), examples of results from a prior- art“watershedding” segmentation method (marked Old segmentation” in Figures 37, 40, 43, 46 and 49), and results using the present disclosure (marked“New segmentation” in Figures 38, 41 , 44, 47 and 50);

Figure 51 is an example flow diagram illustrating operation of the machine learning module in the form of a fully convolutional deep neural network (FCDNN), that may have an encoder-decoder type architecture;

Figure 52 shows an example graph illustrating an indicator (in this case a mantle health indicator) against time, which indicator may be determined by an indicator determining component, and this Figure also shows a corresponding component plot;

Figure 53 shows the example graph of Figure 52 and a diagram illustrating an indication of a contribution or weight of each variable in a multivariate model;

Figure 54 shows the example graph of Figure 52, but against throughput and time respectively, and illustrating a projected replacement point predicted by the system for monitoring mineral processing equipment; Figure 55 shows an example graph of a crusher gap model against time, and also illustrating physical crusher gap measurements made with an aluminium or foil ball;

Figure 56 is a high-level block diagram of a backend that may form part of the system for monitoring mineral processing equipment;

Figure 57 illustrates an example of a computing device in which various aspects of the disclosure may be implemented;

Figure 58 is a high-level flow diagram illustrating an exemplary method for estimating the magnitude of an ore load on an ore carrying vehicle;

Figure 59 is a high-level flow diagram illustrating an exemplary method for determining particle size distribution of ore;

Figure 60 is a high-level flow diagram illustrating an exemplary method for monitoring mineral processing equipment;

Figure 61 is a high-level block diagram of an exemplary controller that may form part of the systems disclosed;

Figures 62-63 show a number of images containing only ore on the left, and examples of responses or outputs of the machine learning module on the right;

Figures 64-65 show examples of blended training images that may be used to train the machine learning module; and

Figures 66-68 show examples of test images and corresponding responses or outputs that may be generated by the machine learning module.

DETAILED DESCRIPTION WITH REFERENCE TO THE DRAWINGS

There is disclosed a system for monitoring mineral processing equipment. The overall monitoring system may include a plurality of subsystems which may function independently of one another, or which may function in conjunction with one another. The system may include a backend server in data communication with a plurality of sensors and/or cameras of the sub-systems. In one embodiment, the system may be configured to monitor a truck which is transporting ore from a mine to a dumping area or into a receiving area of a crusher. In another embodiment, the system may be configured to determine the particle size distribution (PSD) of ore from images of the ore. In yet another embodiment the system may be configured to use data captured by the sensors and/or cameras and/or other data to estimate how much the mineral processing equipment has worn over time, or what their expected remaining lifespan is before maintenance or replacement may be required. Various factors can influence the rate at which a crusher liner wears, including ore hardness and reduction ratio, and other operating parameters. A single variable is not necessarily a good indication of liner wear, due to the complex relationships between many factors such as the feed and product size distribution, throughput, power draw, mantle position, crusher gap, ore composition, and whether the crusher is choke fed or not, to name but a few. Throughout the Figures, like features are referenced with like numerals.

In Figure 1 is shown a high-level block diagram of an ore monitoring system. The system may include a system (100) for monitoring mineral processing equipment. Figure 2 shows a high-level flow diagram illustrating various phases of the system (100). The system (100) may include a plurality of subsystems or further systems, preferably in data communication (in this embodiment over the Internet) with a backend server (1020) including a server (1022) and a database (1034). A system (1000) may be provided for estimating the magnitude of an ore load (101 1 ) on an ore carrying vehicle (1012). The ore carrying vehicle (1012) delivers its load to a crusher (1014) such as a gyratory crusher or jaw crusher, but any type of crusher or particle size reduction apparatus may be used. The crusher reduces particle sizes of ore (1010) as is well known in the art. Another system (2000) may be provided for determining particle size distribution (PSD) of the ore (1010). The ore (1010) may be carried by one or more conveyors (201 1 ) or other conveying mechanisms, or on the ore carrying vehicle (1012) for further processing by a grinder or mill (2021 ) and a froth flotation cell (2031 ). One or more cameras or image capturing devices (1016) may be provided for capturing images during the various stages of mineral processing. A scanner such as a laser scanner, Lidar scanner or three-dimensional (3D) scanner (1018) may be provided to scan the ore carrying vehicle (1012) including its ore load. Example features of the system (1000) for estimating the magnitude of the ore load (101 1 ) are shown in Figures 1 to 29. Example features of the system (2000) for determining particle size distribution (PSD) of ore are shown in Figure 1 and Figures 30 to 51 . Example features of the system (100) for monitoring mineral processing equipment are shown in Figures 1 to 57. Some of the Figures may relate to more than one of the systems, methods or processes disclosed herein.

Referring now to Figure 1 , there is shown the system (1000) for estimating the magnitude of the ore load (101 1 ) on the ore carrying vehicle (1012). Figure 3 shows a block diagram of the backend server (1020) that may form part of the system (1000) and that may include a memory component (1024) for storing computer-readable program code and a processor (1028) for executing the computer-readable program code. The backend of the system (1000) may further include an image receiving component (1030) for receiving image data of the ore carrying vehicle (1012) captured by the image capturing device (1016). The scanner (1018) may be provided for sensing data relating to the ore load (101 1 ) on the ore carrying vehicle (1012). The scanner (1018) may also be in data communication with the backend (1020). The backend (1020) may further include a comparing component (1027) and a magnitude calculation component (1032) for calculating the magnitude of the ore load (101 1 ) by accessing the image data, the data relating to the ore load from the scanner (1018) and data relating to the ore carrying vehicle (1012). The data relating to the ore carrying vehicle may be pre-stored in the database (1034) or storage device at the backend. The backend (1020) may also include a transmitting component (1036) and a receiving component (1038) to provide data communication, and a three-dimensional (3D) model generating component (1039) for generating a 3D virtual model of the ore carrying vehicle (1012) and its ore load (101 1 ) as is discussed in more detail below. It should be appreciated that in the present disclosure (including the described systems, sub-systems, processes and methods) the backend or server computer may be provided remotely, or it may be provided locally. Cloud-based implementations may also be possible.

Figure 4 shows an example of a plurality of data points (1040) captured by the scanner (1018). In an embodiment, the scanner and image capturing device may be provided to capture data of the ore carrying vehicle from above, for example at a tipping area for trucks. An example photograph or image (1042) captured by the image capturing device (in this embodiment a camera, for example a webcam) is shown in Figure 5 and may correspond to the data points in Figure 4, when the ore carrying vehicle (1012) is empty. Image data of images captured by the camera may be sent to the backend (1020). A first set of data points (1040.1 ) are hence on a load carrying region of the load carrying vehicle (in this embodiment a truck) while a second set of data points (1040.2) are elsewhere, for example on the ground next to the truck. The scanner may be configured to scan a plurality of lines of data points as the truck moves underneath it. The camera, in turn, may capture images of the truck. The system (1000) may be configured to scan the truck and its ore load in a contactless or unobtrusive fashion. Other sensors for detecting the presence or absence of the truck may be provided, but these are not shown for the sake of brevity. Once a truck is detected, the camera (1016) and/or scanner (1018) may automatically start capturing data which may be transmitted to the backend (1020), and the camera (1016) and scanner (1018) may continue to capture data until the truck comes to a standstill, or until the truck moves past the scanner (1018) (depending on the physical setup, some trucks might pass straight though the scanner before stopping to tip). The scanner may be configured to scan 180° therefrom. A time- of-flight device may also be used. The output of the scanner (1018) may be a set of data points which may be measured in millimetres (mm) from the scanner (1018). The horizontal and vertical axes in Figure 4 are indexed in mm in the example embodiment.

Figures 6 and 7 are similar to Figures 4 and 5, but show data points captured by the scanner (1018) when the ore carrying vehicle (1012) has an ore load in its ore carrying region. In Figure 6, the first set of data points (1040.1 ) that are on the truck (1012) are indicative of part of the ore load (101 1 ) carried by the truck. Figure 7 shows an image (1044) of the truck (1012) in a position corresponding to the data points (1040) shown in Figure 6.

Referring to Figures 8 and 9, a further image (1046) captured by the camera (1016) is shown. As the ore carrying vehicle (1012) moves underneath the scanner (1018) and camera (1016), a number of the data points (1040) in each set are selected by the system (1000) as“key points” (1044) in the image (1046), a first group (1044.1 ) of which may be located on the truck (1012) and a second group (1044.2) of which may be ignored or removed by the system (1000), because they are not located on the truck (1012). The key points may be selected points in the camera image (1046) that correspond to one or more of the line of data points (1040) captured by the laser scanner (1018). The camera images may be used to track how the truck (1012) moves between respective frames or images captured by the camera. The system (1000) may be configured to find key points and track their movement between frames captured. The distance (which may be measured in pixels) moved by each key point from one frame to the next may then be calculated. The distance from each key point to the camera (1016) or scanner (1018) may also be calculated so that each key point may be assigned a mm/pixel value. Transformation parameters may also be calculated to determine how the laser line (or line of data points) is to be transformed from one frame to the next. These transformation parameters are then used to“stitch” laser profiles (or sensed profiles) together to create a virtual representation or reconstruction such as a 3D point cloud of the ore carrying vehicle (1012) with, or without its ore load (101 1 ). These features are described in more detail below.

In Figure 10 is shown another example of a line of data points (1040) captured by the laser scanner (1018). The first group of data points (1040.1 ) are illustrated as white data points which are relatively closer to the scanner (1018) and camera (1016), and the second group of data points (1040.2) are illustrated as black data points and are relatively further away from the camera and scanner.

Figure 1 1 shows key points (1044.1 ) on the truck (1012) representing data relating to part of the ore load (101 1 ) selected in the image (1046). In the example embodiment, lighter key points are closer to the camera and scanner and darker key points are further away from the camera and scanner.

Referring to Figures 12, 13 and 14, it will be appreciated that the truck (1012) may reverse toward the tipping area in a relatively straight or a relatively curved manner. Figure 13 shows an example sequence of images for a substantially curved approach (1048), and Figure 14 shows an example sequence of images for a substantially straight approach (1050) of the truck (1012). Each key point may be tracked between frames and the distance and angle that each key point moves may be determined in real-world units (two dimensional (2D) image pixels may be converted to 3D real-world millimetres) which can be used to describe movement of the truck (1012) over time as it parks to unload its ore load (101 1 ). In the example image in Figure 12, the part of the truck in the right-hand side of the image (1046) is moving faster than the part of the truck in the left-hand side of the image (1046), because the truck is moving in an anti-clockwise direction. The key points toward the right of Figure 12 that are moving relatively faster are illustrated as lighter key points, whereas the key points toward the left of Figure 12 that are moving relatively slower are illustrated as darker key points.

Referring to Figures 15 and 16, there are shown two sequential frames (1052, 1054) or images captured by the camera (1016). The difference between key points on the truck in each of these images may be calculated to determine how the key points move from one frame to the next. The scanner data and the image data may be combined or superimposed to create the 3D model of the ore carrying vehicle and its ore load as will be described in more detail below. The difference between key points as they move from one image to the next may be measured in pixels, and this difference in pixels may correspond to real-world distances travelled by each of the key points.

Transformation parameters may be calculated by the system (1000) (or any of the other disclosed systems or methods) to describe how the laser line should be transformed for each frame pair (or, stated differently, to calculate parameters to describe the transformation between the data points or selected key points in one frame and in a subsequent frame or image captured by the camera). In Figure 17 is shown a graph (1058) illustrating an example of calculated transformation parameters. The tracked key points are shown as black dots, whereas fitted points are shown as triangles. Figure 18 shows another graph (1060) illustrating a result of the transformation relative to a first set of key points in the earlier frame, where squares indicate key points in the first image or frame, black dots indicate key points tracked between frames and diamonds indicate a rigid transform fit between two sets of key points.

Referring to Figures 13 and 19, when the truck (1012) follows a curved approach (1048), data points representing the curved approach may be captured by the scanner (1018). Examples of sequential lines of data points captured by the scanner are shown in the graph (1062) in Figure 19, and indicate that the truck (1012) moved about a metre to the right (or transversely) in the image (1046) in a generally anti-clockwise direction and turned by about 20°. In the graph (1064) in Figure 20, for the substantially straight approach (1050) the truck does not move or rotate substantially to the right or left and follows a substantially straight line of approach. The horizontal and vertical axes in Figures 19 and 20 are indexed in millimetres in the example embodiment.

The laser profiles (or lines of captured data points) may then be stitched together according to the calculated truck movement and data may be interpolated and binned where necessary. An example of the stitched profiles (1066) is shown in Figure 21 and an example of the interpolated and binned data points (1068) is shown in Figure 22. It will be appreciated that“binning” refers to grouping values or data points into“bins” (categories) to make the data more useful.

Using the above data, a virtual model or point cloud (1070, 1072) may hence be created by the system (1000) (or by any of the other disclosed systems or methods), examples of which are shown in Figures 23 and 24. Figure 23 shows a first point cloud (1070) of the truck (1012) when filled with the ore load (101 1 ) and Figure 24 shows a second point cloud (1072) of the truck (1012) when the loading region of the truck is substantially empty. This point cloud of the empty ore carrying vehicle may be created by scanning the empty vehicle. Many of these point clouds may be created for various different types of load carrying vehicles and data relating to these load carrying vehicles and their respective models or point clouds may accordingly be stored or pre stored in the database (1034) at the backend (1020).

Figures 25 and 26 show an example polygon mesh (1074) generated from the point cloud (1070) of a truck with a large ore load, a corresponding image (1076) of the truck with its large ore load and an ore region (1078) that may be selected by the system (1000) (or any of the other disclosed systems or methods). The ore region (1078) may be separated from the rest of the image (1076) by using a mask, but other intelligent means of separating the ore region from the rest of the image may also be used. The polygon mesh (1074) of the truck with the large load may then be compared or matched by the comparing component (1027) to data of ore carrying vehicles (that may be pre-stored), to match the scanned vehicle with one of the pre-stored point clouds or with known dimensions of the truck so that the type of the ore carrying vehicle (1012) may be determined. The pre-stored data of ore carrying vehicles may be data from vehicles that have been scanned earlier, and this data may be pre-stored in the database, or known data from known ore carrying vehicles may be sourced elsewhere in some circumstances. These pre-stored point clouds (or pre-stored polygon mesh, as the case may be) may be referred to as templates. The polygon mesh (1074) of the loaded truck may then be compared with the point cloud (1072) of the empty truck for the particular type of vehicle and the magnitude of the ore load (101 1 ) may then be calculated by the magnitude calculation component (1032). The magnitude may for example be the volume of the ore load (101 1 ) measured in cubic metres (m³). The volume may for example be calculated by subtracting the point cloud of the empty truck from the point cloud of the full truck, in other words, by calculating a difference between the full point cloud and the empty point cloud. A full, and empty polygon mesh or any other virtual three-dimensional model may alternatively be used. The scanned point cloud may be lined up with the template to calculate the volume of the ore load (101 1 ).

The weight of the ore load (101 1) may then be calculated for a known density (p) of the ore (1010). Figures 27 and 28 are similar to Figures 25 and 26, but show a point cloud (1079) of a relatively smaller load (1080) than the load shown in Figures 25 and 26. The corresponding image and selected ore region (1082) for the smaller load (1080) are shown in Figure 28. Figure 29 shows examples of calculations that may be performed by the magnitude calculation component (1032) where the volume of the smaller ore load (1080) is calculated as 41 m³. In other words, for an example density of ore being p = 2 tonnes per m³, the calculation yields an estimated weight of the ore load of 82 tonnes. For the larger load (1076), the volume is calculated as 52 m³ yielding an estimated weight of 104 tonnes for p = 2 tonnes per m³. These calculated values may also be physically measured on site for accuracy testing of the system (1000) (or any of the other disclosed systems or methods). The system (1000) may be configured to scan a truck in a relatively small amount of time (for example in less than a few minutes or within seconds or less).

In an alternative embodiment, and as is shown in the image (1077) in Figure 26A, the movement of the whole truck (1012) or vehicle may be detected (instead of a subsection of the image). Flence, one or more, or all the pixels associated with the truck (1012) may be selected as key- points (or data points) by the system. A dense velocity image (1077) may be generated with a determined velocity reading for every relevant pixel (that is part of the truck or ore load) instead of the sparser key-points depicted in Figures 8 to 12 and Figures 15 to 16 described above. Each pixel in the image (1077) may correspond to truck movement, where the whiter (or lighter) pixels may indicate more movement (or faster movement) and the darker pixels may indicate less movement (or slower movement). The calculated velocity of the data points and/or other data may then be used to generate the three-dimensional model representative of the ore load.

Referring again to Figure 1 , in another embodiment of the present disclosure, there is provided a system (2000) for determining particle size distribution (PSD) of ore (1010). After the ore load (101 1 ) is dumped or discharged by the truck (1012) at the tipping area, the ore (1010) may be received by a crusher (1014). The ore (1010) may also be deposited in the crusher by other means and the present disclosure is not limited to any specific type of crusher or particle size reduction apparatus. For exemplary purposes, a schematic representation of a gyratory crusher (1014) is shown in Figure 1 . The gyratory crusher (1014) may include a mantle (2010) mounted on a shaft (2012) which may be moveable in an eccentric fashion by a motor (2014) and a bevel gear arrangement (2016) as is well known in the art. Some of the variable parameters of gyratory crushers include the closed side setting (CSS) and/or the open side setting (OSS) or an average gap setting, or stated differently, a crusher gap (2018) or opening at the bottom of the crusher (1014) may be adjusted to vary the maximum sizes of ore particles that may be able to pass therethrough (i.e. particles that are produced by the crusher). The top opening of the crusher may be referred to in the art as the gape (2020) and the bottom opening or crusher gap (2018) of the crusher may be referred to as the“set”. The reduction ratio of the crusher is typically defined as the gape (2020) divided by the set (2018) and this ratio may determine how much the ore (1010) is crushed, or what the maximum size of ore particles produced by the crusher may be. The inner peripheral part of the crusher (1014) may be referred to as the concave liner (2022), but sometimes a bowl liner or other type of liner or surface may be used. Sometimes, the mantle (2010) and the concave (2022) are collectively referred to simply as a liner of the crusher. When the mantle (2010) moves relative to the concave liner (2022), the ore (1010) may be crushed.

Components of the crusher wear over time and need to be maintained and/or replaced when they have worn to a certain degree. The system (2000) for determining PSD may include the backend server (1020) in data communication with one or more image capturing devices (1016) as is diagrammatically illustrated in Figure 1 . Flowever, as shown in Figure 35, in this embodiment (or any of the other disclosed systems or methods), the backend (1020) may also include a machine learning module (1041 ) and a calculating component (1043). The image capturing devices (1016) may capture image data of the ore (1010) and this image data may be transmitted to the backend (1020) for further processing. Image capturing devices may be provided to capture images of the one or more conveyors (201 1 ). The image capturing devices may be provided before, or after crushing by the crusher (1014). For explanatory purposes, various images captured of ore on the conveyor (201 1 ) are shown in the Figures. It should be appreciated that the backend or server computer (1020) may be provided remotely, or it may be provided on-site, or locally. Cloud-based implementations may also be possible.

Figure 30 shows an example image (2024) of ore (1010) on the conveyor (201 1 ) captured by the camera (1016) or image capturing device. The machine learning module (1041 ) of the system (2000) (or any of the other disclosed systems or methods) may be trained with one or more training images to identify an ore region (2026) in the image (2024) and to separate data relating to the ore region (2026) from a remainder (2027) or background of the image (2024). The training images may include ground truth segmentation data and/or a predetermined PSD against which the system may be trained and/or tested for accuracy of prediction or estimation results. The results may also be compared to PSD of samples tested in a laboratory.

Referring to Figures 30 to 35, the machine learning module (1041 ) may be used to recognize ore particles (2028) within the ore region (2026) and to generate and/or output data relating to the ore particles (2028). The calculating component (1043) may, in turn, be used to calculate a particle size distribution (PSD) value of the ore (1010) using the output data relating to the ore particles and the data relating to the ore region (2026). In this embodiment, the ore region (2026) is a region on the conveyor (201 1 ) that has ore on it at the time when the image (2024) is taken. It will be appreciated that a plurality of images may be captured by the camera (1016) (or webcam) in real time, to form a video feed of the ore (1010) which may be transmitted in real time or near real time to the backend (1020) for further processing, so that ore particles may be identified in each frame or image of the video feed, and a PSD may be continually calculated and/or updated by the system (2000). In the example image (2024), the remainder (2027) of the image (2024) does not have a significant amount of ore in it. The system (2000) may be configured to generate a mask (2030), an example of which is shown in Figure 33. Examples of some of the recognized ore particles (2028) are shown in Figures 34, 38, 41 , 44, 47 and 50, where the detected ore particles are shown superimposed onto the respective original ore images. It should be appreciated that some of the features of the various systems and methods disclosed may be used in conjunction with one or more of the other systems and methods. For example, it is envisaged that a mask may applied to the image of the ore carrying vehicle to separate an ore region in the vehicle’s load carrying region from a remainder of the image. The machine learning module may be configured to recognize other objects or foreign material (i.e. non-ore objects) in the image, as is described in more detail below with reference to Figures 62 to 68.

To describe the machine learning module (1041 ) in more detail, a flow diagram (1045) is shown in Figure 51 . The machine learning module (1041 ) may be a neural network (NN), and may have a deep learning network architecture. The deep learning network may be a convolutional neural network (CNN) or in this embodiment a fully convolutional deep neural network (FCDNN or FCNN). At (2032), an input image (such as the image (2024) in Figure 30) may be received by the machine learning module (1041 ). The fully convolutional deep neural network may have an encoder-decoder architecture (2034) with one or more skip connections (2036). These skip connections (2036) may be provided to feed forward data and to skip one or more nodes in the FCDNN. The encoder-decoder architecture (2034) may include multiple blocks of convolutional layers arranged in a bottleneck formation. The FCDNN may include convolutional layers (and/or one or more hidden layers) and pooling or downsampling and upsampling of data may be performed by the FDCNN. The FCDNN may include a plurality of nodes and all of these nodes need not be connected to one another. A spatial (x, y) resolution of the convolutional layers may decrease by a factor of two after each max pooling (downsampling) layer in the encoder. The inverse may occur in the decoder part. The purpose of the encoder-decoder architecture may be to force the model to learn the most important features of the input image. Since a limited amount of information may be able to pass through the bottleneck region, the network may be forced to discard the less important information. This may help it learn the general characteristics of the ore particles and which features to pay attention to. Image noise, for instance, should therefore be discarded as it should not influence the output. Flowever, certain high-resolution information may be very important to propagate forward through the network without getting discarded. This may include the boundaries of the particles which should be determined as precise as possible. To achieve this, the skip connections (2036) may be employed, which may pass the output of their respective convolutional layers in the decoder to the same-sized convolutional blocks in the encoder. The convolution features may be concatenated. Two decoder stages may be included in the network. One of the decoders may be trained to output the edges of the ore particles, while the other may output a distance transform of the ore particles. From these two outputs, a threshold may be applied to each output image to create two binary images (where 1 =edge or blob, 0=non- edge or non-blob, respectively). Then a Boolean subtraction of the edge image may be taken from the blob image, so that the edges may be used to separate adjacent blobs.

In this example embodiment the FCDNN includes two outputs (2038, 2040). Information or data about the image is propagated from the input side toward the output side in the FCDNN. The machine learning module (1041 ) outputs pixels of the image (2024) that are in the ore region (2026) as a first output (2038), and pixels that correspond to each individual ore particle (2028) as a second output (2040). Edge image estimation or prediction and distance transform image estimation or prediction may be used for this purpose. From these two outputs (2038, 2040), a particle size distribution of the ore can be calculated by the calculating component (1043). A loss function may be used to penalize false predictions or estimations made by the machine learning module (1041 ), for example when a PSD is calculated for a training image with a predefined PSD (which may be based on laboratory results) is wrongly calculated by the system. In this manner, the accuracy of particle identification of the system may be increased as the system is trained or used. Error correction may for example be performed repetitively (e.g. millions of times) to increase prediction or estimation accuracy even in complex or difficult lighting conditions of the ore (1010), or where other factors such as water on the conveyor (201 1 ) may be present. Hence, inaccuracies due to loss of spatial information (for example particles being partially buried beneath other particles or beneath finer particles or“fines”) may be alleviated by the system. It will be appreciated that the machine learning module may be used in any of the disclosed systems or methods of the present disclosure.

Referring again to Figures 30 to 35, so-called“blobs” (2042) may be detected in the image (2042). An example of these blobs is shown in Figure 31 . As shown in Figure 32, the edges (2044) of these blobs may also be determined or estimated by the system (2000) (or any of the other disclosed systems or methods). Each blob may be defined as a group of pixels where the system predicts or estimates an ore particle (2028) to be present in the image (2024). The machine learning technique used may include two stages, a first stage where the ore region (2026) is detected, and a second stage where individual particle segmentation is performed. During the first stage, a mask (such as the example shown in Figure 33) is generated to prevent parts of the belt or items in the background to be detected as ore particles by the second stage. Each stage may be trained separately by labelled training images. The machine learning module (1041 ) may include a particle segmentation module trained to simultaneously estimate the centre of mass of each particle as well as an estimated boundary of the particle. Using the estimated centre of mass and the estimated boundary, each particle (2028) may be identified. Predictions or estimations that lie outside the ore region (2026) may then be removed by applying the mask. Remaining particles that cannot be individually segmented, due to the limit of the camera’s resolution, or due to these particles being fines, are identified as a remaining ore region (2048). The remaining ore region (2048) falls outside the predicted or estimated boundaries of the recognized particles (2028), but falls within the ore region (2026), and may hence be classified as a fine ore region (or a region that includes finer ore particles).

In Figure 34 is shown an example of ore particles (2028) and their respective boundaries or outlines identified or estimated by the machine learning module (1041 ) using image data from the image capturing device (1016). These estimated ore particles (2028) and data relating to the ore region (2026) may then be used by the calculating component (1043) to calculate an estimated PSD of the ore (1010).

In Figures 36 to 50 are shown visual comparisons of ore images (marked“Original image” in Figures 36, 39, 42, 45 and 48), examples of results from a known prior-art“watershedding” segmentation method (marked“Old segmentation” in Figures 37, 40, 43, 46 and 49) and results using the present disclosure (marked“New segmentation” in Figures 38, 41 , 44, 47 and 50). Example outlines of estimated ore particles (2028) using the machine learning module (1041 ) (in this embodiment using a FCDNN) are shown in Figures 34, 38, 41 , 44, 47 and 50. A side by side visual comparison of the results yielded by the old watershedding method and the results yielded by the present disclosure reveals that the present disclosure may yield significantly more robust estimations of ore particles and of PSD, which may also be more accurate and may provide resistance to changing light conditions or other factors such as the presence of water. For example, using the old segmentation approach, in Figure 49, a relatively large and incorrect ore particle (2046.1 ) and another incorrect ore particle (2046.2) are revealed, where both of these are in fact not ore, but part of the conveyor. The present disclosure may prevent this from happening or may at least reduce the chances of this happening. As shown in Figure 50, the machine learning module (1041 ) may correctly recognize that there are not any particles in the region where the incorrect particles were shown by the old segmentation of Figure 49 and the system (2000) may correctly identify this region actually being part of the conveyor (201 1 ), for example by using the mask. Detection“confidence” may be provided by the present disclosure. Particles that are barely visible to a human observer may be estimated by the system (2000) (or any of the other disclosed systems or methods) with a lower level of confidence which may mimic human like prediction or estimation. This contrasts with the previous watershedding approach where detection confidence does not feature and particles were revealed by the“Old segmentation” which may be unintuitive to a human observer.

The machine learning module (1041 ) of the system (2000) (or any of the other disclosed systems or methods) may perform visible particle segmentation and may ignore fines (2048) forming part of the ore region (2026). As is evident from Figures 37, 40, 43, 46 and 49, the known watershedding approach on the other hand incorrectly provides segmentation boundaries where it may not have been possible for a human to do so, or where it would have been unintuitive to a human observer. For example, in Figure 43, the Old segmentation provides incorrect ore particles (2050). These incorrect ore particles (2050) and/or the large incorrect ore particle (2046.1 ) and/or the other incorrect ore particle (2046.2) may cause the Old segmentation to yield incorrect results wherein the fines and the particles are not correctly weighted which may cause a bias in PSD. These problems may be alleviated by the present disclosure. The machine learning module (1041 ) may also be trained with training images where the ore is subjected to other environmental factors, such as water on the conveyor or a variety of lighting conditions to increase particle estimation accuracy of the system (2000) (or any of the other disclosed systems or methods). A significant increase in estimation or segmentation accuracy may hence be achieved which may alleviate or eliminate a need to average PSD measurements over a large number of frames. This may provide an advantage because in some circumstances, only a limited number of frames may be available for analysis. Accuracy of sieving process(es) may also be determined using the present disclosure and the PSD of the ore may be determined accurately using the system. This may provide the advantage of reducing breakage of equipment, as the system may enable predictive maintenance to be performed. For example, when the PSD calculated by the calculating component (1043) reveals that the particles produced by the crusher (1014) that are too large, this may require an adjustment of the crusher gap (2018) or perhaps the liner (the concave and mantle collectively), shaft (2012) and/or mantle (2010) and/or concave liner (2022) or any component of the crusher may require maintenance or replacement (e.g. due to excessive wear, mechanical failure etc.)· The predicted or estimated boundaries of the recognized ore particles and data relating to the fine ore region may be used by the system (2000) to calculate the particle size distribution value of the ore (1010).

It will be appreciated that the system (2000) (or any of the other disclosed systems or methods) may be implemented before or after a mineral process or before or after the comminution of ore (1010). For example, image data from the camera (1016) which observes the ore carrying vehicle (1012) (or a conveyor provided up-stream of the crusher (1014)) may also be analysed by the machine learning module (1041 ) to determine the PSD of the ore coming from the mine (or other ore source). This may be indicative of the effectiveness of the mining or ore gathering methods used to produce the ore in the first place. Further down-stream processes may also be monitored by the system (2000) (or any of the other disclosed systems or methods) such as the grinder (2021 ) and/or froth flotation cell (2031 ) and/or conveyors or ore carrying vehicles provided between these processes or after them. It is further envisaged that a plurality of image capturing devices may be provided in the system (2000), with a first image capturing device provided up stream, or at a location before the comminution of the ore and a second image capturing device provided down-stream, or at a location after the comminution of the ore. Particles may then be recognized by the machine learning module (1041 ) in images received from the first and/or second image capturing devices, so that an up-stream particle size distribution (PSD) value (in respect of an up-stream location) and a down-stream particle size distribution value (in respect of a down-stream location) may be calculated by the calculating component (1043). The comparing component (1027) may then be used for comparing the respective up-stream and down-stream particle size distribution values with one another to yield a result which may be indicative of the efficiency of the comminution of the ore by the relevant mineral processing equipment.

Referring again to Figure 1 , in another embodiment of the present disclosure, there is provided a system (100) for monitoring mineral processing equipment. The system may include the system (2000) for determining particle size distribution of ore as defined above, however in this embodiment (or any of the other disclosed systems or methods) the backend server (1020) may also include an indicator determining component (1049) as shown in the block diagram in Figure 56. The indicator determining component (1049) is arranged for determining an indicator (3010) (an example of which is shown in Figure 52) that may estimate the condition of mineral processing equipment by using at least the calculated particle size distribution (PSD) and one or more operating parameters relating to the mineral processing equipment. Multivariate statistics may be applied by the system (100) (or any of the other disclosed systems or methods). The indicator (3010) may be referred to as an indicator of mineral processing equipment“health” or condition.

In an embodiment of the present disclosure, the monitored mineral processing equipment may be the crusher (1014), and the condition that is monitored by the system (100) (or any of the other disclosed systems or methods) may be the wear of a component of the crusher such as the mantle (2010) or the concave (2022) of the crusher (or wear of any other component of the crusher).

The mineral processing equipment may be controlled based on the determined indicator. The indicator may be determined based on multivariate statistical analysis including, but not limited to principal component analysis (PCA) and partial least squares (PLS). A key performance indicator (KPI) may hence be calculated by the indicator determining component (1049) which may be indicative of the amount of wear that the mantle (2010), or another component, has been subjected to during its lifetime. The mantle KPI may be determined based on historical data. The rate at which the indicator changes over time may be extrapolated by the system (100) to make predictions about the remaining expected useful lifetime of the crusher mantle (2010) at any given point in time. The one or more operating parameters relating to the mineral processing equipment may for example include the calculated magnitude of the ore load (or the collective magnitude of a plurality of ore loads received over time by the crusher (1014)) as defined above. The multivariate statistical analysis may be provided by a multivariate data model which may form part of, or be implemented by the indicator determining component (1049).

A mantle health model may hence be determined, which may take into account a feed rate of the crusher (which feed rate may for example be derived from the calculations made by the magnitude calculation component (1038) described above with reference to Figures 1 to 29), as well as the determined PSD as determined by the system (2000) for determining PSD of ore as described above. The operating parameters may also include other measured parameters such as, but not limited to, power draw (amount of electric or other power used by the crusher or other mineral processing equipment), throughput, and ore composition or hardness to name but a few. Data relating to the ore composition or hardness may be sourced from laboratory results if necessary.

Controlling the mineral processing equipment may include: adjusting the crusher gap of the crusher (1014) (in the example embodiment where the controlled equipment is the crusher), or adjusting a relative position of the mantle according to a predefined target crusher gap. The calculated indicator may provide an inferential estimate of the crusher gap (2018) and may be provided at relatively high temporal resolution which may minimise disturbances of down-stream processes, potentially improving throughput and which may reduce operational costs of the mineral processing equipment. The system may provide an inferential estimate of the crusher gap (2018) at high temporal resolution which may allow for much finer control of the crusher product particle size distribution (i.e. to control the size of crushed ore particles produced by the crusher). The equipment may be controlled during operation of the equipment, for example when the crusher (1014) is in the form of a cone crusher such as a hydro-cone crusher. The crusher may be adjusted during use, for example based on the calculated indicator. As described above, the PSD of the ore may be calculated up-stream or down-stream of the relevant mineral processing equipment monitored by the system (100).

Real-time or near real-time monitoring and control of the mineral processing equipment may be implemented using the system (100) and maintenance and/or replacement of components of the mineral processing equipment may be optimised.

In Figure 52 is shown an example graph (3000) illustrating the indicator against time, in this case a mantle health indicator (3010), which may be determined by the indicator determining component (1049). The mantle health indicator (3010) may for example include a value between 0 and 7. The mantle health indicator may be repetitively calculated or updated by the system (100) and the graph (3000) illustrates how the indicator changes over time. The sudden change at (3012) in the indicator which occurred between June 2017 and August 2017 may be indicative of a shaft (2012) and/or mantle and/or concave (or liner) having been replaced, thereby increasing the calculated“health” of the crusher. Further significant changes in the indicator occurred at (3014) and (3016), or between December 2017 and February 2018, and between June 2018 and July 2018. These changes may represent mantle (2010) replacements having been made (but could be indicative of other components of the crusher having been replaced). The equipment health, in this case the mantle health or the overall health of the crusher may be calculated as the Euclidian distance (3018) between a current position (3024) and a previous replacement point (3022) on a corresponding component plot (3020). It should be appreciated that even though an example of the mantle health indicator (3010) is shown in the graph (3000) in Figure 52, an indicator may be determined for any other type of mineral processing component as required. In other words, in the example embodiment a mantle health indicator is shown, but the indicator may be a more general indicator of the equipment health, representative of the overall state of the equipment being monitored by the system.

Figure 53 also shows the graph (3000) of the indicator (3010), however it additionally shows a corresponding diagram (3028) which illustrates a contribution of each variable in the multivariate model to the overall health score or component health indicator. Higher scores indicate that the relevant component is newer and lower scores may indicate that the relevant component is more worn. The variables may each be weighted differently by the indicator determining component (1049).

Figure 54 shows the graph (3000), but illustrates the indicator (3010) against throughput and time respectively. Because throughput may remain relatively constant in the example embodiment, there are not significant differences between the throughput and time graphs in Figure 54. The system (100) may be configured to estimate a projected time (or throughput amount) when the relevant mineral processing equipment may require maintenance or replacement, as indicated by the example estimated replacement point (3030) in Figure 54. The multivariate model forming part of the system (100) may be trained using machine learning techniques (such as, but not limited to NN, CNN, DNN, FCDNN etc.) to improve accuracy of its predictions and/or the accuracy of the determined indicator (3010). The present disclosure may provide the advantage of getting more production out of a crusher mantle before replacement is required, due to the predictions made by the system. This may reduce operational costs and/or may reduce downtime of the mineral processing equipment.

In Figure 55 there is shown an example graph (3050) of a crusher gap model (3052) against time, and also illustrating examples of physical crusher gap measurements (3054) made with an aluminium or foil ball passed through the crusher (1014). The crusher gap model may be generated by the system (100) using the multivariate model.

The operating parameters taken into account may further include the reduction ratio, whether the crusher is choke fed or not, data relating to a froth phase of a flotation cell (2031 ) such as a bubble segmentation image from a digital froth image for estimating the rate of recovery or beneficiation rate of the mineral process. The indicator may be indicative of the efficiency of any of the mineral processes, for example blasting fragmentation in the mine, ROM (Run of Mine) PSD. Oversize detection may also be performed by any one of the systems disclosed herein, so that if an abnormally large sized particle is detected, steps may be taken to adjust the mineral processing equipment, and/or maintenance may be performed to prevent damage to the equipment or to potentially increase efficiency. Even though the PSD is described above as being calculated for the ore on the conveyor, the PSD may be calculated by receiving images from other image capturing devices such as the image capturing device that observes the ore carrying vehicle or the grinder or the flotation cell, etc. Further scanners may also be provided at any stage during the various mineral processes.

The system (1000) described above (or any of the other disclosed systems) may implement a method for estimating the magnitude of an ore load (101 1 ) on an ore carrying vehicle (1012). An exemplary method (5000) for estimating the magnitude of the ore load on the ore carrying vehicle is illustrated in the flow diagram in Figure 58. An image including image data of the ore carrying vehicle may be received (5010) from an image capturing device. Data relating to the ore load on the ore carrying vehicle may be sensed (5012) with a scanner. The data relating to the ore load and the image data may be used to generate (5014) a three-dimensional model (such as a point cloud) of the ore load and/or of the ore carrying vehicle. A difference between the three- dimensional model of the ore load and the three-dimensional model of an empty ore carrying vehicle may be calculated (5016). An estimated magnitude of the ore load may be calculated (5018) by accessing the data relating to the ore load, the image data, the data relating to the ore carrying vehicle and/or the calculated difference.

The system (2000) described above (or any of the other disclosed systems) may implement a method for determining particle size distribution of ore (1010). An exemplary method (6000) for determining particle size distribution of ore is illustrated in the flow diagram in Figure 59. An image of ore may be received (6010) from an image capturing device. A machine-learning module that may be trained with one or more training images may be used (6012) to identify an ore region in the image and to separate data relating to the ore region from a remainder of the image. Ore particles within the ore region may be recognized (6014) and data relating to the ore particles may be output (6016). A particle size distribution value of the ore may be calculated (6018) or determined using the output data relating to the ore particles and the data relating to the ore region. Instead of steps (6014, 6016 and 6018) the method may include using the machine learning module to recognize one or more objects (ore particles and/or non-ore objects or foreign material) within the image and outputting data relating to the recognized objects. The method may include inputting the one or more images into the machine learning module to recognize one or more objects within each image and outputting data relating to the recognized objects.

The system (100) described above (or any of the other disclosed systems) may implement a method for monitoring mineral processing equipment. An exemplary method (7000) for monitoring mineral processing equipment is illustrated in the flow diagram in Figure 60. A method for determining the particle size distribution value of ore as described herein may be used (7010). Data relating to one or more operating parameters of the mineral processing equipment may be received (7012). An indicator estimating the condition of the mineral processing equipment may be determined (7014) by using at least the determined particle size distribution value and the one or more operating parameters.

The described methods (5000, 6000, 7000) (or any of the other disclosed methods) may, at least partially, be performed by one or more computing devices, such as one or more remote and/or local server computers associated with the systems described herein. The one or more computing devices may be in data communication with one or more of the other components of the systems (100, 1000, 2000) (or any of the other disclosed systems). It will be understood that each of the systems and methods disclosed herein, or features of each of these systems and methods may be used separately, or they may be used together with one another.

It is further envisaged that accessing data relating to the ore carrying vehicle may include accessing data relating to one or more identifiers, labels, tags, beacons or indicia that may be provided on the ore carrying vehicle to facilitate the calculations or determinations by the system (1000). The system (1000) (or any of the other disclosed systems or methods) may be enabled to identify the type of vehicle, its load capacity, its dimensions (e.g. the dimensions of the load carrying region of the vehicle) or other data relating to the vehicle, for example when the vehicle comes into proximity with the scanner or camera. The load carrying region may for example be marked by markers or indicia, so that the system may determine the extent or size of the load carrying region by locating or identifying the markers in the image(s) of the ore carrying vehicle.

One or more controllers may be provided to facilitate operation of the systems and methods described. An example of a controller (8000) is shown in Figure 61. The controller may include a processor (8010) for executing the functions of components described, which may be provided by hardware or by software units executing on the controller. The software units may be stored in a memory component (8012) and instructions may be provided to the controller (8000) to carry out the functionality of the described components. In some cases, for example in a cloud computing implementation, software units arranged to manage and/or process data on behalf of the controller may be provided remotely (e.g. by the server computer (1020)). Some or all of the components may be provided by software or firmware downloadable onto and executable on the controller (8000). In some embodiments, the controller (8000) may include, or may be connected to the image capturing device (1016) and/or scanner (1018). The controller (8000) may include a scanner interface component (8018) and an image capturing device interface component (8020). The controller (8000) may include a receiving component (8014) and a transmitting component (8016), and it may be in data communications with the backend or server computer (1020). The receiving component (8014) may also be configured to receive data relating to the ore carrying vehicle, such as an identifier of the ore carrying vehicle, for example to enable the controller and/or the backend to identify a type of the ore carrying vehicle, or its load capacity.

The machine-learning module (1041 ) may be arranged to analyse ore images and it may be arranged to recognize objects such as foreign material in an image and to output data relating to the foreign material. For example, if there is foreign particulate that is not ore (i.e. any non-ore material), or foreign objects such as objects that have inadvertently ended up on the conveyor (201 1 ) or on the ore carrying vehicle (1012), then the machine learning module may be arranged to detect or recognize these objects in the image(s) received from the image capturing device(s). The data relating to the foreign material or foreign particulate may be transmitted to the backend server (1020) or server computer for further processing. The machine-learning module (1041 ) may also be arranged to recognize one or more boundaries or edges of the foreign material. The data relating to the foreign material may be taken into account when calculating the PSD of the ore and/or when calculating an indicator. A notification or error message may also be displayed by the disclosed systems or methods, if foreign material is detected or recognized. This may enable an operator of the system to take corrective action(s), or the corrective action(s) may be performed automatically. This may prevent damage to the mineral processing equipment and may prevent downtime of equipment.

Detection of foreign objects before they enter the crusher (1014) (or other equipment) may enable corrective action, e.g. the truck (1012) can be diverted, or a belt or conveyor (not shown) feeding the crusher (1014) can be stopped (i.e. a conveyor that is provided up-stream of the crusher). A machine learning method to detect or recognize foreign material may be implemented in a similar way to the methods for determining the PSD of ore according to the present disclosure. However, instead of training the machine learning module (1041 ) to detect ore particles, it can be trained to detect or recognize the edges, boundaries, outlines, centres (estimated centre of mass) of foreign objects in the images. These features may be implemented in any of the systems or methods of the present disclosure.

In Figures 62 and 63, there are shown a number of images containing only ore on the left, and examples of corresponding responses or outputs of the machine learning module (1041 ) on the right. No foreign objects were thus detected in these example images.

In Figures 64 and 65, there are shown examples of blended training images that may be used to train the machine learning module (1041 ). On the left are images of non-ore objects, items or foreign material (6400) that are blended or superimposed onto example images of ore. On the right are examples of corresponding outputs or responses that may be generated by the machine learning module (1041 ), indicating that the foreign objects or material may be recognized by the systems according to the present disclosure. The machine learning module may be repetitively trained over time, to increase prediction accuracy of foreign objects. As an alternative to using blended images, images of actual foreign material mixed in ore may also be used to train the machine learning module, if such images are available. Hence, training the machine-learning module with the one or more training images may include training the machine learning module with images that include foreign material. This may enable the machine learning module to recognize foreign material and to distinguish between foreign material or foreign particles and ore particles. These features may be implemented in any of the systems or methods of the present disclosure.

In Figures 66 to 68 are shown examples of test images (on the left-hand side) received by the machine learning module (1041 ) and corresponding responses or outputs (on the right-hand side) that may be generated by the machine learning module (1041 ). Examples of foreign material that may be recognized in these images include bolts, nuts, metal pipes, scrap metal, wood, parts of machinery such as drill bits, shovel teeth, rubber or other non-ore items such as tyres etc. As is evident from these Figures, a profile or outline of the foreign material may be generated by the machine learning module. These features may be implemented in any of the systems or methods of the present disclosure.

Figure 57 illustrates an example of a computing device (5700) in which various aspects of the disclosure may be implemented. The computing device (5700) may be embodied as any form of data processing device including a personal computing device (e.g. laptop or desktop computer), a server computer (which may be self-contained, physically distributed over a number of locations), a client computer, or a communication device, such as a mobile phone (e.g. cellular telephone), satellite phone, tablet computer, personal digital assistant or the like. Different embodiments of the computing device may dictate the inclusion or exclusion of various components or subsystems described below.

The computing device (5700) may be suitable for storing and executing computer program code. The various participants and elements in the previously described system diagrams may use any suitable number of subsystems or components of the computing device (5700) to facilitate the functions described herein. The computing device (5700) may include subsystems or components interconnected via a communication infrastructure (5705) (for example, a communications bus, a network, etc.). The computing device (5700) may include one or more processors (5710) and at least one memory component in the form of computer-readable media. The one or more processors (5710) may include one or more of: CPUs, graphical processing units (GPUs), microprocessors, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs) and the like. In some configurations, a number of processors may be provided and may be arranged to carry out calculations simultaneously. In some implementations various subsystems or components of the computing device (5700) may be distributed over a number of physical locations (e.g. in a distributed, cluster or cloud-based computing configuration) and appropriate software units may be arranged to manage and/or process data on behalf of remote devices.

The memory components may include system memory (5715), which may include read only memory (ROM) and random access memory (RAM). A basic input/output system (BIOS) may be stored in ROM. System software may be stored in the system memory (5715) including operating system software. The memory components may also include secondary memory (5720). The secondary memory (5720) may include a fixed disk (5721 ), such as a hard disk drive, solid state drive, and, optionally, one or more storage interfaces (5722) for interfacing with storage components (5723), such as removable storage components (e.g. magnetic tape, optical disk, flash memory drive, external hard drive, removable memory chip, etc.), network attached storage components (e.g. NAS drives), remote storage components (e.g. cloud-based storage) or the like.

The computing device (5700) may include an external communications interface (5730) for operation of the computing device (5700) in a networked environment enabling transfer of data between multiple computing devices (5700) and/or the Internet. Data transferred via the external communications interface (5730) may be in the form of signals, which may be electronic, electromagnetic, optical, radio, or other types of signal. The external communications interface (5730) may enable communication of data between the computing device (5700) and other computing devices including servers and external storage facilities. Web services may be accessible by and/or from the computing device (5700) via the communications interface (5730).

The external communications interface (5730) may be configured for connection to wireless communication channels (e.g., a cellular telephone network, wireless local area network (e.g. using Wi-Fi™), satellite-phone network, Satellite Internet Network, etc.) and may include an associated wireless transfer element, such as an antenna and associated circuitry.

The computer-readable media in the form of the various memory components may provide storage of computer-executable instructions, data structures, program modules, software units and other data. A computer program product may be provided by a computer-readable medium having stored computer-readable program code executable by the central processor (5710). A computer program product may be provided by a non-transient computer-readable medium, or may be provided via a signal or other transient means via the communications interface (5730).

Interconnection via the communication infrastructure (5705) allows the one or more processors (5710) to communicate with each subsystem or component and to control the execution of instructions from the memory components, as well as the exchange of information between subsystems or components. Peripherals (such as printers, scanners, cameras, or the like) and input/output (I/O) devices (such as a mouse, touchpad, keyboard, microphone, touch-sensitive display, input buttons, speakers and the like) may couple to or be integrally formed with the computing device (5700) either directly or via an I/O controller (5735). One or more displays (5745) (which may be touch-sensitive displays) may be coupled to or integrally formed with the computing device (5700) via a display (5745) or video adapter (5740).

The foregoing description has been presented for the purpose of illustration; it is not intended to be exhaustive or to limit the invention to the precise forms disclosed. Persons skilled in the relevant art can appreciate that many modifications and variations are possible in light of the above disclosure.

Any of the steps, operations, components or processes described herein may be performed or implemented with one or more hardware or software units, alone or in combination with other devices. In one embodiment, a software unit is implemented with a computer program product comprising a non-transient computer-readable medium containing computer program code, which can be executed by a processor for performing any or all of the steps, operations, or processes described. Software units or functions described in this application may be implemented as computer program code using any suitable computer language such as, for example, Java™, C++, or Perl™ using, for example, conventional or object-oriented techniques. The computer program code may be stored as a series of instructions, or commands on a non-transitory computer-readable medium, such as a random access memory (RAM), a read-only memory (ROM), a magnetic medium such as a hard-drive, or an optical medium such as a CD-ROM. Any such computer-readable medium may also reside on or within a single computational apparatus, and may be present on or within different computational apparatuses within a system or network.

Flowchart illustrations and block diagrams of methods, systems, and/or computer program products according to embodiments are used herein. Each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, may provide functions which may be implemented by computer readable program instructions. In some alternative implementations, the functions identified by the blocks may take place in a different order to that shown in the flowchart illustrations.

Some portions of this description describe the embodiments of the invention in terms of algorithms and/or symbolic representations of operations on information. These algorithmic descriptions and representations are commonly used by those skilled in the data processing arts to convey the substance of their work effectively to others skilled in the art. These operations, while described functionally, computationally, or logically, are understood to be implemented by computer programs or equivalent electrical circuits, microcode, or the like. The described operations may be embodied in software, firmware, hardware, or any combinations thereof.

The language used in the specification has been principally selected for readability and instructional purposes, and it may not have been selected to delineate or circumscribe the inventive subject matter. It is therefore intended that the scope of the invention be limited not by this detailed description, but rather by any claims that issue on an application based hereon. Accordingly, the disclosure of the embodiments of the invention is intended to be illustrative, but not limiting, of the scope of the invention set forth in any accompanying claims.

Finally, throughout the specification and any accompanying claims, unless the context requires otherwise, the word‘comprise’ or variations such as‘comprises’ or‘comprising’ will be understood to imply the inclusion of a stated integer or group of integers but not the exclusion of any other integer or group of integers.

Claims

CLAIMS:

1 . A computer implemented method comprising:

receiving an image of ore from an image capturing device;

2. The method as claimed in claim 1 , wherein the method includes one or more of the steps of:

using the machine-learning module to recognize ore particles within the ore region and outputting data relating to the ore particles; and

calculating a particle size distribution value of the ore using the output data relating to the ore particles and the data relating to the ore region.

3. The method as claimed in claim 1 or claim 2, wherein the machine-learning module includes a neural network that has a deep learning network architecture.

4. The method as claimed in claim 2 or claim 3, wherein the method includes one or more of the steps of: training the machine-learning module with the one or more training images labelled according to known particle size distributions of the training images to recognize ore particles; estimating a boundary of each ore particle in the ore region using the machine-learning module; separating a remainder of the ore region that falls outside the estimated boundaries and classifying this region as a fine ore region; and using the estimated boundaries of the recognized ore particles and the fine ore region to calculate the particle size distribution value of the ore.

5. The method as claimed in any one of claims 2 to 4, wherein the method includes one or more of the steps of: implementing the method in a process where comminution of ore is performed; providing a plurality of image capturing devices with a first image capturing device provided up-stream of the comminution of the ore and a second image capturing device provided down-stream of the comminution of the ore; recognizing particles in images received from the first and second image capturing devices; calculating an up stream particle size distribution value and a down-stream particle size distribution value; and comparing the respective up-stream and down-stream particle size distribution values with one another to yield a result.

6. The method as claimed in any one of the preceding claims, wherein the method includes the step of generating a mask to be applied to the received image to separate data relating to the ore region from a remainder of the image.

7. The method as claimed in any one of the preceding claims, wherein the method includes the step of using the machine-learning module to recognize foreign material in the image and outputting data relating to the foreign material.

8. The method as claimed in any one of claims 2 to 7, wherein the method includes one or more of the steps of:

receiving data relating to one or more operating parameters of mineral processing equipment; and

estimating a condition of the mineral processing equipment by using at least the determined particle size distribution value and the one or more operating parameters.

9. The method as claimed in claim 8, wherein the method further includes the step of controlling the equipment based on the estimated condition.

10. The method as claimed in claim 8 or claim 9, wherein the equipment is a crusher and wherein the condition is the wear of a component of the crusher.

1 1. The method as claimed in claim 10, wherein the step of controlling the equipment includes adjusting a gap of the crusher.

12. The method as claimed in any one of claims 9 to 1 1 , wherein the step of controlling the equipment is performed during operation of the equipment and wherein the method includes determining particle size distribution of ore in real-time during operation of the equipment.

13. The method as claimed in any one of claims 8 to 12, wherein the one or more operating parameters are selected from the list comprising: power draw, throughput and an estimated magnitude of an ore load on an ore carrying vehicle.

14. The method as claimed in any one of claims 8 to 13, wherein the method includes the step of determining a rate of change of the estimated condition and extrapolating the rate of change over time to predict a remaining lifetime of the equipment.

15. An ore monitoring system, the system including a memory for storing computer-readable program code and a processor for executing the computer-readable program code, the system comprising:

a machine-learning module that is trained with one or more training images to identify an ore region in the received image, to separate data relating to the ore region from a remainder of the image, wherein the machine-learning module is configured for recognizing one or more objects within the received image and outputting data relating to the recognized objects.

16. The system as claimed in claim 15, wherein the machine learning module is further configured for recognizing ore particles within the ore region and outputting data relating to the ore particles, and wherein the system further includes a calculating component for calculating a particle size distribution value of the ore using the data relating to the ore particles and the data relating to the ore region.

17. The system as claimed in claim 16, wherein the system further includes an indicator determining component for determining an indicator that estimates the condition of the mineral processing equipment by applying multivariate statistics using at least the calculated particle size distribution value and one or more operating parameters relating to the mineral processing equipment.