Method and device for determining the slowness of sound waves

Abstract

Accessing sonic logging data including sonic waveforms associated with a plurality of shot gathers. Applying a transform operator to the sonic logging data to provide a transformed sonic image, the transform operator including at least one of a short-time average long-time average (STA / LTA) operator, a phase shift operator, and a deconvolution operator. Performing a machine learning process using the transformed sonic image to determine a sonic slowness associated with the sonic logging data. Providing the sonic slowness as an output.

Description

[0001] Cross-reference paragraphs

[0002] This application claims the benefit of U.S. Provisional Application No. 62 / 739,580, filed on October 1, 2018, entitled “METHOD AND DEVICE FOR DETERMINING SONIC SLOWNESS,” the disclosure of which is hereby incorporated herein by reference. Background Art

[0003] Acoustic logging can be performed in a subsurface formation using a logging tool, such as a wireline tool and / or a logging while drilling tool. The logging tool is placed in a subsurface formation (e.g., a wellbore) and includes at least one transmitter for transmitting a reference acoustic wave. The logging tool may include a plurality of receivers that receive and register arriving acoustic waves after the source waves propagate through the subsurface formation. The acoustic data provided by acoustic logging can be used to characterize the physical properties of the subsurface formation, such as the properties of the rock inside the subsurface formation. The acoustic data can be used to estimate the acoustic slowness (e.g., the inverse of the velocity) in different parts of the subsurface formation. Slowness can be defined as the amount of time it takes for a wave to travel a certain distance, and can be measured in microseconds per foot.

[0004] Processing sonic logging data has always been an expert-driven process. The results of sonic logging analysis can vary greatly depending on the approach and skill of the expert performing the analysis. Furthermore, the ability and availability of the expert to perform the analysis limits the amount of sonic logging data that can be processed and can also cause significant delays between the time the sonic data is recorded and the time the results are obtained. Improved acoustic data analysis methods are needed to increase the accuracy and usability of sonic logging data analysis. Improved methods are also needed to achieve actual real-time sonic slowness estimation for while drilling operations. Summary of the invention

[0005] Methods and systems for determining acoustic wave slowness are described.

[0006] In one embodiment, a method for determining acoustic slowness is provided. Sonic logging data including acoustic waveforms associated with a plurality of shot gathers is accessed. A transform operator is applied to the sonic logging data to provide a transformed acoustic image, the transform operator including at least one of a short-time average long-time average (STA / LTA) operator, a phase shift operator, and a deconvolution operator. A machine learning process is performed using the transformed acoustic image to determine an acoustic slowness associated with the sonic logging data. The acoustic slowness is provided as an output.

[0007] The sonic logging data may include shot gathers of monopole waveforms, and the transform operator may include a STA / LTA operator.

[0008] The sonic logging data may include shot gathers of dipole waveforms, and the transform operator may include a phase shift operator.

[0009] The sonic logging data may include a two-dimensional image, and the transform may include a deconvolution operator.

[0010] The sonic logging data may include a three-dimensional image, and the transform may include a deconvolution operator.

[0011] The machine learning process may include convolutional neural networks.

[0012] The determined acoustic wave slowness may include at least one of the slowness of longitudinal waves, the slowness of shear waves, the slowness of Stoneley waves, the slowness of leakage P waves, the slowness of Rollie waves, and the slowness of pseudo-Rollie waves.

[0013] The providing may include displaying the slowness of the sound waves.

[0014] The providing may include displaying the slowness of the acoustic waves as a function of depth.

[0015] In one embodiment, a device includes an interface, a memory, and a processor. The interface is configured to obtain sonic logging data including sonic waveforms associated with a plurality of shot gathers. The memory is configured to store computer executable instructions. The processor is operably connected to the interface and the memory. The processor is configured to execute the instructions and cause the device to: apply a transform operator to the sonic logging data to provide a transformed sonic image, the transform operator including at least one of a short-time average long-time average (STA / LTA) operator, a phase shift operator, and a deconvolution operator; perform a machine learning process using the transformed sonic image to determine a sonic slowness associated with the sonic logging data; and provide the sonic slowness as an output.

[0016] The sonic logging data may include shot gathers of monopole waveforms, and the transform operator may include a STA / LTA operator.

[0017] The sonic logging data may include shot gathers of dipole waveforms, and the transform operator may include a phase shift operator.

[0018] The sonic logging data may include a two-dimensional image, and the transform may include a deconvolution operator.

[0019] The sonic logging data may include a three-dimensional image, and the transform may include a deconvolution operator.

[0020] The machine learning process may include convolutional neural networks.

[0021] The determined acoustic wave slowness may include at least one of the slowness of longitudinal waves, the slowness of shear waves, the slowness of Stoneley waves, the slowness of leakage P waves, the slowness of Rollie waves, and the slowness of pseudo-Rollie waves.

[0022] The device may include a display configured to display the slowness of the sound waves.

[0023] The display may be configured to display the slowness of the sound waves as a function of depth.

[0024] In one embodiment, a method for training a machine learning process is provided. A plurality of sonic logging data including sonic waveforms and sonic slownesses associated with a plurality of shot gathers is accessed. A database of the machine learning process is trained using the sonic logging data as a training input and using the sonic slownesses as a training output.

[0025] A transform operator may be applied to the sonic logging data prior to the training step, the transform operator comprising at least one of a short time average long time average (STA / LTA) operator, a phase shift operator, and a deconvolution operator.

[0026] Although some embodiments are described herein as devices, one of ordinary skill will recognize that such devices may be defined in a single integrated device, or alternatively, may be defined in a distributed system having more than one physical component. Therefore, embodiments described as devices may also be identifiable and defined as systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The following drawings form part of this specification and are included to further illustrate certain aspects of the present disclosure. The present disclosure may be better understood by reference to one or more of these drawings in combination with the detailed description of specific embodiments presented herein.

[0028] Figure 1 is a schematic block diagram illustrating an embodiment of a wellsite system.

[0029] Figure 2 is a schematic block diagram illustrating an embodiment of a wellsite system.

[0030] Figure 3 is a graph of an exemplary acoustic wave waveform.

[0031] Figure 4 is a graph of an exemplary shot gather.

[0032] Figure 5 is a graph of exemplary shot gathers and P-wave slowness.

[0033] Figure 6 is a flow chart of an exemplary process for determining acoustic wave slowness.

[0034] Figure 7 is a graph of an exemplary shot gather.

[0035] Figure 8is a graph of exemplary shot gathers and associated features.

[0036] Fig. 9 is a graph of an exemplary transformed shot gather.

[0037] Fig.10 is a graph of an exemplary transformed shot gather.

[0038] Fig.11 is a graph of an exemplary transformed shot gather shown in grayscale.

[0039] Fig.12 is a schematic diagram of an exemplary convolutional neural network.

[0040] Fig.13 is a table of configuration parameters for an exemplary convolutional neural network.

[0041] Fig.14 is a flow chart of an exemplary determination of the slowness of a unipolar waveform.

[0042] Fig.15 is a flow chart of an exemplary determination of the slowness of a unipolar waveform.

[0043] Fig.16 is a flow chart of an exemplary determination of dipole waveform slowness.

[0044] Fig.17 is a block diagram of an exemplary training of a neural network.

[0045] Fig.18 is a block diagram of exemplary training outputs of a neural network with unipolar data.

[0046] Fig.19 is a block diagram of exemplary training outputs of a neural network with dipole data.

[0047] Fig. 20 is a graph of a 2D image that may be used as an exemplary input to a neural network.

[0048] Fig.21 is a graph of a quadrupole inversion that may be used as an exemplary input to a neural network.

[0049] Fig. 22 is an exemplary dispersion curve graph.

[0050] Fig.23 is an example displayed output of a neural network.

[0051] Fig.24 is a block diagram of an exemplary architecture of a geological system.

[0052] Fig.25 is a block diagram of an exemplary architecture for a geological system. DETAILED DESCRIPTION

[0053] Various features and advantageous details are more fully described with reference to the non-limiting embodiments shown in the accompanying drawings and described in detail in the following description. However, it should be understood that the detailed description and specific examples are given by way of illustration only and not by way of limitation. Various substitutions, modifications, additions and / or rearrangements within the spirit and / or scope of the present disclosure will become apparent to those skilled in the art.

[0054] One technical problem solved by embodiments of the present invention relates to the limitations of sonic logging analysis as an expert driven process. Embodiments of the present invention may include a machine learning system, such as deep learning, to process, for example, sonic logging data to determine sonic slowness. Thus, in one example, the present disclosure and the appended claims provide a technical solution to the technical problem of determining sonic slowness of sonic logging data. In some embodiments, embodiments of the present invention may provide accurate results in a sufficiently short period of time to allow real-time logging while drilling decisions and processes to occur.

[0055] The underground strata may be natural or artificial. It should be understood that the underground geological region may be located under land or sea without loss of generality. The underground geological region may include the underground strata in which the wellbore is drilled. In addition to the area in close proximity to the wellbore, the underground geological region may also include any area that affects or may affect the wellbore or where a wellbore may be drilled.

[0056] Exemplary embodiments relate to slowness estimates for formations surrounding a wellbore. Slowness estimates may be used to identify gas entry points in a wellbore. Slowness estimates may also be used to estimate the porosity of a rock or another material forming a wellbore, characterize the induced or natural anisotropy or orientation of a rock, characterize the geomechanical properties of a rock, for example, to define the weight of a fluid used in drilling a wellbore. Slowness estimates may also be used to establish a time / depth relationship for a wellbore, thereby enabling the conversion of seismic data acquired for a wellbore into depth data, and to generate mapping of wellbore properties.

[0057] Figure 1 A wellsite system is shown in which embodiments disclosed herein may be employed. The wellsite may be an onshore wellsite or an offshore wellsite. In this exemplary system, a wellbore 11 is formed in a subterranean formation by rotary drilling. However, the examples described herein may also use directional drilling, as will be described below.

[0058] A drill string 12 may be suspended within the wellbore 11 and have a bottom hole assembly 100 including a drill bit 105 at its lower end. The surface system may include a platform and a derrick assembly 10 positioned above the wellbore 11. The assembly 10 may include a rotary table 16, a kelly 17, a hook 18, and a rotary swivel 19. The drill string 12 may be rotated by the rotary table 16. The rotary table 16 may engage the kelly 17 at the upper end of the drill string 12. The drill string 12 may be suspended from the hook 18, which is attached to a traveling block. The drill string 12 may be positioned through the kelly 17 and the rotary swivel 19, which allows the drill string 12 to rotate relative to the hook 18. A top drive system may be used to impart rotation to the drill string 12. In this example, the surface system also includes a drilling fluid or mud 26 stored in a pit 27 formed at the well site. A pump 29 delivers drilling fluid 26 to the interior of the drill string 12 via a port in the swivel 19, causing the drilling fluid 26 to flow downward through the drill string 12, as indicated by directional arrow 8. The drilling fluid 26 exits the drill string 12 via a port in the drill bit 105 and then circulates upward through the annulus region between the exterior of the drill string 12 and the wall of the wellbore 11, as indicated by directional arrow 9. In this manner, the drilling fluid 26 lubricates the drill bit 105 and carries formation cuttings to the surface as the drilling fluid returns to the pit 27 for recirculation.

[0059] Figure 1 The bottom hole assembly 100 of the illustrated example includes a logging while drilling (LWD) module 120 , a measurement while drilling (MWD) module 130 , a rotary steerable system and motor 150 , and a drill bit 105 .

[0060] The LWD module 120 may be housed in a special type of drill collar and may include one or more logging tools. In some examples, the bottom hole assembly 100 may include additional LWD modules and / or MWD modules. The LWD module 120 may include capabilities for measuring, processing, and storing information and for communicating with surface equipment. The LWD module 120 may include an acoustic wave measurement device.

[0061] The MWD module 130 may also be housed in the drill collar and may include one or more devices for measuring characteristics of the drill string 12 and / or the drill bit 105. The MWD module 130 may include a device for generating electricity for at least a portion of the bottom hole assembly 100. The device for generating electricity may include a mud turbine generator powered by a flow of drilling fluid. However, other power sources and / or battery systems may also be used. In this example, the MWD module 130 includes one or more of the following types of measurement devices: a bit pressure measurement device, a torque measurement device, a vibration measurement device, an impact measurement device, a stick-slip measurement device, a direction measurement device, and an inclination measurement device.

[0062] although Figure 1The components are shown and described as being implemented in a particular transmission type, but the examples disclosed herein are not limited to a particular transmission type, but instead may be implemented in combination with different transmission types, including, for example, coiled tubing, wireline drill pipe, and / or any other transmission type known in the industry.

[0063] Figure 2 A sonic logging while drilling tool that can be used to implement the LWD tool 120 or can be part of the LWD tool kit 120A is shown. An offshore drilling rig 210 with a sonic transmitting source or array 214 can be deployed near the water surface. In at least some embodiments, any other type of wellhead or downhole source or transmitter can be provided to transmit sonic signals. In some examples, a wellhead processor controls the firing of the transmitter 214.

[0064] The wellhead equipment may also include an acoustic receiver and recorder for capturing a reference signal close to the signal source (e.g., transmitter 214). The wellhead equipment may also include telemetry equipment for receiving MWD signals from downhole equipment. The telemetry equipment and recorder may be connected to the processor so that the uphole clock and the downhole clock can be used to synchronize the recording. The downhole LWD module 200 includes one or more acoustic receivers (e.g., 230 and 231) that are connected to the signal processor so that the signal detected by the receiver can be recorded in synchronization with the firing of the signal source.

[0065] In operation, the transmitter 214 transmits signals and / or waves that are received by one or more of the receivers 230, 231. The received signals may be recorded and / or registered to generate associated waveform data. The waveform data may be processed by the processors 232 and / or 234 to determine a slowness value as disclosed herein.

[0066] An exemplary goal of acoustic logging is to estimate the slowness of propagating acoustic waves in a borehole formation. An acoustic tool may include one or more transmitters and one or more receivers. For a certain tool position in the wellbore, the transmitter is fired. The released energy travels through different media (e.g., borehole fluid, borehole formation rock, etc.). A portion of this energy is captured by the receiver available in the acoustic tool. When the transmitter is fired, the receiver starts recording for a certain period of time. In this recording, there may be ambient noise. Along with the ambient noise, at a later time, the energy released by the captured transmitter is recorded. This recording may be called an acoustic waveform. Figure 3 An example of an acoustic waveform is depicted in For a given shot, multiple waveforms associated with multiple receivers may be recorded. Figure 4An example of a shot gather associated with an acoustic tool having thirteen receivers is depicted in . The set of waveforms recorded for a shot may be referred to as a shot gather. Analysis of these waveforms on a shot-by-shot basis helps identify the different onset arrivals of waves (e.g., P-waves, S-waves, Stoneley waves, leakage P-waves). The slowness of these waves may be determined for the rock available at the depth of interest.

[0067] Analysis of the recorded waveforms can provide the travel times of the acquired energy patterns (e.g., longitudinal waves, shear waves, Stoneley waves, leakage P waves) along with associated slowness values. Acoustic slowness can be determined based on arrival time picking (such as by first motion detection algorithms or fast arrival time and slowness estimation) or directly by algorithms such as slowness-time correlation (STC) algorithms, dispersed STC algorithms, dipole inversion, and quadrupole inversion.

[0068] The previously mentioned algorithms can be applied to wireline and LWD data, especially STC. Acoustic waveforms have been acquired in thousands of wells and processed to generate P-, S- and / or Stoneley wave slownesses. Processing such data requires intervention by expert users.

[0069] A workflow that can be implemented in real time will now be discussed. In one example, the workflow determines slowness from recorded borehole acoustic waveforms. In another example, the workflow determines travel time from the waveform. The obtained travel time can then be used to derive a high depth resolution slowness log. Preferably, the workflow includes machine learning techniques.

[0070] A collection of acoustic waveforms and associated processing results are collected and subdivided into shot gathers. The waveforms may undergo preparation steps such as noise reduction or information enhancement, after which the content of interest becomes easier to process and / or interpret. A convolutional neural network may be trained using both the acoustic waveforms and the associated results (e.g., slowness). A relationship may be established between the input acoustic waveforms and the results sought. After this process, a trained neural network may be created. The neural network may derive the slowness result (or travel time) directly from the input waveform. The generated neural network may be used for prediction purposes. When a set of waveforms is input to the network, the associated results (e.g., slowness, travel time) may be directly calculated.

[0071] The trained model can be used for data collected by both wireline and logging while drilling (LWD) acoustic tools. The workflow can be used for real-time embedded applications and post-processing applications. The generated workflow results can provide slowness determination. In addition, the output can be used to verify the results of other processing techniques (e.g., slowness-time correlation method).

[0072] In one example, the method includes a machine learning technique such as a convolutional neural network (CNN). A collection of input and output data can be used to train a CNN model.

[0073] Input data for model training may include one or more of recorded data, synthetic data, or a combination thereof, such as described below.

[0074] Recorded Data: The acoustic waveforms acquired by an acoustic tool in a well or logging environment. The waveforms may be pre-processed (e.g., noise attenuation). Outliers may be removed prior to incorporation into a database, or may be removed.

[0075] Synthetic data: Synthesize sound waveforms. Noise or transformations can be added to increase the amount of data in the database.

[0076] Combination of the above: The recorded waveforms with known slowness values ​​in a shot gather can be time shifted to obtain a new shot gather with a predefined synthetic slowness value. For a given recorded shot gather with known slowness values, a large set of pseudo-synthetic shot gathers can be generated to provide a range of possible slowness values. The moveout / true slowness (e.g., 65 μs / ft) can be corrected by time shifting the waveforms of the shot gathers separately to provide a new moveout (e.g., 80 μs / ft) associated with the synthetic slowness value. The process can be repeated to cover a range of slowness values ​​(e.g., 40 μs / ft to 240 μs / ft for P-wave slowness) and provide a large set of pseudo-synthetic waveforms (e.g., recorded waveforms with associated synthetic slowness values).

[0077] Time shifting of the waveform may include padding the waveform. Padding may include extrapolation, for example, adding time samples to each waveform signal to provide a common start time and end time. Waveform padding may include applying a linear prediction filter, padding the waveform with a constant value, and linear extrapolation of the average value.

[0078] Figure 5 An example of a recorded shot gather is shown. The recorded shot gather is plotted with dotted lines. In this example, the P-wave slowness is 106 μs / ft. The P-wave slowness is given by the slope of the dotted diagonal line. A time-shifted waveform is plotted with dashed lines to provide a pseudo-synthetic shot gather. In the time-shifted waveform, each receiver, except nearby receivers, is time-shifted to produce a waveform with a P-wave slowness of 160 μs / ft. The P-wave slowness is given by the slope of the dashed diagonal line. This process can also be applied to generate gathers centered on P-waves, S-waves, and Stoneley waves.

[0079] refer to Figure 6, at step 302, data is input to the model. Preferably, the type of data input is the type of data that the database was trained on. In the case where the database was trained on a different type of data set, a transformation may be applied to the input data to convert it to the type of data that the database was trained on. For example, a trained full-array monopolar CNN model may be used to process monopolar waveform data acquired by an acoustic wave tool made with a receiver array in full-array mode.

[0080] The input waveform may be represented as a collection of shot gathers. A shot gather may include a set of waveforms recorded by multiple receivers. If the number of waveforms exceeds the number of receivers for which the database was trained, the number of receivers may be reduced automatically or by the user during the data preparation step.

[0081] The input data may be prepared in a data preparation step 304. The data preparation step 304 may include reducing noise by applying a filter such as a classical bandpass filter, a linear prediction filter, an adaptive filter, an adaptive block thresholding filter, a median filter, or any other filter that may enhance the quality of the data of interest.

[0082] The data preparation step 304 may also include identifying and removing outliers in the input data.

[0083] refer to Figure 4 , the waveform recorded in the shot gather can be represented by a "swing" graph. Shot gathers can also be represented as grayscale images, such as Figure 7 Grayscale image shown. The x-axis of the image corresponds to the acquisition time, and the y-axis corresponds to the receiver index. The darkness of the pixel corresponds to the waveform amplitude at each receiver. An exemplary goal of acoustic wave data processing is to estimate the slowness of waves such as P-waves, S-waves, or Stoneley waves in a shot gather.

[0084] The intercepted waves of interest may be head waves. For a given tool position, the processing may treat a set of receivers illuminating a piece of rock as having constant slowness. The processing may look for coherent events in the shot gathers under a linear moveout constraint. The processed receivers may represent a subarray of the receivers available in the tool. These head waves may be observed when the acoustic tool is not positioned at a boundary between geological layers, e.g. Figure 8 The arrival of a longitudinal wave, a shear wave or a Stoneley wave can be represented by a linear time difference. A straight line captures the arrival of these waves. The slope of this line is used to calculate the slowness of the wave type under consideration.

[0085] If the tool is positioned at a geological boundary, lines are preferably determined for each geological layer illuminated by the aperture of the receiver. For example, in the case of two layers, preferably a total of two lines are used to capture the arrival of waves of interest (e.g., P-waves). In one example, the number of receivers used in processing can be reduced to reduce the likelihood of the tool being positioned at a geological boundary. In the case of a small receiver aperture, the time difference of the waves of interest may be linear.

[0086] Identifying head waves in such grayscale images (e.g., Figure 7 and Figure 8 ) can be difficult and requires a significant amount of input, such as from an expert operator, to edit or relabel the data to accurately determine the slowness.

[0087] In one example, machine learning such as a deep neural network is employed to improve the accuracy of image recognition. In a specific example, a convolutional neural network (CNN) is used. The CNN is described by way of example, and it should be understood that other techniques may also be used.

[0088] The goal of the CNN is to identify the lines that capture the arrival of waves of interest (e.g., P-waves). Referring back to Figure 6 , after preparing the data at step 304, the shot gather waveforms can be provided to a deep neural network such as a CNN at step 308. The input data with known slowness or slowness determined by an expert as discussed above can be used to train the CNN for the training CNN database.

[0089] In one example, a data transformation step 306 is performed after the data preparation step 304 and before providing the data to the neural network at step 308. An example of data transformation is applying an operator to the waveform to generate a new signal. Exemplary operators include deconvolution. Preferably, the newly generated signal provides content that is more easily processed by the neural network.

[0090] In a specific example, a short-time average / long-time average (STA / LTA) operator is applied to the input waveform. Using the STA / LTA image can improve the reliability of network processing by reducing the uncertainty associated with time difference identification when processing acoustic waveform images. Fig. 9 is an example of applying the STA / LTA method to generate a CNN input image from the shot gather of monopole waveforms. The exemplary equation for the STA / LTA operator is:

[0091]

[0092] where g represents the Hilbert envelope of the waveform under consideration, 0 < sw ≤ lw, and is a small constant. Then, the transformed data can be provided to the neural network at step 308.

[0093] For waveforms with strong dispersive content, such as dipole waveforms, the transformation may include a phase shift in the frequency domain of the waveform pair.In a shot gather, the phase shift between a receiver's waveform and the waveforms of nearby receivers may be calculated. Fig.10 An exemplary result of determining the phase shift of a dipole waveform recorded by a wireline sonic tool with thirteen inline receivers is shown. In this case, twelve curves were generated in the frequency domain. Fig.10 In the example, the phase shift curves have not been unfolded, but it should be understood that these curves can also be unfolded. The transformed curves can also be displayed as grayscale images, such as Fig.11 The process then proceeds to the neural network at step 308.

[0094] Referring to the neural network processing step 308, in one example, a CNN network is used to process the (transformed) shot waveform input data to output slowness or travel time data. Fig.12 An exemplary schematic diagram of a CNN is shown and Fig.13 Shows Fig.12 . It should be understood that other types of neural networks or architectures may also be used. It should also be understood that the neural network processing step 308 may operate using raw data such as that which has not been transformed in the data transformation step 306.

[0095] As discussed above, a multi-layer CNN is trained using a set of waveforms and associated slownesses, e.g. Fig.12 Multi-layer CNN.

[0096] In the example of input data having a monopole waveform and associated longitudinal wave slowness measurements, the input waveform may be processed for noise reduction.Then, the STA / LTA operator may be applied to the input data.

[0097] refer to Fig.14 , at step 402, a waveform of a shot gather having a monopole waveform is input. In some embodiments, at step 404, outliers are detected and removed from the input data. In some embodiments, at step 406, data quality improvement (e.g., denoising) is performed. In some embodiments, at step 408, the data is transformed, such as by a STA / LTA operator. At step 410, a CNN is applied to the (transformed) input data for slowness determination. At step 412, the determined slowness is output.

[0098] refer to Fig.15, at step 452, a waveform of a shot gather having a monopole waveform is input. In some embodiments, at step 454, outliers are detected and removed from the input data. In some embodiments, at step 456, data quality improvement (e.g., denoising) is performed. In some embodiments, at step 458, the data is transformed, such as by STA / LTA operators. At step 460, a CNN is applied to the (transformed) input data for travel time determination. At step 462, the determined travel time is output. At step 464, the determined travel time is converted to a high depth resolution slowness and output.

[0099] In the example of input data having a dipole waveform and associated shear slowness measurements, the input dipole waveform can be used to calculate the phase shift between the waveform of the corresponding receiver in the shot gather and the waveform of a nearby offset receiver.

[0100] refer to Fig.16 , at step 502, a waveform of a shot gather having a dipole waveform is input. In some embodiments, at step 504, outliers are detected and removed from the input data. In some embodiments, at step 506, data quality improvement (e.g., denoising) is performed. In some embodiments, at step 408, the data is transformed, such as by a pairwise phase shift operator. At step 410, a CNN is applied to the (transformed) input data for slowness determination. At step 412, the determined slowness is output.

[0101] refer to Fig.17 , during the training process of the CNN, a set of reference inputs 552 (e.g., STA / LTA images for a given well at a given depth and given bit size) and associated outputs 554 (e.g., P-wave slowness) are provided. By providing such a large set of training input / output data, the CNN model will update its numerical parameters / weights at 556. In this process, the model calculates the outputs to match the provided reference values. The model updates to reduce the error between its outputs and the provided reference output data.

[0102] The CNN can be trained for more than one output, for example, providing an input waveform along with P-wave slowness values, S-wave slowness values, and Stoneley slowness values ​​for an input image. The neural network training phase provides a trained CNN model that can be exported and implemented in an embedded software application.

[0103] Fig.18The results of training a CNN model uphole are shown. The input monopole waveform is denoised and then processed by the STA / LTA operators. In this example, the processing is performed on the full array, where all waveforms in the shot gather are considered. It should be understood that less than a full array can also be used. The right track shows a comparison between the expected P-wave slowness record (dashed line) and the P-wave slowness record (dashed line) output by the CNN training. The left track shows the gamma ray record 602, the drill bit size (dashed line 604) and the hole diameter (606). The middle track 608 shows the waveform recorded by the first receiver at the well depth interval. It should be understood that the CNN can also be trained using the input waveform and the associated travel times of the P-wave, S-wave, Stoneley waves and other data.

[0104] Fig.19 The results of training a CNN model uphole are shown. The input dipole waveform is transformed by a phase shift operator. In this example, the processing is done in a set of five consecutive receivers. It should be understood that the full array can also be used. The right track shows a comparison between the expected P-wave slowness record (dashed line) and the P-wave slowness record (dashed line) output by the CNN training. The left track shows the gamma ray record 652, the drill bit size (dashed line 654) and the hole diameter (656). The middle track 658 shows the waveform recorded by the first receiver at the well depth interval. Both shear wave slowness records are generated using a multi-shot technique.

[0105] Although in the above discussion, the input waveform has been subdivided into shot gathers to help generate the input image for the CNN, it should be understood that other ways of grouping the input data may also be used. For example, a waveform recorded by one receiver may be used to generate a CNN input image. Travel time and / or slowness values ​​may also be used to train the neural network. In another example, the input data may include a 3D image of a waveform recorded by an acoustic tool at a well depth interval. The 3D image may include a collection of waveforms recorded by different receivers in the acoustic tool. For corresponding tool positions in the well, 2D images (e.g., shot gathers) are recorded. When the acoustic tool is moved in the wellbore to cover the well depth interval, a set of 2D images is generated, thereby producing a 3D image. Along the 3D image, the travel time and / or slowness values ​​can be used to train the neural network.

[0106] In various embodiments, the input provided to the artificial intelligence may include 2D or 3D images generated by transformations such as deconvolution, STA / LTA, and phase shift operators. The input may also include images generated by STC processing (e.g., Fig. 20 , which depicts a 2D coherence image) or such as quadrupole inversion (e.g. Fig.21 2D images generated by alternative processing techniques (as shown).

[0107] It should be understood that combinations of the described inputs may also be used. In some embodiments with several types of inputs such as a combination of different input images (e.g., STC single depth images, STA / LTA images, dispersion analysis images), a CNN network may be used for each type of input. The outputs of the CNN networks may be combined to generate a final output.

[0108] Fig. 22 An example of dispersion analysis is shown in . This image is an example of data that can be used to train a neural network. The image includes data content of an acoustic tool at a depth (e.g., 1584.96m). In this example, the transformed monopole waveform is represented by circle 752. The monopole P-wave slowness can be determined based on the data of circle 752 (the transformed monopole waveform information). The wire rope dipole waveform is represented by square 754 and diamond 756. The shear wave slowness and / or mud slowness can be determined based on the data of square 754 and diamond 756 (the wire rope dipole waveform information). The data of the cyan circle can also be used for the determination of Stoneley slowness and / or the determination of mud slowness. Various aspects of the dispersion data can be used for training. For example, one slowness value (e.g., dipole or quadrupole shear wave slowness) can be selected, or several or all slowness values ​​(e.g., P-wave slowness along dipole / quadrupole shear wave slowness, along Stoneley slowness, and mud slowness) can be selected.

[0109] Re-reference Figure 6 , providing output from the neural network at step 310. The output may be in the form of an interface to other software for further processing of the output. The output may also include displaying the neural network output. Fig.21 is an example format for displaying output. Fig.23 In the figure, the output of the P-wave slowness determination provided by the CNN trained model on a monopolar waveform is shown. The trained model was applied to a monopolar waveform of a well that was not used in the training phase. The determined P-wave slowness is shown in the last track (dashed line). In order to have a reference result, in the same track in purple, the result of the P-wave slowness processing provided by the expert user when running the STC algorithm is shown as a dotted line. The left track shows the gamma ray recordings, drill bit size and hole diameter. The second track shows the waveform recorded by the first receiver on the well depth interval. The third track shows the STC intermediate processing results. The results shown were obtained for the full array processing method and the P-wave slowness recordings predicted by the CNN.

[0110] It should be understood that the present disclosure is not limited to the determination of acoustic slowness. The described methods can also be used to determine any attribute in an acoustic waveform, but are not limited thereto. For example, processing an acoustic waveform can be used to determine attributes including travel time of a mode of interest (e.g., P-wave, S-wave), fast shear wave orientation, stiffness tensor parameters, etc.

[0111] Fig.24 An exemplary geological system 1000 is depicted in accordance with some embodiments. System 1000 may be a single system 1101A or an arrangement of distributed systems. System 1101A includes one or more geoscience analysis modules 1102 configured to perform various tasks in accordance with some embodiments, such as one or more methods disclosed herein. To perform these various tasks, geoscience analysis modules 1102 execute independently or in conjunction with one or more processors 1104, which are connected to one or more storage media 1106A. Processor 1104 is also connected to network interface 1108 to allow system 1101A to communicate with one or more additional systems and / or systems such as 1101B, 1101C, and / or 1101D via data network 1110 (note that systems 1101B, 1101C, and / or 1101D may or may not share the same architecture as system 1101A and may be located in different physical locations, for example, systems 1101A and 1101B may be located on a ship sailing at sea or at a well site while communicating with one or more data centers located on shore, other ships, and / or one or more systems (such as 1101C and / or 1101D) located in different countries on different continents). Note that data network 1110 may be a private network, which may use portions of a public network, which may include remote storage and / or application processing capabilities (e.g., cloud computing).

[0112] A processor may include a microprocessor, a microcontroller, a processor module or subsystem, a programmable integrated circuit, a programmable gate array, or another control or computing device.

[0113] Storage medium 1106 may be implemented as one or more computer-readable or machine-readable storage media. Figure 1In the exemplary embodiment of , storage medium 1106 is depicted as being located within computer system 1101A, but in some embodiments, storage medium 1106 may be distributed within and / or across multiple internal and / or external packages of computing system 1101A and / or additional computing systems. Storage medium 1106 may include one or more different forms of memory, including: semiconductor memory devices, such as dynamic or static random access memory (DRAM or SRAM), erasable and programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), and flash memory; magnetic disks, such as fixed disks, floppy disks, and removable disks; other magnetic media, including tape; optical media, such as compact disks (CDs) or digital video disks (DVDs), BluRays, or any other type of optical media; or other types of storage devices. It is noted that the instructions discussed above may be provided on one computer-readable or machine-readable storage medium, or alternatively, may be provided on multiple computer-readable or machine-readable storage media distributed in a large system with potentially multiple nodes and / or non-transitory storage devices. Such computer-readable or machine-readable storage media are considered part of an article (or product). An article or product may refer to any manufactured single component or multiple components. One or more storage media may be located in a machine running machine-readable instructions, or at a remote site where machine-readable instructions can be downloaded over a network for execution.

[0114] It should be understood that system 1101A is only one example and that system 1101A may have more or fewer components than shown, may be combined Figure 1 Additional components not depicted in the exemplary embodiment of the present invention, and / or system 1101A may have Figure 1 Different configurations or arrangements of the depicted components. Figure 1 The various components shown may be implemented in hardware, software, or a combination of both hardware and software, including one or more signal processing and / or application specific integrated circuits.

[0115] It should also be understood that the system 1000 may include user input / output peripherals such as a keyboard, a mouse, a touch screen, a display, etc. The system 1000 may include a desktop workstation, a laptop computer, a tablet computer, a smart phone, a server computer, etc.

[0116] In addition, the steps in the processing method described herein can be implemented by running one or more functional modules in an information processing device, such as a general-purpose processor or a dedicated chip, such as an ASIC, FPGA, PLD or other suitable device. These modules, combinations of these modules and / or their combination with hardware are all included in the scope of the present disclosure.

[0117] The data collection system 1130 may include systems, sensors, user interface terminals, etc., configured to receive data corresponding to records collected at a petroleum service facility, such as an exploration unit, an oil rig, an oil or gas production system, etc. The collected data may include acoustic data, such as the sonic logging data discussed above, as well as other sensor data, employee recorded data, computer generated data, etc.

[0118] refer to Fig.25 , the multi-client system 1200 may include a centralized service system 1202, which may be implemented on a cloud service system. In such an embodiment, the centralized service system 1202 may include one or more cloud data storage systems 1201 and one or more computing nodes 1203. If it is such an embodiment, the system 1200 may include multiple client networks, including a first client network 1206, a second client network 1208, and a third client network 1210. Each client network 1206-1210 may communicate with the centralized service system 1202 via a system communication network 1204, which may be the Internet or a dedicated WAN connection.

[0119] In such embodiments, each of the client networks 1206-1210 may include Fig.24 1A-D and data acquisition system 1130. Such devices may also be connected via an internal network 1110. In one embodiment, first client network 1206 may be operated by a first customer of a data analysis system provider. In another embodiment, second customer network 1208 and third customer network 1210 may both be operated by a second customer, but at separate geographic locations. One of ordinary skill will recognize the various client / customer relationships that may be established.

[0120] In such an embodiment, each of the client networks 1206-1210 can communicate with the centralized service system 1202 for data storage and to implement certain centralized data processing and analysis processes. Advantageously, the centralized service system 1202 can be configured for large-scale data storage and data processing.

[0121] Embodiments of the present invention have been described in a manner that is particularly beneficial to geological systems and services. Various aspects and ordered combinations provide unique and improved solutions to provide accurate determination of acoustic slowness, in some cases automatically, which in some embodiments can facilitate actual real-time decision-making processes based on the determined acoustic slowness. Although these benefits to geological systems and services have been emphasized, it should be understood that additional fields that can benefit from embodiments of the present invention include archaeology, marine biology, etc. Although the embodiments described herein are useful in any of these geological fields, embodiments of the present invention are described primarily with reference to petroleum services.

[0122] It should also be understood that the methods described cannot be performed mentally. For example, neural networks have image processing capabilities that cannot be achieved on any reasonable time scale. In addition, machine learning techniques are performed by specially programmed machines, for example.

[0123] Although the present invention is described herein with reference to specific embodiments, various modifications and changes may be made without departing from the scope of the present disclosure. Therefore, the specification and drawings are to be regarded as illustrative rather than restrictive, and such modifications are intended to be included within the scope of the present disclosure. Any benefits, advantages, or solutions to problems described herein with respect to specific embodiments are not intended to be construed as key, required, or essential features or elements of any or all the claims.

[0124] Unless otherwise specified, terms such as "first" and "second" are used to arbitrarily distinguish the elements described by such terms. Therefore, these terms are not necessarily intended to indicate the time or other priority of such elements. The term "connected" or "operably connected" is defined as connected, although not necessarily directly and not necessarily mechanically. The term "a / an" is defined as one or more, unless otherwise specified. The terms "comprise" (and any form of include, such as "comprise / comprising"), "have" (and any form of have, such as "has / having"), and "contain" (and any form of contain, such as "contains / containing") are open-ended linking verbs. Therefore, a system, device or equipment that "comprises / includes", "has", or "contains" one or more elements has such one or more elements, but is not limited to having only such one or more elements. Similarly, a method or process that “comprises / includes,” “has,” or “contains” one or more operations possesses those one or more operations, but is not limited to possessing only those one or more operations.

Claims

1. A method for determining the slowness of an acoustic wave, wherein include: accessing sonic logging data including sonic waveforms associated with a plurality of shot gathers; applying a transform operator to the sonic logging data to provide a transformed sonic image, the transform operator comprising at least one of a short-time average, a long-time average operator, a phase shift operator, and a deconvolution operator; performing a machine learning process using the transformed sonic image to determine sonic slowness associated with the sonic logging data, the machine learning process comprising a convolutional neural network that has been trained using input data having known slowness or slowness determined by an expert, the input data comprising one or more of recorded sonic waveforms, synthesized sonic waveforms, or a combination of recorded and synthesized sonic waveforms, and wherein a transformation operation is selected to convert the sonic logging data into a type of data used to train the convolutional neural network; and The acoustic wave slowness is provided as an output.

2. The method of claim 1, wherein the sonic logging data comprises shot gathers of monopole waveforms, and the transform operator comprises the short-time average long-time average operator.

3. The method of claim 1, wherein the sonic logging data comprises shot gathers of dipole waveforms and the transform operator comprises the phase shift operator.

4. The method of claim 1, wherein the sonic logging data comprises a two-dimensional image and the transform comprises the deconvolution operator.

5. The method of claim 1, wherein the sonic logging data comprises a three-dimensional image and the transform comprises the deconvolution operator.

6. The method of claim 1, wherein the determined acoustic wave slowness comprises at least one of a slowness of a longitudinal wave, a slowness of a shear wave, a slowness of a Stoneley wave, a slowness of a leakage P wave, a slowness of a Rollie wave, and a slowness of a pseudo-Rollie wave.

7. The method of claim 1, wherein said providing comprises displaying said acoustic wave slowness.

8. The method of claim 7, wherein said providing comprises displaying said acoustic wave slowness as a function of depth.

9. A device for determining the slowness of sound waves, wherein include: an interface configured to obtain sonic logging data including sonic waveforms associated with a plurality of shot gathers; a memory configured to store computer-executable instructions; and a processor operably coupled to the interface and the memory, the processor configured to execute the instructions and cause the device to: applying a transform operator to the sonic logging data to provide a transformed sonic image, the transform operator comprising at least one of a short-time average, a long-time average operator, a phase shift operator, and a deconvolution operator; performing a machine learning process using the transformed sonic image to determine sonic slowness associated with the sonic logging data, the machine learning process comprising a convolutional neural network that has been trained using input data having known slowness or slowness determined by an expert, the input data comprising one or more of recorded sonic waveforms, synthesized sonic waveforms, or a combination of recorded and synthesized sonic waveforms, and wherein a transformation operation is selected to convert the sonic logging data into a type of data used to train the convolutional neural network; and The acoustic wave slowness is provided as an output.

10. The apparatus of claim 9, wherein the sonic logging data comprises shot gathers of monopole waveforms and the transform operator comprises the short-time average long-time average operator.

11. The apparatus of claim 9, wherein the sonic logging data comprises shot gathers of dipole waveforms and the transform operator comprises the phase shift operator.

12. The apparatus of claim 9, wherein the sonic logging data comprises a two-dimensional image and the transform comprises the deconvolution operator.

13. The apparatus of claim 9, wherein the sonic logging data comprises a three-dimensional image and the transform comprises the deconvolution operator.

14. The apparatus of claim 9, wherein the determined acoustic wave slowness comprises at least one of a slowness of a longitudinal wave, a slowness of a shear wave, and a slowness of a Stoneley wave.

15. The device of claim 9, wherein the device comprises a display configured to display the slowness of the sound waves.

16. The apparatus of claim 15, wherein the display is configured to display the acoustic wave slowness as a function of depth.

17. A method for training a machine learning process, wherein include: accessing a plurality of sonic logging data including sonic waveforms and sonic slownesses associated with a plurality of shot gathers; as well as Using a convolutional neural network, a database of the machine learning process is trained using the sonic logging data as a training input and using the sonic slowness as a training output, wherein the sonic logging data has a known slowness or a slowness determined by an expert, and the training input includes one or more of a recorded sonic waveform, a synthesized sonic waveform, or a combination of a recorded sonic waveform and a synthesized sonic waveform.

18. The method of claim 17, further comprising applying a transform operator to the sonic logging data prior to the training step, the transform operator comprising at least one of a short-time average long-time average operator, a phase shift operator, and a deconvolution operator.

Images