CA3035734C - A system and method for estimating permeability using previously stored data, data analytics and imaging - Google Patents

Abstract

A system and method are provided for estimating permeability of a formation from a formation sample. The method includes acquiring an image of the formation sample and comparing the acquired image with one or more stored images stored in an image database to associate the acquired image with one of the one or more stored images and corresponding stored data. The method can also include determining a permeability measure associated with the formation sample from the stored data and the acquired image by identifying an extracted parameter from the acquired image and using a predetermined permeability model and predetermined porosity data, or determining a permeability measure associated with the formation sample from the stored data and the acquired image by identifying a class and using a predetermined permeability-porosity relationship for that class and predetermined porosity data.

Description

A SYSTEM AND METHOD FOR ESTIMATING PERMEABILITY USING PREVIOUSLY
STORED DATA, DATA ANALYTICS AND IMAGING
TECHNICAL FIELD
[0001] The following relates to systems and methods for estimating permeability using previously stored data, data analytics and imaging, particularly in analyzing core samples to determine permeability of a formation.
BACKGROUND

[0002] Permeability is a measure of the ability of a porous material to allow fluids to pass through the material. Permeability is a difficult parameter to estimate in areas such as the McMurray oil sands in Canada, since it has been found that traditional measurement methods either do not work, do not work well, or are otherwise unsuitable. Typical well tests carried out in the conventional oil and gas business where a fluid is either produced or injected into the reservoir do not work in the oil sands because the bitumen is solid at reservoir conditions.
Current well log-based methods applied in the conventional oil and gas industry are also found to be unsuitable because the logs typically cannot detect variations in grain size, which control permeability. Permeability estimates taken from core floods in the lab are considered to be very expensive and are problematic because of sample disturbance during core recovery and from cleaning of bitumen from the core. Permeability can also be estimated from particle size distribution using the Carman-Kozeny equation. However, taking many physical samples along the length of the drill core and from a large number of drill cores present at a typical SAGD field site is expensive and time consuming. The current approach involves obtaining physical core samples, carrying out a Particle Size Distribution (PSD) analysis, and calculating permeability from the PSD results.

[0003] The PSD of a soil, granular material, or reservoir includes a list of values, or, a mathematical function that defines the relative amount, typically by mass, of particles present according to size. The PSD of a material can be important in understanding physical and chemical properties such as permeability. The PSD can affect the strength, load-bearing properties and the reactivity of rocks and soils.

[0004] There are multiple techniques known to measure the PSD of a given rock formation such as a sieve, laser diffraction, and a photoanalysis. A sieve analysis is considered to be a common method of determining PSD, in which the granules are separated based on sieves of different sizes. Laser diffraction is also considered to be a common method where a stream of 23594333.1 - 1 -particles in air or liquid are "streamed" past a laser, and the PSD is measured from the diffraction pattern of the laser beam.

[0005] PSD is typically defined in terms of discrete size ranges: e.g.
"percent/portion of samples between 0 pm and 50 pm, percent/portion of samples between 50 pm and 100 pm, etc." The PSD is usually determined over a list of size ranges that covers nearly all the sizes present in the sample, and PSD data is generally represented with a histogram.

[0006]

[0007] Permeability can be estimated from the PSD results using a form of the Carman-Kozeny equation (as suggested by Panda and Lake (1994)):
1000*DpO3 K 7--- or Kc* SF* r*(1-02Y
k= ________________________ TYA53 [(yap + 3c2;,p + 1)2 I
72- r Fs(1 ¨ 0)2 (1 + ap)2

[0008] In the equation above, Dp is the mean diameter, 4) is the porosity, r is the tortuosity, Kc is the Kozeny constant, SF is the shape factor, K is the permeability, y is the skewness, and CDp is the coefficient of variation, which is the standard deviation divided by the mean grain size.
Porosity (4)) is the ratio of the pore volume to the bulk volume of porous media; tortuosity (r) is the ratio of the average flow path length to the sample length, assumed constant; Kc is assumed to be 72; and SF accounts for angularity of the particles.

[0009] One method of calculating the permeability of a material uses bulk physical properties such as porosity, and particle size distribution. Specifically, media with different particle size distributions have different permeability-porosity relationships, and a medium's particle size distribution and porosity can be used to estimate its permeability. The permeability of a medium depends on both its porosity and its particle size distribution, and thus the particle size distribution is required to identify its permeability-porosity relationship. The particle size distribution of a given material, such as mean grain size and standard deviation, can be used with the porosity to calculate the permeability of the medium.

[0010] This method requires statistical values to calculate permeability and thus, a full statistical analysis of the particle size distribution is required. A full statistical analysis can be 23594333.1 completed by conducting a full PSD analysis which can be computationally intensive, time consuming and expensive particularly for the large number of cored wells at a typical SAGD
field.
SUMMARY

[0011] By analyzing an image of a drill core or rock sample, and comparing to previously stored data, permeability can be determined. The sample image can be analyzed using a mathematical technique to identify an image in a database with at least one similar property.
Permeability can be estimated using a previously determined permeability-porosity relationship associated with that database image and porosity data associated with the image, or by performing a permeability prediction directly based on an extracted parameter from the image.
The permeability-porosity relationship can be determined using, for example, machine learning techniques.

[0012] In one aspect, there is provided a method of estimating permeability of a formation from a formation sample, the method comprising: acquiring an image of the formation sample;
comparing the acquired image with one or more stored images stored in an image database to associate the acquired image with one of the one or more stored images and corresponding stored data; and determining a permeability measure associated with the formation sample from the stored data and the acquired image by identifying an extracted parameter from the acquired image and using a predetermined permeability model and predetermined porosity data.

[0013] In another aspect, there is provided a method of estimating permeability of a formation from a formation sample, the method comprising: acquiring an image of the formation sample; comparing the acquired image with one or more stored images stored in an image database to associate the acquired image with one of the one or more stored images and corresponding stored data; and determining a permeability measure associated with the formation sample from the stored data and the acquired image by identifying a class and using a predetermined permeability-porosity relationship for that class and predetermined porosity data.

[0014] In other aspects, there are provided computer readable media storing computer executable instructions for estimating permeability of a formation from a formation sample according to the above methods.

23594333.1

[0015] In other aspects, there are provided systems for estimating permeability of a formation from a formation sample, the systems comprising an image acquisition device; an image database; and an image processing module executing computer executable instructions for performing the above methods.

[0016] The systems and methods described can provide an alternative to techniques for estimating the permeability of a rock sample that require completing a full PSD analysis. The systems and methods can identify sample image data using various machine learning and pattern recognition techniques and use the identified data to estimate the permeability of the rock formation.
BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Embodiments will now be described with reference to the appended drawings wherein:

[0018] FIG. 1 illustrates a system for obtaining an image of a drill core using a handheld device for subsequent analysis to estimate permeability of the drill core;

[0019] FIG. 2 is a flowchart illustrating operations used in estimating permeability from an image of a drill core and from a porosity estimate from well log data;

[0020] FIG.3 illustrates a conversion from an original image to an image showing geological textures for comparison against images from an image database;

[0021] FIG. 4 is a permeability versus porosity chart in which classes are defined;

[0022] FIG.5 is a permeability versus porosity chart in which each sample class is identified;

[0023] FIG. 6 is a flow chart illustrating operations used in determining a prediction model for classifying sample permeability;

[0024] FIG. 7 illustrates schematically the generation of the image database;

[0025] FIG. 8 illustrates schematically the determination of permeability either directly based on an extracted parameter from the image, or by a permeability-porosity model and porosity estimated by well log data; and

[0026] FIG. 9 illustrates a gamma log in identifying clustering of neighbours.

23594333.1 DETAILED DESCRIPTION

[0027] A system and method are described herein for estimating permeability of a sample of a rock formation. The method includes image capturing, image comparison, porosity estimation, and permeability estimation directly based on an extracted parameter from the image or through image class assignment. The system can identify a parameter such as large particles from an image of a sample of the rock formation, compare the identified parameter to a database of known parameters, and either perform a permeability prediction directly from the data based on an extracted parameter from the image, or apply the permeability-porosity relationship to estimate the permeability of the rock formation using that relationship and porosity estimated by well logs. It may be noted that permeability-porosity models described herein represent different geological facies. Each geological facies has a characteristic grain size, characteristic permeability-porosity relationship, and characteristic sedimentary structure.
As such, the permeability-porosity models described herein may differ for different geological facies.

[0028] The system described can provide an alternative to techniques for estimating the permeability of a rock sample that require completing a full PSD analysis. The system can identify sample image data using various machine learning and pattern recognition techniques and use the identified data to estimate the permeability of the rock formation, as described in greater detail below.
System Configuration

[0029] Turning now to the figures, FIG. 1 illustrates a system 50 that is used to analyze a drill core 101 by imaging the drill core 101. In this example, the drill core 101 includes particles 102 having a varied PSD. In this illustrative example, the drill core 101 is drilled from the surface 100 of a formation 80 and can be transported to another location for imaging (as illustrated) or analyzed in situ. A handheld camera, hyperspectral imaging, or any other imaging-capable device 103 can be used to take an image 104 of the drill core 101. The image 104 is electronically sent, saved or otherwise transferred to a data collection module 105. The data collection module 105 transfers the image 104 to an image processing module 106, where the image 104 is compared to data stored in an image database 109. As will be discussed in greater detail below, data regarding porosity and class is predicted using pre-defined permeability models, and permeability can be determined and provided as an output, e.g., via an output module 107 to a data display device 108. The data display device 108 can include 23594333.1 the display of an electronic device providing a visual representation of the permeability determined or can also represent a report or other textual output. The system 50 can be configured as shown to provide a data processing system 110 that includes the data collection module 105, the image processing module 106, the image database 109, the output module 107 and the data display device 108. The data processing system 110 can be embodied as a custom device, an embedded device, a software program executed on a general purpose computing device, or any other digital device, system or electronic service.

[0030]
FIG. 2 illustrates operations performed by the system 50 in determining permeability from an image 104, including pre-processing operations. A drill core image 104 is captured at stage 200 by the imaging device 103, e.g., using high-resolution hyperspectral imaging or a camera, or both. The image 104 is obtained or provided to the data processing system 110 to be compared with images from the image database 109 at stage 202. This includes applying one or more mathematical techniques characterizing the image 104 to identify a parameter, such as what proportion of large particles exist (as explained by way of example below), textural information in the images 104, patterns in the grains from characteristic sedimentary structures, or other mathematical patterns including those difficult to observe. Porosity can be estimated at stage 203, from geophysical well logs that would have been acquired along with the drill core or rock sample having an image in the image database 109. A classifier can be assigned to each sample using pre-generated classification models at stage 204. The term classifier as described herein refers to a mathematical function that is implemented by a classification algorithm. Classification refers to splitting a data set into subclasses. The permeability of the sample can then be determined at stage 206, by conducting a permeability estimation operation using the permeability-porosity relationship associated with that database image and estimation of porosity, or by predicting permeability directly using a developed model based on an extracted parameter from the image 104. Optionally, a Kv/Kh ratio can also be determined at stage 216, that is, the ratio of between vertical and horizontal permeability associated with the permeability-porosity model. It can be appreciated from FIG. 2 that stage 202 operates using data from the image database 109, which is populated during a first pre-processing stage 208.
Likewise, stage 204 operates using data obtained through generating permeability models, during a second pre-processing stage 210. Moreover, the permeability determined at stage 206 can optionally be refined at stage 212 using other data such as gamma logs. As such, a third pre-processing stage 214 can be implemented wherein such other data is obtained. Further 23594333.1 details concerning the stages 200-216 are provided below. It can also be appreciated that the gathering of the well log data, used in stage 203, and from which porosity can be estimated, can also be considered a pre-processing stage.
Image Capture

[0031] The process of determining the permeability of a rock formation begins by taking at least one image 104 of a sample of the rock formation, such as a drill core 101 illustrated by way of example in FIG. 1. The image 104 can be a high-resolution image or an image 104 having a standard or typical resolution, e.g., from a standard camera or camera-equipped electronic device. Images of the drill core, drill cuttings, or other rock sample can be taken at any specified depth interval. It can be appreciated that the sample of the rock formation as described herein can include both drill core samples and drill cuttings. As is known in the art, a drill core sample is a cylindrical section of a naturally occurring substance such as sediment or rock. Core samples are obtained by drilling into the substance using a core drill. Drill cuttings refer to the smaller pieces of rock expelled as a result of drilling a core.
Either drill core samples or drill cuttings can be used for imaging and estimating the permeability of the rock formation from which they are extracted.

[0032] Techniques which can be used to take an image of a drill core 101 can include, for example: a Computed Tomography (CT) scan, a Computer Axial Tomography (CAT) scan, hyperspectral imaging, drone scan imaging, a digital image camera, downhole image logs, etc.
Scanning technology can be valuable in providing information about the internal structure and saturation distributions within core materials. CT scanning and CAT scanning involves the use of computer-processed combinations of many X-ray measurements taken from different angles to produce cross-sectional (tomographic) images of specific areas of a scanned object, allowing the user to observe inside the object without cutting. Hyperspectral imaging, like other spectral imaging, collects and processes information from across the electromagnetic spectrum. Drone scan imaging is an imaging technique that can be accomplished using aerial drone image capturing. A digital image is a hand-held imaging technique in which the digitally encoded visual characteristics of an object are represented. The use of handheld imaging devices or downhole imaging tools can be particularly advantageous due to their portability and ease of use in the field.

[0033] The image type or format generated by the particular imaging technique may require conversion to a type or format which can be recognized by the computer matching 23594333.1 software being applied in stage 202. The techniques used to scan the drill cores 101 can output image types such as: binary (black or white pixels), indexed, grayscale or truecolour (RGB) images. Image types which can be recognized by the software can be any one or more of these image types, as well as other image types.
Image Comparison

[0034] After the image 104 of the drill core 101 (or drill cutting or other rock sample) has been captured and converted to an acceptable format, the image 104 is compared to images which exist in the image database 109. The image database 109 is populated during the first pre-processing stage 208 as indicated above.

[0035] Each input image 104 and image in the image database 109 is acquired from a sample of rock formation and thus the images 104, 109 include one or more parameters that can be characterized using one or more mathematical techniques. In the example provided in FIG. 3, the images 104, 109 illustrate a plurality of particles of varying particle sizes that can be used as a basis for comparison. In general, the particle size distribution of the particles in the images can be normally distributed. Each database image 109 can also have a class with which it is associated.

[0036] Computerized software can be used to compare the converted image 104 (input image) with an image in the image database 109. Matching techniques or comparison methods can be used to identify which image from the database 109 is most similar to any given drill core image 104. Matching techniques used by the computer software can include, for example:
pattern recognition, neural networks, Scale-Invariant Feature Transform (SIFT), and keypoint matching.

[0037] For retrieving the images 104 by query, a library of core plug images 109 are prepared as discussed above and can be correlated with measured PSDs (or permeability) and/or interpreted rock categories which are obtained from an analysis of previous (analogue) wells. This analysis and interpretation can provide a rich source of information that can assist the system 50 in classifying various core plugs 101 by extracting the hidden features in their images 104 (e.g., photographs), which may or may not be visible to the human eye. That is, the system 50 can train one or more machine learning algorithms by learning from previous analyses. However, rather than compare the pixels of the images 104 which may not be possible when the pictures have different resolutions, size and quality, in the comparison stage 23594333.1 202, several methods can be used to extract the features of interest. These methods can include, without limitation, Gabor filter, local binary patterns, and gray level co-occurrence matrix methods. These methods can explain a given image 104 in terms of a vector with a set of values that can highlight the particular features of any core image 104.

[0038] Having constructed the database of images 109, the system 50 can pass a new query image 104 where no information about its permeability is known. This example comparison method at stage 202 can decompose the query image 104 using similar feature extraction methods as those used in generating the database of images 109.
Therefore, the query image 104 can efficiently search the database of images 109 and find the similar images 109 using a number of similarity measures such as L2 Norm, L1 Norm, Mutual Information etc.

[0039] This set of similar images 109 can be used to estimate the average permeability and predict the class that this new image belongs to, e.g., as shown in FIG. 3 and/or as described in greater detail below. This prediction can be achieved either by using the similarity measures (i.e., an unsupervised algorithm) or by finding a classification function that can relate the extracted features with their interpreted categories (i.e., a supervised algorithm). In the former case, the system 50 compares the extracted features with each other without looking at to which category the images 104, 109 belong. However, for the latter case, the system 50 can attach a label (i.e., a prior interpreted category) to each image 104, 109 and attempt to use both features and labels to train a general function that can be used in future to predict (and attach) a label for any new image 104. Several methods such as Support Vector Machine, Naïve Bayes, and Discriminant analysis can be used for this purpose.

[0040] The system 50 can also include a clustering module to provide a high-level general classification to a pool of new data and visualize the new data in a scatter 2D graph. In doing so, the features can be extracted for every core plug image in the database 109. The feature vector for any image 109 can have a high dimension (e.g., around 180 elements). Therefore, a set of linear and non-linear dimension-reduction techniques such as principle component analysis (PCA) can be implemented to reduce the dimension of the feature vector to a lower dimension (e.g., 2). As a result, the feature vector can be described in terms of only a few numbers. After this, some clustering algorithms such as k-means clustering can be used to represent the pool data in terms of a few clusters (or categories). In this way, the clustering module can be used to efficiently find various categories in the data.

23594333.1

[0041] It can be appreciated that the comparison techniques described above can also be used for estimating permeability from image logs rather than using the core plug images 104.
The patterns extracted from image logs could also be used to correlate with permeability by performing a classification, potentially reducing the cost of coring.

[0042] FIG.3 schematically illustrates a process of image comparison in which the parameter being analyzed is the number of coarse grains. In this example, an original RGB
image 301 is converted to an input image 302. The input image 302 can be used to compare with images 304, 305, and 306 from the image database 109 (three images being shown for ease of illustration). The system 50 uses the image processing module 106 to compare the converted image 302 with the images in the image database 109. The image processing module 106 can make use of matching techniques as described above. The matching techniques can be used to identify which image from the database 109 is sufficiently similar (e.g., most similar) to any given drill core image 104. In one implementation, the software uses keypoint matching to identify common features 307 (e.g., coarse grains) between the sample image 302 and an image existing in the database 109.

[0043] Pattern recognition is a well-known technique which can be used for matching.
Pattern recognition can automatically detect regularities in the image data using computer algorithms and subsequently classify the data into different categories. Data regarding the porosity, saturation, particle size distribution, and specific surface area (parameters) can be obtained for each sample using pattern recognition techniques.

[0044] Neural networks can be used to detect features in an image by mapping the pixels within the image with respect to the other pixels in the image.

[0045] SIFT can be used for feature detection in computer vision to detect and describe local features in images. SIFT keypoints of objects are first extracted from a set of reference images and stored in a database. An object can be recognized in a new image by individually comparing each feature from the new image to this database and finding candidate matching features based on Euclidean distance of their feature vectors. From the full set of matches, subsets of keypoints that agree on the object and its location, scale, and orientation in the new image are identified to filter out good matches.

[0046] It can be appreciated that comparing a given image 104 to an existing database of images 109 in the example shown in FIG. 3 can be a less intensive process since in this 23594333.1 example, only a portion of the particles would need to be considered. This can reduce processing time and costs associated with estimating permeability.
Additionally, it is found that detecting only the large particle sizes can reduce resolution errors associated with processes that require detecting small particles. Similar principles can apply to a comparison based on other parameters in the images 104, 109, e.g., textural information in the images 104, patterns in the grains from characteristic sedimentary structures, or other mathematical patterns including those difficult to observe.
Permeability Estimation Through Classification

[0047] Once an input image is matched to an image in the image database, the image processing module 106 can assign the features of the matched image from the image database 109, to the input image 104, particularly, the class of the image from the image database 109.
The class of the image 104 is associated with the features or parameters of the image referring to data which can be extracted from an image of a sample of the rock formation, which can include sieve size, diameter and PSD.

[0048] Based on the available data, a plurality of units (classes) can be defined, where each unit/class has its unique mean, Cop, and skewness, etc. For each input image, the permeability-porosity relationship can be calculated from a set of, for example, six standardized units. The unit that provides closer permeability to the one calculated from the associated data is assigned as the unit/classifier of the sample. The goal of the system is to predict the permeability unit, without looking at the whole particle size distribution.
The permeability-porosity relationship is then used with an estimation of porosity determined from, e.g., well log data, to estimate the permeability.

[0049] The computer algorithms which can be used to classify data are called classification algorithms. Commonly used classification algorithms include, but are not limited to: linear discriminant analysis, quadratic discriminant analysis, logistic regression, neural networks and machine learning.

[0050] Sample data can be used to train the system 50 to find the correct permeability class of a future sample. This can be completed during the pre-processing stages 208 and 210.
Sample data on permeability can be calculated using a known method such as the Carman-Kozeny equation.

23594333.1

[0051] The sample data can be used to find the permeability-porosity relationships in the pre-processing stages 208, 210. The relationships between permeability and porosity can be formed during the pre-processing steps 208, 210, and modelling techniques can be used to determine the correlation between permeability and porosity.

[0052] FIG. 4 illustrates permeability (k) versus calculated porosity in a chart in which each of the classes/units are defined. Permeability is shown in a logarithmic scale. Each black dot represents the permeability of a well core training sample at a given porosity. For each sample, the permeability has been calculated using its own PSD data and also the PSD
from each of these six units. The unit that provides closer permeability to the one calculated from the sample PSD is assigned as the unit of the sample.

[0053] FIG.5 illustrates permeability (k) versus calculated porosity in a chart in which each sample class/unit is identified. The unit that provides closer permeability to the one calculated from the sample PSD has been identified as the unit of the sample, and thus each sample has a unit assigned to it.

[0054] With the porosity data, and the permeability-porosity relationship available, the system 50 can estimate the permeability of the input image 104.
Permeability Estimation from Extracted Parameter

[0055] It can be appreciated that, as discussed in more detail below, in some cases the permeability can also be predicted directly based on an extracted parameter from the image, without relying on the permeability-porosity relationship and classification method. For example, a regression technique can be used to perform the prediction.
Refining Permeability Estimate

[0056] As illustrated in FIG. 2, other data obtained at stage 214 can be used to refine the permeability estimate at stage 212. This other data can be used to refine which permeability-porosity relationship is associated with a particular image 104. The other data can include, for example, gamma logs, the distance from the bottom of the reservoir, and the similarity of proximal samples with a particular permeability-porosity relationship. It can be appreciated that considering such additional data can not only improve a prediction made using a high resolution or high-quality image but also permit the use of standard lower resolution core photos while accurately predicting the permeability-porosity relationship for a given image 104. Since wells 23594333.1 typically have standard resolution core photos, this can enable permeability to be predicted using existing data.

[0057] Referring to FIG. 9, it can be appreciated that alluvial systems have been observed to have a fining upwards sequence, wherein the coarsest sands are found at the bottom and the finest sands at the top. As such, the gamma log 900 in general increases with height measured above the bottom of the reservoir. The permeability-porosity models in this example correspond to the PSD units 902 shown in FIG. 9. It can be seen that units 5 and 6 are found most commonly at the bottom at 904, and units 1 and 2 are most commonly found at the top at 906.
It can also be observed that units 5 and 6 tend to occur together, and likewise units 1 and 2 commonly occur together. It can therefore be appreciated that while unit 6 can be easier to predict with standard core photos, clean sands that are associated with units 5, 3, 2 can be more difficult. However, if a difficult to predict image 104 is located near a larger number of images 104 that are confidently predicted as unit 6, the odds of that image 104 being a 5, for example, can be considered higher than it being a unit 2. As such, the use of additional data such as the gamma log 900 can reduce uncertainty, particularly for standard low resolution core photos.

[0058] Each permeability-porosity relationship relates to a fades model or a geological model. For example, a sandy deposit in the bottom of a channel can differ from an included heterolithic strata that is deposited where the water moves more slowly. This can occur because each facies model, or permeability-porosity model typically has a particular Kv/Kh ratio associated with it, that is, the ratio between vertical and horizontal permeability. When this Kv/Kh ratio is determined, it can also be output at stage 216 as illustrated in FIG. 2.
Pre-Processing

[0059] The system 50 as described above can take advantage of a pre-determined correlation or similarity to perform a permeability prediction directly, using modelling techniques as described in the pre-processing stages 208, 210. Pre-processing may be required prior to executing the image processing and permeability estimation steps. The pre-processing steps involve the formation of the image database 109 in the first stage 208, the development of permeability-porosity models in the second stage 210, as well as storing well log data that can be used to estimate porosity. Moreover, pre-processing at stage 214 can be performed to obtain other data such as gamma logs, which can be used at stage 216 to refine the permeability estimate.

23594333.1

[0060] FIG. 6 illustrates operations that can be performed in implementing the pre-processing in stages 208 and 210 shown in FIG. 2. Samples are collected at step 600 and data is extracted at step 602 from these samples for training and testing the model. Actual permeability is calculated using an existing technique such as the modified Carman-Kozeny equation for each sample 604 and the data available is used at 606 to determine a classifier and/or permeability predictor to facilitate subsequent analyses as shown in FIGS. 1 and 2. For example, single or multi-variate relationships can be analyzed to find suitable variables for classification such as porosity, sieve size, d90, etc. Discriminant functions (i.e. models) can also be developed for the parameters which are strongly correlated to permeability.
The models can also be tested against the actual permeability for accuracy, such that if the model does not accurately estimate permeability, a new set of functions are to be developed.
In such a case, once the model is accurate enough, different parameter combinations can be compared to find the best fitting model, e.g., via an error rate analysis.
Database Generation

[0061] To complete the image comparison steps, the input image can be compared to the existing image database 109. The image database 109 can be created prior to the system processing steps as discussed below.

[0062] FIG. 7 illustrates the image database 109 populated by a set of images 700, with corresponding data, including permeability-porosity 701 and porosity 702 (e.g., via well logs). A
data input process 703 can be used to populate the database 109 from third-party data 704, self-collected data 705 or, historical data 706 from past studies.

[0063] That is, the database 109 can be formed from third-party sources as well as from existing and/or otherwise available data. This database 109 can also evolve over time as more drill cores 101 are extracted, imaged, and associated well logs stored. Each image in the image database 109 can have certain parameters associated with the image including:
sieve size, diameter, porosity, the number (or proportion) of particles that have large grain sizes, etc.

[0064] As shown in FIG. 6, a plurality of drill core samples can be taken and parameter data is collected to form a training set. Parameters such as porosity, tortuosity, PSD, and specific surface area are used as inputs to the Carman-Kozeny equation. The permeability values calculated using the Carman-Kozeny equation as well as the other parameters form the training set and build the model to be leveraged in subsequent analyses. The training set is 23594333.1 stored in the database 109. A higher number of drill core samples analyzed and stored in the image database 109 can equate to a higher accuracy of permeability estimation of drill core images 104.

[0065] Since the database 109 can source data from a variety of sources including third-party data, self-collected data, historical data, etc., the database can be periodically updated and therefore can increase in size as new data becomes available, including data which is learned via machine learning.

[0066] Applying machine learning techniques can help reduce the classification error with a small number of parameters. A model is trained based on the images 104 of cores or cuttings to determine a variety of parameters (e.g. sieve sizes, porosity, d90) and then porosity-permeability classes. On the other hand, a model can be trained to directly recognize the porosity-permeability classes from the images. Depending on the results, the most accurate method to identify permeability-porosity classes can be selected for future use.
Permeability Model Development

[0067] The goal of the permeability model development is to find a suitable classifier to identify which permeability-porosity class each sample can be assigned to, or a permeability predictor to directly predict permeability (or both). The relationship between permeability and parameters such as sieve sizes, porosity and diameter can be of interest in determining an accurate model to predict permeability.

[0068] Developing the parameter-permeability relationships is required for the permeability estimation step and thus, the models used to estimate the permeability are developed during the pre-processing stages 208, 210. The parameter-permeability relationships can be developed using a permeability prediction algorithm as described in FIG. 6, specifically as illustrated in step 606.

[0069] FIG 8 illustrates a process of developing models to estimate permeability. The parameters assigned to each input image include d. (diameter) 810, sy (sieve size) 812, and porosity 814. The inputs 800 are used in the various modelling techniques at 802, for example, single or multi-variate analyses, classification trees, random forest, discriminant analysis, neural networks, or regression, to name a few. The modelling technique chosen can output either a permeability model or a classification. Neural networks, regression or random forest are typically used to develop equations that predict permeability. Classification trees, random forest, and 23594333.1 discriminant analyses are typically used to solve a classification problem to define permeability units. Once the permeability correlations and prediction is found at 806, or the permeability class and permeability-porosity relationship are found at step 804, the estimated permeability can be determined and output at step 808. The estimate of permeability can be displayed to a user or operator, or analyst.

[0070] Various methods can be used to determine the correct permeability correlations and prediction methods or permeability-porosity classes, such as discriminant analyses, classification trees, random forest analysis, regression analysis and neural networks.

[0071] Machine learning can train a computer to classify future data by looking for patterns/relationships between the known data. Supervised learning feeds a computer a training set, where the data has a known classification, to help the computer build a model.
Classification error can be compared to determine the accuracy of the model.
Confusion matrices visually show the performance of a classifier. One axis displays the predicted classifications and the other axis displays the actual classifications. The diagonal shows the number of correct predictions and the off-diagonal are the number of misclassified cases.
Classification error can be found by dividing the number of misclassified data by the total number of data samples.

[0072] Certain modelling techniques require training sets and test sets to correctly predict permeability. Neural networks, random forest, discriminant analysis and classification trees can require a training set and a test set. Machine learning can be used to train a computer to perform a task by looking for patterns/relationships between the data. A
computer can be fed a training set in order to help the computer learn, called supervised learning.
Training sets can help the computer build a model to predict future data.

[0073] Model development can be used to find a classifier to identify to which permeability class a sample belongs. The data set can be split into subclasses, a process known as classification. Finding a classifier to correctly identify these classifications can be quite difficult.
Classifier methods include single variable analysis, box plots, discriminant analysis, classification trees, random forests and neural networks.

[0074] Single variable analysis can be used to correlate parameters and permeability to find correlation coefficients, the strength of the relationship between the relative movements of two variables. Crossplots can be used to determine if classification of the data can be achieved.

23594333.1

[0075] Discriminant analysis techniques analyze data and develop discriminant functions to classify the data from parameter inputs. The discriminant functions can be a linear combination of parameters that will discriminate between classes. Common analyses include, for example:
linear discriminant analysis (LDA), and quadratic discriminant analysis (QDA).
LDA makes a classification decision based on the value of a linear combination of parameters. LDA assumes the data set has a normal distribution and the covariance of each class is identical. Covariance measures the degree to which two variables are linearly associated. QDA also assumes the data set has a normal distribution; however, it derives a quadratic discriminant function used to classify the data.

[0076] Classification trees use a decision tree to predict an outcome based on parameters.
The predicted output can be the class to which the set of data belongs. Using statistics and machine learning, the tree learns from the training data where the data set should be partitioned into subsets. The decisions can be based on single or multiple parameters.

[0077] The random forest technique combines several decision trees that can be built based on a subset of features and samples to determine what class the data should be. Feature selection methods can determine what parameters are relevant for model construction. The random forest learns how to map data to the output during the training phase.

[0078] Neural Networks can be used, which include linear combinations of nonlinear transformations of data that recognize relationships in a data set to output a classification. The network can be composed of layers of neurons, computational units, with connections in different layers. During the learning phase for neural networks, the network trains by adjusting the weights associated with the input parameters to predict the correct class for the data. The network can adapt as more information is fed. The training data can be processed many times as the weights get tweaked and refined.

[0079] Once the methods are tested for accuracy and can achieve low error rates, the models are generated and stored for future adoption. It was found that the parameters porosity and sieve size correlate best with permeability; however, other parameters can also be used.

[0080] Any combination of inputs at step 800 can be used from the database 109 to develop the permeability-porosity model. Any combination of output modelling techniques at step 802 can be used to develop the permeability-porosity model.

23594333.1 Example Implementations

[0081] In one example, the image processing and classification algorithms can identify the permeability-porosity relationship by three different sieve sizes (e.g. 5125, s425, sum). Sy represents a percentage of particles smaller than sieve size "y"; s250 represents a percentage of particles smaller than 250 microns, etc.

[0082] In another example, the image processing and classification algorithms can identify the permeability-porosity relationship by three different diameter sizes (e.g.
d50, d90, d95). dx is the diameter of the particle that x% of a sample's mass is smaller than and (100-x)% of a sample's mass is larger than.

[0083] In another example, the image processing and classification algorithms can identify the permeability-porosity relationship by a variety of diameter sizes, sieve sizes and porosity (e.g. d50, dsch d95, 5125, 5425, or sio00,)=

[0084] The parameters listed above can be input to the classification/prediction models to determine classes or permeability values. With these determinations made, subsequent image comparison operations, e.g., in the field or in a lab can be performed with much less processing power and time required when compared to previous techniques.

[0085] For simplicity and clarity of illustration, where considered appropriate, reference numerals may be repeated among the figures to indicate corresponding or analogous elements.
In addition, numerous specific details are set forth in order to provide a thorough understanding of the examples described herein. However, it will be understood by those of ordinary skill in the art that the examples described herein may be practiced without these specific details. In other instances, well-known methods, procedures and components have not been described in detail so as not to obscure the examples described herein. Also, the description is not to be considered as limiting the scope of the examples described herein.

[0086] It will be appreciated that the examples and corresponding diagrams used herein are for illustrative purposes only. Different configurations and terminology can be used without departing from the principles expressed herein. For instance, components and modules can be added, deleted, modified, or arranged with differing connections without departing from these principles.

[0087] It will also be appreciated that any module or component exemplified herein that executes instructions may include or otherwise have access to computer readable media such 23594333.1 as storage media, computer storage media, or data storage devices (removable and/or non-removable) such as, for example, magnetic disks, optical disks, or tape.
Computer storage media may include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules, or other data. Examples of computer storage media include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by an application, module, or both. Any such computer storage media may be part of the system 50, any component of or related to the system 50, etc., or accessible or connectable thereto. Any application or module herein described may be implemented using computer readable/executable instructions that may be stored or otherwise held by such computer readable media.

[0088] The steps or operations in the flow charts and diagrams described herein are just for example. There may be many variations to these steps or operations without departing from the principles discussed above. For instance, the steps may be performed in a differing order, or steps may be added, deleted, or modified.

[0089] Although the above principles have been described with reference to certain specific examples, various modifications thereof will be apparent to those skilled in the art as outlined in the appended claims.

23594333.1

Claims (53)

Claims:

1. A method of estimating permeability of a formation from a formation sample, the method comprising:
acquiring an image of the formation sample;
comparing the acquired image with one or more stored images stored in an image database to associate the acquired image with one of the one or more stored images and corresponding stored data; and determining a permeability measure associated with the formation sample from the stored data and the acquired image by identifying an extracted parameter from the acquired image and using a predetermined permeability model and predetermined porosity data.

2. The method of claim 1, wherein the formation sample comprises a drill core sample.

3. The method of claim 1, wherein the formation sample comprises a drill cutting sample.

4. The method of any one of claims 1 to 3, wherein the image is acquired using one or more of the following techniques: Computed Tomography (CT), Computer Axial Tomography (CAT), hyperspectral imaging, drone scan imaging, digital camera imaging, hand held imaging, and downhole imaging.

5. The method of any one of claims 1 to 4, wherein the comparing comprises using one or more of: pattern recognition, a Scale-Invariant Feature Transform (SIFT), and keypoint matching.

6. The method of any one of claims 1 to 4, wherein the comparing comprises using one or more of: L2 Norm, L1 Norm, and Mutual Information.

7. The method of any one of claims 1 to 6, wherein the acquired image is associated with the stored image by identifying similar large grain sizes in the compared images.

8. The method of any one of claims 1 to 6, wherein the acquired image is associated with the stored image by identifying textural information in the compared images

9. The method of any one of claims 1 to 6, wherein the acquired image is associated with the stored image by identifying patterns in grains from characteristic sedimentary structures.

10. The method of any one of claims 1 to 6, wherein the acquired image is associated with the stored image by identifying one or more mathematical patterns.

11. The method of any one of claims 1 to 10, wherein the predetermined permeability model considers at least one of sieve size, porosity, and diameter.

12. The method of any one of claims 1 to 11, wherein the image database is generated by:
collecting data from each sample; and calculating permeability for each sample.

13. The method of claim 12, further comprising generating a permeability model from the collected data.

14. The method of claim 12 or claim 13, wherein the permeability is calculated using a Carman-Kozeny equation.

15. The method of claim 1, wherein the comparing comprises comparing pixels in the acquired image and the one or more stored images.

16. The method of claim 1, wherein the comparing comprises extracting at least one feature of interest and defining the images by a feature vector with a set of values to highlight each feature.

17. The method of claim 16, wherein the extracting comprises applying a Gabor filter, local binary pattern, or gray level co-occurrence matrix method.

18. The method of claim 16 or claim 17, further comprising clustering feature vector data to generate a plurality of clusters or categories.

19. The method of claim 18, wherein the clustering comprises applying a k-means clustering algorithm.

20. The method of any one of claims 16 to 19, further comprising applying a dimension reducing technique to the feature vectors to reduce the dimension of the feature vector.

21. The method of any one of claims 1 to 20, further comprising obtaining other data to refine the estimated permeability measure.

22. The method of claim 21, wherein the other data comprises at least one of: gamma log data, a distance measure from the bottom of a corresponding reservoir, and a similarity of proximal samples with a particular permeability-porosity relationship.

23. The method of any one of claims 1 to 22, further comprising determining a Kv/Kh ratio defining a ratio between vertical and horizontal permeability; and outputting the Kv/Kh ratio.

24. A computer readable medium storing computer executable instructions for estimating permeability of a formation from a formation sample, the computer executable instructions comprising instructions for performing the method of any one of claims 1 to 23.

25. A method of estimating permeability of a formation from a formation sample, the method comprising:
acquiring an image of the formation sample;
comparing the acquired image with one or more stored images stored in an image database to associate the acquired image with one of the one or more stored images and corresponding stored data; and determining a permeability measure associated with the formation sample from the stored data and the acquired image by identifying a class and using a predetermined permeability-porosity relationship for that class and predetermined porosity data.

26. The method of claim 25, wherein the formation sample comprises a drill core sample.

27. The method of claim 25, wherein the formation sample comprises a drill cutting sample.

28. The method of any one of claims 25 to 27, wherein the image is acquired using one or more of the following techniques: Computed Tomography (CT), Computer Axial Tomography (CAT), hyperspectral imaging, drone scan imaging, digital camera imaging, hand held imaging, and downhole imaging.

29. The method of any one of claims 25 to 28, wherein the comparing comprises using one or more of: pattern recognition, a Scale-Invariant Feature Transform (SIFT), keypoint matching, L2 Norm, L1 Norm, and Mutual Information.

30. The method of any one of claims 25 to 29, wherein the comparing comprises using one or more of: pattern recognition, a Scale-Invariant Feature Transform (SIFT), and keypoint matching.

31. The method of any one of claims 25 to 29, wherein the comparing comprises using one or more of: L2 Norm, L1 Norm, and Mutual Information.

32. The method of any one of claims 25 to 31, wherein the acquired image is associated with the stored image by identifying similar large grain sizes in the compared images.

33. The method of any one of claims 25 to 31, wherein the acquired image is associated with the stored image by identifying textural information in the compared images

34. The method of any one of claims 25 to 31, wherein the acquired image is associated with the stored image by identifying patterns in grains from characteristic sedimentary structures.

35. The method of any one of claims 25 to 31, wherein the acquired image is associated with the stored image by identifying one or more mathematical patterns.

36. The method of any one of claims 25 to 35, further comprising determining a classifier to identify the class by applying at least one of: a discriminant analysis, classification trees, a random forest analysis, a regression analysis, and a neural network.

37. The method of any one of claims 25 to 36, wherein the predetermined permeability model considers at least one of sieve size, porosity, and diameter.

38. The method of any one of claims 25 to 37, wherein the image database is generated by:
collecting data from each sample; and calculating permeability for each sample.

39. The method of claim 38, further comprising determining a permeability classifier and at least one variable for classification, from the collected data.

40. The method of claim 38 or claim 39, further comprising generating a permeability model from the collected data.

41. The method of any one of claims 38 to 40, wherein the permeability is calculated using a Carman-Kozeny equation.

42. The method of claim 25, wherein the comparing comprises comparing pixels in the acquired image and the one or more stored images.

43. The method of claim 25, wherein the comparing comprises extracting at least one feature of interest and defining the images by a feature vector with a set of values to highlight each feature.

44. The method of claim 43, wherein the extracting comprises applying a Gabor filter, local binary pattern, or gray level co-occurrence matrix method.

45. The method of claim 43 or claim 44, further comprising clustering feature vector data to generate a plurality of clusters or categories.

46. The method of claim 45, wherein the clustering comprises applying a k-means clustering algorithm.

47. The method of any one of claims 43 to 46, further comprising applying a dimension reducing technique to the feature vectors to reduce the dimension of the feature vector.

48. The method of any one of claims 25 to 47, further comprising obtaining other data to refine the estimated permeability measure.

49. The method of claim 48, wherein the other data comprises at least one of: gamma log data, a distance measure from the bottom of a corresponding reservoir, and a similarity of proximal samples with a particular permeability-porosity relationship.

50. The method of any one of claims 25 to 49, further comprising determining a Kv/Kh ratio defining a ratio between vertical and horizontal permeability; and outputting the Kv/Kh ratio.

51. A computer readable medium storing computer executable instructions for estimating permeability of a formation from a formation sample, the computer executable instructions comprising instructions for performing the method of any one of claims 25 to 50.

52. A system for estimating permeability of a formation from a formation sample, the system comprising:
an image acquisition device;
an image database; and an image processing module executing computer executable instructions for performing the method of any one of claims 1 to 23.

53. A system for estimating permeability of a formation from a formation sample, the system comprising:
an image acquisition device;
an image database; and an image processing module executing computer executable instructions for performing the method of any one of claims 25 to 50.