Modelling geological features

Abstract

A method of modelling a geological structure comprises obtaining geological data for a geological feature, the geological data comprising (i) first measurements corresponding to a first known location of the geological feature, and (ii) second measurements corresponding to at least one of a second known location of the geological feature, an orientation of the geological feature, and an overall orientation of the geological feature, and processing the geological data, including the first and second measurements, with a probabilistic model to generate an uncertainty model comprising a probabilistic representation of other locations of the geological feature in three-dimensional space between the known locations.

Description

MODELLING GEOLOGICAL FEATURESFIELD

[0001] The present application relates to modelling geological features.BACKGROUND

[0002] Safe and economic operation of an open pit mine requires ongoing assessment of the stability of the slopes created during the mine development. The reliability of this assessment is significantly dependent on the geological, geotechnical, groundwater and material property models (i.e., Ore Body Knowledge) and on the ability to utilise them for accurate slope stability analysis and then pit design. These models have inherent uncertainties caused by data sparsity, measurement errors, modelling scale / bias and dynamic changes in the environment.

[0003] Current surface interpretation practices are deterministic with a broad notion of confidence optionally assigned to the interpretations based on input data. There is a need for alternative modelling techniques.SUMMARY

[0004] An example embodiment describes a method of modelling a geological structure, the method comprising:obtaining geological data for a geological feature, the geological data comprising (i) first measurements corresponding to a first known location of the geological feature, and (ii) second measurements corresponding to at least one of a second known location of the geological feature, an orientation of the geological feature, and an overall orientation of the geological feature; andprocessing the geological data, including the first and second measurements, with a probabilistic model to generate an uncertainty model comprising a probabilistic representation of other locations of the geological feature in three-dimensional space between the known locations.

[0005] In an embodiment, the probabilistic model is a probabilistic field model.

[0006] In an embodiment, processing the geological data with a probabilistic field model comprises generating a potential field and at least one uncertainty value for the potential field.

[0007] In an embodiment, the method comprises generating the uncertainty model from the potential field and the at least one uncertainty value.

[0008] In an embodiment, the geological data comprises a plurality of instances of additional measurements, each instance of the plurality of instances additional measurements corresponding to at least one of a known location of the geological feature, an orientation of the geological feature, and an overall orientation of the geological feature.

[0009] In an embodiment, the method comprises outputting a visualization of the generated uncertainty model to a display.

[0010] In an embodiment, outputting the visualization includes outputting a representation of a mean position of the geological feature and a representation of the uncertainty of the mean position of the geological feature.

[0011] In an embodiment, the representation of the uncertainty includes at least one line or surface representing a defined level of uncertainty.

[0012] In an embodiment, the representation of the uncertainty comprises a heat map representing different levels of uncertainty.

[0013] In an embodiment, the method comprises outputting the visualization together with a visualization of at least one other geological feature.

[0014] In an embodiment, the method comprises outputting the visualization together with a representation of a mine to which the geological feature relates.

[0015] In an embodiment, the probabilistic field model is a Gaussian Process model.

[0016] In an embodiment, the probabilistic field model is a Bayesian Machine Learning model.

[0017] Another example embodiment describes a method of modelling a plurality of geological features, the method comprising:obtaining, for each of a plurality of geological features, geological data defining the geological feature, the geological data comprising (i) first measurements corresponding to a first known location of the respective geological feature, and (ii) second measurements corresponding to at least one of a second known location of the respective geological feature, an orientation of the geological feature, and an overall orientation of the geological feature; andprocessing, for each of a plurality of geological features, the geological data of the respective geological feature with a probabilistic model to generate anuncertainty model comprising a probabilistic representation of other locations of the geological feature in three-dimensional space between the known locations; and combining the generated uncertainty models into a combined uncertainty model.

[0018] Another example embodiment describes a method of generating a warning in respect of a geological feature, the method comprising:comparing an alternative representation of a geological feature with an uncertainty model generated by the method described above; andoutputting a warning upon the alternative representation of the geological feature being at least partially outside the uncertainty model.

[0019] Another example embodiment describes a system for modelling a geological feature, the system comprising:a processor; andmemory storing instructions, which when executed by the processor cause the processor to:obtain geological data for a geological feature, the geological data comprising (i) first measurements corresponding to a first known location of the geological feature, and (ii) second measurements corresponding to at least one of a second known location of the geological feature, an orientation of the geological feature, and an overall orientation of the geological feature; andprocess the geological data, including the first and second measurements, with a probabilistic model to generate an uncertainty model comprising a probabilistic representation of other locations of the geological feature in three-dimensional space between the known locations.BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Embodiments of the technology are described in relation to the following drawings, in which:

[0021] FIGs. 1A to FIG. 1D illustrate the steps of a modelling process of the technology using synthetic data;

[0022] FIGs. 2A to 2F illustrate example behaviours of uncertainty envelopes;

[0023] FIGs. 3A to 3E illustrate the impact of incorporating local orientation information on the modelling process using synthetic data;

[0024] FIGs. 4A to 4C illustrate the propagation of local orientation information across layers;

[0025] FIGs. 5A and 5B illustrate a model of a surface before and after introduction of a fault that cuts through the surface;

[0026] FIG. 6 illustrates combining uncertainties from a plurality of features of interest;

[0027] FIG. 7 is a block diagram of an example system architecture of the technology;

[0028] FIG. 8 is a block diagram of an example workflow;

[0029] FIGs. 9A to 9D illustrate an example of the impact of a parameter guiding uncertainty increase with distance from data;

[0030] FIGs. 10A to 10D illustrate the evolution of uncertainty envelopes as data becomes available corresponding to deterministic interpretations;

[0031] FIGs. 11 A and 11 B show another example of changes of uncertainty envelopes as data becomes available;

[0032] FIGs. 12A to 12C illustrate the evolution of uncertainty heatmaps as orientation data becomes available;

[0033] FIG. 13 illustrates the impact of data quality on uncertainty envelopes;

[0034] FIG. 14 illustrates an example feature interactions chart

[0035] FIGs. 15A to 15D illustrate fault truncation in three and two dimensions;

[0036] FIGs. 16A and 16B show the impact of faults cutting through a feature on the resulting uncertainty heat map;

[0037] FIGs. 17A to 17C illustrate the intersection of fault surface uncertainty envelopes with a pit wall design;

[0038] FIGs. 17A to 17C illustrate the intersection of fault surface uncertainty envelopes with a pit wall design; and

[0039] FIGs. 18A to 18C illustrate the change in an uncertainty model with the addition of data before and after a failure.DETAILED DESCRIPTION

[0040] Example embodiments of the technology are described which use probabilistic modelling to enable near-real time identification, reporting and visualisation of uncertainty envelopes around modelled geological features which may be geological structures or other features. The modelled features may include faults,rock joints, geologic contacts of interest lithologic contacts, bedding planes, alteration zones, etc. In some examples, the uncertainty envelopes are formed by modelling, for any given feature, three surfaces: mean surface (the most likely location of the feature); and two surfaces that demonstrate how much that surface is likely to vary (e.g. one and two standard deviations around the mean surface). In some examples, embodiments enable live three-dimensional visualisation, web-based reporting of orebody knowledge confidence and issuing of alerts, for example, in high-risk situations.

[0041] Example embodiments provide the ability to systematically model and convey the uncertainties in geological features such as fault and geologic contact surfaces for better informed decision making. Example embodiments may provide one or more of the following functionalities, as discussed in more detail below:• Evolution of uncertainty envelopes over time and alerts as interpretations move outside of envelopes.• Multi-source fusion of data with different levels of data quality.• Handling high amounts of geological complexity.• Projecting potential structural locations on pit surface.• A unified uncertainty map• Simulation of realisations of the features representative of the uncertainty envelopes. These realisations can be used in downstream processes, e.g. for computation of the histogram I statistics of the angles (and locations) at which different features meet; or Factor of Safety (FoS) calculation (by accounting for feature variation inside the uncertainty envelope)• Transferring data between related features (e.g. parallel or near parallel features) for better informed modelling (especially when one of the features is well-informed by available data and the other lacks data). Accounting for such data can improve the produced uncertainty model.• Estimating uncertainty in respect of already existing deterministic (i.e. without any uncertainty attached to it) feature interpretation produced using existing industry practices and tools. Existing deterministic interpretations may incorporate expert domain knowledge (e.g. information not explicitly captured in the data) Adding uncertainty modelling to them brings in information on how much variation / uncertainty could be expected around the accepted feature interpretation.• Data-only uncertainty models that estimate the expected location of the feature together with the uncertainty or variation around it. An advantage is that this can be done without having to wait for the current process of generating the interpretations to be completed (e.g. as new data becomes available).

[0042] In some examples, this enables the provision of a model that projects and visualises uncertainty around geological features, communicating risks and guiding design and operational decision making; and / or generates surface models that may be used in stability modelling tools.Gaussian Processes based Machine Learning

[0043] Some embodiments of the technology use machine learning via Gaussian Processes (GPs). Other examples, may use other forms of Bayesian machine learning, conditional random fields or probabilistic extensions of neural networks. Machine learning via GPs is a supervised learning problem that utilises a given training set D = {x^yt}^, consisting of N input points x£e IRLand the corresponding outputs y£e IRLto compute the predictive distribution (%*) at a new test point x5. The relationship between the actual value f and the noisy observation y can be expressed as:y(x) = (x) + E(X) (1) where E is the observation noise.

[0044] A GP model uses a multivariate Gaussian distribution over the space of function variables (x) mapping the input space to the output space. A GP is fully specified by its mean function / z(x) and covariance function fc(x,x'), which is often written as (x) ~ j’( / z(x),fc(x,x')). Using X,f,y) = ({x£}, {£}, {y )^! for the training set and (X*, / *,y*) = ({x*£}, {f^}, {y*£})^i for test points, the joint Gaussian distribution with jtz(x) = 0 becomes:fxl n rKGW + tf2 / K(X, X.)1\ M H K(X, X.(2)In Eq. (2) J\r ( / / , ) is a multivariate Gaussian distribution with mean and covariance and K is the covariance matrix computed between all the corresponding points in the data set. By conditioning on the observed training points, the predictive distribution for new points can be obtained by p(£|;, ) = where= K(X*, X)[K(X, X) + cr2 / ]-^ (3)andX* = K XM - K X„X)[K X, X) + < J2 / ]-1 / <(X, X*) + a21 (4)Probabilistic Modelling of Surfaces

[0045] In order to facilitate understanding of the technology, examples of the technology using synthetic data are described in order to convey properties of the models without complicating the picture with multidimensional complexities present in real-world data.

[0046] In an example, probabilistic modelling of structural surfaces is achieved through 3D modelling of a potential field. The potential field is modelled probabilistically using GPs with a predefined constant value (such as (x) = 0) assigned to the points of the surface being modelled, with values above and below the predefined constant being used for the different sides of the surface (e.g., (x) > 0 above the surface and (x) < 0 below the surface). FIGs. 1 A to 1 D demonstrate the GP-based modelling pipeline on a synthetically created example of two parallel surfaces in order to indicate properties of the model of an example.

[0047] As can be observed from FIGs 1 A to 1 D, in an example, the modelling process uses input data 110 in the form of discrete data points 112, 114 in FIG. 1A corresponding to first and second surfaces. The modelling process estimates the potential field 120 (FIG. 1B). In FIG. 1B, the different shades 121-126 represent different ranges of field values ranging from lowest 121 to highest 126. While shades are used in the figures for illustration, colours would typically be used in practice in order to enable generation of heat maps. FIG. 1 C shows the corresponding uncertainty 130 estimated by the modelling process with the different shades 131-137 representing different ranges of uncertainty from lowest 131 to highest 137. The modelling process then combines the mean and uncertainty of the potential field to produce probabilistic estimates 140 of the surfaces of interest as shown in FIG. 1D. It will be observed that in FIG. 1 D, that for each surface 112, 114 uncertainty is greater in the respective region 144,148 corresponding to the absence of data points in FIG.1 A and increases as the plot goes away from data points.

[0048] The choice of the covariance function in Gaussian Processes combined with the data (and the data quality) defines the shape and behaviour of the estimated function and its uncertainty profile away from the available (known) data. An exampleof probabilistic modelling of structural surfaces of this technology uses a combination of linear and stationary covariance functions. In general terms, the linear component drives the transition of the potential field through the surface and the stationary covariance function drives the behaviour along the surface.

[0049] Modifications to the parameters of the stationary covariance function are used to guide the behaviour of the uncertainty envelopes as shown in FIGs. 2A to 2F, which show respectively: a small observation noise uncertainty envelope 210; a higher observation noise uncertainty envelope 220; a faster growing uncertainty envelope 230; a faster growing at small distances from data and slower growing at larger distances from datauncertainty envelope 240; a near-linear uncertainty envelope 250; and a constant uncertainty envelope 260.

[0050] In some examples, local orientation information (such as data obtained by mapping with orientation or TeleViewer data obtained from instruments run down drilled holes) is incorporated into the model. In an example, this is achieved by making the value of the potential field constant across a short interval along the direction of interest. Another way to achieve similar result may be to set the derivative of the potential field to zero along the provided orientation at the location where the orientation information is provided.

[0051] FIGs. 3A to 3E demonstrate the impact of incorporating local orientation information on probabilistic modelling of surfaces (using 2D synthetic data for demonstration purposes). In FIGs. 3A to 3E, circles represent existing contact data points 301-307 - that is, data defining known locations or, put another way, locations for which there are measurements indicative of location. In this respect, it will be appreciated a “known location” is usually only known with some degree of error and / or uncertainty and is not perfectly known. The degree of error and / or uncertainty is dependent on the quality of the data.

[0052] In FIGS. 3A to 3E, central lines represent the mean of the field generated by the probabilistic modelling process between the contact data points indicate and each of the two upper and lower lines represent uncertainty envelopes of defined, different levels of uncertainty, in this example, one and two standard deviations from the mean surface. It will be appreciated that a different number of uncertainty envelopes could be generated, for example, one or three uncertainty envelopes. In another example, quantiles could be used instead of standard deviations.

[0053] In FIG. 3A, first uncertainty model 310 is based solely on the contact data. FIG. 3B is an enlarged view of a portion 350 of the first uncertainty model, showing a first section 311 of the uncertainty envelope between contact points 301 and 302, and a second section 312 of the uncertainty envelope between contact points 302 and 303.

[0054] In FIG. 3A, second uncertainty model 320 incorporates four instances of local orientation data 324-327 that are processed together with the location data when the uncertainty model is generated. FIG. 3C is an enlarged view of a portion 360 of the second uncertainty model, showing a modified first section 321 of the uncertainty envelope between contact points 301 and 302, and a modified second section 322 of the uncertainty envelope between contact points 302 and 303. A comparison of the first and second uncertainty models, reveals that incorporating local orientation information 324, 325 increases the level of certainty around contact points 302,303 and hence reduces the extent of the uncertainty envelopes 321,322. Further the mean surface shape changes in response to the additional information.

[0055] Third uncertainty model 330 incorporates four instances 334-337 of local orientation data outside (but near to) the contact data 301-307. That is where orientation information is provided from nearby but not at the same 3D location from which the location information for the feature is obtained. For example, data may be available from a feature parallel (but different) to the one of interest and take the observed orientation of the parallel feature to improve the feature of interest. FIG. 3D is an enlarged view of a section 370 of the second uncertainty model. A comparison of the first and third models, shows that incorporating local orientation information 334,335, even when outside the contact data, still increases the level of certainty around the contact points 302, 303 and hence reduces the extent of the uncertainty envelopes 331,332. Further the mean surface shape changes in response to the additional information.

[0056] Fourth uncertainty model 340 incorporates local orientation data 344-347 outside (but near to) the contact data 301-307 but with a larger amount of local orientation data for at least one contact point 344 (for example, from multiple observations). FIG. 3E is an enlarged view of a section 380 of the fourth uncertainty model. A comparison of the first and fourth models, shows that incorporating additional local orientation information further increases the level of certainty around the relevant contact points (here contact point 303) and hence reduces the extent ofthe uncertainty envelopes 341,342. Further the mean surface changes shape in response to the additional information.

[0057] As shown in FIGs. 4A to 4C, orientation data incorporated into one surface can be propagated to neighbouring surfaces. In this respect, the first model 420 of FIG. 4A shows three parallel surfaces, namely top 401, middle 402 and bottom 403 surfaces. As shown in the second model 430 of FIG. 4A, local orientation information 431-433 has been obtained for the top surface and incorporated into a modified model 401 A of the top surface. Detailed views of the first model 400 and second model 450 of are shown side by side in FIGs. 4B and 4C respectively, from which it will be observed that incorporating local orientation data 432 into the upper surface, not only changes, for example, a first section 441 of the uncertainty envelope of the top surface as shown by modified first section 451 of the top surface but also propagates to the middle and lower layers. For example, a first section 442 of the uncertainty envelope of the bottom surface 402 has changed as shown in FIG. 4C to a modified second section 452 the bottom surface 403A, where there is more certainty in the region below the local orientation data 411, and the shape of the mean estimate of the surface has changed.

[0058] In some examples of the technology, modelling in the presence of faults cutting through surfaces is achieved by ensuring the correlations do not propagate along those fault surfaces. This is presented on a synthetic example in FIGs. 5A and 5B. The first probabilistic model 510 of FIG. 5A is a reference model, and the second probabilistic model of FIG. 5B 520 shows a fault surface 525cutting through the surface between first and second contact points 501, 502. As shown in the second model, there is a discontinuity of the mean estimated surface 533, 543 in the region of the fault. For example compare the mean surface 533 on the left hand side of the fault with the mean surface 543 in the right hand side of the fault. There is also an increase in uncertainty around the fault surface as shown by the lines representing one standard deviation above 532, 542 and below 534, 544 the mean, and two standard deviations above 531, 541 and below 535, 545 the mean to the left and right of the fault 525 respectively.

[0059] In some examples, probabilistic models may model multiple datasets with different levels of quality I trust assigned to each of them. In an example, this is achieved by replacing the term a21 (which has the same value of < J2on the entire diagonal of the matrix and zeros everywhere else in the matrix) in Eqs. (2)-(4) by amatrix that has the sai2on the diagonal (and zeros elsewhere) where s ai represents the level of noise (related to data quality I trust) of each data point.

[0060] In some examples, the technology combines uncertainty models from multiple features of interest into a single representation, in order to provide a unified representation of the global uncertainty from the multiple features, thereby conveying areas of high / low risk for better informed decision making. In an example of the technology, this is achieved by analysing the volume before the surface constrained by a pre-defined depth (e.g. 1 / 3 of the depth of the pit) and for each point on the surface computing the proportion of the volume below it that is located within an uncertainty envelope. An example of how a unified uncertainty map 600 is computed from two-dimensional data is shown in FIG. 6. In FIG. 6, first 601 and second 602 surfaces are separated by a distance 605, for example, 45 metres. Mean surfaces are shown for first 611, second 621 and third 631 faults which intersect the first surface 601 at first 610, second 620 and third 630 contact points respectively. Uncertainty envelopes having upper 612, 622, 632 and lower 613, 622, 633 bounds are shown for the first 611, second 622 and third 633 faults. In some examples, the representation of the mean surfaces may be coloured to represent certainty of the location if the mean surface - e.g. coloured blue in areas of higher certainty and changing through a preset range of colours as the certainty decreases, for example, to red in areas of lowest certainty.Example System 700 Architecture

[0061] Figure 7 shows an example system architecture 700. In an example, a computational backend 710 is provided on a cloud computing platform, such as Microsoft Azure, to enable the use of infrastructure and computational capabilities. This enables tools implemented by the computational backend 710 to have access to a dynamically controllable amount of computational resources that can be adjusted with the real-time computational demand, e.g. such that N computational instances 711-713 can be executed concurrently.

[0062] In an example, user access 730 is via a web service 720. In some examples, user access may be provided by a web browser 732. In other examples, user access may be provided via a bespoke portal 734 (e.g. by being integrated into an existing user tool via an API or a plug-in). System storage 740 stores files and a database 744 and again can be provided by a cloud computing platform such asAzure. A connection is established to relevant client databases 750, so that tools running on computational backend 710 have access to relevant data corresponding to features to be modelled and / or monitored for updating existing models, for example, to read and analyse data and update existing probabilistic models as soon as new data become available.

[0063] Accordingly, it will be appreciated that in order to implement the technology, data is obtained. The data may be obtained in a number of different ways, including by retrieving it from a database, retrieving it from a local memory store, receiving it via an upload, etc.Work Flow

[0064] FIG. 8 illustrates an example work flow 800 in which the system 700 is employed.Project setup

[0065] The project setup step 810 of the workflow 800 provides information to the system 700 about the input data sources, defines the physical features to be modelled and the parameters to be used for probabilistic estimation.a) Input data

[0066] Data access: In an example, the system 700 has access data sources defining the geology to be modelled and updates as they become available over time. In some examples, this is achieved through uploading relevant data files, or by providing access to the relevant databases.

[0067] Data format: The system 700 utilises data converters (implemented in the software code) to translate the required data sources into surface location, orientation (optional), data quality / trust (optional) and association between the input data and the corresponding feature to be modelled. The data converters take into consideration the formats in which the information is represented in the data sources.b) Definition of the features and probabilistic modelling parameters

[0068] The parameters for probabilistic modelling are divided into two groups: project-level parameters and parameters at the level of individual features.

[0069] Project-level parameters: The project-level parameters apply to all the features in the project. In an example, the user access 730 to the system 700 provides the capability to override some of those parameters at the level of individualfeatures as needed (see below for further discussion). Example project-level parameters include:• Project name: a label for the project for further reference• Pit wall mapping files: e.g. DXF format files that contain the pit wall mapping polylines, and information that links the polylines to the corresponding features. Any number of pit wall mapping files (e.g., representing different RLs) can be present in a project. RL is the vertical coordinate (that is added to Easting and Northing to define a 3D point).• Mesh resolution: which represents the approximate distance (in meters) between the vertices of the surface meshes to be generated.• Distance (in meters) to extend around data: defines the extents of the surface to be modelled (in the sense of the distance from its data),• A rate parameter guiding the uncertainty increase with distance from data: guides how fast the uncertainty envelope should widen as going away from the data points.

[0070] In this respect, as shown in FIGs 9A to 9D, larger values of the rate parameter result in slower increase in uncertainty. Different values for the rate parameter guide the uncertainty increase (with distance from data) and the corresponding cross-section plot of the uncertainty envelope here shown for one standard deviation 912, 914 and two standard deviations 913, 915 relative to the mean 911. For example, FIG. 9A shows the one standard deviation 912A, 914A and two standard deviations 913A, 915A where the rate parameter = 210; FIG. 9B shows the one standard deviation 912B, 914B and two standard deviations 913B, 915B where the rate parameter = 140; FIG. 9C shows the one standard deviation 912C, 914C and two standard deviations 913C, 915C where the rate parameter = 70; and FIG. 9D shows the one standard deviation 912D, 914D and two standard deviations 913D, 915D where the rate parameter = 35.

[0071] Feature-level parameters: The feature-level parameters define the probabilistic modelling task for each individual feature’s surface and the uncertainty envelope around it. Example feature-level parameters include:• Feature’s name: a label for the feature for further reference• Interpreted surface: the corresponding triangulated surface representing the client’s interpretation of the feature.Feature orientation: the general orientation of the surface represented by the rotation around the axes Z and Xo Note: if the general orientation is not specified then it is automatically calculated using the provided client’s interpretation.• Mesh resolution: represents the approximate distance (in meters) between the vertices of the surface meshes to be generated.• Polylines: represents the polyline indexes in the pit wall mapping files (using the pit wall mapping files provided in the project-level parameters). Any number of polyline indexes from any number of pit wall mapping files can be specified.• Truncator surfaces: the names of the features that truncate the current surface being modelled.• Run modelling?: a parameter defining whether a probabilistic modelling task needs to be executed for this surface.

[0072] In this embodiment, the parameter “Mesh resolution” is present at both the project level and feature level. In an example, the project level specification of the mesh resolution is used as the default value, with the possibility to overwrite it for any individual feature. In other examples, other project level parameters which can may be overwritten by a feature level parameter such as the rate parameter guiding the uncertainty increase with distance from data.Data pre-processing and QA / QC

[0073] During the data pre-processing step 820, the system 700 utilises access to the input data sources (such as databases and / or data files) to obtain input datasets and transform them using the available converters (as discussed above). The outcome is that the data is converted from an original format (that can vary between different data sources and between different mine sites) into a pre-defined format that can be directly utilised by the system 700 for probabilistic modelling. If there are any data format inconsistencies, those issues will be identified and brought to the user’s attention (e.g. in the form of an irregularity list presented on the user interface). It will be appreciated that data converters are not needed in the case that data formats are standardized.

[0074] In some examples, selected data quality checks 830 can be executed on the pre-processed data. Suspect data identified during this automated process (such as contradictory data, isolated data points, etc.) may be brought to the user’s attention via the user interface for further analysis and decision making. The subject matter experts (SMEs) can also utilise the user interface 730 to interactively interrogate the pre-processed data within the context of other information sources (such as existing geological interpretations and other data sources) to identify any suspicious data. The identified erroneous data may then be removed from the modelling process and / or low-quality data may be marked as high uncertainty to reduce their impact on the probabilistic surface model to be produced.

[0075] The decisions made on data quality and erroneous data may be stored 835 in the database 744 and fed back into the data pre-processing step 820. The outcome will be the availability of validated data (in the format required by the system 700) available for probabilistic modelling. As the information about any identified erroneous and low-quality data are be stored in the database 744, those identifications can be accessed and utilised in other projects that use the same data sources.

[0076] Manual data validation will only be needed if the input data sources contain erroneous data that cannot be identified through the available autonomous means, or, there are low quality data points that need to be manually identified and assigned their level of quality I trust.

[0077] In situations where the input data sources are well maintained I cleaned (and therefore don't contain erroneous data) and the level of quality I trust of the data are known, there is no need for steps 830 and 835.Probabilistic modelling and computation of the variables of interest

[0078] The pre-processed data is fed into the automated process 840 of generating probabilistic models for the features of interest. For each feature of interest, the system 700 will produce its mean triangulated surface (marked to indicate the estimated uncertainty, e.g. by using a heatmap having different colours) and two uncertainty envelopes representing one and two standard deviations from the mean triangulated surface, respectively. In an example, a heatmap presented on the mean surface, represents the width of the uncertainty envelope around that mean surface.

[0079] Once the mean surfaces and uncertainty envelopes for all the features of interest are calculated, the truncation rules will be applied to them.

[0080] The resulting probabilistic estimates of all the features are then utilised to compute 860 the user selected (or default) variables of interest 850, such as• an overall combined uncertainty heatmap shown on the pit shell,• statistics of the intersection angle between the features of interest and the pit shell,• statistics of the intersection angle between different features, and• other quantities of interest as specified by the user.3D visualisation and reporting

[0081] The system 700 provides an interactive web-based 3D visualization 874 capability for the purpose of quick visualisation and validation of the results as well as enabling the results to be exported for reporting 872. For example, the web browser 732 may display the pit shell and the probabilistic models of any number of features (coloured by the uncertainty heatmap) so that they can be visualised together. In some examples, transparent visualisation may be provided that enables a user to see through the features and their corresponding uncertainty envelopes, so that, for example, the mean surface is not occluded by the uncertainty envelopes. In some examples, a plugin may be provided to load the resulting surfaces (including the uncertainty envelopes) into other software toa) interrogate the results within the context of a wider range of data sources, and,b) feed the outcomes back into the workflows.Decision making and closing the loop

[0082] In some examples, the analysis of the produced models by the SMEs and their utilisation within workflows may enable better informed decision making 880. Due to the automated nature of this capability, in some examples, updated models can be automatically generated as mining progresses and new data becomes available. In some examples, probabilistic models can help improve the decision making in various ways such as:• providing live information on the areas of higher risk,• issuing alerts if new data conflicts with the model, and• helping better utilise the data collection and modelling efforts by focusing them on high-risk areas.

[0083] Information and data resulting from the analysis and utilisation of the produced models is fed back into the pre-processing stage and may be utilised together with other data sources to produce better models.Example FunctionalityFunctionality 1: Evolution of uncertainty envelopes over time.

[0084] FIGs 10A to 10D illustrate an example of the evolution of the uncertainty envelopes over time as more pit wall mapping data becomes available. FIGs. 10A to 10D show (in a NW-SE section) the evolution over time of the uncertainty envelopes for a mine fault, and a comparison of those uncertainty envelopes to interpretations of the same surface produced using conventional techniques during the years 2019, 2020 and 2021. FIG. 10A shows the pit shell 1002 and toe 1004 and a first conventional interpretation 1012 from May 2019, a second conventional interpretation 1014 from April 2020, and a third conventional interpretation 1016 from February 2020.

[0085] FIG. 10B shows an uncertainty envelope developed using the above example of the technology based on the data available at the time of the first conventional interpretation, including the mean estimate 1021, upper bound 1022 and lower bound 1023 of the uncertainty envelope. In FIG. 10B all three conventional interpretations 1012,1014,1016 are contained within the upper and lower bounds of the uncertainty envelope 1022,1023, which is wide due to lack of data and conveys the high uncertainty in the fault model.

[0086] As new data becomes available over time, the uncertainty envelopes become tighter. FIG. 10C shows an example of an updated uncertainty envelope based on the data available at a later time, including the mean estimate 1031, upper bound 1032 and lower bound 1033 of the uncertainty envelope.

[0087] FIG. 10D shows a further example of an updated uncertainty envelope based on the data available at a still later, including the mean estimate 1041, upper bound 1042 and lower bound 1043 of the uncertainty envelope. It will be observed that, in FIG. 10D, the April 2020 conventional interpretation 1014 is now outside of the envelope, thus highlighting the need for an update of the corresponding interpretation, including the angle of the fault and the point of its intersection with the toe 1004. The uncertainty of the estimate going into the walls of the pit is also significantly reduced. Accordingly, it will be appreciated that early knowledge of the structural uncertaintiescan enable more optimal mine operation with an improved risk profile. In this respect, in some examples, the models will be updated automatically as new data becomes available - helping identify any inconsistencies (e.g. the current conventional model getting outside of the uncertainty envelope) early on, rather than waiting for a lengthy period of time until enough data is collected (and people have enough available time) to generate a complete new interpretation. FIGs. 11 A and 11 B show a three-dimensional example of updating uncertainty models for a pit design 1110. FIG. 11A shows a first model 1130 of and uncertainty envelope relative to the pit design 1110, here showing two standard deviations above the mean 1111, one standard deviation above the mean 1112, the mean 1113, one standard deviation below the mean 1114, and two standard deviations below the mean 1115. Additional data is then acquired to generate a second, updated model 1120A as shown in FIG. 11B. Updated model 1120A shows updated uncertainty envelopes 1111 A-1115A relative to the pit design 1110A, here two standard deviations above the mean 1111 A, one standard deviation above the mean 1112A, the mean 1113A, one standard deviation below the mean 1114A, and two standard deviations below the mean 1115A.Functionality 2: Multi-source fusion with different levels of data quality.

[0088] In some examples, system 700 may be used to model surfaces by fusing data from different sources. FIGs. 12A to 12C show the evolution of a probabilistically estimated surface 1220 and its uncertainty heatmap as the model 1210 is initially produced using location data, then orientation data is acquired and incorporated into an updated model 1210A (FIG. 12B) and then the model 1210B is further updated as more orientation data is acquired (FIG. 12C). Note the reduction of uncertainty around the newly incorporated data 1231, 1232, 1233, and the dynamic evolution of the shape of the surface 1220A, 1220B, honouring both location and orientation data.

[0089] The ability of example embodiments to account for the data quality is shown in FIG 13 which shows uncertainty envelopes for low-quality data 1310, mid-quality data 1320 and high-quality data 1330. As it can be observed in FIG. 13, low quality data 1310 results in wider uncertainty envelopes (e.g. in region 1312A) while increasing the data quality leads to tighter envelopes (e.g. in regions 1312B, 1312C). This behaviour is in-line with the requirements, as lower quality data provides less specificity about the location of the feature while higher quality data must be honoured more precisely by the tools. Examples of high-quality data include televiewer data withorientations mapped to faults with a high degree of certainty, and pit wall mapping data collected in the presence of good visibility of the exposed features. Examples of low-quality I less-precise data include pit wall mapping data collected in the presence of significant amount of dust on the walls, and RC drilling-based identifications where the surface is identified to pass through a 2 m sample (without more specific knowledge of the location within that sample)Functionality 3: Handling high amounts of geological complexity.

[0090] In some embodiments, the system 700 can handle high amounts of geological complexity. In this respect, as shown in FIG.14, in some examples, system 700 can map interactions and chronology of geological features via an interactive chart visualization 1400 where each node (e.g. node 1401) represents a fault and the arrows (e.g. arrow 1420) represents the relationship between the faults.

[0091] As shown in FIGs. 15A to 15D, in some examples, system 700 can take into account the truncation rules between all the faults. FIGs. 15A and 15B show a fault 1520 extending through the mean estimated surface and the corresponding uncertainty envelope in three-dimensions 1510 (FIG. 15A) and two-dimensions 1510A respectively (FIG. 15B). FIGs. 15C and 15D show the truncated estimated surface and the corresponding uncertainty envelopes in three-dimensions 1530 (FIG. 15C) and two-dimensions 1530A respectively (FIG. 15D).

[0092] In some examples, system 700 can model geologic contacts in the presence of one or more faults cutting through them. For example, FIG. 16A shows a model 1610 of a geologic contact without faults. FIG. 16B shows an updated model 1620 of the same geologic contact with the addition of faults 1621-1624.Functionality 4: Projecting potential structural locations on pit surface.

[0093] In some examples, system 700 can project potential fault locations on the pit surface as shown in FIGs. 17A - 17C, which enables the calculation of the statistical distribution of angles of intersection between the faults of interest and the pit wall, and angles at which different faults meet each other.

[0094] Knowledge of the statistics of these angles can be a key contributor to the pit wall failure risk evaluation.

[0095] Figure 17A illustrates an example model 1700 of a pit surface and an intersection of fault surface uncertainty envelopes of a first fault 1710 and a secondfault 1720 in region 1730 (mean, ±1 and ±2 standard deviations) with the pit wall design. FIGs. 17B and 17C show the intersection of the fault simulations (defined by the uncertainty envelopes) with the pit wall design in (a) a cross-section 1740 (FIG.17B) which shows the uncertainty envelopes 1710A and the surface realisations 1712; and (b) in three-dimensions which shows region 1730 in more detail where the first fault 1710 and second fault 1720 intersect.Functionality 5: Unified uncertainty map.

[0096] As mine pits can have a large number of fault surfaces, in some examples, the system combines uncertainties from all the structural surfaces of interest into a single unified representation. This unified uncertainty is visualised as a heatmap on the pit surface. As in the above examples, shading or colours can be used to represent different levels of uncertainty. Ffor example, where blue, green, red and grey colours represent a low, medium, high uncertainty and absence of a fault model, respectively. However, FIGs. 18A and 18C have been simplified by using a single type of shading to show a single certainty level of data in order to show how data can evolve.

[0097] FIG. 18B shows an example location 1822 of a failure in a mine pit. An example of unified uncertainty heatmap model 1810 generated using the data available before the failure is presented in FIG. 18A, and demonstrates the lack of data for area 1813 where the failure compared to other areas 1811,1812. In contrast, the updated unified uncertainty heatmap model 1830 of FIG. 18C produced using data available after the failure shows a significant increase in the acquired information in the failure area 1813A since the failure took place. In this examples, viewing the heatmap model of FIG. 18A would have revealed the lack of data at the location where the failure took place before it actually took place and acted as a trigger or warning to obtain further data which may have enabled prevent or mitigate it.

[0098] Availability of the system functionality before the failure took place would have highlighted the lack of knowledge in the area where the pit wall failure occurred 1813 compared to other areas 1811,1812, prompting the need to address the corresponding risks, e.g., through acquisition of relevant data in the high-risk area to better understand the risks and operate accordingly.

[0099] The unified uncertainty map model thus represents a visual summary of complex fault interactions, which allows for the communication of risks to guide design and operational decision making in:• Design optimisation• Definition of drilling targets• Definition of mapping targets• Monitoring and operational risk management strategies• Dewatering strategies.

[0100] Unless the context clearly requires otherwise, throughout the description and the claims, the words “comprise”, “comprising”, and the like, are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense, that is to say, in the sense of “including, but not limited to”.

[0101] Reference to any prior art in this specification is not, and should not be taken as, an acknowledgement or any form of suggestion that that prior art forms part of the common general knowledge in the field of endeavour in any country in the world.

[0102] Where in the foregoing description reference has been made to integers or components having known equivalents thereof, those integers are herein incorporated as if individually set forth.

[0103] It should be noted that various changes and modifications to the presently preferred embodiments described herein will be apparent to those skilled in the art. Such changes and modifications may be made without departing from the spirit and scope of the technology and without diminishing its attendant advantages. It is therefore intended that such changes and modifications be included within the present technology.

Claims

CLAIMSWhat is claimed is:

1. A method of modelling a geological feature, the method comprising:obtaining geological data for a geological feature, the geological data comprising (i) first measurements corresponding to a first known location of the geological feature, and (ii) second measurements corresponding to at least one of a second known location of the geological feature, an orientation of the geological feature, and an overall orientation of the geological feature; andprocessing the geological data, including the first and second measurements, with a probabilistic model to generate an uncertainty model comprising a probabilistic representation of other locations of the geological feature in three-dimensional space between the known locations.

2. The method as claimed in claim 1, wherein the probabilistic model is a probabilistic field model.

3. The method as claimed in claim 2, wherein processing the geological data with the probabilistic field model comprises generating a potential field and at least one uncertainty value for the potential field.

4. The method as claimed in claim 3, further comprising generating the uncertainty model from the potential field and the at least one uncertainty value.

5. The method as claimed in any one of claims 1 to 4, wherein the geological data comprises a plurality of instances of additional measurements, each instance of the plurality of instances additional measurements corresponding to at least one of a known location of the geological feature, an orientation of the geological feature, and an overall orientation of the geological feature.

6. The method as claimed in any one of claims 1 to 5, further comprising outputting a visualization of the generated uncertainty model to a display.

7. The method as claimed in claim 6, wherein outputting the visualization includes outputting a representation of a mean position of the geological feature and a representation of the uncertainty of the mean position of the geological feature.

8. The method of claim 7, wherein the representation of the uncertainty includes at least one of a line and surface representing a defined level of uncertainty.9 The method of claim 7 or claim 8, where the representation of the uncertainty comprises a heat map representing different levels of uncertainty.

10. The method of any one of claims 6 to 9, comprising outputting the visualization together with a visualization of at least one other geological feature.

11. The method of any one of claims 6 to 10, comprising outputting the visualization together with a representation of a mine to which the geological feature relates.

12. The method of any one of claims 1 to 11, wherein the probabilistic field model is a Gaussian Process model.

13. The method of any one of claims 1 to 11, wherein the probabilistic field model is a Bayesian Machine Learning model.

14. A method of modelling a plurality of geological features, the method comprising:obtaining, for each of a plurality of geological features, geological data defining the geological feature, the geological data comprising (i) first measurements corresponding to a first known location of the respective geological feature, and (ii) second measurements corresponding to at least one of a second known location of the respective geological feature, an orientation of the geological feature, and an overall orientation of the geological feature; andprocessing, for each of the plurality of geological features, the geological data of the respective geological feature with a probabilistic model to generate anuncertainty model comprising a probabilistic representation of other locations of the geological feature in three-dimensional space between the known locations; and combining the generated uncertainty models into a combined uncertainty model.

15. A method of generating a warning in respect of a geological feature, the method comprising:comparing an alternative representation of a geological feature with an uncertainty model generated by the method of any one of claims 1 to 14; and outputting a warning upon the alternative representation of the geological feature being at least partially outside the uncertainty model.

16. A system for modelling a geological feature, the system comprising:a processor; andmemory storing instructions, which when executed by the processor cause the processor to:obtain geological data for a geological feature, the geological data comprising (i) first measurements corresponding to a first known location of the geological feature, and (ii) second measurements corresponding to at least one of a second known location of the geological feature, an orientation of the geological feature, and an overall orientation of the geological feature; andprocess the geological data, including the first and second measurements, with a probabilistic model to generate an uncertainty model comprising a probabilistic representation of other locations of the geological feature in three-dimensional space between the known locations.

17. A tangible, computer readable medium comprising instructions which when executed by a processor, implement the method of any one of claims 1 to 15.

Images