In-situ real-time borehole acoustic treatment

Abstract

Aspects described herein provide methods and apparatus for interpreting borehole acoustic dispersion data using a method based on data driven machine learning. A training data set is generated from two possible sources. First, application of machine learning enabled automatic dipole interpretation (MLADI) and / or machine learning enabled automatic quadrupole interpretation (MLAQI) methods on field data processing will naturally create a large amount of tagged data, i.e., pairing dispersion data with dispersion patterns tagged by MLADI and MLAQI. Secondly, a large amount of synthetic frequency dispersion data may also be generated from known model parameters. The two types of tagged data may be used alone or in combination to train a neural network model. These models may more efficiently map dispersion data to modal dispersion.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to and the benefit of U.S. Provisional Patent Application Serial No. 63 / 586,763, filed on September 29, 2023, and entitled “In-Situ Real-Time Borehole Sonic Processing,” and U.S. Provisional Patent Application Serial No. 63 / 632,845, filed on April 11, 2024, and entitled “In-Situ Real-Time Borehole Sonic Processing,” which are incorporated herein by reference in their entirety for all purposes. Technical Field

[0003] Aspects of the present disclosure relate to acoustic wellbore data processing. More specifically, aspects of the present disclosure provide for interpreting borehole acoustic dispersion data using data-driven machine learning based techniques. Background Art

[0004] This section is intended to introduce the reader to various aspects of the art that may be related to various aspects of the present technology described and / or claimed below. It is believed that this discussion helps to provide the reader with background information to facilitate a better understanding of various aspects of the present disclosure. Therefore, it should be understood that these statements should be read from this perspective, rather than as an admission of any kind.

[0005] Borehole acoustic measurements play an integral role in the industry for formation evaluation and subsurface characterization. Various interpretation methods have been developed to derive basic formation properties such as compressional and shear slowness, anisotropy, radial variations, etc. from the measured acoustic waveforms. The quality of interpretation using these methods depends on expert experience, involving parameter adjustments based on prior information and domain knowledge. The duration from data acquisition conversion to interpretation delivery can vary from days to weeks, based on the availability and experience of the data analyst. Recent Machine Learning Enabled Automatic Dipole Interpretation (MLADI) and Machine Learning Enabled Automatic Quadrupole Interpretation (MLAQI) solutions enable fully automated acoustic data processing, but involve relatively high computational costs and are relatively slow. Summary of the invention

[0006] An overview of certain embodiments described herein is set forth below. It should be understood that these aspects are presented merely to provide the reader with a brief overview of these particular embodiments, and these aspects are not intended to limit the scope of the present disclosure.

[0007] Embodiments described herein provide techniques for processing acoustic dispersion data in situ substantially in real time. For example, the systems and methods described herein are configured to train a neural network model using training acoustic dispersion data to map the training acoustic dispersion data to one or more dispersion mode curves in a plurality of dispersion mode curves, receive real-time acoustic dispersion data from an acoustic logging tool substantially in real time when the acoustic logging tool is deployed in a borehole extending through a geological formation, and analyze the real-time acoustic dispersion data using the trained neural network model substantially in real time when the acoustic logging tool is deployed in a borehole extending through a geological formation to predict a dispersion mode curve in a plurality of dispersion mode curves to which the real-time acoustic dispersion data relates and / or directly calculate a parameter of a geological formation.

[0008] Various improvements of the above features may be made with respect to various aspects of the present disclosure. Additional features may also be incorporated into these various aspects. These improvements and additional features may exist alone or in any combination. For example, the various features discussed below with respect to one or more illustrated embodiments may be incorporated into any of the above aspects of the present disclosure alone or in any combination. The brief overview presented above is intended to familiarize the reader with certain aspects and contexts of the embodiments of the present disclosure without limiting the claimed subject matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Various aspects of the present disclosure may be better understood upon reading the following detailed description and referring to the accompanying drawings, in which:

[0010] Figure 1 is a graphical representation of an exemplary logging while drilling (LWD) quadrupole dispersion plot according to an embodiment of the present disclosure;

[0011] Figure 2 depicts a schematic diagram of a system for processing sonic and ultrasonic wellbore measurements according to an embodiment of the present disclosure;

[0012] Figure 3 Depicting example operations of an acoustic downhole tool according to an embodiment of the present disclosure;

[0013] Figure 4 is a graphical representation of LWD quadrupole dispersion data and interpreted quadrupole dispersion mode curves according to an embodiment of the present disclosure;

[0014] Figure 5 is a diagram of converting raw dispersion data into a two-dimensional image according to an embodiment of the present disclosure;

[0015] Figure 6 is an example of a neural network model training workflow according to an embodiment of the present disclosure;

[0016] Figure 7is an example of a general workflow for applying a trained neural network model to sound data processing according to an embodiment of the present disclosure;

[0017] Figure 8 is a graphical representation of a comparison of a neural network result and a true dispersion modal curve according to an embodiment of the present disclosure; and

[0018] Fig. 9 An example method utilizing the in situ real-time acoustic dispersion data processing techniques described herein is described in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0019] One or more specific embodiments of the present disclosure will be described below. These described embodiments are merely examples of the currently disclosed technology. In addition, in order to provide a concise description of these embodiments, all features of the actual implementation may not be described in the specification. It should be understood that in the development of any such actual implementation, as in any engineering or design project, many implementation-specific decisions must be made to achieve the developer's specific goals, such as meeting system-related and business-related constraints, which may vary from implementation to implementation. In addition, it should be understood that such development efforts may be complex and time-consuming, but for ordinary technicians who benefit from the present disclosure, it is still a routine task of design, production and manufacturing.

[0020] When introducing elements of various embodiments of the present disclosure, the articles "a," "an," and "the" are intended to indicate that there are one or more elements. The terms "comprising," "including," and "having" are intended to be inclusive, and mean that there may be additional elements in addition to the listed elements. Additionally, it should be understood that references to "one embodiment" or "an embodiment" of the present disclosure are not intended to be interpreted to exclude the existence of additional embodiments that also include the recited features.

[0021] As used herein, the terms "connect," "connection," "connected," "in connection with," and "connecting" are used to mean "directly connected with" or "connected with via one or more elements"; and the term "group" is used to mean "one element" or "more than one element." In addition, the terms "couple," "coupling," "coupled," "coupled together," and "coupled with" are used to mean "directly coupled together" or "coupled together via one or more elements." As used herein, the terms "upper" and "lower," "uphole" and "downhole," "upper" and "lower," "top" and "bottom," and other similar terms indicating relative positions to a given point or element are used to more clearly describe some elements. In general, whether the well (e.g., wellbore, borehole) is vertical, horizontal, or inclined relative to the ground, these terms refer to a reference point in the ground from which the drilling operation is initiated as the top (e.g., uphole or upper) point, and the total depth along the drilling axis is the lowest (e.g., downhole or lower) point. In addition, the term "interval" with respect to coiled tubing (CT) pipe is used to refer to a specific axial portion along the axial length of the CT pipe. In addition, the term "a priori data" is used to refer to data determined based on theoretical derivation rather than empirical measurement.

[0022] Additionally, as used herein, the terms "real time," "real-time," or "substantially real-time" are used interchangeably and are intended to describe operations (e.g., computing operations) that are performed without any human-perceivable interruptions between operations. For example, as used herein, data associated with the systems described herein may be collected, transmitted, and / or used in control calculations "substantially in real time," such that data reading, data transmission, and / or data processing steps occur once per second, once per 0.1 second, once per 0.01 second, or even more frequently during operation of the system (e.g., while the system is operating). Additionally, as used herein, the terms "automatic," "automatically," and "automated" are intended to describe, for example, operations that are performed or caused to be performed by a data processing system (i.e., performed solely by the data processing system without human intervention). Additionally, as used herein, the term "approximately equal to" may be used to represent values ​​that are relatively close to each other (e.g., within 5%, within 2%, within 1%, within 0.5%, or even closer to each other).

[0023] A major problem in dispersion analysis is to correctly label the desired modes in order to perform a model-based inversion. For example, Figure 1A graph 10 of logging while drilling (LWD) quadrupole dispersion data is shown, where several modes can be excited simultaneously. In particular, Figure 1 The following sets of acoustic wave dispersion data are shown:

[0024] Stoneley pattern (data point 12)

[0025] Drilling quadrupole mode (data point 14)

[0026] Collar quadrupole mode (data point 16)

[0027] Modeled dispersion curve for the borehole quadrupole mode for the shear slowness output from the inversion (solid line 18)

[0028] Shear-head or pseudo-Rayleigh mode (dashed line 20)

[0029] Dispersive second-order quadrupole mode (data point 22)

[0030] Compression first wave (data point 24)

[0031] Spectra of a single pole (solid line 26) and a quadrupole (solid line 28)

[0032] The borehole quadrupole mode 14 is the desired mode for inversion of formation shear slowness. Traditionally, it is necessary to manually set appropriate frequency filters in order to mark the correct mode and avoid mislabeling other contaminating modes, such as the collar quadrupole mode 16 or the high-order quadrupole mode 22.

[0033] As discussed above, machine learning enabled automatic dipole interpretation (MLADI) and machine learning enabled automatic quadrupole interpretation (MLAQI) methods have been developed to interpret such borehole acoustic dispersion data of formation elastic properties. Such techniques provide a fully automated solution, but are relatively computationally expensive. Therefore, there is a need to further increase the speed of data processing and reduce the computational cost.

[0034] Embodiments described herein include methods for marking various patterns and estimating formation slowness from borehole acoustic slowness-frequency data. Specifically, embodiments described herein implement a fully automated solution with high computational efficiency, which enables real-time field processing of borehole acoustic measurements. The data-driven machine learning (ML) techniques described herein can be deployed at the well site or downhole for real-time data processing.

[0035] The embodiments described herein include a framework for interpreting borehole acoustic dispersion data using a data-driven ML-based approach. In particular, training data sets from two possible sources can be generated. First, the application of MLADI and MLAQI methods to field data processing will naturally create a large amount of labeled data (e.g., pairing dispersion data with dispersion modes labeled by MLADI and MLAQI). Second, a large amount of synthetic dispersion data can also be generated from known model parameters. These two types of labeled data can be used alone or in combination to train neural network models. These models can map dispersion data to modal dispersion more efficiently than MLADI and MLAQI solutions. In this way, the embodiments described herein can complete data processing in just a few seconds using only a single CPU, making it suitable for deployment on a computer at the well site or even downhole. In some embodiments, the training models and processing algorithms described herein can be further implemented on an FPGA or ASIC to obtain even better performance.

[0036] As an introduction, Figure 2 A schematic diagram of a system 30 for processing sonic and ultrasonic borehole measurements is depicted. In particular, Figure 2 Surface equipment 32 is shown above a geological formation 34. Figure 2 In the example of , drilling operations have previously been performed to drill a wellbore 36. In the illustrated embodiment, cement 38 has been used to seal an annulus 40 (ie, the space between the wellbore 36 and the casing joint 42 and collar 44).

[0037] like Figure 2 As shown, several casing joints 42 (hereinafter also referred to as casing 42) are connected together by casing collars 44 to stabilize the wellbore 36. The casing joints 42 represent the length of the conductive pipe, which can be formed by steel or similar materials. In certain embodiments, the casing joints 42 can each include an external thread (male thread form) connection at each end. The corresponding internal thread (female thread form) connection in the casing collar 44 can connect two nearby casing joints 42. Connected in this way, the casing joints 42 can be assembled to form the casing string 54 to a suitable length and specification for the wellbore 36. The casing joints 42 and / or the collars 44 can be made of carbon steel, stainless steel or other suitable materials to withstand various forces, such as collapse, burst and tensile failure and chemically aggressive fluids.

[0038] The surface equipment 32 may perform various logging operations to detect conditions of the wellbore 36, as described in more detail herein. The logging operations may measure parameters of the geological formation 34 (e.g., resistivity or porosity) and / or parameters of the wellbore 36 (e.g., temperature, pressure, fluid type, or fluid flow rate). As will be discussed in more detail below, some of these measurements may be obtained at various points in the design, drilling, and completion of the well, and may be used to evaluate the properties of the geological formation 34 and / or the wellbore 36.

[0039] As described in further detail herein, the acoustic logging tool 46 may obtain at least some of these measurements. Figure 2 The example of FIG. 4 shows an acoustic logging tool 46 conveyed through a wellbore 36 by a cable 48. Such a cable 48 may be a mechanical cable, an electrical cable, or an electro-optical cable including an optical fiber line protected from the harsh environment of the wellbore 36. However, in other embodiments, the acoustic logging tool 46 may be conveyed using any other suitable conveying device, such as a coiled tubing. The acoustic logging tool 46 may obtain measurements of amplitude and variable density from acoustic waves, measurements of acoustic impedance from ultrasonic waves, and / or measurements of bend attenuation and velocity from third interface echoes. The availability of these independent measurements may be used to increase the accuracy and confidence of the well and / or formation evaluation and interpretation achieved by the acoustic logging tool 46.

[0040] The acoustic logging tool 46 may be deployed within the wellbore 36 by the surface equipment 32, which may include a vehicle 50 and a deployment system, such as a drilling rig 52. Data collected by the acoustic logging tool 46 relating to the geological formation 34 or the wellbore 36 may be transmitted to the surface, and / or stored in the acoustic logging tool 46 for later processing and analysis. In certain embodiments, the vehicle 50 may be equipped with or may communicate with a computer and software to perform data collection and analysis.

[0041] Figure 2Also schematically shown is an enlarged view of a portion of the wellbore 36. As described above, the acoustic logging tool 46 can obtain acoustic measurements related to the presence of solids or liquids behind the casing 42. When the acoustic logging tool 46 provides such measurements to the surface equipment 32 (e.g., via the cable 48), the surface equipment 32 can pass the measurements as data 56 to a data processing system 58, which includes a processor 60, a memory 62, a storage device 64, and / or a display 66. In other examples, the data 56 can be processed by a similar data processing system 58 at any other suitable location (e.g., a remote data center). The data processing system 58 can collect the data 56 and determine one or more indices and indicators, which, as described in more detail herein, can objectively indicate various properties within the geological formation 34 and the wellbore 36 itself. Additionally or alternatively, the data processing system 58 can associate various data obtained throughout the creation of the well (e.g., design, drilling, logging, completion, etc.) that can help evaluate the wellbore 36 and / or geological formation 34. That is, the processor 60 can calculate indicators and / or indices using instructions stored in the memory 62 and / or storage device 64, and / or can collect and associate other data into the well and / or formation evaluation. In this way, the memory 62 and / or storage device 64 of the data processing system 58 can be any suitable article that can store instructions. The memory 62 and / or storage device 64 can be a ROM, a random access memory (RAM), a flash memory, an optical storage medium, or a hard drive, to name just a few examples. The display 66 can be any suitable electronic display that can display logs, indices, and / or indicators related to the wellbore 36 and / or geological formation 34.

[0042] With this in mind, Figure 3 Depicted is an example operation of an acoustic logging tool 46 in a wellbore 36. Specifically, in certain embodiments, a transducer 72 in the acoustic logging tool 46 can emit an acoustic wave 74 outwardly toward the casing 42 and the geological formation 34. Reflected acoustic waves 76, 78 can correspond to the interfaces at the casing 42, cement 38, and the geological formation 34, respectively. The acoustic logging tool 46 can use any suitable number of different techniques, including measuring the amplitude and variable density from the acoustic wave, the acoustic impedance from the ultrasonic wave, and / or the bending attenuation and velocity from the third interface echo. There is a mathematical model for each method of calculating the properties of the contents in the annulus 40. As a mathematical model, many assumptions and simplifications of the real world can be adopted. These may involve correction and calibration or reference point checks.

[0043] As described above, an acoustic wave source (e.g., a monopole acoustic wave source) can be used to excite a cased well (e.g., a double cased well) environment, and an array of acoustic wave sensing elements can be used to detect reflected acoustic waves corresponding to various interfaces (e.g., interfaces at casing 42, cement 38, and geological formation 34, respectively).

[0044] As described in more detail herein, to facilitate automation of data processing, an ML-assisted method, referred to as MLADI, was developed for wireline (WL) borehole acoustic dipole interpretation. The method was subsequently extended to include LWD borehole acoustic quadrupole interpretation, referred to as MLAQI. Using MLADI and MLAQI, a neural network model is trained on synthetic data (generated by modeling such as pattern search) to map from model parameters to dispersion data (i.e., slowness-frequency domain). During acoustic data processing, the waveform can first be transformed into dispersion data. A combination of clustering algorithms and an inversion process, in which the neural network model is used as a forward solver, is used to automatically label and invert corresponding dispersion modes (e.g., dipole modes and quadrupole modes). These innovative methods enable fully automated data processing without the need for user intervention while maintaining a level of quality comparable to traditional expert interpretation. As a result, they can significantly reduce the time from data acquisition to interpretation delivery.

[0045] An important consideration for such methods is that they rely on relatively high performance computing platforms to achieve timely results. For example, due at least in part to the computationally intensive nature of the processing, data processing times can range from minutes to hours, depending on the amount of data being processed. The embodiments described herein provide a data-driven ML-based approach to map directly from acoustic dispersion data to various dispersion modes (e.g., dipoles and quadrupoles, as well as Figure 1 other dispersion modes described in ) and simultaneously obtain the most sensitive formation parameters, such as slowness.

[0046] The embodiments described herein involve training a neural network model that maps directly from acoustic wave dispersion data 80 to dispersion modal curves 82, such as Figure 4 . A first source of training data sets may come from field data processing (e.g., from field data collected by acoustic logging tool 46). In practice, many borehole acoustic measurements may be processed using MLAQI / MLADI or other methods, from which a large amount of labeled data may be obtained for machine learning. A second source of training data sets may come from numerical modeling. Various model scenarios may be created and simulated by numerical solvers to generate synthetic acoustic data with natural labels. In general, acoustic dispersion data 80 may be used as input, and dispersion mode curves 82 and / or slowness values ​​may be used as output.

[0047] In some embodiments, the training data set can be pre-processed before being used for machine learning. For example, for a data set generated from field data processing, a quality control (QC) threshold can be selected to filter out those data with low QC scores. Figure 5 A data sample with a relatively high QC score is shown. In addition, during the conversion, some additional thresholds can be applied to remove some scattered points to suppress noise. For example, Figure 5 As shown, certain scattered data points 80 scattered with low coherence can be removed from consideration to obtain a transformed image 84. In addition, during pre-processing, the slowness and frequency can be normalized using predefined lower and upper limits, such as Figure 5 As shown. In addition, in some embodiments, the normalized dispersion data 80 can be converted to a two-dimensional (2D) image 84, or in fact can be transformed into any type of image 84 (e.g., a binary image, an image with more color scale, etc.). In some embodiments, other attributes such as amplitude in the dispersion map can be ignored, and the dispersion data 80 can be converted to a binary image (i.e., black and white, as shown in the converted image 84). However, in other embodiments, additional attributes can also be considered and the dispersion data 80 can be converted to an image 84 with a fine color scale.

[0048] The next step is to feed the preprocessed labeled dataset (e.g. Figure 5 The transformed image 84 shown is used as an input array 86 to train the neural network model, such as Figure 6 As shown in the neural network model training workflow 88. Various neural network structures 90 can be selected for this ML problem. Depending on the specific neural network structure 90 used, the converted 2D image 84 can be transformed into different types of arrays as input. In addition, in some embodiments, the neural network model parameters can be optimized to ensure that the output matches the labeled dispersion mode curve 82 or variant (for example, the true formation shear slowness can be linked to the one-dimensional (1D) array of the dispersion mode curve 82). In some embodiments, a fully connected neural network model 90 can be used, but it can be replaced by other structures such as a convolutional neural network model 90. As discussed above, the dispersion data 80 can be transformed into an image 84, which is a binary image or an image with more color scales. In order to use a fully connected neural network structure, the 2D image 84 (i.e., a 2D input array) can be transformed into a 1D array as an input array 86. The output can be a 1D array representing a dispersion mode curve 82 composed of slownesses at predefined frequencies.

[0049] Once trained (for example, using Figure 6 The neural network model training workflow 88), the trained neural network model 90 can be used for real-time field data processing. Figure 7 A general high-level workflow 92 for utilizing a trained neural network model 90 is shown, and Figure 8 The results when applied to a sample data set are shown. In particular, acoustic waveforms 94 can be collected in substantially real time (e.g., using an acoustic logging tool 46), and real-time dispersion data 96 (e.g., similar to the training dispersion data 80 discussed above) can be extracted from the acoustic waveforms 94 and converted into a 2D image 98 (e.g., similar to the training 2D image 84 discussed above) or other type of image, which can be fed into the trained neural network model 90 to predict dispersion mode curves 100 and / or directly calculate slowness. It should be understood that the predicted dispersion mode curves 100 can, in turn, be used to determine certain parameters of the geological formation 34, including but not limited to slowness.

[0050] As described above, the workflow 92 can be executed on different infrastructures, including a CPU, FPGA, or ASIC on a wellsite or downhole computer. In other words, instead of utilizing the data processing system 58 to perform all of the data processing functions described herein, in other embodiments, other data processing systems (e.g., an FPGA or ASIC integrated into the acoustic logging tool 46) can be utilized to perform the data processing functions described herein. In addition, in other embodiments, the data processing system 58 can perform some of the data processing functions described herein, while other data processing systems (e.g., an FPGA or ASIC integrated into the acoustic logging tool 46) can perform other data processing functions described herein. As just one non-limiting example, in some embodiments, the data processing system 58 can perform the training of the neural network model 90 described herein, while other data processing systems (e.g., an FPGA or ASIC integrated into the acoustic logging tool 46) can analyze real-time data based on the trained neural network model 90, also as described herein.

[0051] Fig. 9 An example method 102 is shown that utilizes the in situ real-time acoustic dispersion data processing techniques described herein. Fig. 9As shown, in some embodiments, the method 102 may include training the neural network model 90 using the training acoustic dispersion data 80 to map the training acoustic dispersion data 80 to one or more dispersion mode curves 82 in the plurality of dispersion mode curves 82 (box 104). In addition, in some embodiments, the method 102 may also include receiving real-time acoustic dispersion data 96 from the acoustic logging tool 46 in substantially real time while the acoustic logging tool 46 is deployed in the wellbore 36 extending through the geological formation 34 (box 106). Additionally, in certain embodiments, the method 102 may further include analyzing the real-time acoustic dispersion data 96 substantially in real time using the trained neural network model 90 while the acoustic logging tool 46 is deployed in a wellbore 36 extending through the geological formation 34 to predict a dispersion mode curve 82 (as a predicted dispersion mode curve 100) among a plurality of dispersion mode curves 82 associated with the real-time acoustic dispersion data 96 and / or directly calculating at least one parameter (e.g., compression slowness, shear slowness, anisotropy, radial variation, etc.) of the geological formation 34 (block 108). As described above, it should be understood that the predicted dispersion mode curve 100 may in turn be used to determine certain parameters of the geological formation 34, including, but not limited to, compression slowness, shear slowness, anisotropy, radial alteration, etc.

[0052] Furthermore, in some embodiments, training the neural network model 90 using the training acoustic wave dispersion data 80 may include converting the training acoustic wave dispersion data 80 into an image 84, and using the image 84 as an input array 86 to train the neural network model 90. In some embodiments, the image 84 may be a 2D image 84 of the training acoustic wave dispersion data 80. Furthermore, in some embodiments, the method 102 may further include preprocessing the training acoustic wave dispersion data 80 before training the neural network model 90 using the training acoustic wave dispersion data 80. For example, in some embodiments, preprocessing the training acoustic wave dispersion data 80 may include filtering out one or more training acoustic wave dispersion data sets 80 having a quality control (QC) score below a QC threshold. Additionally, in some embodiments, preprocessing the training acoustic wave dispersion data 80 may include normalizing parameters of the training acoustic wave dispersion data 80 using predefined lower and upper limits.

[0053] Additionally, in some embodiments, the method 102 may include utilizing the trained neural network model 90 to simultaneously interpret the multiple modes associated with the multiple dispersion modal curves 82. Additionally, in some embodiments, the method 102 may include utilizing the trained neural network model 90 to simultaneously interpret the compression slowness and the shear slowness of the geological formation 34. Additionally, in some embodiments, at least one parameter of the geological formation 34 may include the shear slowness of the geological formation 34, and the method 102 may include using the formation shear slowness as an initial guess for a machine learning enabled automatic dipole interpretation (MLADI) analysis or a machine learning enabled automatic quadrupole interpretation (MLAQI) analysis to further refine the shear slowness of the geological formation 34.

[0054] Additionally, in some embodiments, the plurality of dispersion modal curves 82 may include a modeled dispersion curve 18 of a Stoneley mode 12, a borehole quadrupole mode 14, a collar quadrupole mode 16, a borehole quadrupole mode for shear slowness output, a shear first wave or pseudo-Rayleigh mode 20, a dispersion second-order quadrupole mode 22, a compression first wave 24, and a monopole 26 or a quadrupole 28.

[0055] In general, the solutions described herein require minimal computing resources, enabling real-time data processing in downhole conditions, which is very attractive for LWD scenarios due to the limited bandwidth for data transmission during drilling. This means that sonic measurements can be interpreted essentially in real time while drilling, which can help optimize the drilling process with reduced risk (e.g., real-time pore pressure and / or stress prediction).

[0056] Furthermore, the framework can be extended to other modes, such as Stoneley modes, dipole modes, pseudo-Rayleigh modes, leaky P modes, etc. (e.g., as discussed above with respect to Figure 1 The proposed hybrid solution is a novel method for interpreting multiple modes simultaneously. The framework can also be extended to interpret compression and shear slowness simultaneously. In addition, the solution can be used to obtain formation shear slowness, which can then be used as an initial guess for MLADI / MLAQI analysis for further refinement. This hybrid solution is much faster than a typical MLADI / MLAQI analysis while maintaining good accuracy.

[0057] The above specific embodiments have been shown by way of example, and it should be understood that these embodiments may allow various modifications and alternative forms. It should be further understood that the claims are not intended to be limited to the specific forms disclosed, but to cover all modifications, equivalents and substitutes that fall within the spirit and scope of the present disclosure.

[0058] The technology presented and claimed herein is cited and applied to material objects and specific examples of a practical nature that demonstrably improve the art and is therefore not abstract, intangible, or purely theoretical. In addition, if any claim appended at the end of this specification contains one or more elements designated as "means for [performing] ... [function]" or "steps for [performing] ... [function]," it is intended that such elements be interpreted under 35 U.S.C. §112(f). However, for any claim containing elements designated in any other manner, it is intended that such elements not be interpreted under 35 U.S.C. §112(f).

Claims

1. A method comprising: Using the training acoustic wave dispersion data to train a neural network model to map the training acoustic wave dispersion data to one or more dispersion mode curves among a plurality of dispersion mode curves; receiving real-time acoustic dispersion data from the acoustic logging tool in substantially real time while the acoustic logging tool is deployed in a wellbore extending through a geological formation; and When the acoustic logging tool is deployed in a wellbore extending through the geological formation, the trained neural network model is used to analyze the real-time acoustic wave dispersion data substantially in real time to predict a dispersion mode curve among multiple dispersion mode curves related to the real-time acoustic wave dispersion data and / or directly calculate at least one parameter of the geological formation.

2. The method of claim 1 , wherein using the training acoustic wave dispersion data to train the neural network model comprises converting the training acoustic wave dispersion data into a two-dimensional image, and using the two-dimensional image as an input array to train the neural network model.

3. The method of claim 1 , comprising preprocessing the training acoustic wave dispersion data before using the training acoustic wave dispersion data to train the neural network model. 4 . The method of claim 3 , wherein preprocessing the training acoustic wave dispersion data comprises filtering out one or more samples of the training acoustic wave dispersion data having a quality control (QC) score below a QC threshold. 5 . The method of claim 3 , wherein preprocessing the training acoustic wave dispersion data comprises normalizing parameters of the training acoustic wave dispersion data using predefined lower and upper limits.

6. The method of claim 1, comprising utilizing a trained neural network model to simultaneously interpret multiple modes associated with a plurality of dispersion modal curves.

7. The method of claim 1, comprising utilizing a trained neural network model to simultaneously interpret compression slowness and shear slowness of the geological formation.

8. A method according to claim 1, wherein at least one parameter of the geological formation includes the shear slowness of the geological formation, and wherein the method includes using the shear slowness of the geological formation as an initial guess for a machine learning enabled automatic dipole interpretation (MLADI) analysis or a machine learning enabled automatic quadrupole interpretation (MLAQI) analysis to further refine the shear slowness of the geological formation.

9. The method of claim 1, wherein the plurality of dispersion mode curves include Stoneley modes.

10. The method of claim 1, wherein the plurality of dispersion mode curves include borehole quadrupole modes.

11. The method of claim 1 , wherein the plurality of dispersion mode curves include a collar quadrupole mode.

12. The method of claim 1, wherein the plurality of dispersion mode curves include a dipole mode.

13. The method of claim 1, wherein the plurality of dispersion mode curves include shear head waves or pseudo-Rayleigh modes.

14. The method of claim 1, wherein the plurality of dispersive modal curves include a compression head wave.

15. A data processing system, the data processing system being configured to: Using the training acoustic wave dispersion data to train a neural network model to map the training acoustic wave dispersion data to one or more dispersion mode curves among a plurality of dispersion mode curves; receiving real-time acoustic dispersion data from the acoustic logging tool in substantially real time while the acoustic logging tool is deployed in a wellbore extending through a geological formation; and When the acoustic logging tool is deployed in a wellbore extending through the geological formation, the trained neural network model is used to analyze the real-time acoustic wave dispersion data substantially in real time to predict a dispersion mode curve among multiple dispersion mode curves related to the real-time acoustic wave dispersion data and / or directly calculate at least one parameter of the geological formation.

16. The data processing system of claim 15, wherein using the training acoustic wave dispersion data to train the neural network model comprises converting the training acoustic wave dispersion data into a two-dimensional image, and using the two-dimensional image as an input array to train the neural network model.

17. The data processing system of claim 15, wherein the plurality of dispersion mode curves include Stoney modes, borehole quadrupole modes, collar quadrupole modes, borehole dipole modes, shear head waves or pseudo-Rayleigh modes, and compression head waves.

18. The data processing system of claim 15, wherein the data processing system is configured to utilize a trained neural network model to simultaneously interpret other modes associated with a plurality of dispersion mode curves, wherein the other modes include Stoney modes, dipole modes, pseudo-Rayleigh modes, leaky P modes, or some combination thereof.

19. The data processing system of claim 15, wherein the data processing system is configured to utilize a trained neural network model to simultaneously interpret compression slowness and shear slowness of the geological formation.

20. A data processing system according to claim 15, wherein at least one parameter of the geological formation includes the shear slowness of the geological formation, and wherein the data processing system is configured to use the shear slowness of the geological formation as an initial guess for a machine learning enabled automatic dipole interpretation (MLADI) analysis or a machine learning enabled automatic quadrupole interpretation (MLAQI) analysis to further refine the shear slowness of the geological formation.

Images