DETERMINING BOUNDARIES FOR SUBSURFACE FEATURES THROUGH PIXEL-WISE INFERENCING OF INVERSION IMAGES

Description

BACKGROUND OF THE DISCLOSURE

Many natural resources are located underground, including water reservoirs and hydrocarbon reservoirs, such as natural gas and oil. To access these resources, downhole drilling systems drill a wellbore along a trajectory away from a surface location to a target location, formation, or geological feature. Modern drilling systems use measurements underground to determine geological features along the trajectory. However, in many cases it may be difficult to ascertain from measurement data the location, shape, orientation and/or boundaries of subsurface features.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to describe the manner in which the above-recited and other features of the disclosure can be obtained, a more particular description will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. For better understanding, the like elements have been designated by like reference numbers throughout the various accompanying figures. While some of the drawings may be schematic or exaggerated representations of concepts, at least some of the drawings may be drawn to scale. Understanding that the drawings depict some example embodiments, the embodiments will be described and explained with additional specificity and detail through the use of the accompanying drawings in which:

FIG. 1 is an example of a downhole system, according to at least one embodiment of the present disclosure;

FIG. 2-1 illustrates an example environment in which a boundary detection system is implemented, according to at least one embodiment of the present disclosure;

FIG. 2-2 illustrates an example implementation of a boundary detection system as described herein, according to at least one embodiment of the present disclosure;

FIG. 3 illustrates an example workflow of generating inversion images from downhole measurement data, according to at least one embodiment of the present disclosure;

FIG. 4-1 illustrates an example workflow for generating simulated inversion images and simulated boundary masks corresponding to inversion images, according to at least one embodiment of the present disclosure;

FIG. 4-2 illustrates an example workflow for generating pixel data sets for the inversion images, according to at least one embodiment of the present disclosure;

FIG. 5-1 illustrates an example workflow for training the subsurface boundary machine learning model, according to at least one embodiment of the present disclosure;

FIG. 5-2 illustrates an example workflow for using the subsurface boundary machine learning model to generate target boundary information, according to at least one embodiment of the present disclosure;

FIG. 6 illustrates an example workflow for applying target boundary masks to target inversion images to generate a three-dimensional subsurface object, according to at least one embodiment of the present disclosure;

FIG. 7-1 illustrates a flow diagram for a method or a series of acts for identifying subterranean features as described herein, according to at least one embodiment of the present disclosure;

FIG. 7-2 illustrates a flow diagram for a method or a series of acts for identifying subterranean features as described herein, according to at least one embodiment of the present disclosure; and

FIG. 8 illustrates certain components that may be included within a computer system.

DETAILED DESCRIPTION

This disclosure describes a drilling system that uses a boundary detection system to identify boundaries of subsurface features. For example, the boundary detection system uses a subsurface boundary machine learning model that is generated to process individual pixels of input inversion images to identify subsurface features. The boundary information may facilitate generating a boundary mask for the input inversion images to facilitate identifying boundaries of subsurface features in the inversion images. In this way, the boundary detection system can efficiently and accurately locate subsurface features within inversion images that may not otherwise be identifiable in the inversion images.

In particular, this disclosure relates to devices, systems, and methods for determining boundary information using machine learning models, training data, and/or real-time inputs. In this disclosure, these devices, systems, and methods are described in the context of a boundary detection system, which may automatically identify boundary information and/or generate boundary masks for inversion images for identifying subsurface features within the inversion images.

To illustrate, in some embodiments, the boundary detection system identifies inversion training images indicating a portion of a subsurface feature, and obtains corresponding simulated inversion images from a subsurface model corresponding to a wellbore position of the inversion images. Based on generating simulated boundary masks for the simulated inversion images and based on the inversion training images, the boundary detection system generates a subsurface boundary machine learning model that processes pixels of the inversion training images through a decision-based architecture to generate predicted boundary masks indicating boundaries of the subsurface feature.

As will be discussed in further detail below, the present disclosure includes a number of practical applications having features described herein that provide benefits and/or solve problems associated with identifying subsurface features and/or boundaries of subsurface features. Some example benefits are discussed herein in connection with various features and functionalities provided by a boundary detection system implemented on one or more computing devices. It will be appreciated that benefits explicitly discussed in connection with one or more embodiments described herein are provided by way of example and are not intended to be an exhaustive list of all possible benefits of the boundary detection system.

For example, the boundary identification system described herein implements a subsurface boundary machine learning model that processes each pixel of an inversion image individually. This simplified manner of processing inversion images facilitates generating boundary information for the inversion images efficiency and quickly. For example, in contrast, some conventional image-processing techniques consider images as a whole and process many or all of the pixels simultaneously resulting in significant latency due to the high dimensionality and complexity of the image-processing operations. Additional efficiency benefits arise from the fact that the subsurface boundary machine learning model processes pixel number objects (e.g., pixel values as described herein) representative of each pixel of an inversion image as opposed to processing more complex image data directly from the pixels of target images. Further, the subsurface boundary machine learning model is generated based on a tree-or decision-based architecture which can process the target pixels through the trained decision tree(s) much more quickly than, for example, machine learning models having deep learning or neural network architectures, which due to their complexity, may be computationally expensive and inefficient, resulting in generating outputs with a significant delay.

These efficiency benefits not only facilitate the subsurface boundary machine learning model being implemented in real time to facilitate real-time identification of subsurface features, but may also advantageously facilitate generating or training the subsurface boundary machine learning model in real time or near real time. For example, tree-based models may typically have a simple and intuitive structure composed of a series of decision nodes and leaf nodes which, during training, are recursively split based on feature thresholds in the training data. These relatively straightforward computations may be trained and tuned considerably faster than the more complex operations often involved in the training of neural networks and other deep learning models, which involves the optimization of millions or even billions of parameters. Indeed, the simplicity of the tree-based model, which drives the efficiency of the subsurface boundary machine learning model, is facilitated by the simplicity of the input data being implemented on a pixel-by-pixel basis as well as the input data being simpler, number objects which are more computationally manageable. Thus, the subsurface boundary machine learning model may be trained in a manner of seconds as opposed to a manner of hours or days for deep learning and other machine learning models.

In addition to providing quick and timely boundary indications, the efficient manner in which the subsurface boundary machine learning model is implemented facilitates a more efficient use of computing resources. For example, some conventional machine learning models, such as neural networks and other deep learning models require implementation on GPU hardware, for example, to facilitate parallel processing operations. In contrast, the subsurface boundary machine learning model may be implemented entirely on CPU hardware, for example, based on the simplicity of the tree-based architecture and input data allowing for implementation through serial computations. Thus, the boundary detection system, and more specifically the subsurface boundary machine learning model, may realistically be implemented on any computing device without requiring robust, specialized, or excess computing and hardware components, allowing for ease and flexibility of deployment of the boundary detection techniques described herein.

Further, the pixel-wise manner in which the boundary detection system processes inversion images facilitates identifying any number of subsurface features in inversion images, including any number of boundaries for each subsurface feature. For example, because the boundary detection system is trained to determine on a pixel-by-pixel basis whether each pixel corresponds to a subsurface feature (e.g., reservoir) or whether it does not, the boundary detection system may generate a boundary mask that may indicate any and every subsurface feature, including multiple features, that may be depicted in an underlying inversion image. In contrast, some conventional methods may be limited to only detecting a singular reservoir, and/or only detecting a singular (e.g., top or bottom) boundary of a reservoir at a time.

The pixel-wise implementation of the boundary detection system may additionally facilitate training the boundary detection system accurately and efficiently based on a very limited set of inversion training images. For example, image-processing techniques that evaluate images as a whole (or large parts of an image) often require hundreds, thousands, or more images in order to generate an accurately trained model. Because the boundary detection system processes individual pixels of inversion images as independent inputs or samples, each inversion image contains thousands of unique training samples for tuning the parameters of the decision-tree architecture. Thus, the subsurface boundary machine learning model may be trained to be highly accurate based on as few as 20 inversion images, whereas some conventional models may require at least 300 or more inversion images for less accurate results.

As illustrated in the following discussion, this disclosure uses a variety of terms to describe the features and advantages of one or more implementations described in this disclosure. Additional details are provided to clarify the meaning of some of these terms, while details regarding other terms may be provided later in the document.

As used herein, “wellbore measurement data,” “wellbore data,” and “measurement data” refer to data which each describe an aspect, value, rate, property, state, etc. of some feature detected, observed, or otherwise measured with respect to a downhole operation. For example, the wellbore data may include formation evaluation data such as resistivity data, porosity data, gamma-ray data, density data, acoustic data, seismic data, electromagnetic data, etc. The wellbore data may include drilling parameter data such as flow rate, temperature, pressure, speed, torque (TOR), rate of penetration (ROP), and weight on bit (WOB). The wellbore measurement data may include measurements of formation evaluation, wellbore stability, mud properties, survey data, and equipment health and status. Indeed, the measurement data may include any measurement, metric, or value relevant to a downhole operation, and combinations thereof. The wellbore measurement data may include measurements taken from various downhole and/or surface sensors and/or measurements received from one or more computing devices.

As used herein, a “log” such as a data log, wellbore data log, or downhole data log may refer to data contained or documented within an operation report or log for a downhole operation. For example, the wellbore data logs may document the various measurements taken during or in pursuit of one or more downhole operations. The logs may document relevant times and/or depths of the measurements. In some cases, the downhole logs may be generated or aggregated manually, such as by a drilling engineer compiling various measurement data for a downhole operation. The logs may be generated while downhole (e.g., drilling) activities are being conducted, or may be generated after the completion of one or more activities such as part of an upload or transmission of various data for one or more downhole activities.

As used herein, a “feature” such as a geological feature, downhole feature, or a subsurface feature may be any element of a geological formation. A geological feature may include a reservoir, pay zone, subterranean target, or any other underground feature for which it may be desirable to know its location, orientation, position, etc. For instance, a geological feature may include a geological structure, such as a formation. A feature may include the entire geological structure. A downhole feature may include a volume of space, including one or more structures, rock types, material types, and so forth. In some embodiments, a feature may include a boundary between two geological structures, such as a boundary between strata. In some embodiments, a feature may include a boundary between rock types. In some embodiments, a downhole feature may include a specific structure of a set of structures, such as a fluid reservoir. A feature may be three-dimensional. For example, a feature may include a three-dimensional surface having variations in latitude, longitude, and depth. In some embodiments, a feature may be a reservoir, pay zone, or underground resource, such as an oil, gas, or water reservoir, a source of geothermal energy, or any other subterranean target.

As used herein, “machine-learning model” refers to a computer model or computer representation that may be trained (e.g., optimized) based on inputs to approximate unknown functions. For instance, a machine-learning model may include, but is not limited to, a neural network (e.g., a convolutional neural network (CNN), LSTM, graph neural network, or deep learning model), a decision tree (e.g., a gradient-boosted decision tree), a linear regression model, a logistic regression model, Dirichlet allocation (LDA) model, multi-arm bandit model, random forest model, support vector machine (SVM) model, or a combination of these models.

As used herein, “resistivity image” refers to an image that includes resistivity measurements of subsurface geological features. A resistivity image may include a geosphere inversion image and/or electromagnetic (EM) field measurements and/or mappings of a surrounding area of a wellbore measured using a downhole resistivity sensor. In some instances, the measurements are in a specific direction, typically along the length of the borehole or drill hole, which is done to gather information about subsurface structures and geological formations.

As used herein, “inversion image,” “inversion result,” and the like refers to an image representation of subsurface formations encountered by or from a wellbore. Inversion images are generated by processing raw measurement data collected from downhole tools such as logging-while-drilling tools or wireline tools. Generating inversion images involves the mathematical transformation of the measured data to an image of the subsurface formation indicating one or more details, aspects, or other qualities of the formation as sensed or observed by the underlying measurements. For example, inversion includes forward modeling to simulate the response of underlying measurement data and then solving an inversion problem to adjust model parameters by inferring the subsurface properties from the measurement data. Once the model converges, the model parameters are used to generate an inversion image of the subsurface formations. For example, inversion images may depict information about lithology, fluid content, structural features, reservoir characteristics, or other subsurface features.

In some cases, inversion images may be described as being generated from resistivity data. In some instance, inversion images may be generated from any downhole measurement data such as, for example, porosity data, density data, acoustic data, seismic data, and the like. Inversion images are made up of an array of pixels, with each pixel having a value (e.g., color) corresponding to a specific value or magnitude of an aspect of the underlying measurement data. For example, an inversion image may include an array of 128×128 pixels. In some embodiments inversion images have a square resolution, or may have any other resolution.

As used herein, a “mask,” “boundary mask” and the like refers to a binary image corresponding to an identified feature of an associated inversion image. For example, a mask may be a binary labelling of a reservoir and/or boundary depicted in an inversion image. For instance, a given feature of interest may be demarked within an inversion image by setting pixels corresponding to the feature to a certain value (e.g., white or 1) while other pixels are set to another value (e.g., black or 0). The mask may be applied to (e.g., overlaid on) the inversion image, or referenced separately, to provide visual cues about a feature, for example by isolating the feature within the inversion image. In this way, a mask highlights the spatial locations of a given feature within an inversion image.

As used herein, “resistivity data,” “resistivity measurements,” and the like refer to electromagnetic (EM) field measurements and/or mappings of a surrounding area of a wellbore measured using a downhole resistivity sensor. In some instances, the measurements are in a specific direction, such as transverse or perpendicular to the length of the wellbore, which is done to gather information about the surrounding subsurface structures and geological formations.

Additional details will now be provided regarding systems described herein in relation to illustrative figures portraying example implementations. For example, FIG. 1 shows one example of a downhole system 100 for drilling an earth formation 101 to form a wellbore 102. The downhole system 100 includes a drill rig 103 used to turn a drilling tool assembly 104 which extends downward into the wellbore 102. The drilling tool assembly 104 may include a drill string 105, a bottomhole assembly (“BHA”) 106, and a bit 110, attached to the downhole end of the drill string 105.

The drill string 105 may include several joints of drill pipe 108 connected end-to-end through tool joints 109. The drill string 105 transmits drilling fluid through a central bore and transmits rotational power from the drill rig 103 to the BHA 106. In some embodiments, the drill string 105 further includes additional downhole drilling tools and/or components. The drill pipe 108 provides a hydraulic passage through which drilling fluid is pumped from the surface to the bit 110.

The BHA 106 may include other downhole drilling tools and components. Examples of additional BHA components include measurement-while-drilling (“MWD”) tools, logging-while-drilling (“LWD”) tools, and measurement sensors.

To elaborate, while performing downhole (e.g., drilling) activities, wellbore measurement data may be taken, measured, or observed through a variety of (e.g., downhole and/or surface) sensors. In this way, various information may be collected related to the wellbore and/or downhole activity in order to facilitate the techniques described herein. Additionally, in some cases, reports or logs may be generated for documenting various downhole activities or operations.

The BHA 106 may further include a rotary steerable system (RSS). The RSS may include directional drilling tools that change a direction of the bit 110, and thereby the trajectory of the wellbore. At least a portion of the RSS may maintain a geostationary position relative to an absolute reference frame, such as one or more of gravity, magnetic north, or true north. Using measurements obtained with the geostationary position, the RSS may locate the bit 110, change the course of the bit 110, and direct the directional drilling tools on a projected trajectory. The RSS may steer the bit 110 in accordance with or based on a trajectory for the bit 110. For example, a trajectory may be determined for directing the bit 110 toward one or more subterranean targets such as an oil or gas reservoir.

The downhole system 100 may include or may be associated with a client device 112 with a boundary detection system 120 implemented thereon (e.g., or with a client application implemented thereon for accessing the boundary detection system 120 as described herein). The boundary detection system 120 may facilitate identifying subsurface features and/or boundaries of subsurface features, for example, to facilitate directing or steering the bit 110 to access and/or remain within a subsurface feature.

FIG. 2-1 illustrates an example environment 200 in which a boundary detection system 120 is implemented in accordance with one or more embodiments describe herein. As shown in FIG. 2-1, the environment 200 includes a server device 114. The server device 114 may include one or more computing devices (e.g., including processing units, data storage, etc.) organized in an architecture with various network interfaces for connecting to and providing data management and distribution across one or more client systems. As shown in FIG. 2-1, the server device 114 may be connected to and may communicate with (either directly or indirectly) a client device 112 through a network 116. The network 116 may include one or multiple networks and may use one or more communication platforms and/or technologies suitable for transmitting data. The network 116 may refer to any data link that enables transport of electronic data between devices of the environment 200. The network 116 may refer to a hardwired network, a wireless network, or a combination of a hardwired network and a wireless network. In one or more embodiments, the network 116 includes the internet. The network 116 may be configured to facilitate communication between the various computing devices via well-site information transfer standard markup language (WITSML) or similar protocol, or any other protocol or form of communication.

The client device 112 may be representative of one or multiple client devices, and may refer to various types of computing devices. For example, the client device 112 may include a mobile device such as a mobile telephone, a smartphone, a personal digital assistant (PDA), a tablet, a laptop, or any other portable device. Additionally, or alternatively, the client device 112 may include one or more non-mobile devices such as a desktop computer, server device, surface or downhole processor or computer (e.g., associated with a sensor, system, or function of the downhole system), or other non-portable device. In one or more implementations, the client device 112 includes graphical user interfaces (GUI) thereon (e.g., a screen of a mobile device). In addition, or as an alternative, one or more of the client device 112 may be communicatively coupled (e.g., wired or wirelessly) to a display device having a graphical user interface thereon for providing a display of system content. The server device 114 may similarly refer to various types of computing devices. Each of the devices of the environment 200 may include features and/or functionalities described below in connection with FIG. 8.

As shown in FIG. 2-1, the environment 200 may include a boundary detection system 120 implemented on the server device 114. While shown on the server device 114, the boundary detection system 120 may be implemented wholly or in part on the client device 112, across the server device 114 and the client device 112, or on or across one or more additional devices, such that different portions or components of the boundary detection system 120 are implemented on different computing devices in the environment 200.

The client device 112 may include a client application 118. The client application 118 may include an application or interface for interacting with and/or receiving the features of the boundary detection system 120 as described herein. In some embodiments, one or more of the functionalities or features of the boundary detection system 120 may be carried out or performed on or by the client application 118. In this way, the environment 200 may be a cloud computing environment, and the boundary detection system 120 may be implemented across one or more devices of the cloud computing environment in order to leverage the processing capabilities, memory capabilities, connectivity, speed, etc., of such cloud computing environments in order to facilitate the features and functionalities described herein.

FIG. 2-2 illustrates an example implementation of the boundary detection system 120 as described herein, according to at least one embodiment of the present disclosure. The boundary detection system 120 may include a downhole data manager 122, a machine learning manager 124, and a boundary mask manager 126. The boundary detection system 120 may also include a data storage 130 having inversion images 132, a subsurface boundary machine learning model 134, and boundary masks 136 stored thereon. While one or more embodiments described herein describe features and functionalities performed by specific components 122-126 of the boundary detection system 120, it will be appreciated that specific features described in connection with one component of the boundary detection system 120 may, in some examples, be performed by one or more of the other components of the boundary detection system 120.

By way of example, one or more of the data receiving, gathering, or storing features of the downhole data manager 122 may be delegated to other components of the boundary detection system 120. As another example, while boundary masks may be generated by the boundary mask manager 126, in some instances, some or all of these features may be performed by the machine learning manager 124 (or other component of the boundary detection system 120). Indeed, it will be appreciated that some or all of the specific components may be combined into other components and specific functions may be performed by one or across multiple components 122-126 of the boundary detection system 120.

Additionally, while FIG. 1, for example, depicts the boundary detection system 120 implemented on a client device 112 of the downhole system, it should be understood that some or all of the features and functionalities of the boundary detection system 120 may be implemented on or across multiple client devices 112 and/or server devices 114. For example, data may be input and/or received by the downhole data manager 122 on a (e.g., local) client device, and the subsurface features and/or boundaries may be identified as described herein on one or more of a remote, server, or cloud device. Indeed, it will be appreciated that some or all of the specific components 122-126 may be implemented on or across multiple client devices 112 and/or server devices 114, including individual functions of a specific component being performed across multiple devices.

The boundary detection system 120 may be implemented to identify subsurface features in inversion images, which may include identifying one or more boundaries of the subsurface feature. Identifying subsurface features in this way may inform the directing or steering of the wellbore in order to access the subsurface feature. For example, one or more downhole drilling parameters or steering parameters may be adjusted based on subsurface features identified by the boundary detection system 120. In some embodiments, the boundary detection system 120 may adjust one or more downhole drilling parameters automatically and without user input. For example, based on the subsurface features and/or boundaries of the subsurface features identified in the inversion images and/or based on the boundary masks, the boundary detection system 120 may identify one or more steering parameters for maintaining a trajectory of a downhole tool within the boundaries of the subsurface feature, and may automatically send, transmit, or apply the steering parameters to the downhole tool in order to steer the downhole tool, without use input or intervention.

In some embodiments, the downhole data manager 122 may receive, generate, and access a variety of types of data. For example, the downhole data manager 122 may receive formation evaluation data, such as resistivity data, and may generate inversion images from the formation evaluation data. The downhole data manager 122 may also receive a three-dimensional model of the formation which may depict subsurface features with increased accuracy, precision, resolution, etc., over that of the inversion images. The downhole data manager 122 may generate simulated, two-dimensional inversion images from the three-dimensional model that correspond to the inversion images, such as by slicing a cross-section of the three-dimensional model. Based on the simulated inversion images, the downhole data manager 122 may generate simulated boundary masks that indicate subsurface features in the simulated inversion images and the corresponding inversion images. In some embodiments, the downhole data manager 122 may identify various information associated with the inversion images and, along with the underlying measurement data, may generate a table or pixel data set for the inversion images.

In some embodiments, the machine learning manager 124 may utilize the simulated boundary masks and the associated pixel data sets as training data for training a subsurface boundary machine learning model 134 to predict boundary information based on the inversion images. Once trained, the subsurface boundary machine learning model 134 may be implemented to generate accurate boundary information for identifying subsurface features in target inversion images. For example, the boundary mask manager 126 may generate boundary masks from the boundary information for identifying the position and locations of subsurface features. In some embodiments, the boundary mask manager 126 may generate a three-dimensional model from several consecutive inversion images and associated boundary masks. In this way, the boundary detection system 120 may facilitate characterizing subsurface formations and identifying features of interest in order that the wellbore may be directed to access these features. For example, the boundary detection system 120 may be implemented in real time or near real time to receive inversion images, train and implement the subsurface boundary machine learning model 134, and produce boundary masks for the inversion images.

Turning now to FIG. 3, this figure illustrates an example workflow 300 of generating inversion images from downhole measurement data, according to at least one embodiment of the present disclosure. The workflow 300 may be performed by the boundary detection system 120. For example, the boundary detection system 120 may implement the downhole data manager 122 (and/or other components) to perform some or all of the workflow 300.

As described herein, a drilling tool assembly may be implemented in a wellbore to form the wellbore along a trajectory 182 in order to access an underground target. In some embodiments, the drilling tool assembly includes a measurement tool 180, such as an LWD tool for taking formation evaluation measurements. For instance, the measurement tool 180 may measure resistivity data, neutron density data, gamma-ray data, acoustic data, and the like. The formation evaluation measurements may include measurements of resistivity, density, porosity, or any other property or aspect of the formation. The resistivity measurements may be recorded in a downhole data log 138. The downhole data log 138 may record any number and any type of downhole measurement data, such as any formation evaluation measurements, drilling parameters, or any other measurements.

In some embodiments, the formation evaluation measurements and/or the downhole data log may include measurements from sensors and/or sources other than those included in the measurement tool 180, such as measurements from surface sensors, other downhole tools, or other devices. For example, some or all of the downhole data log 138 may be accessed from libraries, databases, user input, other devices, etc.

In some embodiments, inversion images 132 are generated from the downhole data log 138. For example, resistivity data of the downhole data log 138 may be inverted to generate the inversion images 132. The inversion images 132 may be generated to correspond with periodic positions of the trajectory 182. For example, the inversion images 132 may be generated at 10-30 ft intervals (or any other interval) along the trajectory 182. In some embodiments, the inversion images 132 are generated at intervals of 1 ft, 2 ft, 3 ft, 4 ft, 5 ft, 10 ft, 15 ft, 20 ft, 30 ft, 50 ft, or any other interval and combinations thereof. In some embodiments, the inversion images 132 may be generated at various different intervals. For example, a longer interval (e.g., 10-30 ft) may be used for performing the boundary detection techniques described herein in real time, while a shorter interval (e.g., 1-5 ft) may be utilized for post-operation boundary detection from the inversion images.

In some embodiments, the inversion images 132 may be two-dimensional images that are transvers to the wellbore and the trajectory 182. For example, the inversion images may capture and/or characterize the formation surrounding the wellbore at a given position of the trajectory 182. In some embodiments, the inversion images may be one-dimensional, such as a ribbon indicating formation properties in only 1-dimension. In some embodiments, the inversion images may be three-dimensional, representing formation properties in all dimensions. The dimensionality of the inversion images 132 may be determined based on one or more properties of a subsurface feature of interest. For example, one-dimensional inversion images may be implemented based on a subsurface features that are known to change or fluctuate in only one dimension, two-dimensional inversion images may be implemented based on subsurface features that are known to change in two dimensions, etc. In this way, inversion images of different dimensionalities may be implemented to facilitate flexibility with different situations. Indeed, while the boundary detection techniques may be described herein with respect to two-dimensional, transvers inversion images, it should be understood that the boundary detection system may be implemented with respect to inversion images of any dimensions, such as one-dimensional and/or three-dimensional inversion images.

In some embodiments, the boundary detection system 120 generates the inversion images 132. For example, the downhole data manager 122 may receive the downhole data log 138 and may invert some or all of the information included therein to generate the inversion images 132. In some embodiments, the boundary detection system 120 receives or accesses the inversion images 132. For example, another system may generate the inversion images 132 and may provide the inversion images 132 to the boundary detection system 120.

In some embodiments, the inversion images may depict or indicate at least a portion of a subsurface feature and/or one or more boundaries of a subsurface feature. In some embodiments, it may be difficult to identify the subsurface feature from the inversion images and/or it may be desirable to automatically identify boundaries of the subsurface feature in the inversion images through computer implemented techniques and without user input.

FIG. 4-1 illustrates an example workflow 400 for generating simulated inversion images and simulated boundary masks corresponding to inversion images, according to at least one embodiment of the present disclosure. The workflow 400 may be performed by the boundary detection system 120. For example, the boundary detection system 120 may implement the downhole data manager 122 (and/or other components) to perform some or all of the workflow 400.

In some embodiments the boundary detection system 120 receives a three-dimensional subsurface model of downhole formations. In some embodiments, the three-dimensional subsurface model 140 may include higher quality, higher fidelity, higher resolution, and/or more detailed, complete, or robust information about the formation than, for example, the inversion images 132. For instance, the three-dimensional subsurface model 140 may be generated based on more refined measurement data and/or more sources of measurement data. For instance, the three-dimensional subsurface model 140 may be generated from an assembly of several formation evaluation measurements, including seismic data measurements. The three-dimensional subsurface model 140 may be generated from measurements taken at the surface, measurements taken by a wireline tool, measurements taken from a nearby or offset wellbore, and/or measurements taken at an earlier stage (e.g., uphole) of the same wellbore.

Based on the three-dimensional subsurface model 140, one or more subsurface features (and boundaries of the features) may be identifiable, for example, with a higher confidence than based on the inversion images 132. For instance, one or more automatic and/or computer-implemented techniques may be suited for identifying subsurface features from the three-dimensional model which may otherwise not be possible or practical based on the inversion images 132. In many cases, however, a detailed, three-dimensional subsurface model may not be generated or otherwise available for a given location of the wellbore and/or trajectory 182. For example, it may be computationally expensive and/or slow to generate a three-dimensional subsurface model such that these models may only be generated periodically and/or for select portions of the trajectory 182. In another example, generating a three-dimensional subsurface model may be performed after drilling operations are completed or paused. Thus, it may not be possible to rely on models like the three-dimensional subsurface model 140 to identify subsurface feature and inform the steering of the wellbore.

In some embodiments, the boundary detection system 120 may generate simulated inversion images 142 from the three-dimensional subsurface model 140. For instance, as shown in box 410, based on the inversion images 132, the boundary detection system 120 may identify the locations or portions of the trajectory 182 corresponding to the inversion images 132. The boundary detection system 120 may accordingly identify the corresponding portions of the three-dimensional subsurface model 140 that align with or correspond to the inversion images 132 and may generate the simulated inversion images 142. For example, the boundary detection system 120 may segment, slice, section, or otherwise isolate transverse portions (e.g., cross-sections) of the three-dimensional subsurface model 140 that correspond to the inversion images 132. In some embodiments, the simulated inversion images 142 may be an entire cross-section of the three-dimensional subsurface model 140, or may be a cropped portion of a cross-section of the three-dimensional subsurface model 140. The simulated inversion images 142 may have the same dimensionality as the inversion images 132 (e.g., one-dimensional, two-dimensional, or three-dimensional). In some embodiments, the simulated inversion images 142 may have the same vertical and/or horizontal dimensions and/or resolution as the inversion images 132. In some embodiments, the simulated inversion images 142 may have a higher resolution than the inversion images 132. In this way, the simulated inversion images 142 may capture substantially the same view or window of the formation as the inversion images 132, but may include more detail, more information, and/or a higher resolution of the formation.

Based on the simulated inversion images 142, the boundary detection system 120 may generate simulated boundary masks 144. For example, the boundary detection system 120 may identify changes or contrasts in the simulated inversion images 142 to identify boundaries of a subsurface feature identified in the simulated inversion images 142. For example, pixels (e.g., colors) of the simulated inversion images 142 may correspond to measured values of one or more underlying measurements of the three-dimensional subsurface model 140 and/or simulated inversion images 142. In some embodiments, the simulated boundary masks 144 may be generated based on a threshold as applied to the pixels of the simulated inversion images 142. For instance, pixels having a color or value above a given threshold may correspond with a 1 or white value of the simulated boundary masks 144, and pixels having a color or value below a (same or different) threshold may correspond with a 0 or black value of the simulated boundary masks 144. In this way, the simulated boundary masks 144 may be generated based on identifying, pixel by pixel, which pixels correspond to an identified feature and which pixels do not (or correspond with some other feature).

Accordingly, the simulated boundary masks 144 may be generated to indicate one or more subsurface features depicted in the simulated inversion images 142, including boundaries of the features. Moreover, the simulated boundary masks 144, and the subsurface features and/or boundaries identified in the simulated boundary masks 144, may similarly correspond to the inversion images 132 based on the simulated inversion images 142 being generated to align and correspond with the inversion images 132. As described herein, the simulated boundary masks 144 along with the inversion images 132 may facilitate training data to facilitate training the subsurface boundary machine learning model 134 as described herein.

FIG. 4-2 illustrates an example workflow 401 for generating pixel data sets 146 for the inversion images 132, according to at least one embodiment of the present disclosure. The workflow 401 may be performed by the boundary detection system 120. For example, the boundary detection system 120 may implement the downhole data manager 122 (and/or other components) to perform some or all of the workflow 401.

In some embodiments, the boundary detection system 120 may receive or identify additional information associated with the inversion images 132 and may associated specific information with each of the inversion images 132. The pixel data sets 146 may be generated to include or represent this additional information, along with underlying pixel and/or measurement data information for each of the inversion images 132. For example, as shown in box 412, the boundary detection system 120 may identify each pixel of an inversion image and may generate a series of one or more pixel values to represent each pixel. For instance, each inversion image may have pixels 1−n, and the boundary detection system 120 may assemble a pixel-wise collection of pixel values V_i−V_n for each pixel 1−n. The pixel data sets 146 may comprise this pixel-wise collection of pixel values V₁−V_n for each of the pixels 1−n of the inversion images 132. For example, the pixel data sets 146 may include a table, matrix, or other collection of values for representing the pixel-wise information for the inversion images 132. The pixel data sets 146 may be in the form of a text file (or other data file) for recording the pixel-wise information.

In some embodiments, one or more pixel values of the pixel-wise collection of pixel values V₁−V_n for each pixel 1−n in the pixel data sets 146 may indicate a measurement value for an underlying measurement of the inversion image. For example, one or more pixel values may indicate a resistivity measurement on which the pixel (e.g., pixel color) of the inversion image was based. In some embodiments, several pixel values may indicate several underlying measurement values for each pixel, such as pixel values for multiple different types of measurements upon which the inversion images were based.

In some embodiments, one or more pixel values may indicate additional information for each pixel. For example, one or more pixel values may indicated a measurement depth (MD) for the inversion image, corresponding to a MD where the underlying measurement data for the inversion image was taken. One or more pixel values may indicate coordinates of the pixel within the inversion images. In some embodiments, one or more pixel values may indicate a global position or location of the inversion image, such as a latitude and longitude corresponding with the inversion image 132. In some embodiments, one or more pixel values may indicate information about a measurement tool that was used to sense, record, or observe the underlying measurement data for the inversion image. For example, one or more pixel values may indicate a type of tool, a calibration of the tool, a resolution of the tool, or any other information about the measurement tool. In some embodiments, one or more pixel values may indicate channel information about the data channel of the underlying measurement values for the inversion image. For example, the pixel values may indicate a resolution or sampling frequency of the data channel. The pixel values may indicate a frequency of the underlying measurement, such as a waveform or signal frequency that was utilized to collect the underlying measurement for the inversion image. In this way, the pixel data sets 146 may be generated to represent a variety of information associated with (e.g., each pixel of) the inversion images 132, for example, in addition to the magnitudes, values, or colors of the inversion images 132. In some embodiments, the pixel data sets 146 may be provided to facilitate training the subsurface boundary machine learning model 134, for example, based on the additional information of the pixel data sets 146.

FIG. 5-1 illustrates an example workflow 500 for training the subsurface boundary machine learning model 134, according to at least one embodiment of the present disclosure. The boundary detection system 120 may facilitate training the subsurface boundary machine learning model 134, for example, by implementing the machine learning manager 124.

The boundary detection system 120 may train the subsurface boundary machine learning model 134 based on training data 148. For example, the training data 148 may include the pixel data sets 146, and the simulated boundary masks 144 which correspond to the inversion images upon which the pixel data sets 146 were based. In some embodiments, the training data 148 may include the inversion images, for example, in addition to or as an alternative to the pixel data sets 146.

The subsurface boundary machine learning model 134 may be trained to determine predicted boundary information 150 for the inversion images based on the pixel data sets 146. The predicted boundary information 150 may include predicted information for each pixel of the pixel data sets 146 and/or inversion images. For example, the subsurface boundary machine learning model 134 may predict, for each pixel, whether the pixel corresponds to a subsurface feature of interest in the inversion image, or whether the pixel does not (or corresponds to another feature). The predicted boundary information 150 may include a predicted probability or confidence of whether or not each pixel corresponds to a subsurface feature of interest. The predicted boundary information 150 may include a boundary mask for each inversion image of the pixel data sets 146. For example, based on the predicted probabilities, one or more thresholds may be applied to each pixel for generating a boundary mask. In this way, the subsurface boundary machine learning model 134 may predict identifiable features in the inversion images based on generating the predicted boundary information 150.

In various implementations, such as the one shown, the subsurface boundary machine learning model 134 includes a tree-based architecture having one or more hierarchical, tree-like structures. In various instances, the subsurface boundary machine learning model 134 is a decision tree model, a random forest or ensemble of decision trees, or a gradient boosting model.

To elaborate, in some instances, the subsurface boundary machine learning model 134 is a tree-based model that includes a hierarchical structure of decision nodes, branches, and leaf nodes. Each node represents a decision based on a specific feature and each leaf represents a candidate subsurface feature type (e.g., a type indicating that the pixel is a subsurface feature of interest and a type indicating the pixel is not the subsurface feature of interest) for which the subsurface boundary machine learning model 134 may classify each pixel of pixel data sets 146 based on the pixel values. The tree-based structure is built through the recursive splitting of the decision nodes into further decision nodes through training of the subsurface boundary machine learning model 134 as described below.

In some instances, the boundary detection system 120 utilizes a leaf-based model for the subsurface boundary machine learning model 134. A leaf-based model allows the boundary detection system 120 to use unbalanced growth and/or splitting among nodes to minimize losses of nodes. In various implementations, the boundary detection system 120 implements a gradient boosting machine learning architecture (e.g., LightGBM or XGBoost) along with the leaf-based model. For example, the boundary detection system 120 uses an LGBM classifier to predict boundary information for the pixels of the inversion images.

In various implementations, the boundary detection system 120 processes each input pixel data set and generates an input feature vector or a dimensional array containing numerical values each representing one of the various attributes of a pixel of an inversion image. The subsurface boundary machine learning model 134 then processes the feature vector through a sequence of decision nodes, where the subsurface boundary machine learning model 134 evaluates a specific feature of the input vector and makes a decision based on the feature's value. In this way, the subsurface boundary machine learning model 134 recursively navigates through the tree structure making decisions at each node based on different features until a leaf node is reached, providing the final output prediction of the predicted boundary information for the pixel.

In some embodiments, the subsurface boundary machine learning model 134 includes an ensemble architecture that utilizes multiple decision trees. For example, the subsurface boundary machine learning model 134 may process the input vector through separate decision trees independently. The outputs are then combined, such as through voting or averaging, to generate a final ensemble prediction of the predicted boundary information for each pixel.

In some embodiments, the subsurface boundary machine learning model 134 may determine a probability or confidence of the boundary information for a given pixel. For example, the probability may be determined based on the voting of the various trees, based on a logistic function such as a LGBM classifier, through calibration techniques such as Platt scaling or isotonic regression, or any other suitable technique for determining multi-classification probabilities.

In some implementations, the subsurface boundary machine learning model 134 is implemented as another type of machine learning model, such as a neural network architecture. For example, the subsurface boundary machine learning model 134 may be a Monte Carlo Dropout prediction model, a U-Net neural network, or a U-Net++ neural network.

As mentioned above, the boundary detection system 120 may train the subsurface boundary machine learning model 134 based on the training data 148. For example, the boundary detection system 120 may provide the pixel data set 146 for a given inversion image to the subsurface boundary machine learning model 134, and the subsurface boundary machine learning model 134 may predict or determine estimated boundary information for each pixel of the inversion image. The predicted boundary information 150 along with the associated simulated boundary mask 144 for the inversion image may be provided to the loss model 152 to evaluate the performance of the subsurface boundary machine learning model 134 during the training process. The predicted boundary information 150 may be an assembly of all the predicted boundary information for all of the pixels of a given pixel data set and inversion image. In some embodiments, the boundary detection system 120 provides the (e.g., assembled) predicted boundary information 150 to the loss model 152 for comparison to the (e.g., entire) associated simulated boundary mask 144.

In some embodiments, boundary detection system 120 provides predicted boundary information of a single pixel or a group of (less than all of the) pixels to the loss model 152 for comparison to some or all of an associated simulated boundary mask 144. For example, as described herein, the subsurface boundary machine learning model 134 may operate based on processing individual pixels (e.g., pixel values for individual pixels) of an inversion image. In some cases, the boundary detection system 120 may compare the output of the subsurface boundary machine learning model 134 to the simulated boundary masks 144 a pixel at a time for training the subsurface boundary machine learning model 134 (e.g., as well as when later inferencing via the subsurface boundary machine learning model 134), for example, rather than assembling the predicted boundary information 150 for an entire inversion image worth of pixels. This may achieve the benefits of providing orders of magnitude more training samples (e.g., each pixel as a training sample) for training the subsurface boundary machine learning model 134, for example, as compared to training the subsurface boundary machine learning model 134 based on images in their entirety, or even partial images.

In various implementations, the loss model 152 implements one or more loss functions or techniques such as cross-entropy loss, Gini impurity, deviance, etc. to determine and estimated transition type error amount. In various implementations, the boundary detection system 120 provides the error or loss amount back to the subsurface boundary machine learning model 134 as label feedback 154 to train and fine-tune the subsurface boundary machine learning model 134.

Additionally, in one or more implementations, the boundary detection system 120 uses the label feedback 154 to train, optimize, and/or fine-tune the decision tree(s) of the subsurface boundary machine learning model 134 through techniques such as recursive partitioning and/or boosting. For example, the boundary detection system 120 may use the loss model 152 to facilitate selecting a feature and corresponding threshold for separating the pixel values of the pixel data sets 146 into homogeneous subsets and generating or splitting corresponding decision nodes to generate one or more decision trees.

As another example, the boundary detection system 120 may use the loss model 152 to facilitate generating trees sequentially, with each tree correcting the errors of the previous tree. The boundary detection system 120 may iteratively train the subsurface boundary machine learning model 134 in this way with respect to many pixels of the training data 148 to further fine-tune the subsurface boundary machine learning model 134 for a set number of iterations, until it converges, until the training data is exhausted, or until a satisfactory level of accuracy is otherwise achieved.

As described earlier, in some embodiments, the boundary detection system 120 generates training data 148 that includes the pixel data sets 146 generated by the boundary detection system 120 for including a variety of information associated with each pixel of an inversion image. In one or more implementations, the boundary detection system 120 does not generate or does not use the pixel data sets 146 with the subsurface boundary machine learning model 134. In these implementations, the boundary detection system 120 trains the subsurface boundary machine learning model 134 directly based on the inversion images, and/or based solely on the pixel values of the pixel data sets 146 corresponding to the underlying measurement data of the inversion images. For example, the boundary detection system 120 provides the individual pixels of the inversion images directly to the subsurface boundary machine learning model 134 rather than pre-processing the inversion images into the pixel data sets 146.

To elaborate, in various implementations, the subsurface boundary machine learning model 134 is trained to determine predicted boundary information 150 for the pixels of an inversion image based on the identifying patterns, relationships, statistical characterizations, and/or other attributes directly from the individual inversion image pixels. In some cases, the boundary detection system 120 may generate an input feature vector representing these various pixel features identified from the inversion images and may process the input feature vector through the series of nodes of the subsurface boundary machine learning model to determine predicted boundary information for each pixel individually.

In these implementations, the transition identification machine learning model may recursively process the pixels of the inversion images as input through the tree and/or leaf architecture to provide a final output prediction of boundary information. Then, the boundary detection system 120 may use the loss model 152 to fine-tune the predictions of the subsurface boundary machine learning model 134. In this way, the subsurface boundary machine learning model 134 may be trained to predict identified subsurface features and/or boundaries for inversion images based directly on the pixels of the inversion images.

Once trained, the boundary detection system 120 uses the subsurface boundary machine learning model 134 to automatically generate target boundary information 166 for target inversion images 162 for identifying subsurface features of interest which may not be known or verified. To illustrate, FIG. 5-2 shows an example workflow 501 for using the subsurface boundary machine learning model 134 to generate target boundary information 166, according to at least one embodiment of the present disclosure. In this figure, the subsurface boundary machine learning model 134 represents a trained model with a tuned decision-based network and other trained components.

The subsurface boundary machine learning model 134 may generate the target boundary information 166 from target pixel data sets 164. For instance, the target pixel data sets 164 may be generated for target inversion images 162. The target inversion images 162 may be inversion images for a formation, location, and/or wellbore position in which it may be desirable to identify one or more subsurface features. For example, the target inversion images 162 may be inversion images for a wellbore, wellbore location, formation, angle of view, window, etc. for which a corresponding three-dimensional subsurface feature model is not generated or is otherwise not available. Thus, the subsurface boundary machine learning model 134 may be applied to the target inversion images 162 (e.g., either directly or via the target pixel data sets 164 as described herein) to generate target boundary information 166 identifying subsurface features depicted in the target inversion images 162.

As shown, the subsurface boundary machine learning model 134 may be applied to the target pixel data sets 164. The target pixel data sets 164 may be generated from the target inversion images 162 as part of a pre-processing step by the boundary detection system 120, or in some cases, the subsurface boundary machine learning model 134 may receive the target inversion images 162 and may perform processing of the target inversion images 162 to generate the target pixel data sets 164. Alternatively, the subsurface boundary machine learning models 134 may receive the target inversion images 162 and may by applied directly to the (e.g., pixels of the) target inversion images 162 as described herein.

The subsurface boundary machine learning model 134 generates target boundary information 166. As described herein, the target boundary information 166 may include a predicted probability or confidence as to whether each pixel corresponds to a formation of interest or not. For example, the target boundary information 166 may be utilized to generate a target boundary masks 168 for the target inversion images 162. The target boundary masks may provide a visual representation of the location, shape, orientation, dimensions, boundaries, etc. of a subsurface feature. In some embodiments, the boundary detection system 120 generates the target boundary masks 168 based on the target boundary information 166 output from the subsurface boundary machine learning model 134, and in some cases the subsurface boundary machine learning model 134 may generate and output the target boundary masks 168, for example, in addition to or as an alternative to the target boundary information 166.

In some embodiments, the target inversion images 162 may be received and/or generated by the boundary detection system 120 in real time, based on real-time measurement data. The boundary detection system 120 may implement the subsurface boundary machine learning model 134 to generate the target boundary information 166 and/or target boundary masks 168 in real time based on the real time target inversion images 162. In this way, the boundary detection system 120 may provide a real-time indication of subsurface features to inform timely decisions regarding wellbore steering.

In some embodiments, the target inversion images 162 may include multiple realizations of inverted measurement data (e.g., multiple realization of resistivity inversion). For example, the target inversion images 162 may include multiple inversion images for a given point in the trajectory based on different inversion results from the same or different measurement data. In some embodiments, the boundary detection system 120 may implement the subsurface boundary machine learning model 134 to predict boundary information for the multiple different instances of inversion for each point in the trajectory and may generate a final prediction of the boundary information based on an ensemble of the predictions from the various inversion results. This ensembling may be performed by the subsurface boundary machine learning model 134, or may be performed by the boundary detection system 120, for example, after implementing the subsurface boundary machine learning model 134. In some embodiments, this ensembling may be utilized for generating an uncertainty estimate for the boundary information of a given point in the trajectory.

In some embodiments, the target boundary information 166 and/or target boundary masks 168 may be verified and/or adjusted. For example, other downhole measurements may be utilized to compare against the predictions made by the subsurface boundary machine learning model 134 to validate the target boundary information 166 and/or the target boundary masks 168. For instance, a user such as a well planning engineer may manually check and/or adjust the output predictions of the target boundary information 166 and/or the target boundary masks 168. In another instance, more information and/or measurements may be collected, for example, after a downhole operation or as part of another downhole operation, which may facilitate validating the predictions of the subsurface boundary machine learning model 134. These checked and/or adjusted outputs of the subsurface boundary machine learning model 134 may be provided to the subsurface boundary machine learning model 134 as feedback for further tuning of the subsurface boundary machine learning model 134. In this way, the subsurface boundary machine learning model 134 may be continually tuned and/or updated (e.g., after the training process) to maintain a high level of accuracy.

Additionally, because of the pixel-wise training and implementation of the subsurface boundary machine learning model 134, because of the decision-or tree-based architecture of the subsurface boundary machine learning model 134, and because the subsurface boundary machine learning models 134 is applied to number objects of the target pixel data sets 164 (e.g., rather than image data), the boundary detection system 120 may not only implement the subsurface boundary machine learning model 134 for target inversion images in real time, but may also generate or train the subsurface boundary machine learning model 134 in real time or near real time. For example, the subsurface boundary machine learning models 134 may be trained in a manner of seconds, such as less than 5 or 10 seconds. In contrast, other conventional machine learning techniques may train models in a manner of hours or even days. Thus, the subsurface boundary machine learning model 134 may be trained and implemented almost immediately as inversion training images (and a corresponding three-dimensional subsurface model) are received in order to provide efficient and timely identification of subsurface features from real-time target inversion images.

FIG. 6 illustrates an example workflow 600 for applying the target boundary masks 168 to the target inversion images 162, according to at least one embodiment of the present disclosure. The workflow 600 may be performed by the boundary detection system 120. For example, the boundary detection system 120 may implement the boundary mask manager 126 (and/or other components) to perform some or all of the workflow 600.

In some embodiments, the target boundary masks 168 may be applied to the target inversion images 162 to indicate and/or isolate the identified subsurface features in the target inversion images 162. For example, the boundary detection system 120 may generate masked inversion images 170 by overlaying or augmenting information onto the target inversion images 162. In some embodiments, the masked inversion images 170 may include the target boundary masks 168 overlaid thereon, or may include an indication of one or more boundaries of a subsurface feature included thereon. In some embodiments, the masked inversion images 170 may be generated by updating or modifying pixels of the target inversion images 162 to indicate a subsurface feature. For example, one or more pixels of the masked inversion images 170 may be hidden, removed, or otherwise modified such that only the subsurface feature is depicted in the masked inversion images 170.

In some embodiments, the boundary detection system 120 may generate a three-dimensional subsurface model 172 based on the masked inversion images 170. For example, as shown in box 606, the boundary detection system 120 may assemble several, consecutive masked inversion images 170 in order to create a three-dimensional surface or model. The boundary detection system 120 may align the masked inversion image 170 based on a detected subsurface feature in the masked inversion images 170 and/or based on an associated trajectory on which the underlying target inversion images 162 were generated.

As discussed herein, inversion images may be generated periodically and/or may be spaced apart along a trajectory. In some embodiments, the boundary detection system 120 may generate intermediate data for the spaces between the masked inversion image 170 in order to generate the three-dimensional subsurface model 172. For example, the intermediate data may be generated based one or more identified subsurface features or boundaries as indicated by the target boundary masks 168 in consecutive images. For instance, a line or curve may be created between features of consecutive masked inversion images 170. In other instances, the intermediate data may be generated based on interpolating based on identified features of consecutive masked inversion images. For example, the intermediate data may be interpolated based on a linear, polynomial, logarithmic, quadratic, exponential, inverse distance weighting, nearest neighbor, or spline interpolation, or any other form or method for interpolation. In this way the boundary detection system 120 may generate the three-dimensional subsurface model 172 from the masked inversion images 170 in order to provide a three-dimensional visualization of one or more subsurface features.

FIG. 7-1 illustrates a flow diagram for a method 700 or a series of acts for identifying subterranean features as described herein, according to at least one embodiment of the present disclosure. While FIG. 7-1 illustrates acts according to one embodiment, alternative embodiments may add to, omit, reorder, or modify any of the acts of FIG. 7-1. In some embodiments, the method 700 may be performed by a system. in some embodiments, the method 700 may be implemented as instructions stored on a computer-readable storage medium.

In some embodiments, the method 700 includes an act 710 of receiving an inversion image. For example, the act 710 may include receiving an inversion image indicating a portion of a subsurface feature.

In some embodiments, the method 700 includes an act 720 of determining boundary information using a subsurface boundary machine learning model. For example, the act 710 may include determining boundary information using a subsurface boundary machine learning model that is generated to process pixels of input inversion images through a decision-based architecture to identify boundaries of subsurface features.

In some embodiments, the method 700 includes an act 730 of generating a boundary mask for the inversion image. For example, the act 730 may include, based on the boundary information, generating a boundary mask for the inversion image

In some embodiments, the method 700 includes an act 740 of providing the boundary mask for adjusting one or more downhole parameters. For example, the act 740 may include providing the boundary mask for adjusting one or more downhole parameters based on the boundary mask.

In some embodiments, the subsurface boundary machine learning model is generated to process individual pixels of the input inversion images through the decision-based architecture to identify boundaries of subsurface features.

In some embodiments, the boundary mask indicates at least one boundary of the subsurface feature of the inversion image.

In some embodiments, the method 700 includes updating the inversion image to indicate the at least one boundary of the subsurface feature based on the boundary mask.

In some embodiments, adjusting the one or more downhole parameters includes adjusting one or more steering parameters for maintaining a downhole tool within identified boundaries of the subsurface feature indicated by the boundary mask.

In some embodiments, the steering parameters are adjusted automatically and without user input.

In some embodiments, the inversion image is generated based on downhole resistivity data.

In some embodiments, the inversion image is a two-dimensional inversion result.

In some embodiments, the method 700 includes automatically determining the boundary information and generating the boundary mask based on real-time inversion results.

In some embodiments, wherein determining the boundary information includes determining an indication of boundary uncertainty for each pixel of the inversion image, and generating the boundary mask is based on a boundary uncertainty threshold for each pixel of the inversion image.

In some embodiments, the method 700 includes determining a pixel data set for each pixel of the inversion image and providing the pixel data set for each pixel to the subsurface boundary machine learning model to process the pixel data set through the decision-based architecture and determine an indication of boundary uncertainty for each pixel of the inversion image.

in some embodiments, determining the pixel data set for each pixel includes determining values that indicate, for each pixel, one or more of a measurement value of an associated downhole measurement, a downhole tool type, a measurement depth, pixel coordinates of the pixel within the inversion image, or latitude and longitude of the inversion image.

In some embodiments, the inversion image indicates a top boundary and a bottom boundary of the subsurface feature, and determining the boundary information includes identifying each of the top boundary and the bottom boundary of the subsurface feature in the inversion image.

In some embodiments, the inversion image indicates a portion of the subsurface feature and additionally indicates a portion of an additional subsurface feature, and determining the boundary information includes identifying a boundary of each of the subsurface feature and the additional subsurface feature in the inversion image.

In some embodiments, the method 700 includes receiving a plurality of additional inversion images indicating portions of the subsurface feature, determining additional boundary information for each of the plurality of additional inversion images using the subsurface boundary machine learning model, based on the additional boundary information, determining an additional boundary mask for each of the plurality of additional inversion images, generating a three-dimensional subsurface object of the subsurface feature based on assembling the inversion image and the plurality of additional inversion images and based on the boundary mask and the plurality of additional boundary masks, and providing the three-dimensional subsurface object for adjusting the one or more downhole parameters based on identified boundaries of the subsurface feature in the three-dimensional subsurface object.

In some embodiments, wherein generating the three-dimensional subsurface object includes interpolating boundary information between consecutive inversion images of the assembled inversion images.

In some embodiments, a computer-readable storage medium includes instruction that, when executed by at least one processor, cause the processor to receive an inversion image indicating a portion of a subsurface feature, determine boundary information using a subsurface boundary machine learning model that is generated to process individual pixels of input inversion images through a decision-based architecture to identify boundaries of subsurface features, based on the boundary information, generating a boundary mask for the inversion image, and update the inversion image to indicate a boundary of the subsurface feature based on the boundary mask.

In some embodiments the computer-readable storage medium and the processor are located in a downhole tool positioned downhole in a wellbore.

FIG. 7-2 illustrates a flow diagram for a method 750 or a series of acts for identifying subterranean features as described herein, according to at least one embodiment of the present disclosure. While FIG. 7-2 illustrates acts according to one embodiment, alternative embodiments may add to, omit, reorder, or modify any of the acts of FIG. 7-2. In some embodiments, the method 750 may be performed by a system. in some embodiments, the method 750 may be implemented as instructions stored on a computer-readable storage medium.

In some embodiments, the method 750 includes an act 760 of identifying training images indicating a portion of a subsurface feature.

In some embodiments, the method 750 includes an act 770 of obtaining simulated inversion images from a subsurface model. For example, the act 770 may include obtain simulated inversion images that identify the subsurface feature from a subsurface model corresponding to a wellbore position of the inversion training images.

In some embodiments, the method 750 includes an act 780 of generating simulated boundary masks for the simulate inversion images. For example, the act 780 may include generate simulated boundary masks for the simulated inversion images based on detected boundaries of the subsurface feature in the simulated inversion images.

In some embodiments, the method 750 includes an act 790 of generating a subsurface boundary machine learning model based on the inversion training images and the simulated boundary masks. For example, the act 790 may include generate a subsurface boundary machine learning model based on the inversion training images and the simulated boundary masks to process individual pixels of the inversion training images through a decision-based architecture and generate predicted boundary masks indicating boundaries of the subsurface feature within the inversion training images.

In some embodiments, the inversion training images are two-dimensional inversion results, the subsurface model is a three-dimensional formation model spanning at least a portion of the wellbore corresponding to the inversion training images, and generating the simulated inversion images includes creating two-dimensional transverse sections of the three-dimensional formation model corresponding to the inversion training images.

In some embodiments, generating the subsurface boundary machine learning model includes providing feedback to the subsurface boundary machine learning model based on comparing the predicted boundary masks to the simulated boundary masks to further tune the decision-based architecture of the subsurface boundary machine learning model.

In some embodiments, generating the subsurface boundary machine learning model is performed automatically based on receiving the inversion training images in real time.

In some embodiments, the subsurface boundary machine learning model is generated based on a small sample of 50 inversion training images or less.

Turning now to FIG. 8, this figure illustrates certain components that may be included within a computer system 800. One or more computer systems 800 may be used to implement the various devices, components, and systems described herein.

The computer system 800 includes a processor 801. The processor 801 may be a general-purpose single-or multi-chip microprocessor (e.g., an Advanced RISC (Reduced Instruction Set Computer) Machine (ARM)), a special purpose microprocessor (e.g., a digital signal processor (DSP)), a microcontroller, a programmable gate array, etc. The processor 801 may be referred to as a central processing unit (CPU). Although just a single processor 801 is shown in the computer system 800 of FIG. 8, in an alternative configuration, a combination of processors (e.g., an ARM and DSP) could be used.

The computer system 800 also includes memory 803 in electronic communication with the processor 801. The memory 803 may include computer-readable storage media and can be any available media that can be accessed by a general purpose or special purpose computer system. Computer-readable media that store computer-executable instructions are non-transitory computer-readable media (device). Computer-readable media that carry computer-executable instructions are transmission media. Thus, by way of example and not limitations, embodiment of the present disclosure can comprise at least two distinctly different kinds of computer-readable media: non-transitory computer-readable media (devices) and transmission media.

Both non-transitory computer-readable media (devices) and transmission media may be used temporarily to store or carry software instructions in the form of computer readable program code that allows performance of embodiments of the present disclosure. Non-transitory computer-readable media may further be used to persistently or permanently store such software instructions. Examples of non-transitory computer-readable storage media include physical memory (e.g., RAM, ROM, EPROM, EEPROM, etc.), optical disk storage (e.g., CD, DVD, HDDVD, Blu-ray, etc.), storage devices (e.g., magnetic disk storage, tape storage, diskette, etc.), flash or other solid-state storage or memory, or any other non-transmission medium which can be used to store program code in the form of computer-executable instructions or data structures and which can be accessed by a general purpose or special purpose computer, whether such program code is stored or in software, hardware, firmware, or combinations thereof.

Instructions 805 and data 807 may be stored in the memory 803. The instructions 805 may be executable by the processor 801 to implement some or all of the functionality disclosed herein. Executing the instructions 805 may involve the use of the data 807 that is stored in the memory 803. Any of the various examples of modules and components described herein may be implemented, partially or wholly, as instructions 805 stored in memory 803 and executed by the processor 801. Any of the various examples of data described herein may be among the data 807 that is stored in memory 803 and used during execution of the instructions 805 by the processor 801.

A computer system 800 may also include one or more communication interfaces 809 for communicating with other electronic devices. The communication interface(s) 809 may be based on wired communication technology, wireless communication technology, or both. Some examples of communication interfaces 809 include a Universal Serial Bus (USB), an Ethernet adapter, a wireless adapter that operates in accordance with an Institute of Electrical and Electronics Engineers (IEEE) 802.11 wireless communication protocol, a Bluetooth® wireless communication adapter, and an infrared (IR) communication port.

The communication interfaces 809 may connect the computer system 800 to a network. A “network” or “communications network” may generally be defined as one or more data links that enable the transport of electronic data between computer systems and/or modules, engines, or other electronic devices, or combinations thereof. When information is transferred or provided over a communication network or another communications connection (either hardwired, wireless, or a combination of hardwired or wireless) to a computing device, the computing device properly views the connection as a transmission medium. Transmission media can include a communication network and/or data links, carrier waves, wireless signals, and the like, which can be used to carry desired program or template code means or instructions in the form of computer-executable instruction or data structures and which can be accessed by a general purpose or special purpose computer.

A computer system 800 may also include one or more input devices 811 and one or more output devices 813. Some examples of input devices 811 include a keyboard, mouse, microphone, remote control device, button, joystick, trackball, touchpad, and lightpen. Some examples of output devices 813 include a speaker and a printer. One specific type of output device that is typically included in a computer system 800 is a display device 815. Display devices 815 used with embodiments disclosed herein may utilize any suitable image projection technology, such as liquid crystal display (LCD), light-emitting diode (LED), gas plasma, electroluminescence, or the like. A display controller 817 may also be provided, for converting data 807 stored in the memory 803 into one or more of text, graphics, or moving images (as appropriate) shown on the display device 815.

The various components of the computer system 800 may be coupled together by one or more buses, which may include one or more of a power bus, a control signal bus, a status signal bus, a data bus, other similar components, or combinations thereof. For the sake of clarity, the various buses are illustrated in FIG. 8 as a bus system 819.

The techniques described herein may be implemented in hardware, software, firmware, or any combination thereof, unless specifically described as being implemented in a specific manner. Any features described as modules, components, or the like may also be implemented together in an integrated logic device or separately as discrete but interoperable logic devices. If implemented in software, the techniques may be realized at least in part by a non-transitory processor-readable storage medium comprising instructions that, when executed by at least one processor, perform one or more of the methods described herein. The instructions may be organized into routines, programs, objects, components, data structures, etc., which may perform particular tasks and/or implement particular data types, and which may be combined or distributed as desired in various embodiments.

Further, upon reaching various computer system components, program code in the form of computer-executable instructions or data structures can be transferred automatically or manually from transmission media to non-transitory computer-readable storage media (or vice versa). For example, computer executable instructions or data structures received over a network or data link can be buffered in memory (e.g., RAM) within a network interface module (NIC), and then eventually transferred to computer system RAM and/or to less volatile non-transitory computer-readable storage media at a computer system. Thus, it should be understood that non-transitory computer-readable storage media can be included in computer system components that also (or even primarily) utilize transmission media.

INDUSTRIAL APPLICABILITY

The following description from ¶¶[0140]-[0162] includes various embodiments that, where feasible, may be combined in any permutation. For example, the embodiment of ¶[0140] may be combined with any or all embodiments of the following paragraphs. Embodiments that describe acts of a method may be combined with embodiments that describe, for example, systems and/or devices. Any permutation of the following paragraphs is considered to be hereby disclosed for the purposes of providing “unambiguously derivable support” for any claim amendment based on the following paragraphs. Furthermore, the following paragraphs provide support such that any combination of the following paragraphs would not create an “intermediate generalization.”

In some embodiments, a method of identifying subterranean features, includes receiving an inversion image indicating a portion of a subsurface feature, determining boundary information using a subsurface boundary machine learning model that is generated to process individual pixels of input inversion images through a decision-based architecture to identify boundaries of subsurface features, based on the boundary information, generating a boundary mask for the inversion image, and providing the boundary mask for adjusting one or more downhole parameters based on the boundary mask.

In some embodiments, the subsurface boundary machine learning model is generated to process individual pixels of the input inversion images through the decision-based architecture to identify boundaries of subsurface features.

In some embodiments, the boundary mask indicates at least one boundary of the subsurface feature of the inversion image.

In some embodiments, the method further includes updating the inversion image to indicate the at least one boundary of the subsurface feature based on the boundary mask.

In some embodiments, adjusting the one or more downhole parameters includes adjusting one or more steering parameters for maintaining a downhole tool within identified boundaries of the subsurface feature indicated by the boundary mask.

In some embodiments, the one or more steering parameters are adjusted automatically and without user input.

In some embodiments, the inversion image is generated based on downhole resistivity data.

In some embodiments, the inversion image is a two-dimensional inversion result.

In some embodiments, the method further includes automatically determining the boundary information and generating the boundary mask based on real-time inversion results.

In some embodiments, determining the boundary information includes determining an indication of boundary uncertainty for each pixel of the inversion image, and generating the boundary mask is based on a boundary uncertainty threshold for each pixel of the inversion image.

In some embodiments, the method further includes determining a pixel data set for each pixel of the inversion image and providing the pixel data set for each pixel to the subsurface boundary machine learning model to process the pixel data set through the decision-based architecture and determine an indication of boundary uncertainty for each pixel of the inversion image.

In some embodiments, determining the pixel data set for each pixel includes determining values that indicate, for each pixel, one or more of a measurement value of an associated downhole measurement, a downhole tool type, a measurement depth, pixel coordinates of the pixel within the inversion image, or latitude and longitude of the inversion image.

In some embodiments, the inversion image indicates a top boundary and a bottom boundary of the subsurface feature, and determining the boundary information includes identifying each of the top boundary and the bottom boundary of the subsurface feature in the inversion image.

In some embodiments, the inversion image indicates a portion of the subsurface feature and additionally indicates a portion of an additional subsurface feature, and determining the boundary information includes identifying a boundary of each of the subsurface feature and the additional subsurface feature in the inversion image.

In some embodiments, the method further includes receiving a plurality of additional inversion images indicating portions of the subsurface feature, determining additional boundary information for each of the plurality of additional inversion images using the subsurface boundary machine learning model, based on the additional boundary information, determining an additional boundary mask for each of the plurality of additional inversion images, generating a three-dimensional subsurface object of the subsurface feature based on assembling the inversion image and the plurality of additional inversion images and based on the boundary mask and the plurality of additional boundary masks, and providing the three-dimensional subsurface object for adjusting the one or more downhole parameters based on identified boundaries of the subsurface feature in the three-dimensional subsurface object.

In some embodiments, generating the three-dimensional subsurface object includes interpolating boundary information between consecutive inversion images of the assembled inversion images.

In some embodiments, a system includes at least one processor, memory in electronic communication with the at least one processor, and instructions stored in the memory, the instructions being executable by the at least one processor to identify inversion training images indicating a portion of a subsurface feature, obtain simulated inversion images that identify the subsurface feature from a subsurface model corresponding to a wellbore position of the inversion training images, generate simulated boundary masks for the simulated inversion images based on detected boundaries of the subsurface feature in the simulated inversion images, and generate a subsurface boundary machine learning model based on the inversion training images and the simulated boundary masks to process individual pixels of the inversion training images through a decision-based architecture and generate predicted boundary masks indicating boundaries of the subsurface feature within the inversion training images.

In some embodiments, the inversion training images are two-dimensional inversion results, the subsurface model is a three-dimensional formation model spanning at least a portion of the wellbore corresponding to the inversion training images, and generating the simulated inversion images includes creating two-dimensional transverse sections of the three-dimensional formation model corresponding to the inversion training images.

In some embodiments, generating the subsurface boundary machine learning model includes providing feedback to the subsurface boundary machine learning model based on comparing the predicted boundary masks to the simulated boundary masks to further tune the decision-based architecture of the subsurface boundary machine learning model.

In some embodiments, generating the subsurface boundary machine learning model is performed automatically based on receiving the inversion training images in real time.

In some embodiments, the subsurface boundary machine learning model is generated based on a small sample of 50 inversion training images or less.

In some embodiments, a computer-readable storage medium includes instruction that, when executed by at least one processor, cause the processor to receive an inversion image indicating a portion of a subsurface feature, determine boundary information using a subsurface boundary machine learning model that is generated to process individual pixels of input inversion images through a decision-based architecture to identify boundaries of subsurface features, based on the boundary information, generating a boundary mask for the inversion image, and update the inversion image to indicate a boundary of the subsurface feature based on the boundary mask.

In some embodiments, the computer-readable storage medium and the processor are located in a downhole tool positioned downhole in a wellbore.

The embodiments of the boundary detection system have been primarily described with reference to wellbore drilling operations; the boundary detection system described herein may be used in applications other than the drilling of a wellbore. In other embodiments, the boundary detection system according to the present disclosure may be used outside a wellbore or other downhole environment used for the exploration or production of natural resources. For instance, the boundary detection system of the present disclosure may be used in a borehole used for placement of utility lines. Accordingly, the terms “wellbore,” “borehole” and the like should not be interpreted to limit tools, systems, assemblies, or methods of the present disclosure to any particular industry, field, or environment.

One or more specific embodiments of the present disclosure are described herein. These described embodiments are examples of the presently disclosed techniques. Additionally, in an effort to provide a concise description of these embodiments, not all features of an actual embodiment may 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 embodiment-specific decisions will be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which may vary from one embodiment 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.

Additionally, it should be understood that references to “one embodiment” or “an embodiment” of the present disclosure are not intended to be interpreted as excluding the existence of additional embodiments that also incorporate the recited features. For example, any element described in relation to an embodiment herein may be combinable with any element of any other embodiment described herein. Numbers, percentages, ratios, or other values stated herein are intended to include that value, and also other values that are “about” or “approximately” the stated value, as would be appreciated by one of ordinary skill in the art encompassed by embodiments of the present disclosure. A stated value should therefore be interpreted broadly enough to encompass values that are at least close enough to the stated value to perform a desired function or achieve a desired result. The stated values include at least the variation to be expected in a suitable manufacturing or production process, and may include values that are within 5%, within 1%, within 0.1%, or within 0.01% of a stated value.

A person having ordinary skill in the art should realize in view of the present disclosure that equivalent constructions do not depart from the spirit and scope of the present disclosure, and that various changes, substitutions, and alterations may be made to embodiments disclosed herein without departing from the spirit and scope of the present disclosure. Equivalent constructions, including functional “means-plus-function” clauses are intended to cover the structures described herein as performing the recited function, including both structural equivalents that operate in the same manner, and equivalent structures that provide the same function. It is the express intention of the applicant not to invoke means-plus-function or other functional claiming for any claim except for those in which the words ‘means for’ appear together with an associated function. Each addition, deletion, and modification to the embodiments that falls within the meaning and scope of the claims is to be embraced by the claims.

The terms “approximately,” “about,” and “substantially” as used herein represent an amount close to the stated amount that is within standard manufacturing or process tolerances, or which still performs a desired function or achieves a desired result. For example, the terms “approximately,” “about,” and “substantially” may refer to an amount that is within less than 5% of, within less than 1% of, within less than 0.1% of, and within less than 0.01% of a stated amount. Further, it should be understood that any directions or reference frames in the preceding description are merely relative directions or movements. For example, any references to “up” and “down” or “above” or “below” are merely descriptive of the relative position or movement of the related elements. Additionally, as used herein, the term “and/or” includes any and all combinations of one or more of the associated listed items.

The present disclosure may be embodied in other specific forms without departing from its spirit or characteristics. The described embodiments are to be considered as illustrative and not restrictive. The scope of the disclosure is, therefore, indicated by the appended claims rather than by the foregoing description. Changes that come within the meaning and range of equivalency of the claims are to be embraced within their scope.