Formation porosity estimation from digital images

Abstract

A method for estimating porosity of a rock sample includes acquiring a calibrated digital image of the rock sample, extracting color and texture features from the digital image of the rock sample, and estimating the porosity of the rock sample using a model to evaluate the extracted color and texture features. The model is configured to correlate digital image color features and textural features with porosity.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 369,028, filed on July 21, 2022, and U.S. Provisional Patent Application No. 63 / 379,941, filed on October 18, 2022. Each of the above applications is incorporated herein by reference in its entirety. Background Art

[0003] Formation porosity and permeability are among the most important physical properties of a reservoir and are often used to provide an estimate of the volume of recoverable fluid in a reservoir. Porosity is generally understood as a measure of the ratio of the pore volume to the total volume of formation rock, while permeability is generally understood as a measure of the connectivity of individual pores to one another.

[0004] Formation porosity can be measured in a variety of ways. For example, porosity is often inferred from neutron porosity logs acquired during logging while drilling (LWD) and / or wireline logging operations. Although neutron porosity measurements are used commercially, some assumptions may need to be made to arrive at an estimated formation porosity value. In addition, neutron porosity logs typically require extensive calibration to arrive at a reliable porosity estimate.

[0005] Formation porosity measurements are also typically performed on sidewall core plug samples or entire core plug samples of formation rock (e.g., obtained during coring operations when constructing a wellbore). For example, the porosity of the obtained core sample can be measured using a porosimeter. Although it is generally believed that such porosimeter measurements provide accurate porosity values, the measurements can be very time consuming and laborious. There is a need in the industry for improved methods for performing formation porosity measurements, particularly methods that automate or partially automate the measurement process. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] For a more complete understanding of the disclosed subject matter and advantages thereof, reference is now made to the following description taken in conjunction with the accompanying drawings, in which:

[0007] Figure 1 An example drilling rig including an example system for estimating formation porosity is depicted.

[0008] Figure 2A and Figure 2B (collectively, FIG. 2 ) depicts a flow chart of an example method for estimating formation porosity.

[0009] Figure 3 An example method for acquiring a calibrated digital image of drill cuttings particles is depicted.

[0010] Figure 4Example clustering of lithic particles in a synthetic 2D representation of the multidimensional hyperspace depicting the extracted color and texture features.

[0011] Figure 5 A flow chart of an example method for constructing a porosity model that relates color and texture features to rock porosity is depicted.

[0012] Figure 6 A block diagram of an example system for estimating rock porosity is depicted.

[0013] Fig. 7A , Figure 7B and Figure 7C (collectively, FIG. 7 ) depicts a plot of predicted porosity versus ground truth porosity generated in a first porosity model validation test using limestone rock cuttings particles.

[0014] Fig. 8A , Figure 8B and Figure 8C (collectively, FIG. 8 ) depicts a plot of predicted porosity versus ground truth porosity generated in a second porosity model validation test using sandstone rock cuttings particles. DETAILED DESCRIPTION

[0015] Embodiments of the present disclosure include systems and methods for estimating formation porosity. An example method includes obtaining a calibrated digital image of rock cuttings particles separated from drilling fluid circulating in a wellbore. Color features and / or texture features are extracted from the rock cuttings particles in the digital image and evaluated using a model to estimate porosity. In an example embodiment, multiple individual rock cuttings particles are evaluated to estimate multiple corresponding porosity values. The estimated porosity values ​​can be further processed to calculate an average porosity, for example, including a mean porosity, a median porosity, or a porosity distribution. The disclosed embodiments can advantageously enable accurate porosity estimation in a near real-time manner in essentially any drilling operation (including operations where logging measurements cannot be performed). In addition, the disclosed embodiments can enable formation porosity measurements to be performed in operations where conventional logging measurements cannot be performed (e.g., continuous tubing or dry drilling operations).

[0016] Figure 1An example drilling rig 20 is depicted, which includes a system 70 for automatically estimating formation porosity based on drill cuttings extracted from circulating drilling fluid. The drilling rig 20 can be positioned above a subterranean formation (not shown). The drilling rig 20 can include, for example, a derrick and a jack-up device (also not shown) for raising and lowering a drill string 30, as shown, which extends into a wellbore 40 and includes, for example, a drill bit 32 and one or more downhole measurement tools 38 (e.g., logging while drilling tools or measurement while drilling tools) in a bottom hole assembly (BHA) above the drill bit 32. Suitable drilling systems, such as those including drilling, steering, logging, and other downhole tools, are well known in the art.

[0017] The drilling rig 20 also includes a surface system 50 for controlling the flow of drilling fluid used on the drilling rig (e.g., used when drilling out the wellbore 40). In the depicted example drilling rig, the drilling fluid 35 is pumped downhole (as depicted at 92) via a conventional mud pump 57, for example. The drilling fluid 35 may be pumped, for example, through a riser 58 and a mud hose 59 on its way to the drill string 30. The drilling fluid 35 typically emerges from the drill string 30 at or near the drill bit 32 and creates an upward flow 94 of mud through the wellbore annulus 42 (the annular space between the drill string and the wellbore wall). The drilling fluid 35 then flows to the mud pit system 56 through the return pipe 52, where it can be circulated. It will be appreciated that the terms drilling fluid and mud are used synonymously herein.

[0018] The circulating drilling fluid 35 is used to perform a variety of functions during the drilling operation, one of which is to carry drill cuttings 45 to the surface (in the upflow 94). The drill cuttings 45 are typically removed from the returning mud via a shale shaker 55 (or other similar solids control equipment) in a return conduit (e.g., directly upstream of the mud pit 56). Formation gases released during drilling may also be brought to the surface in the circulating drilling fluid. These gases are typically removed from the drilling fluid, for example, via a degasser or degasser 54 located in or near the head tank 53, which in the example description is directly upstream of the shale shaker 55. The drill cuttings 45 can be evaluated to estimate the porosity of the underground formation layers through which the well was drilled, as described in more detail below.

[0019] The drilling rig 20 may include a system 70 configured to capture and automatically evaluate digital images of drill cuttings and estimate formation porosity, as described in more detail herein. The system 70 may be deployed at a drilling site (e.g., in a field laboratory 60) or off-site. The disclosed embodiments are not limited in this respect. The system 70 may include computer hardware and software configured to automatically or semi-automatically evaluate cuttings images. To perform these functions, the hardware may include one or more processors (e.g., microprocessors) that may be connected to one or more data storage devices (e.g., hard drives or solid-state memory). As known to those of ordinary skill in the art, the processor may also be connected to a network, for example, to receive images from a networked camera system (not shown) or another computer system. Of course, it will be understood that the disclosed embodiments are not limited to the use or configuration of any particular computer hardware and / or software.

[0020] although Figure 1 An onshore drilling rig 20 is depicted, but it will be appreciated that the disclosed embodiments are equally applicable to onshore or offshore drilling rigs. As known to those of ordinary skill in the art, an offshore drilling rig typically includes a platform deployed on top of a riser extending from the seafloor to the surface of the water. A drill string extends downward from the platform, through the riser, and into the wellbore through a blowout preventer (BOP) located on the seafloor. The disclosed embodiments are not limited in these respects.

[0021] Figure 2A and Figure 2B (collectively, FIG. 2 ) depicts a flow chart of example methods 100 and 120 for estimating formation porosity. Figure 2A In the method 100, at least one calibrated digital image of a rock sample (such as a sidewall or entire core sample or one or more drill cuttings particles) is acquired. The calibrated digital image may be processed to extract color features and / or texture features from the rock sample at 104. The extracted color features and texture features may then be evaluated (processed) with a model at 106 to estimate the porosity (e.g., previously unknown porosity) of the rock sample. In an example embodiment, the model may be an empirical model that relates digital image color features and digital image texture features to porosity. Example model embodiments are described in more detail below.

[0022] exist Figure 2B In the method 120, at 122, the method includes acquiring at least one calibrated digital image of rock cutting particles separated from the drilling fluid. For example, rock cutting particles may be generated while drilling, and the rock cutting particles may be transported to the surface in the circulating drilling fluid, as described above with respect to Figure 1At 124, the calibrated digital image may be processed with a segmentation algorithm to obtain a segmented image. The segmentation algorithm may be configured, for example, to identify a plurality of individual rock cuttings particles in the calibrated image. The segmented image may be processed at 126 to extract color features and texture features from each of a plurality of selected rock cuttings particles in the image. For example, color features and / or texture features may be extracted from each segmented particle or a subset of the segmented particles (e.g., only particles having a size greater than a threshold). The selected particles may be evaluated, for example, particle by particle, to extract color features and texture features.

[0023] In an example embodiment, the color features and texture features may be processed at 128 to classify each selected rock fragment particle according to a corresponding lithology type (such as sandstone, limestone, shale, etc.). The extracted color features and / or texture features may then be further processed with a model (a porosity model) at 130 to estimate the porosity (e.g., a previously unknown porosity) of each selected particle. As described in more detail below, the model may be selected, for example, based on the lithology classification at 128. For example, a first porosity model may be used for sandstone, a second porosity model may be used for limestone, a third porosity model may be used for shale, etc. It will be appreciated that the processing at 130 may generate a plurality of porosity values ​​(e.g., one porosity value for each selected rock fragment particle). These porosity values ​​estimated at 130 may be further processed to estimate the formation porosity at 132, for example, by calculating a mean or median value of the estimated porosity values ​​for each selected particle.

[0024] 2, it will be appreciated that methods 100 and 120 can advantageously achieve automatic (or semi-automatic) estimation of formation porosity without (or with only limited) human intervention. For example, in method 100, processing the calibrated image to extract color and texture features at 104 and evaluating the extracted features with the model at 106 to estimate porosity can be performed automatically. In method 120, segmenting the image at 124, classifying the lithology of the selected particles at 126, extracting color features and texture features at 128, and estimating rock fragments and formation porosity at 130 and 132 can be performed automatically without human intervention.

[0025] Now go to Figure 3 , depicts an example method 140 for separating rock cuttings from drilling fluid and acquiring calibrated digital images at 102 and 122 of FIG. 2 . For example, using the above description of Figure 1 The example drilling rig 20 is shown drilling through a subsurface formation of interest to form a borehole at 142. The rock cuttings generated while drilling are transported to the surface by an upwardly flowing drilling fluid (e.g., Figure 1). The rock cuttings may be collected at 144, for example, using a shale shaker or other solid separation / control equipment. In an example embodiment, the rock cuttings obtained from the shale shaker may be prepared for image analysis at 146, for example, by washing and then drying in an oven and / or sieving or meshing the rock cuttings to remove large particles and / or small particles (e.g., removing a portion of the rock cutting particles). One purpose of the preparation process is to reveal the color and texture of the rock by removing liquids and other residues on the rock surface that may interfere with color and texture extraction.

[0026] The prepared rock cuttings may also be placed in a tray with a high contrast (bright) background color to enhance subsequent particle identification and segmentation in the acquired image, for example, pure magenta (e.g., RGB values ​​of 255, 0, 255), pure blue (e.g., RGB values ​​of 0, 0, 255), pure green (e.g., RGB values ​​of 0, 255, 0), etc. In general, such colors do not exist in nature and therefore help the instance segmentation model avoid detecting the background of the tray as part of the particles. The tray may be placed in front of a digital camera, and at least one digital image may be captured at 148. For example, a white light image (where the rock cuttings are illuminated by white light) may be captured at 148. In other example embodiments, multiple digital images may be captured at 148, for example, a first white light image and a second infrared or ultraviolet image (where the rock cuttings are illuminated by infrared or ultraviolet light), or even a first white light image, a second infrared image, and a third ultraviolet image. The disclosed embodiments are not limited in these respects; however, it will be appreciated that acquiring multiple images may be advantageous because certain textural features and / or color features are more easily discernible under infrared or ultraviolet light than under white light (particularly under ultraviolet light).

[0027] It will be appreciated that acquiring a calibrated digital image can advantageously utilize standardized and / or calibrated lighting, color enhancement, magnification, and / or focus / resolution settings. For example, in certain embodiments, color / illumination calibration is obtained by using a colorimetric algorithm on previously analyzed photos and the current photo of interest, while resolution calibration can be based on the lens focal length, focus distance, and sensor size / resolution of the current photo of interest (compared to those of the previously analyzed photos). Images can be taken when the cuttings are wet or dry, with humidity typically controlled to obtain dry cuttings images.

[0028] Reference again Figure 2BIn an example embodiment, the segmentation algorithm may employ a mask region based convolutional neural network (Mask R-CNN), such as disclosed in U.S. Patent Application Serial No. 17 / 647,407, which is incorporated herein by reference in its entirety. Mask R-CNN may be configured to identify individual rock debris particles in a digital image and thereby generate a segmented image at 124. For example, Mask R-CNN may generate, for example, a bounding box and a mask image. A bounding box may be defined as a set of xy coordinates in an image that indicates an image region containing an object of interest. The bounding box may include a confidence score for each object of interest, ranging from 0 to 1 (e.g., a larger value indicates a higher associated confidence). The mask image may indicate (e.g., highlight or otherwise constrain) regions of interest having confidence scores exceeding a threshold.

[0029] It will be appreciated that Mask R-CNN is a model architecture that belongs to the category of supervised learning, which means that it generally requires a training dataset that includes images and corresponding labels. For example, the model can be trained using images that include rock debris particles of various sizes, shapes, colors, and lithology types. It will also be appreciated that the R-CNN model can be continuously retrained during drilling operations or between drilling operations in an oil field. For example, segmentation errors can be identified and corrected, and then used to generate labeled training images that can be used to retrain (or further train) R-CNN.

[0030] In an example embodiment, the segmented image may depict a plurality of rock cuttings particles on a high contrast tray. Individual particles may be identified, for example, by particle outlines or other demarcations. Furthermore, each identified particle may be identified by a corresponding set of pixels in the image. In other words, the segmented image may include a pixel-by-pixel segmentation, where each pixel in the image is assigned to either background or a single individual particle.

[0031] Further references Figure 2A and Figure 2B, methods 100 and 120 include extracting color features and texture features from the calibrated digital image or the segmented digital image at 104 and 126. The extracted color features and texture features may include color-related features, texture-related features, particle size-related features, and shape and size-related features for each of the plurality of individual rock debris particles. The color-related features may include, for example, average (such as mean, median, or mode) red, green, and blue intensities or distributions or standard deviations of the red, green, and blue intensities and / or average brightness of each particle. The color-related features may also include histograms, variances, skewness, and / or kurtosis of the red, green, and blue intensities. In addition, for infrared images and / or ultraviolet images, the color-related features may include average (such as mean, median, or mode) infrared or ultraviolet intensities or distributions or standard deviations of the infrared or ultraviolet intensities and / or average infrared or ultraviolet brightness of each particle. The color-related features may also include histograms, variances, skewness, and / or kurtosis of the infrared or ultraviolet intensities. In addition, for infrared images and / or ultraviolet images, the color-related features may include average (such as mean, median, or mode) infrared or ultraviolet intensities or distributions or standard deviations of the infrared or ultraviolet intensities and / or average infrared or ultraviolet brightness of each particle. The color-related features may also include histograms, variances, skewness, and / or kurtosis of the infrared or ultraviolet intensities.

[0032] Texture-related features can quantify various spatial relationships and / or directional changes of pixel color and / or brightness in each particle. The extracted texture-related features may include, for example, edge detection, pixel-to-pixel contrast, correlation and / or entropy. In addition, in certain embodiments, texture-related features may be extracted with techniques such as image texture filters (e.g., Gabor filters, etc.), autoencoders or other deep learning-based techniques. In addition, directional changes may be evaluated, for example, about symmetry and used to generate a spectrum, which may be further compared with a reference spectrum to assign a texture classification to each particle, such as homogeneous, inhomogeneous, granular, laminated particles, etc. Texture features may be evaluated, for example, further to characterize the particle size or particle size distribution of particles in a stratum or rock debris particle. For example, particle size may be identified as fine, medium, coarse or having an average size and size distribution.

[0033] In an example embodiment of method 120, the segmented image can be further processed to extract geometric features of the rock debris particles, such as size and shape features. Size and shape related features can include, for example, particle diameter, area, perimeter, maximum axis, minimum axis, particle aspect ratio, and internal angle measurements. In addition, geometric related features can also include the spatial relationship of pixels grouped in each particle to extract particle circularity, solidity, elongation, roundness, and / or convex hull area. Individual particles can be further classified, for example, as plate-like, fibrous, round or elliptical particles, sharp-angle particles, etc., and as one of various size boxes (e.g., based on the diameter, cross-sectional area, and / or perimeter of the particle).

[0034] It will be appreciated that color, texture, and geometric feature extraction may utilize a trained machine learning algorithm or any other deep learning algorithm. Such an algorithm may be trained, for example, using extracted color and texture features of different rock cutting particle types (e.g., lithology types), sizes, shapes, colors, etc., and may utilize an image database comprising visible light images, infrared images, and / or ultraviolet images of rock cutting particles. Such a database may be maintained on-site (e.g., at a drilling rig location) or off-site (e.g., at an off-site processing center or other location).

[0035] Still further reference Figure 2B , the extracted color features and / or texture features may be processed as appropriate at 128 to classify at least one of the identified rock fragments according to a corresponding lithology type, for example, as described in U.S. Patent Application Serial No. 17 / 647,412, which is incorporated herein by reference in its entirety. In general, the lithology of a rock or rock fragment is a description of its physical properties as seen in an outcrop, in a handheld or core sample, or visible with a low-power microscope. Lithology may refer to a detailed description of these physical properties, or an overview of the overall physical properties of the rock. In a second sense, the lithology of a rock (or rock fragment) refers to the type of rock or the overall (or macroscopic) identification or classification of the rock. Example lithologies or lithology types in this second sense include sandstone, limestone, slate, shale, basalt, coal, anhydrite, dolomite, gypsum, clay, flint, granite, and the like. Thus, the color and / or texture features represent physical characteristics of the lithology that can be measured and quantified from a digital image and evaluated to determine the lithology (in the second sense) of the formation. For example, a trained model may be used to evaluate color and / or texture features extracted from selected grains in a segmented image to classify the lithological type of those selected grains.

[0036] Formation porosity can be estimated (or identified) using a model at 106 and 130, 132 of methods 100 and 120, for example, based on the "position" of the particle in the multidimensional space of extracted color and texture features. For example, as described above, a set of color and texture features can be calculated (e.g., for each particle in the selected rock fragment particles). The set of calculated color and texture features can include a large number of features, such as at least 16 features (e.g., at least 32, 48, 64, 80, 96, 112, or 128 features). The model can be configured to correlate color and texture features with formation porosity, so that in practice, the model assigns a porosity value to a rock (or rock fragment particle) based on a set of values ​​of which features (or in other words, based on the position of the rock in the aforementioned multidimensional color / texture feature space). As described in more detail below, the model can include multiple models, each of which can be built or trained for a specific lithology type. For example, in method 120 , an appropriate model may be selected at 128 for the particular lithology type identified at 126 .

[0037] Figure 4 Depicted are example clustering of rock debris particles in a synthetic two-dimensional representation of a multidimensional hyperspace of extracted color and texture features that would otherwise be difficult (or impossible) to represent. In an example embodiment, a porosity for each particle can be assigned based on the clustering. In such an embodiment, groups of particles that are in the same cluster (or local region of hyperspace) can be assigned the same porosity value.

[0038] One aspect of the disclosed embodiments is the recognition that some of the same rock properties that control or influence porosity also influence the macroscopic visual appearance of the rock. For example, it is recognized that the rock's fabric (how the individual particles or grains of the rock are organized), mineralogy (the relative amounts and types of mineral components), diagenesis (the transformation of sediments or sedimentary rocks into different rocks), and grain size (the size of the individual grains of sediment in the rock) influence both the rock's porosity and the rock's macroscopic visual appearance. It is further recognized that the rock's macroscopic visual appearance can be quantified using a set (or set of) of color and texture properties from calibrated digital images of the rock, and that these color and texture property characteristics can then be correlated to the rock's porosity using a trained model.

[0039] Now go to Figure 5 , depicts a flow chart of an example method 150 for generating (or, in an example embodiment, training) a porosity model that relates digital image features (such as color and texture features) to rock porosity. For example, the method 150 may be configured to generate Figure 2A and Figure 2BThe porosity model used at 106 and 130 of the present invention can be generated, for example, using drill cuttings samples obtained at 154 from a group of one or more selected wells (e.g., Well 1, Well 2, ..., Well X shown at 152). These wells may be drilled in the same local area (e.g., within the same oil field or reservoir) and pass through the same or similar rock formations and formation lithologies.

[0040] At 156, porosity measurements are taken of the formation rock (e.g., rock formations and various lithologies) penetrated by the well. These porosity measurements can be taken using substantially any suitable porosity measurement technique, including, for example, while drilling and wireline logging measurements and porosimeter measurements of core samples obtained during or after the drilling operation. In an example embodiment, the porosity measurements can enable the development of a porosity log that quantifies the formation porosity as a function of wellbore depth and / or formation strata. Such porosity measurement and logging techniques are well known in the industry.

[0041] Continue to refer Figure 5 , a calibrated digital image of the acquired rock cuttings is taken at 160, for example, as described above with respect to Figure 3 The process 160 may include establishing and verifying a rock cutting preparation procedure at 162, a calibration digital photograph acquisition procedure at 164, and acquiring (or photographing) a calibrated digital image at 166. The procedure may be verified and fine-tuned as indicated to ensure adequate image-to-image consistency. Rock cuttings depicted in the calibrated digital image may be assigned corresponding measured porosity values ​​at 168 based on the porosity measurements and / or porosity records at 156.

[0042] Then at 170, a digital image of the rock fragments with known porosity values ​​(e.g., measured as described above) is processed to establish a correlation between color and texture features and known porosity and thereby generate and / or train the model. As shown, the model building and / or training at 170 may include processing the image at 172 to extract color and texture features. These color and texture features may include a plurality of features described above with respect to the methods 100 and 120 depicted in FIG. 2. Model building may also include marking the porosity category at 174 and / or the relationship between the extracted features and the known porosity. In an example embodiment, the color and texture features that are best correlated with the formation porosity may be selected manually and / or automatically, for example, by regression analysis or using a machine learning algorithm. The model may be repeatedly constructed and verified at 176 and 178 as indicated. The model may be further verified, updated and / or retrained, for example, by field testing at 180 and 182. Field testing may include, for example, comparing the porosity values ​​obtained with the model with the porosity values ​​obtained by conventional measurements.

[0043] Continue to refer Figure 5 , the processing at 170 may include, for example, nearest neighbor regression based on the distances (e.g., Euclidean distances) between cuttings in a multidimensional color and texture space of particles with known porosity to develop multidimensional correlations and build a model. It will be appreciated that there is no training step when using the nearest neighbor regression method. In such an embodiment, the local area may be defined by the number of K nearest neighbor reference cuttings, where K may be essentially any number of nearest neighbors (such as 5, 10, 20, or 50). In other embodiments, the processing at 170 may include (or further include) using a machine learning algorithm at 170 to build and train the model.

[0044] It will be appreciated that even if the porosity model is successfully field tested, sometimes the formation porosity may not be accurately estimated. In such cases, conventional measurement techniques may be used to measure or verify the formation porosity, and the estimated particles may be re-labeled with the measured porosity. These re-labeled images (including the labeled particles) may then be used to further update and / or train the porosity model in a manner similar to that shown at 180 and 182.

[0045] Continue to refer Figure 5 It will be appreciated that it may be advantageous to develop multiple different porosity models and / or reference data sets for an oil field or a local area. For example, different porosity models (e.g., a first model for sandstone, a second model for limestone, a third model for shale, etc.) may be developed for each formation lithology (or lithology type) within the oil field. In such an embodiment, different models may utilize the same or different sets of color and / or texture features. For example, in one embodiment, the first model and the second model developed for the corresponding first and second lithologies may utilize the same set of color and texture features. In this embodiment, a specific position in the multidimensional feature space may be mapped to a first porosity value in the first model and a different second porosity value in the second model. In other words, the first model may include a first correlation between the multidimensional feature space and porosity, and the second model may include a different second correlation between the multidimensional feature space and porosity. In a second exemplary embodiment, the first model and the second model corresponding to the first and second lithologies may utilize different first and second sets of color and / or texture features so that the porosity correlation is based on different multidimensional feature spaces. The disclosed embodiments are not limited in these respects.

[0046] Referring again to FIG. 2 , it will be appreciated that the generated model can be used to estimate unknown porosity values ​​for rock samples and / or rock cuttings particles. In an example embodiment, a model (or models) can be generated using rock cuttings particles obtained from a first well. The generated model can then be used to estimate unknown porosity for rock cuttings obtained when drilling a different second well. Thus, it will be appreciated that method 150 ( Figure 5 ) can be implemented in methods 100 and 120 (Figure 2).

[0047] Figure 6 A block diagram of an example system 200 for estimating the porosity of a rock sample or a sample (or samples) of rock cutting particles is depicted. The system 200 includes a digital camera system 210 configured to capture one or more calibrated digital images of a rock sample of rock cutting particles separated from a drilling fluid. The digital camera system may include substantially any suitable digital camera (or cameras) sensitive to infrared light, visible light, and / or ultraviolet light. The system also includes a digital image processing system 220, for example, including one or more processors or computing systems. The digital image processing system 220 may include a segmentation module 222 configured to process the calibrated digital images to identify individual particles of the rock cutting particles depicted in the calibrated images. The digital image processing system 220 also includes a color and texture feature extraction module 224 configured to extract color and / or texture features (as described above) from selected particles of the calibrated digital images or from the identified solid particles in the segmented images. The digital image processing system 220 also includes a porosity estimation module 226, which includes a model or data set that relates the extracted color and / or texture features to estimated porosity values ​​of rock or rock fragment particles. The digital image processing system 220 may also include a merging module 228, which is configured to calculate an average (e.g., mean or median) porosity value for selected particles in the digital image and report the average as the formation porosity value.

[0048] Now go to Fig. 7A , Figure 7B and Figure 7C (collectively referred to as Figure 7) and Fig. 8A , Figure 8B and Figure 8C (collectively, FIG8 ), depicts the summary results of the first validation test and the second validation test. In FIG7 , limestone rock cuttings particles having known porosity values ​​were tested. In FIG8 , sandstone rock cuttings particles having known porosity values ​​were tested. In each case, the rock cuttings particles having known porosity values ​​were obtained by measuring the porosity of multiple rock cores and then crushing each rock core to obtain simulated rock cuttings particles.

[0049] In FIG7 , a first porosity model is trained and constructed using a first set of limestone rock fragments particles (comprising 8,529 particles) with known porosity values. The model is described above with respect to Figure 5 The model was generated as described. Specifically, 64 color and texture features were extracted from each marked particle in the first group (marking means that the particles were marked with known porosity). A machine learning algorithm (regression model based on the open source LightGBM) was then used to generate a multidimensional correlation between the extracted features and the known porosity values. A different second group of limestone cuttings particles with known porosity values ​​(comprising 4,594 different particles) were then tested to validate the model. These cuttings particles were obtained from 13 different core samples with known (measured) porosity values ​​ranging from about 0 to about 39%. Each core sample was crushed to produce simulated cuttings particles, which were dispersed on corresponding trays and prepared for image acquisition.

[0050] Continuing with FIG. 7 , a calibrated white light digital image of each tray particle is obtained, as described above with respect to Figure 3 The digital image is segmented using the Mask R-CNN segmentation algorithm described above. The segmented image is evaluated particle by particle to extract 64 different color and texture features from each particle. The color and texture features are processed to determine the location of the particle in the 64-dimensional feature space. These extracted features are then processed using the model to obtain the porosity value of each individual particle.

[0051] Fig. 7A Depicted is a graph of the predicted porosity (using the methods disclosed herein) versus the ground truth porosity (the known porosity value for each particle) for each of the 4,594 evaluated particles. As shown, there is a clear spread of particles within the particles of each tray. Figure 7B and Figure 7C A graph of predicted porosity versus ground truth porosity is depicted, plotting the median porosity value for each tray of particles (7B) and the mean porosity value for each tray of particles (7C). As shown, the median and mean values ​​for each tray of particles are very close to the ground truth (known) porosity value, with an average error of 3.3%.

[0052] In FIG8 , a first set of sandstone rock cutting particles with known porosity values ​​(comprising 14,209 particles) is used to train and construct a second porosity model. The model is trained and constructed as described above with respect to FIG7 . A different second set of sandstone rock cutting particles with known porosity values ​​(comprising 7,652 different particles) is then tested to validate the model. These rock cutting particles are obtained from 19 different core samples with known (measured) porosity values ​​ranging from about 4% to about 30%. Each core sample is crushed to produce simulated rock cutting particles, which are dispersed on corresponding trays and prepared for image acquisition.

[0053] Continuing to refer to FIG. 8, as described above with respect to Figure 3 A calibrated white light digital image of the particles of each tray is obtained as described. The digital image is segmented using the Mask R-CNN segmentation method described above. The segmented image is evaluated particle by particle to extract 64 different color-related features and texture-related features from each particle, thereby defining a 64-dimensional feature space. These extracted features are then processed using the model to obtain the porosity value of each individual particle.

[0054] Fig. 8A Depicted is a graph of the predicted porosity (using the methods disclosed herein) versus the ground truth porosity (the known porosity value for each particle) for each of the 7,652 evaluated particles. As shown, there is a clear spread of particles within the particles of each tray. Figure 8B and Figure 8C A graph of predicted porosity versus ground truth porosity is depicted, plotting the median porosity value for each tray of particles (8B) and the mean porosity value for each tray of particles (8C). As shown, the median and mean values ​​for each tray of particles closely match the ground truth (known) porosity values, with an average error of within 2.8%.

[0055] It will be understood that the present disclosure includes many embodiments. These embodiments include but are not limited to the following embodiments.

[0056] In a first embodiment, a method for estimating the porosity of a subsurface formation includes acquiring a calibrated digital image of rock cuttings particles produced by drilling a wellbore in the subsurface formation; extracting a plurality of color and texture features from the calibrated digital image of the rock cuttings particles; and estimating the porosity of the subsurface formation using a model, wherein the model includes correlations between the plurality of color and texture features and the porosity.

[0057] A second embodiment may include the first embodiment, wherein said acquiring said calibrated digital image comprises drilling a subterranean wellbore; collecting said rock cuttings particles from a circulating drilling fluid; preparing said rock cuttings particles; and taking a calibrated digital photograph of said prepared rock cuttings particles.

[0058] A third embodiment may include any one of the first to second embodiments, the third embodiment also comprising segmenting the calibrated digital image to generate a segmented image, the segmented image identifying individual particles of the rock debris particles depicted in the calibrated digital image, wherein the extracting comprises extracting a plurality of color and texture features corresponding to selected particles of the identified individual rock debris particles in the segmented image; and the estimating comprises using the model to estimate the porosity of each of the selected particles and calculating a mean or median value of the estimated porosity of each of the selected particles to estimate the porosity of the underground formation.

[0059] A fourth embodiment may include the third embodiment, wherein said segmenting, said extracting, and said estimating are performed automatically.

[0060] A fifth embodiment may include any one of the third to fourth embodiments, wherein the use of the model to estimate the porosity of each of the selected particles comprises: determining the position of each of the selected particles in a multidimensional color and texture feature space; and estimating the porosity of each of the selected particles based on the position of each of the selected particles in the multidimensional color and texture feature space.

[0061] A sixth embodiment may include any one of the third to fifth embodiments, wherein the use of the model to estimate the porosity of each of the selected particles includes: clustering each of the selected particles in a multidimensional color and texture feature space; and estimating the porosity of each of the selected particles based on the clustering.

[0062] A seventh embodiment may include any one of the first to sixth embodiments, wherein the using a model to estimate the porosity of each of the selected particles comprises: marking each of the identified rock fragments in the segmented image with a corresponding lithology type based on the multiple color and texture features; selecting the model corresponding to the marked lithology type of each of the selected particles among the identified rock fragments from a plurality of models; and estimating the porosity of each of the selected particles using the selected model, wherein the selected model is configured to correlate the multiple color and texture features with the porosity of rock fragments particles of the marked lithology type.

[0063] An eighth embodiment may include the seventh embodiment, wherein the rock debris particles include at least first and second different lithology types.

[0064] A ninth embodiment may include any one of the first to eighth embodiments, wherein the model is generated by: acquiring a plurality of calibrated digital images of rock fragments having known porosity values; marking the rock fragments having known porosity values ​​with corresponding known porosity values; extracting a plurality of color and texture features from the marked rock fragments; and determining a correlation between the plurality of extracted color and texture features and the porosity values ​​to generate the model.

[0065] A tenth embodiment may include the ninth embodiment, wherein determining the correlation comprises nearest neighbor regression based on distances between the rock fragments particles having known porosity values ​​in a multi-dimensional color and texture space.

[0066] An eleventh embodiment may include the ninth embodiment, wherein determining the correlation comprises constructing and training the model using a machine learning model that determines which of the color and texture features best correlate with the known porosity value.

[0067] In a twelfth embodiment, a system for estimating the porosity of an underground formation comprises: a digital camera system configured to capture calibrated digital images of rock particles from the underground formation; and a digital image processing system comprising a plurality of modules, the modules comprising: a color and texture feature extraction module configured to extract color features and texture features from the calibrated digital images; and a porosity estimation module configured to estimate the porosity of the underground formation using a model that relates the extracted color features and texture features to porosity.

[0068] A thirteenth embodiment may include the twelfth embodiment, wherein the module further comprises: a segmentation module configured to process the calibrated digital image to identify individual rock fragments in the calibrated digital image; a merging module configured to calculate an average porosity value based on the estimated porosity values ​​of selected ones of the identified rock fragments; wherein the color and texture feature extraction module is configured to extract color features and texture features from the selected ones of the identified rock fragments; and wherein the porosity estimation module is configured to estimate the porosity value of the selected ones of the identified rock fragments using the model.

[0069] A fourteenth embodiment may include the thirteenth embodiment, wherein the porosity estimation module is configured to: determine the position of each of the selected ones of the identified rock fragments in a multidimensional color and texture feature space; and estimate the porosity based on the position of each of the selected ones of the identified rock fragments in the multidimensional color and texture feature space.

[0070] A fifteenth embodiment may include the thirteenth embodiment, wherein the porosity estimation module is configured to: cluster each of the selected ones of the identified rock debris particles in a multi-dimensional color and texture feature space; and estimate the porosity based on the clustering.

[0071] In a sixteenth embodiment, a method for generating a model that relates color and texture features of rock particles to known porosity of the rock particles includes: acquiring a plurality of calibrated digital images, each calibrated digital image including a plurality of rock particles having known porosity values; marking the plurality of rock particles with corresponding known porosity values; extracting a plurality of color and texture features from the marked plurality of rock particles; and determining a correlation between the plurality of extracted color and texture features and the known porosity values ​​and thereby generating the model.

[0072] A seventeenth embodiment may include the sixteenth embodiment, wherein determining the correlation comprises nearest neighbor regression based on distances between cuttings in a multi-dimensional color and texture space.

[0073] An eighteenth embodiment may include any one of the sixteenth to seventeenth embodiments, wherein the determining the correlation comprises constructing and training the model using a machine learning model that determines which of the color and texture features best correlate with the known porosity value.

[0074] A nineteenth embodiment may include any one of the sixteenth to eighteenth embodiments, wherein the plurality of rock particles are obtained as rock cutting particles when drilling an underground wellbore; and the known porosity value is obtained by porosity logging measurements performed when drilling the underground wellbore or porosity measurements performed on core samples obtained from the underground wellbore.

[0075] The twentieth embodiment may include any one of the sixteenth to nineteenth embodiments, the twentieth embodiment further comprising acquiring another calibrated digital image of rock cuttings particles generated while drilling a subterranean wellbore, the rock cuttings particles having unknown porosity; processing the another calibrated digital image to generate a segmented image, the segmented image identifying individual ones of the rock cuttings particles depicted in the calibrated digital image; extracting a plurality of color and texture features from selected ones of the identified rock cuttings particles depicted in the segmented image; and estimating the unknown porosity of each of the selected ones of the identified rock cuttings particles using a model to evaluate the plurality of color and texture features based on the selected ones of the identified rock cuttings particles.

[0076] In a twenty-first embodiment, a method for estimating the porosity of a rock sample comprises: acquiring a calibrated digital image of the rock sample; processing the digital image to extract color features and texture features from the rock sample; and processing the extracted color features and texture features with a model to estimate the porosity of the rock sample, wherein the model relates the digital image color features and texture features to porosity.

[0077] The twenty-second embodiment may include the twenty-first embodiment, wherein the rock sample comprises a core plug sample or drill cuttings particles.

[0078] Although formation porosity estimation from digital images has been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the disclosure as defined by the appended claims.

Claims

1. A method for estimating the porosity of a subterranean formation, the method comprising: acquiring a calibrated digital image of rock debris particles generated by drilling a wellbore in said subterranean formation; extracting a plurality of color and texture features from the calibrated digital image of the rock cuttings particles; as well as The porosity of the subsurface formation is estimated using a model, wherein the model includes correlations between the plurality of color and texture features and the porosity.

2. The method of claim 1, wherein acquiring the calibrated digital image comprises: Drilling underground wells; collecting the rock cutting particles from the circulating drilling fluid; preparing the rock cuttings particles; as well as Calibrated digital photographs of the prepared rock cuttings particles were taken.

3. The method of claim 1 , further comprising segmenting the calibrated digital image to generate a segmented image, the segmented image identifying individual ones of the rock debris particles depicted in the calibrated digital image, wherein: The extracting includes extracting a plurality of color and texture features corresponding to selected ones of the identified respective rock debris particles in the segmented image; and The estimating includes estimating the porosity of each of the selected particles using the model and calculating a mean or median value of the estimated porosity of each of the selected particles to estimate the porosity of the subsurface formation. The method of claim 3 , wherein said segmenting, said extracting and said estimating are performed automatically.

5. The method of claim 3, wherein said using said model to estimate said porosity of each of said selected particles comprises: determining a position of each of the selected particles in a multidimensional color and texture feature space; as well as The porosity of each of the selected particles is estimated based on the position of each of the selected particles in the multi-dimensional color and texture feature space.

6. The method of claim 3, wherein said using said model to estimate said porosity of each of said selected particles comprises: clustering each of the selected particles in a multidimensional color and texture feature space; as well as The porosity of each of the selected particles is estimated based on the clustering.

7. The method of claim 1, wherein said using said model to estimate said porosity of each of said selected particles comprises: labeling each of the identified rock debris particles in the segmented image with a corresponding lithology type based on the plurality of color and texture features; selecting, from a plurality of models, the model corresponding to the labeled lithology type for each of the selected ones of the identified rock fragments particles; as well as The porosity of each of the selected particles is estimated using the selected model, wherein the selected model is configured to correlate the plurality of color and texture features with the porosity of rock fragment particles of the marked lithology type.

8. The method of claim 7, wherein the rock debris particles include at least a first lithology type and a second lithology type that are different.

9. The method of claim 1, wherein the model is generated by: acquiring a plurality of calibrated digital images of rock cutting particles having known porosity values; marking the rock cuttings particles having the known porosity values ​​with the corresponding known porosity values; extracting a plurality of color and texture features from the labeled rock cuttings particles; as well as Correlations between the plurality of extracted color and texture features and the porosity values ​​are determined to generate the model.

10. The method of claim 9, wherein said determining correlation comprises nearest neighbor regression based on distances between said rock fragments particles having known porosity values ​​in a multi-dimensional color and texture space.

11. The method of claim 9, wherein said determining the correlation comprises constructing and training said model using a machine learning model that determines which of said color and texture features best correlates with said known porosity value.

12. A system for estimating the porosity of a subsurface formation, the system comprising: a digital camera system configured to capture calibrated digital images of rock particles from the subsurface formation; as well as A digital image processing system comprising a plurality of modules, wherein the modules include: a color and texture feature extraction module configured to extract color features and texture features from the calibrated digital image; as well as A porosity estimation module is configured to estimate the porosity of the subsurface formation using a model that relates the extracted color features and texture features to porosity.

13. The method of claim 12, wherein the module further comprises: a segmentation module configured to process the calibrated digital image to identify individual rock debris particles in the calibrated digital image; a merging module configured to calculate an average porosity value based on the estimated porosity values ​​of selected particles among the identified rock fragment particles; wherein the color and texture feature extraction module is configured to extract color features and texture features from the selected particles among the identified rock debris particles; and The porosity estimation module is configured to estimate the porosity value of the selected ones of the identified rock fragments particles using the model.

14. The system of claim 13, wherein the porosity estimation module is configured to: determining a position of each of said each of said selected ones of said identified rock debris particles in a multi-dimensional color and texture feature space; and The porosity is estimated based on the position of each of the selected ones of the identified rock fragments particles in the multi-dimensional color and texture feature space.

15. The system of claim 13, wherein the porosity estimation module is configured to: clustering each of said selected ones of said identified rock debris particles in a multi-dimensional color and texture feature space; and The porosity is estimated based on the clustering.

16. A method for generating a model relating color and texture characteristics of rock particles to the known porosity of the rock particles, the method comprising: acquiring a plurality of calibrated digital images, each calibrated digital image including a plurality of rock particles having known porosity values; labeling the plurality of rock particles with corresponding known porosity values; extracting a plurality of color and texture features from the labeled plurality of rock particles; as well as Correlations between the plurality of extracted color and texture features and the known porosity values ​​are determined and the model is generated therefrom.

17. The method of claim 16, wherein determining the correlation comprises nearest neighbor regression based on distances between rock fragments in a multi-dimensional color and texture space.

18. The method of claim 16, wherein said determining the correlation comprises constructing and training said model using a machine learning model that determines which of said color and texture features best correlates with said known porosity value.

19. The method of claim 16, wherein: The plurality of rock particles are captured as rock cuttings particles while drilling a subterranean wellbore; and The known porosity value is obtained by porosity logging measurements made while drilling the subterranean wellbore or by porosity measurements made on core samples obtained from the subterranean wellbore.

20. The method of claim 16, further comprising: acquiring another calibrated digital image of rock cuttings particles generated while drilling a subterranean wellbore, the rock cuttings particles having an unknown porosity; processing the other calibrated digital image to generate a segmented image that identifies individual ones of the rock cuttings particles depicted in the calibrated digital image; extracting a plurality of color and texture features from selected ones of the identified rock debris particles depicted in the segmented image; as well as The unknown porosity of each of the selected ones of the identified rock fragments is estimated using a model to evaluate the plurality of color and texture characteristics based on the selected ones of the identified rock fragments.

Images