METHODS AND SYSTEMS FOR BOREHOLE TEXTURE ANALYSIS

Description

CROSS REFERENCE TO RELATED APPLICATION

The present application claims priority to U.S. Provisional Patent Application No. 63/641,135, filed on May 1, 2024, which is incorporated herein by reference in its entirety for all purposes.

BACKGROUND

The present disclosure relates generally to methods and systems for performing borehole texture analysis of borehole images. More specifically, the present disclosure is related to analyzing borehole image data to identify different types of textures presented therein. Interpretation of borehole images may be utilized for depositional environment analysis and to identify characteristics of a borehole, such as natural or drilling-induced fractures, formation heterogeneity, and sedimentary structure. Different lithology categories may be used to delineate different reservoir types, and each basin may have specific terminologies corresponding to their rock types. However, interpreting borehole images to identify and classify texture features based on certain characteristics of the borehole images may be difficult. Further, the identification and classification process may be inefficient with respect to time and resources (e.g., computing resources, energy).

This section is intended to introduce the reader to various aspects of art that may be related to various aspects of the present disclosure, which are described and/or claimed below. This discussion is believed to be helpful in providing the reader with background information to facilitate a better understanding of the various aspects of the present disclosure. Accordingly, it should be understood that these statements are to be read in this light, and not as admissions of prior art.

BRIEF DESCRIPTION

In certain embodiments, a method including segmenting a borehole image of a first well into a first plurality of zones based on pixel data, segmenting the borehole image of the first well into a second plurality of zones based on covariance data, merging the first plurality of zones and the second plurality of zones to generate an updated borehole image, clustering one or more sets of features of the updated borehole image into one or more clusters based on a classification algorithm, and generating a borehole texture model representative of expected properties of an additional borehole based on the one or more clusters.

In certain embodiments, a system including a controller having a processor, a memory, and instructions stored on the memory and executable by the processor to segment a borehole image of a first well into a first plurality of zones based on pixel data, segment the borehole image of the first well into a second plurality of zones based on covariance data, merge the first plurality of zones and the second plurality of zones to generate an updated borehole image, cluster one or more sets of features of the updated borehole image into one or more clusters based on a classification algorithm, and generate a borehole texture model representative of expected properties of an additional borehole based on the one or more clusters.

In certain embodiments, a tangible and non-transitory machine readable medium including instructions that, when executed by a processor, causes the processor to perform operations including segmenting a borehole image of a first well into a first plurality of zones based on pixel data, segmenting the borehole image of the first well into a second plurality of zones based on covariance data, merging the first plurality of zones and the second plurality of zones to generate an updated borehole image, clustering one or more sets of features of the updated borehole image into one or more clusters based on a classification algorithm, and generating a borehole texture model representative of expected properties of an additional borehole based on the one or more clusters.

Certain embodiments commensurate in scope with the originally claimed invention are summarized below. These embodiments are not intended to limit the scope of the claimed invention, but rather these embodiments are intended only to provide a brief summary of possible forms of the invention. Indeed, the invention may encompass a variety of forms that may be similar to or different from the embodiments set forth below.

DRAWINGS

These and other features, aspects, and advantages of the present disclosure will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:

FIG. 1 is a schematic view of a drilling system including a downhole tool string, in accordance with aspects of the present disclosure;

FIG. 2 is a flow chart depicting a method of generating a borehole texture model from sensor data;

FIGS. 3A to 3E illustrate a high-resolution borehole image and

corresponding histograms related to kernel density estimation and a frequency of pixels corresponding to data related to a borehole image;

FIGS. 4A to 4D illustrate a discretized borehole image, frequency curves, and zone boundaries corresponding to data related to a borehole image;

FIGS. 5A to 5D illustrate variograms, covariance, and zone boundaries corresponding to data related to a borehole image;

FIG. 6 illustrates zone boundaries computed from frequencies in FIGS. 3A-3E, covariance in FIGS. 5A-5D, and merged;

FIG. 7 illustrates a variogram with geometrical parameters corresponding to the merged zone boundaries;

FIG. 8 illustrates an auto-correlogram in the Y-direction of the borehole image corresponding to the merged zone boundaries;

FIG. 9 illustrates a dendrogram plot of a hierarchical tree corresponding to the variogram and auto-correlogram;

FIG. 10 illustrates samples of a training dataset with columns including, global dynamically equalized samples, local dynamically equalized samples, Gamma Ray as an image, and the class each sample belongs to;

FIG. 11 illustrates convergence of a conventional neural network (e.g., SqueezeNet) model trained from scratch;

FIG. 12 illustrates a method operating a drill based on laveled borehole image data retrieved from a borehole texture model;

FIG. 13 illustrates an example of classification performed on samples of a new well that has been already zoned and clustered;

FIG. 14 illustrates a conventional neural network (e.g., SqueezeNet) model predicted lithofacies labeling and mismatching statistics;

FIG. 15 illustrates an example of a texture analysis in a high clay content formation;

FIG. 16 illustrates an example of a texture analysis in a low clay content formation; and

FIG. 17 illustrates an example of a texture prediction with the conventional neural network (e.g., SqueezeNet) model.

DETAILED DESCRIPTION

One or more specific embodiments of the present disclosure will be described below. In an effort to provide a concise description of these embodiments, all features of an actual implementation may not be described in the specification. It should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which may vary from one implementation to another. Moreover, it should be appreciated that such a development effort might be complex and time consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure.

When introducing elements of various embodiments of the present disclosure, the articles “a,” “an,” “the,” and “said” are intended to mean that there are one or more of the elements. The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements. Any examples of operating parameters and/or environmental conditions are not exclusive of other parameters/conditions of the disclosed embodiments.

Borehole images may provide information related to texture features that may assist in performing natural fracture identification, drilling-induced feature classification, depositional environment analysis, and the like. Different lithology categories may delineate different reservoir types that may be present within the borehole. With this in mind, the present disclosure details a method in which a data processing system may analyze borehole image data to identify different types of textures that may be present in the respective subsurface region of the earth.

By way of example, the data processing system may receive borehole image data associated with a well or subsurface region and segment the borehole image based on the frequency of pixel data and the difference in pixel value (e.g., image density change). The data processing system may then cluster the resulting image segments based on similarities between different segments. The clustered group of image segments may then be classified with respect to certain categories (e.g., pixel frequency, image density) based on different purposes (e.g., rock texture, crevice depth, geological material, borehole structures). In addition to classifying the image segments, the data processing system may retrieve petrophysical logs that may be associated with the respective borehole, well, subsurface region, or the like. The data processing system may then store the classifications of the segmented borehole images along with the corresponding petrophysical logs within the respective clustered groups. Using the segmented borehole images, clustered groups, and labeled data, the data processing system may train a deep learning model based on patterns, correlations, and associations between the petrophysical logs related to certain clustered groups with respect to different borehole image segments.

After training the model, the data processing system may receive additional borehole image data related to other borehole and identify the image features that may assist the data processing system in determining the different classes and categories of rock properties through the length of the respective borehole. In this way, the data processing system may perform continuous facies analysis of the newly acquired image data based on similar characteristics identified in the depositional environments of the newly received borehole image data and the borehole image data represented in the model. In other words, by performing the techniques described herein, the data processing system may classify sedimentary facies (e.g., bodies of rock) in a continuous spectrum that may highlight gradual changes in the depositional environment that more accurately represents the actual rock properties of the respective borehole. Rock properties may include mechanical strength, hydraulic conductivity, rock texture, crevice depth, geological material, borehole structures, and the like. Additional details with respect to performing borehole image analysis to determine different textures present therein will be discuss below with reference to FIGS. 1-17.

By way of introduction, FIG. 1 illustrates a drilling system 10 that may employ the systems and methods of this disclosure. The drilling system 10 may be used to drill a borehole 12 into a geological region 14. In the drilling system 10, a drilling rig 18 may rotate a drill string 20 within the borehole 12. As the drill string 20 is rotated, a drilling fluid pump 22 may be used to pump drilling fluid, which may be referred to as “mud” or “drilling mud,” downward through the center of the drill string 20, and back up around the drill string 20, as shown by reference arrows 24. At the surface, return drilling fluid may be filtered and conveyed back to a mud pit 26 for reuse. The drilling fluid may travel down to the bottom of the drill string 20 known as the bottom-hole assembly (BHA) 28. The drilling fluid may be used to rotate, cool, and/or lubricate a drill bit 30 that may be a part of the BHA 28. The fluid may exit the drill string 20 through the drill bit 30 and carry drill cuttings away from the bottom of the borehole 12 back to the surface. One or more surface sensors 31 may record a variety of different data points associated with the drilling system 10, including the rotations per minute (RPM) of the drill string 20 and/or the drill bit 30. For example, the set of sensors 31 may determine the surface RPM of the drilling system 10. In addition, the sensors 31 may be positioned within the drill string 20 to capture data related to properties of the drill string 20, the drill bit 30, and the like while inside the borehole 12.

The BHA 28 may include the drill bit 30 along with various downhole tools, such as one or more logging tools 32. The BHA 28 may thus convey the one or more logging tools 32 through the geological region 14 via the borehole 12. As described in greater detail herein, the one or more logging tools 32 may be any suitable downhole tool that emits electromagnetic waves within the borehole 12 (e.g., a downhole environment). The downhole tools, which may include the one or more logging tools 32, may collect a variety of information relating to the geological region 14 and the state of drilling in the borehole 12. For instance, the downhole tools may be logging-while drilling (LWD) tools that measure physical properties of the geological region 14, such as density, porosity, resistivity, lithology, and so forth. Likewise, the downhole tools may be measurement-while-drilling (MWD) tools that measure certain drilling parameters, such as the temperature, pressure, orientation of the drill bit 30, mapping-while-drilling tools, and so forth.

The one or more logging tools 32 may receive energy from an electrical energy device or an electrical energy storage device, such as an auxiliary power source 34 or another electrical energy source to power the tool. In some embodiments, the one or more logging tools 32 may include a power source within the one or more logging tools 32, such as a battery system or a capacitor, to store sufficient electrical energy to emit and/or receive electromagnetic waves.

The one or more logging tools 32 may also include image acquisition tools that may obtain image data related to the borehole 12. The image acquisition tools may include an acoustic borehole imager, resistivity imaging tools, and the like. In some embodiments, the acquired image data may present information related to the geological region, such as bedding plans, fractures, sediment structures, breakouts, borehole shape, and the like.

The drilling system 10 may include a controller 35 to control different components of the drilling system 10 and collect identified data from the one or more logging tools 32 and/or the one or more sensors 31. The controller 35 may be any electronic data processing system that can be used to carry out the systems and methods of this disclosure. That is, the controller 35 may monitor and regulate various operational parameters of the drilling system 10. The controller 35 may receive data from the sensors (e.g., the sensors 31 and/or logging tools 32 measuring parameters such as drilling depth, rotational speed, torque, pressure, image data, and vibration). Based on these inputs, the controller 35 may perform various actions adjusting drilling variables, including bit rotation speed, feed rate, and fluid flow.

Communications 36, such as control signals, may be transmitted from a data processing system 38 (processing system 38) to the controller 35 and the communications 36, such as data signals related to the results/measurements of the sensors 31 and/or one or more logging tools 32, may be returned to the data processing system 38 via the controller 35. The data processing system 38 may be any electronic data processing system that can be used to carry out the systems and methods of this disclosure. For example, the data processing system 38 may include one or more processors 40, which may execute instructions stored in memory 42 and/or storage 44. The memory 42 and/or the storage 44 of the data processing system 38 may be any suitable article of manufacture that can store the instructions. In certain embodiments, the one or more processors 40 may include a microprocessor, a microcontroller, a processor module or subsystem, a programmable integrated circuit, a programmable gate array, a digital signal processor (DSP), or another control or computing device. In certain embodiments, the one or more processors 40 may include machine learning and/or artificial intelligence (AI) based processors.

In certain embodiments, the memory 42 and storage 44 may be implemented as one or more non-transitory computer-readable or machine-readable storage media. In certain embodiments, the memory 42 may include one or more different forms of memory, including semiconductor memory devices such as dynamic or static random-access memories (DRAMs or SRAMs), erasable and programmable read-only memories (EPROMs), electrically erasable and programmable read-only memories (EEPROMs) and flash memories. The storage 44 may include solid state drives, magnetic disks such as fixed, floppy and removable disks; other magnetic media including tape; optical media such as compact disks (CDs) or digital video disks (DVDs); or other types of storage devices. Note that the computer-executable instructions and associated data of the analysis module(s) may be provided on one computer-readable or machine-readable storage medium of the memory 42 or the storage 44, or alternatively, may be provided on multiple computer-readable or machine-readable storage media distributed in a large system having possibly plural nodes. Such computer-readable or machine-readable storage medium or media are considered to be part of an article (or article of manufacture), which may refer to any manufactured single component or multiple components. In certain embodiments, the storage 44 may be located either in the machine running the machine-readable instructions or may be located at a remote site from which machine-readable instructions may be downloaded over a network for execution.

As illustrated, the data processing system 38 may optionally also include a display 46, which may be any suitable electronic display, and may display images generated by the processor 40. The data processing system 38 may be a local component of the drilling system 10 (i.e., at the surface), within the one or more logging tools 32 (i.e., downhole), a device located proximate to the drilling operation, and/or a remote data processing device located away from the drilling system 10 to process downhole measurements in real time or sometime after the data has been collected. In some embodiments, the data processing system 38 may be a portable computing device (e.g., tablet, smart phone, or laptop) or a server remote from the drilling system 10. In some embodiments, the one or more logging tools 32 may store and process collected data in the BHA 28 or send the data to the surface for processing via communications 36 described above, including any suitable telemetry (e.g., electrical signals pulsed through the geological region 14 or mud pulse telemetry using the drilling fluid).

It should be noted that, although the discussion above relates to a drilling system, other downhole equipment or systems may employ the systems and methods of this disclosure. For example, a downhole tool with an acoustic tool conveyed by slickline, coiled tubing, wireline, or other delivery systems, may utilize the disclosed systems and methods.

Keeping the foregoing in mind, borehole image data may be acquired using some of the equipment described above with respect to FIG. 1. The borehole image data may then be analyzed to determine the texture and facies that may be part of the borehole. For example, FIG. 2 illustrates a flow chart of a method 50 for generating a borehole texture model based on borehole image data. Although the following description of the method 50 will be described as being performed by the data processing system 38, it should be noted that the method 50 may be performed by any suitable computing device in any suitable order. Moreover, at times, the description of the method 50 below may refer to the components described above in FIG. 1; however, it should be understood that these references are merely provided as examples to further clarify the discussion of the method 50.

Referring now to FIG. 2, at block 52, the data processing system may receive image data associated with a borehole. The data processing system may receive the image data from the logging tools 32, sensors disposed inside the borehole, or the like. The image data may include still images of the borehole around the circumference of the borehole. As such, the image data may provide insight as to a mapping and texture of the borehole.

In some embodiments, the logging tools 32 may send associated position and accelerometer data along with the image data. The acceleration data and position data may provide information about the location within the borehole from which the image data originates. By way of example, FIG. 3a illustrates an example image data of a borehole including raw high resolution dynamic image of the borehole. The example image data of FIG. 3a may be analyzed to generate histograms and derivatives of the same, such as those illustrated in FIGS. 3b, 3c, 3d, and 3e. These derivatives of the example image data may be used to determine properties and characteristics of the borehole and train a model that may be used to more efficiently determine properties of other boreholes in an efficient manner. In some embodiments, the model may also provide insight into the expected rock properties of the borehole and may be used to control a drilling system based on the expected properties and characteristics of the borehole during the drilling operation. That is, drilling parameters (e.g., speed, revolutions-per-minute, technique, direction) may be adjusted in view of the expected properties of the rock within the borehole. The assorted horizontal lines on the high-resolution image may indicate structures present within the borehole, which an operator may need to address through adjusting drilling operations.

At block 54, the data processing system may segment the borehole image data received at block 52 based on pixel data. That is, the borehole image may be represented as a composition of successive sedimentary zones. The different sedimentary zones of a borehole image may have different statistical properties (e.g., pixel properties). The statistical properties may be used to characterize portions of the borehole image and generate zonation (e.g., segmented zones). In some embodiments, the data processing system may utilize kernel density estimation (KDE), a variogram, and the like to extract certain features of a borehole image. For example, the data processing system may utilize KDE and variogram features to segment the entire borehole image into different sedimentary zones.

With this in mind, in some embodiments, to segment the borehole image data based on pixel data, the data processing system may generate a histogram to describe the distribution of pixel values (e.g., frequency of pixels) in the borehole image in a normalized pixel domain of [0, 1]. For example, FIG. 3b illustrates an example that indicates a number of pixels (Y-axis) presented as columns that have a corresponding normalized pixel intensity (e.g., brightness or color value) (X-axis). The histogram composed of columns, however, may have limited precision due to the value domain being discretized.

As such, the data processing system may increase the precision of the histogram using kernel density estimation (KDE). The kernel density may correspond to a Gaussian distribution with the sample value as μ and a variance controlling the bandwidth σ²: (μ, σ²). As a result, the data processing system may produce a distribution of pixel values of the borehole image, such that the pixel value may be provided at a higher precision than the histogram. That is, contrary to the usual “discrete” histogram where each sample from input data increases a by count of one to the column corresponding to the sample value, the data processing system may form a “continuous” histogram, such that each sample contributes a density distribution with a given bandwidth called kernel density. That is, the KDE generated from the borehole image (e.g., whole image) may provide information about a distribution of pixel values (e.g., frequency of pixels) within the borehole image.

By performing the KDE, the data processing system may generate a smooth curve, which may lend it to analytical methods that may be used to find peaks and troughs. As will be detailed below, the data processing system may focus on each of the expected pixel value ranges to identify the frequency of pixels falling into those ranges. In this way, the frequency of pixels may provide one of the features used by the data processing system to zonate and classify the borehole image.

For instance, the data processing system may calculate a KDE of the whole borehole image with a variance, σ=0.15 (for normalized pixel value domain [0, 1]). By way of example, FIG. 3c illustrates an example continuous final distribution determined using the KDE, thereby providing information at multiple precise values. Indeed, the KDE generated from the borehole image provides information about distribution of pixel values within the borehole image.

After calculating the KDE of the borehole image, the data processing system may then calculate a second derivative of the KDE, resulting in a continuous distribution of pixel frequencies. For example, FIG. 3d illustrates the second derivative of the KDE represented in FIG. 3c.

The data processing system may then analyze the borehole image with respect to pixel data to determine statistical properties of the borehole image. For example, the data processing system may apply an analytical method to the continuous histogram of FIG. 3c to determine peaks and troughs. The data processing system may analyze the frequency of pixels within different ranges of pixel values to determine segments of the borehole image. As such, the data processing system may use frequency as one of the features to classify each segment of the borehole image.

Using the second derivative of the KDE illustrated in FIG. 3d, the data processing system may determine a number n maxima of the second derivative, as illustrated in FIG. 3e. The n maxima may indicate the troughs of the KDE. The data processing system may annotate the positions of the identified troughs in a value domain as {Xi|i=1, . . . , n}. The data processing system may also utilize a smart searching algorithm to annotate the positions in the value domain. The smart searching algorithm may take into account the maxima separated by a negative minima. For instance, FIG. 4 illustrates an example in which the retained maxima are {X_i}={0.13, 0.64, 0.88}. Keeping this in mind, the data processing system may use the retained maxima illustrated in FIG. 3e serve to cut off the pixel value domains into 4 intervals 66a, 66b, 66c, and 66d. The frequency of all of the pixel values sum up to 1. As such, the data processing system may utilize some or all intervals as independent features. Specifically, the data processing system may characterize the image utilizing only the low intervals (e.g., below a lower threshold value) and high intervals (e.g., above an upper threshold value).

By way of example, FIG. 4 illustrates a process for determining the frequency of the pixels and the KDE. In the illustrated example, the data processing system uses {Xi}={0.13, 0.88}. FIG. 4a illustrates a discretized borehole image. In the illustrated example, to discretize the image by pixel values {X_i}, for each pixel value v the data processing system may utilize formula 1.

X i <= v < X i + 1 , then ⁢ v = ( X i + X i + 1 ) / 2 ( 1 )

Formula 1 may sharpen the features found by the histogram. In the illustrated embodiment, the borehole image has n+1 discrete values as shown in formulas 2-4 below.

C 1 ≡ X 1 / 2 , ( 2 ) C i ≡ ( X i + X i + 1 ) / 2 , I = 1 , … , n , ( 3 ) C n + 1 ≡ ( X n + 1 ) / 2. ( 4 )

As a result of the discretization process, the embodiment illustrated in FIG. 4a has three values {C_i}={0.065, 0.505, 0.96}.

After the data processing system discretizes the image, the data processing system may determine a frequency curve as illustrated in FIG. 4b. To determine the frequency curve, the data processing system may first, for each depth z, count the frequency of each possible pixel values {C_i, i=1, . . . , n+1}, in a sliding window [z+h, z−h], where the half height h of the window may be chosen to be 20, for example. In the illustrated embodiment, the result is n+1 frequency curves as a function of depth. As discussed above, at a given depth, all frequencies sum up to 1. In embodiments of 3 values, such as the illustrated embodiment, one frequency curve, for example the one computed at middle value C₁, may be ignored without issue. The two frequency curves generated in the present embodiment are illustrated in FIG. 4b. The data processing system may normalize the frequency by number of pixels in horizontal direction. In the present embodiment, the range is [0, 1].

The data processing system may then create a graph of the first derivative of frequencies based on the frequency curve from the previous step. FIG. 4c illustrates a first derivative of frequencies curve. Once the data processing system determines a first derivative of the frequencies, it may determine minima and maxima of the frequency curves. The minima 78 and maxima 80 may indicate the steepest change in frequency, which may suggest possible zone boundaries 82. FIG. 4e illustrates such located boundaries 82 superposed on the original borehole image from FIG. 3a.

At block 56, the data processing system may segment the borehole image based on covariance data. That is, the data processing systems may utilize variograms in place of histograms to zonate the borehole image. A variogram may measure covariance of pairs of data points separated by a given distance. The data processing system may utilize diverse types of variogram to characterize borehole images. An omni-directional variogram, for example, describes the homogeneity of an image. As such, an omni-directional variogram may be sensitive to the texture, while single directional variogram may detect image density change in a direction. In particular, when crossing a zone boundary in the borehole image, difference in pixel value may be larger than some threshold. Thus, the covariance (e.g., square of difference) may be larger than the threshold at certain locations (e.g., expected to reach a maximum at the zone boundary).

FIG. 5 illustrates the zonation process using vertical directional variograms. As an example, the vertical variogram of borehole image corresponding to FIG. 3 is as follows in FIG. 5a. As illustrated in FIG. 5a, the covariance illustrated on the Y axis depends on distance separating data points (e.g., range) illustrated on the X axis. At a distance of zero, covariance vanishes as a point value is equal to itself. In some instances in which the range is greater than some threshold, the data points may become uncorrelated. As such, the covariance saturates and reaches the variance of data. To determine a range that may be best suited for describing the covariance change in a borehole image, the data processing system may take the second derivative of the variogram. To take the second derivative of the variogram, the data processing system may calculate the vertical variogram of the whole borehole image as illustrated on the plot in FIG. 5a. The data processing system may then calculate the second derivative of the vertical variogram as illustrated on the plot in FIG. 5b. Once the data processing system calculates the second derivative, it may then determine the first negative minimum of the second derivative (R_m), which gives a standard range value in the variogram before reaching the saturated plafond (e.g., maximum emissions). In the illustrated example, R_m=40 (pixel). For each depth (z), calculate the value at which the covariance equals the variogram value at range R_m, in a sliding window [z+h, z−h], where the half height h of the window equals to R_m. The covariance curve as a function of depth is illustrated in FIG. 5c. The data processing system may then determine the maxima 94 of the above covariance, suggesting possible zone boundaries. FIG. 5d illustrates such found boundaries 96 superimposed on the original image.

Referring back to the method 50 of FIG. 2, after performing the borehole image segmentation based on the pixel data (block 54) and the covariance data (block 56), at block 58, the data processing system may merge borehole boundary data determined at blocks 54 and 56. As illustrated in FIG. 6, the merged borehole boundary data may be determined based on a combination of the boundary locations 82 identified based on the pixel data as described in block 54 and illustrated in borehole image 170a and the boundary data 96 identified based on the covariance data as described in block 56 and illustrated in borehole image 170b. When the borehole boundary data 82 of the borehole image 170a is combined with the boundary data 96 of the borehole image 170b, the data processing system may generate merged borehole image 170c with merged boundary data 72.

After the borehole boundary data is merged at block 58, at block 60, the data processing system may cluster borehole image features of the segmented borehole image based on a classification algorithm. The merged boundary data may provide an indication of different sedimentary zones within the borehole.

For high resolution borehole images, the data processing system may perform the unsupervised classification (e.g., without any prior knowledge) utilizing the sedimentary zones obtained from the segmentation and merging steps described above. For each segmented borehole image, the data processing system may extract a given number N of statistic properties (e.g., features). The data processing system may represent each segment of the borehole image as a point in a space of an N dimension (e.g., feature space). Each axis of the N dimension may correspond to a statistic property (e.g., feature). The classification algorithm (e.g., clustering algorithm) used by the data processing system may group the points into No clusters, which verify that the distance of points inside each cluster is minimal and distance between clusters is maximal.

In the feature space of high dimensionality (e.g., greater than 5, greater than 10), the local optima may be higher than some threshold value making it difficult to analyze. As such, it may be desirable to utilize a clustering method in which the local optimum does not depend on the random initialization and the divergence of determined local minima does not appear arbitrary. To minimize stochastic (e.g., random) results, the data processing system may utilize agglomerative clustering, which may provide a robust result via a dendrogram (e.g., tree-like representation). Agglomerative clustering is a method of hierarchical clustering algorithm based on a similarity known as agglomerative nesting (AGNES), which uses a bottom-up approach. During the AGNES method, an object is initially considered a single-element cluster. All clusters are successively merged based on the similarities between the clusters. Newly formed clusters (e.g., composed of single-element clusters) are linked to other newly formed clusters to create bigger clusters. The clustering process is iterated until all points are a member of a single big cluster (e.g., root).

In addition to the features used for zonation, the data processing system may utilize more statistical properties in its classification process, which may provide more opportunities for the data processing system to characterize the image with thin layers or with a highly hetero-homogeneous texture. For example, the data processing system may utilize a variogram to determine statistical properties that can be used in classifying a borehole. Different image textures may generate variograms with different shapes. The data processing system may extract more parameters from a variogram to use as classification features. FIG. 7 illustrates an example variogram 120 that exemplifies a shoulder height 122, a sill 124 (e.g., the value that the variogram model attains at the range), and a range 126 (e.g., distance after which the variogram levels off).

The data processing system may also utilize a correlogram to determine statistical properties that can be used in classifying a borehole. The data processing system may utilize an auto-correlogram to characterize the vertical variability of borehole images. The auto-correlogram in a Y direction may provide the data processing system with methods to compute autocorrelation of each trace of zone image and determine a sum of the autocorrelation of all the traces. Once the data processing system has determined the sum of the autocorrelation of all the traces, the data processing system may generate an auto-correlogram 130 (e.g., curve of averaged autocorrelation vs. vertical lag distance 132) as illustrated in FIG. 8.

Using the auto-correlogram, the data processing system may extract a maximum lag (e.g., furthest time interval at which the autocorrelation function is calculated) and a correlation strength 134 from the second peak in the auto-correlogram 130. Both the maximum lag and correlation strength 134 may be classification features that may be used in embodiments described herein. Furthermore, the data processing system may utilize any extra logging curve with any statistical feature extracted from any borehole image. For instance, the data processing system may compute a mean value of the logging curve within the depth ranges of image zonation and assign the mean value to that image segment as an additional feature.

After the data processing system extracts the desired statistical properties from the variogram and correlogram, the data processing system may complete a similarity computation. In some embodiments, the data processing system may compute a distance metric or a measure of similarity between clusters. For a single point, the data processing system may utilize Euclidean (e.g., straight-line distance between two points in Euclidean space) or Manhattan (e.g., distance between two points by summing the absolute distances of their coordinates) distances. Conversely, for a set of points, the data processing system may utilize a linkage criterion such as Single-linkage (e.g., shortest distance between any two members of two clusters, one from each cluster) or Ward linkage (e.g., analyzes the variance of clusters). In some embodiments, the data processing system may utilize the Euclidean distance and Ward linkage. In other embodiments, the data processing system may utilize the Manhattan distance or the Single-linkage. The data processing system may output the result of the clustering as a binary tree, which may show the sequence in which clusters were merged and the distance at which each merge took place. By way of example, FIG. 9 illustrates an embodiment of a hierarchical tree presented as a dendrogram plot 140. The length of branches 142 represents the distance between clusters before they are merged. As such, longer branches imply more separated clusters, which may stop the merging process at this stage. However, the long branches underneath the root may be excluded to avoid too few clusters.

In the illustrated embodiment, the longest branch with a number of clusters bigger than four is depicted with a red dash 144. As such, the data processing system may use the longest branch with a number of clusters larger than four to cut the hierarchical tree in the dendrogram plot 140 and enable a first guess at the number of clusters. A user may select the next n^th longest branch, which may cause the data processing system to provide the user with more clusters if the user desires. The n is a parameter exposed to the users. The n may default as 0 for the initial cluster number. Further, negative number values for n may result in fewer cluster numbers than the previous n^th longest branch may determine. Once the data processing system has determined the initial and adjusted cluster number, the data processing system may improve the zonation results by merging any adjacent zones having the same class.

By way of example, the data processing system may use features such as a frequency at low pixel value (the same one used for zonation), a frequency at high pixel value (the same one used for zonation), a sill of variogram in X direction, a ratio of shoulder height to sill of variogram in X direction, maximum lag of auto-correlogram in X direction, maximum correlation strength of auto-correlogram in Y direction, and gamma ray logging curve as inputs into the clustering algorithm or process described above. Indeed, the data processing system may employ the agglomerative clustering algorithm with these features to identify zones based on classes identified using the features. In the same manner, the data processing system may identify other classes or clusters using other combinations of features.

At block 62, the data processing system may generate a borehole texture model to associate borehole image data to well log data based on clusters. That is, the data processing system may use the clusters identified in the borehole image as training data for the borehole texture model. In this way, newly acquired borehole images or sensor data related to other boreholes may be analyzed with respect to the borehole texture model to more efficiently identify the segments in accordance with the embodiments described herein.

For each zone of each borehole image, the agglomerative clustering algorithm may assign one class from the inferred classes determined in block 60 described above. However, the agglomerative clustering algorithm may not be capable of learning or gaining insights from the data. As such, instead of using the agglomerative clustering algorithm to classify unseen data, the data processing system may design a texture modeling system (e.g., supervised classifier) to detect and learn the association between inferred classes determined in the clustering step of block 60 and the properties of inputs, such as borehole image samples and conventional petrophysical logs. Indeed, the data processing system may generate a texture model (e.g., supervised classifier) that may be used to learn or gain insights based on the results of the clustering step of block 60. Specifically, results or categorizations of the clustering step of block 60 may be labelled inputs. In some embodiments, the results may be verified and cleaned (e.g., remove outliers, anomalies, repetitive data) to efficiently train the texture model (e.g., supervised classifier).

In some embodiments, the data processing system may generate the texture model (e.g., supervised classifier) by implementing a multimodal deep learning classifier neural network. The multimodal deep learning classifier neural network may utilize two-dimensional borehole images and one-dimensional conventional logs as inputs to determine one classification answer among the number of inferred classes determined in block 60 as described above. The architecture of the neural network and the data preparation performed on any inputs to the neural network may include various architectures, various input types, and various data preparation processing steps, all of which may provide varying results.

In some embodiments, the data processing system may utilize a deep neural network red-green-blue (RGB) image general purpose classifier (e.g., SqueezeNet or a similar architecture) based on convolutional neural networks (CNN) as its architecture to generate the texture model. In some embodiments, inputs may include a resistive or conductive borehole image and a matching Gamma Ray curve. As such, the data processing system may adjust the inputs of the texture model to align with the CNN architecture, the native architecture of which utilizes equally sized inputs. In embodiments where the inputs include a resistive or conductive borehole image and a matching Gamma Ray curve, the data processing system may transform the multimodal input of the texture model into three separate channels as illustrated in the chart 150 FIG. 10, which shows eight samples of the training dataset, with the columns illustrating, from left to right: the samples global dynamically equalized 152, the samples local dynamically equalized 154, Gamma Ray as an image 156, and the class each sample belongs to 158. The separate channels may include a grayscale image with global histogram equalization applied (for formation matrix information), a grayscale image with local histogram equalization applied (for textural information), and a Gamma Ray (GR) curve normalized and turned into a 2D image by repeating its values (for its relevant value). These input adjustments may expand in a similar manner to include more 1D or 2D channels as inputs.

The data processing system may also utilize inputs of a predetermined size. In some embodiments, the column count for image samples may be set to be the full column count of the original image. In some embodiments, the row count for image samples may be arbitrary. In some embodiments, the row count may be arbitrary for conventional logs. However, the row count for conventional logs may also be set to the same number of rows and repeated over the same number of columns of an image sample for consistency. In embodiments in which a zone features a row count larger than the image sample row count, the zone may be cut into multiple samples to increase the dataset sample count. While the row count may be arbitrary, it may be beneficial for the row count to be larger than some threshold to ensure that information pertaining to one zone is available. Simultaneously, it may be advantageous for the row count to be small enough to generate a sizeable training dataset. The starting size of a row count may be the smallest row size of a zone. As such, the data processing system may create a training dataset and a validation dataset for the texture model from one single well, according to this specification.

As an example in of the texture model as described above, FIG. 11 illustrates an example convergence of a trained CNN (e.g., SqueezeNet) model. The loss criterion of the trained CNN (e.g., SqueezeNet) model is cross entropy and the optimizer of the trained CNN (e.g., SqueezeNet) model is stochastic gradient descent (SGD) (e.g., updates model parameters iteratively based on the gradient of the loss function, calculated from a random subset of training data). The illustrated convergence represents a CNN (e.g., SqueezeNet) model trained using one training dataset. Based on a validation dataset, the illustrated embodiment achieves 80% accuracy.

Once the texture model (e.g., supervised classifier) is trained based on borehole image data, the data processing system may analyze other borehole images using the texture model and determine drilling commands based on the analysis. Turning now to FIG. 12, FIG. 12 illustrates a method 168 of utilizing a texture model to influence drilling operations based on borehole image data received by the texture model (e.g., supervised classifier). Although the following description of the method 168 is described as being performed by the data processing system and in a particular order, it should be noted that the method 168 may be performed by any suitable processing system and in any suitable order.

Referring now to FIG. 12, at block 172, the data processing system may receive borehole image data associated with a borehole. At block 174, the data processing system may retrieve the texture model based on the borehole image data. The texture model may be related to a particular area that corresponds to the borehole image data.

After the data processing system has the borehole image data and the texture model, the data processing system may move to block 174, where the data processing system may identify and label one or more facies of borehole image data based on the texture model.

The data processing system may apply the trained texture model to infer a classification to a new drilled well. In one embodiment, the texture model may use a sliding window, in which the row count matches the row count of images used to train the texture model. On each sliding window, the data processing system may use the texture model to perform a class inference. The texture model may be used to make this inference on each window, thereby enabling the data processing system to provide an automatic classification of the borehole image data received at block 172.

In an alternate embodiment, the data processing system may first perform segmentation (e.g., zonation) to determine reliable zone boundaries as described in relation to block 54 and block 56. Next, the data processing system may perform clustering using variograms and auto-correlograms as described in block 60. This alternate methodology may result in consistent groups of zones belonging to the same class on the new well. The data processing system may cut all zones of one same class of the new well into smaller samples. Next, the data processing system may employ the texture model to infer a classification for each sample. The data processing system may distribute samples unevenly among the inferred classes. As an example, if the data processing system infers most samples belonging to a same class of the new well A to belong to one same inferred class B, then the class A of the new well may be equivalent to the inferred class B. This alternate embodiment may adjust for errors in the texture model, which may enhance the robustness of the automatic facies classification.

An example of the classifications of the texture model are illustrated in FIG. 13. In FIG. 13, the chart 200 illustrates the classification results from the classification the texture model performed on samples of a new well which the texture model has already zoned and clustered. Each class of the new well may have a significant part of the new well's samples (e.g., greater than 40%) assigned to one training class. In the illustrated embodiment, the classes having a significant part of the new well's samples are highlighted in yellow. The texture model has identified classes highlighted in green as having been classified correctly.

As another example, FIG. 14 illustrates a CNN (e.g., SqueezeNet) predicted lithofacies labeling and mismatching statistics. The left side 212 of FIG. 14 illustrates the zonation numbers and labeling. The first column 212a and second column 212b numbers are single well image-based lithofacies clusters for Oil-based mud (OBM) and Water-based mud (WBM) respectively. The third column 212c is CNN predicted lithofacies numbers with labeling for WBM. Conversely, the right side 214 of the chart 210 in FIG. 14 illustrates mismatching statistics of the CNN lithofacies. Track 1 illustrates depth data. Track 2 illustrates gamma ray logging data. Track 3 illustrates static borehole image data. Track 4 illustrates the statistics data (green: matched, blue: empty, red: mismatched).

After the data processing system completes the this automatic texture analysis using the texture model, the data processing system may provide outputs for sedimentary analysis. One of the outputs may be the details related to texture zonation (e.g., the contiguous two zones have different texture features). Another potential output may be the clustered zonation based on texture similarity. If two contiguous zones belong to the same class, the data processing system may merge the zones within the texture model. The output may also include the texture classification for borehole image in clustered zones. The texture model may classify texture features in a high clay content formation, such as a shale formation, as illustrated in the chart 220 in FIG. 15. Similarly, FIG. 16 illustrates a chart 230 of texture features besides facies analysis in a low clay content feature like sandstone or conglomerate. Outputs provided by the texture model may further include another new drilled well, clustered zonation, and texture classification, as illustrated in chart 240 of FIG. 17. FIG. 17 illustrates a texture prediction with CNN. Further, in FIG. 17, track 1 illustrates depth data, track 2 illustrates gamma ray logging data, track 3 illustrates static borehole image data; track 4 illustrates image-based zonation and clustering data, track 4 illustrates predicted texture zonation and clustering from training model, and track 5 illustrates facies labeling.

After the texture model is used to identify and label the facies, the data processing system may present labeled borehole image data via a graphical user interface (GUI) at block 178. In some embodiments, the GUI may present the data to an end user in the form of charts as illustrated in FIG. 14-17. In other embodiments, the GUI may include methods for filtering the data or the results further. In cases in which an end user selects to filter out certain data, the data processing system may run again with only the selected data, and regenerate results based on the new data set.

At block 180, the data processing system may generate drill commands for a drill associated with a borehole based on labeled borehole image data. The drill commands may be based on the expected texture of the borehole wall, expected sedimentary information, the location of the borehole, the goals of the drilling operation and the like. The drill commands may include commands to stop or start drilling, drill at a faster or slower rate, steering the drill in a particular direction, drill to a deeper depth or a shallower depth, and the like. In some embodiments, the data processing system may provide the end user with an alert indicating the command the data processing system may send to a drill system, an actuator, a controller, or the like. In some embodiments, the user may approve or disapprove of the command. Once the data processing system generates drill commands for the drill, at block 182, the data processing system may send drill commands to the drill system for actuation or implementation. The data processing system may send the commands automatically or await user approval. In some embodiments, the data processing system may automatically send the command to the drill but may include an override system for the user to stop the command or revert the drill to a previous drilling setting. The override options may appear on the GUI on the display. In some embodiments, the data processing system may require single or dual authentication for the user to override the drill commands.

A method including segmenting a borehole image of a first well into a first plurality of zones based on pixel data, segmenting the borehole image of the first well into a second plurality of zones based on covariance data, merging the first plurality of zones and the second plurality of zones to generate an updated borehole image, clustering one or more sets of features of the updated borehole image into one or more clusters based on a classification algorithm, and generating a borehole texture model representative of expected properties of an additional borehole based on the one or more clusters.

The method of the preceding clause, wherein the borehole image is segmented based on kernel density estimation (KDE), a variogram, or both.

The method of any preceding clause, wherein the variogram corresponds to a measurement of a covariance of a pair of data points associated with borehole image.

The method of any preceding clause, wherein the updated borehole image comprises one or more boundaries between zone segments identified based on the first plurality of zones and the second plurality of zones.

The method of any preceding clause, wherein clustering the one or more sets of features comprises an agglomerative clustering method.

The method of any preceding clause, including receiving an additional borehole image associated with a second well and applying the borehole texture model to classify one or more textures of the second well.

The method of any preceding clause, including determining one or more commands for drilling the second well based on the one or more textures and sending the one or more commands to a drilling system configured to adjust operations of a drill in response to receiving the one or more commands.

A system including a controller having a processor, a memory, and instructions stored on the memory and executable by the processor to segment a borehole image of a first well into a first plurality of zones based on pixel data, segment the borehole image of the first well into a second plurality of zones based on covariance data, merge the first plurality of zones and the second plurality of zones to generate an updated borehole image, cluster one or more sets of features of the updated borehole image into one or more clusters based on a classification algorithm, and generate a borehole texture model representative of expected properties of an additional borehole based on the one or more clusters.

The system of the preceding clause, wherein the processor is configured to segment the borehole image based on kernel density estimation (KDE), a variogram, or both

The system of any preceding clause, wherein the variogram corresponds to a measurement of a covariance of a pair of data points associated with borehole image

The system of any preceding clause, wherein the updated borehole image comprises one or more boundaries between zone segments identified based on the first plurality of zones and the second plurality of zones.

The system of any preceding clause, wherein clustering the one or more sets of features comprises an agglomerative clustering method.

The system of any preceding clause, wherein the processor is further configured to receive an additional borehole image associated with a second well, and apply the borehole texture model to classify one or more textures of the second well.

The system of any preceding clause, wherein the processor is further configured to determine one or more commands for drilling the second well based on the one or more textures and send the one or more commands to a drilling system configured to adjust operations of a drill in response to receiving the one or more commands.

A tangible and non-transitory machine readable medium including instructions that, when executed by a processor, causes the processor to perform operations including segmenting a borehole image of a first well into a first plurality of zones based on pixel data, segmenting the borehole image of the first well into a second plurality of zones based on covariance data, merging the first plurality of zones and the second plurality of zones to generate an updated borehole image, clustering one or more sets of features of the updated borehole image into one or more clusters based on a classification algorithm, and generating a borehole texture model representative of expected properties of an additional borehole based on the one or more clusters.

The medium of the preceding clause, wherein the borehole image is segmented based on kernel density estimation (KDE), a variogram, or both.

The medium of any preceding clause, wherein the variogram corresponds to a measurement of a covariance of a pair of data points associated with borehole image.

The medium of any preceding clause, wherein the updated borehole image comprises one or more boundaries between zone segments identified based on the first plurality of zones and the second plurality of zones.

The medium of any preceding clause, wherein clustering the one or more sets of features comprises an agglomerative clustering method.

The medium of any preceding clause, including receiving an additional borehole image associated with a second well, and applying the borehole texture model to classify one or more textures of the second well.

While only certain features have been illustrated and described herein, many modifications and changes will occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the disclosure.

The techniques presented and claimed herein are referenced and applied to material objects and concrete examples of a practical nature that demonstrably improve the present technical field and, as such, are not abstract, intangible or purely theoretical. Further, if any claims appended to the end of this specification contain one or more elements designated as “means for (perform)ing (a function) . . . ” or “step for (perform)ing (a function) . . . ”, it is intended that such elements are to be interpreted under 35 U.S.C. 112(f). However, for any claims containing elements designated in any other manner, it is intended that such elements are not to be interpreted under 35 U.S.C. 112(f).