Method and system for machine learning-based processing of underground acoustic signals

Abstract

A method for training a machine learning model to identify the arrival time of an acoustic signal in an underground receiver is disclosed. The method comprises acquiring acoustic data, where the acoustic data represents an acoustic signal recorded by a receiver of a probe located within an underground ground volume, and the acoustic signal is formed by an active signal transmitter. The method further comprises acquiring a target arrival time of the acoustic signal in the receiver and feeding the acoustic data to a machine learning model to acquire an output of the machine learning model, where the output of the machine learning model includes a determined arrival time of the acoustic signal in the receiver. The method further comprises adjusting the parameters of the machine learning model to reduce the error between the determined arrival time and the target arrival time. The present invention further relates to uncovering insights from geographic data to improve sustainability and environmental development, and together we create a safer and more livable world.

Description

[Technical Field]

[0001] This disclosure relates to a method and system for analyzing a target area or volume below the Earth's surface. More specifically, this disclosure relates to a method and system for determining the arrival time of an acoustic signal at one or more underground receivers located within a ground volume using machine learning. The arrival time can be used to determine the velocity of the acoustic signal within the ground volume, from which one or more geographic data parameters associated with the ground volume relevant to infrastructure planning and engineering can be determined. The invention further relates to unleashing insights from geographic data to improve sustainability and environmental development, so that together we can create a safer and more livable world.

[0002] background There is a general and ongoing need for systems and methods to determine subsurface ground characteristics through the acquisition and analysis of geological data (also referred to as geographic data or geodata). In particular, there is a need for systems and methods that can be used to model the characteristics of a target volume below the surface and provide useful information for infrastructure planning. Determining subsurface ground characteristics in the early planning stages of a construction project reduces uncertainty during project location, foundation design, and construction phases. This, in turn, reduces construction delays, overspending, and unnecessary use of material resources (e.g., concrete). Furthermore, a complete understanding of subsurface characteristics enables the proper location and installation of infrastructure projects, thereby improving safety.

[0003] Infrastructure planners are particularly interested in the behavior of acoustic signals (also called audio signals, acoustic waves, seismic waves, sound signals, or sound waves) as they propagate through a target volume within the ground. The behavior of these signals can provide insights into the composition and structure of the ground. In the context of such acoustic signals, two main wave types that are often investigated are compression waves and shear waves. Compression waves, more commonly referred to as P-waves (primary waves), cause particles within a volume to vibrate back and forth in a compressive motion parallel to the direction of wave propagation. Shear waves, more commonly referred to as S-waves (secondary waves), cause particles within a volume to vibrate back and forth in a transverse motion perpendicular to the direction of wave movement. The propagation speed of these waves depends on the material through which the waves travel. In particular, the propagation speed of waves through the ground is an important indicator of certain subsurface properties, such as the elastic or shear properties of soil / rock. Therefore, by measuring the wave velocity and working in the reverse direction, engineers and infrastructure planners can determine the material properties of the volume through which the signal has traveled.

[0004] In some cases, the signal under investigation is a passive audio signal formed by uncontrolled ambient background noise. That is, the signal receiver is configured to detect the wave field present due to the background noise. The background noise can be natural (e.g., from overlapping ocean waves, wind, and other naturally occurring vibrations) or cultural (e.g., from human activities including traffic and machinery). In other examples, the signal under investigation may be created in a more deliberate and controlled manner, such as by creating a controlled signal using an active acoustic signal source, e.g., a hammer drop or explosion. This disclosure focuses on the latter approach, namely the processing and analysis of acoustic signals caused by an active signal source or transmitter.

[0005] When investigating acoustic signals and their behavior within underground soil volumes, both the downhole and crosshole methods can be used for field measurements. In both of these methods, a receiver placed in a borehole measures signals received from an active acoustic signal source located elsewhere. In the downhole method, the signal source may be located in the same borehole as the receiver, or alternatively, on the surface. In the crosshole method, the source is located in a first borehole and the receiver is located in a second borehole. In both the downhole and crosshole methods, the propagation of the received waves is investigated to infer the properties of the material through which the signal from the signal source has traveled. The methods of this disclosure relate to both the downhole and crosshole methods, but the downhole method forms the main focus of the implementation described.

[0006] As will be described in more detail below, the methods and systems of this disclosure address the need for an improved mechanism for determining the time of signal arrival in an underground receiver. As mentioned above, accurate determination of the time of signal arrival in the receiver is crucial for accurately determining the speed at which the signal traveled through the target subsurface volume. Furthermore, accurate determination of the signal speed is crucial for accurate determination of the geographic data parameters of the subsurface volume. Existing methods and systems for determining the time of signal arrival in a receiver have many shortcomings. Existing methods lack precision and often require additional quality checks. Because existing methods lack precision, the measurement results are only for reference or require independent verification if possible. In addition, existing methods are generally slow, often leading to delays in geographic data deliverables. The use of additional time and resources leads to inefficient operation, lack of reliable data, and increased subsurface uncertainty, which can lead to over-parameterization of designs. Consequently, this lack of reliability and reproducibility in determining geographic data parameters can also increase the environmental burden of design work that relies on these geographic data parameters. This disclosure provides methods and systems that address these issues and enable improvements in the determination of the above-mentioned geographic data parameters, thereby improving the quality and processing efficiency of the geographic data parameters, and subsequently improving project efficiency and sustainability.

[0007] overview A first aspect of this disclosure provides a computer implementation for training a machine learning model to identify the time of arrival of an acoustic signal in an underground receiver. In this context, the term “underground” simply means that the receiver is located below the surface of the earth and is configured to detect acoustic signals in such an environment. In this context, “receiver” means any suitable acoustic receiver configured to record acoustic signals (also known as audio signals) below the surface of the earth. Such a receiver may be a simple accelerometer, a geophone, or a more sophisticated underground receiver. “Time of arrival” simply means the time when the acoustic signal in question first arrives or is first detected by the receiver.

[0008] As described above, the arrival time of acoustic signals can be used to determine useful engineering geographic data parameters that provide insights into the characteristics of the underground ground volume in which the receiver is located. By training a machine learning model to identify the arrival time, these important parameters can be determined more efficiently and accurately with reduced bias and improved reproducibility. As a result, an improved mechanism is provided for determining geographic data parameters associated with a given ground volume.

[0009] This method involves acquiring first acoustic data, wherein the first acoustic data represents an acoustic signal recorded by a first receiver of a probe located within the underground ground volume, and the acoustic signal is formed by an active signal transmitter. In other words, an acoustic signal formed by an active signal transmitter, such as a hammer drop or a controlled explosion, is detected and recorded by the first underground receiver, and the recorded data is stored (locally or remotely in a database) as first acoustic data.

[0010] The method further includes obtaining a target arrival time of an acoustic signal at a first receiver. This target arrival time is a known or estimated arrival time of the acoustic signal determined by a user or a different machine learning model. Most typically, obtaining a target arrival time of an acoustic signal involves receiving an indication of the arrival time (e.g., a label), typically via user input or from a database of pre-labeled signal data. In other words, the method may include receiving a labeled version of the first acoustic data, where the label indicates the arrival time of the acoustic signal at the first receiver. The machine learning model is then trained to replicate the target arrival time (i.e., the label), in other words, the labeled arrival time can be used as validated data in the training method. Thus, the method includes feeding the first acoustic data to a machine learning model to obtain the output of the machine learning model, where the output of the machine learning model includes the determined arrival time of the acoustic signal at the first receiver, and adjusting the parameters of the machine learning model to reduce the error between the determined arrival time and the target arrival time. In other words, the arrival time estimated by the machine learning model is compared to, for example, the target arrival time supplied via labeled acoustic signal data. The model's parameters are then adjusted to minimize the difference between the estimated arrival time and the target (labeled, validated data) arrival time, and as a result, through successive iterations, the model is trained to more accurately predict the arrival time of the acoustic signal at the receiver.

[0011] The disclosed method provides a significant improvement in determining the arrival time of acoustic signals at a receiver. Currently, this process has to be performed manually by human experts. The process is not only time-consuming but also requires advanced skills and experience, as will be more fully explained below. Since the process currently requires a skilled human operator, there is a significant limitation on the number of signals that can be processed within a given time. The operator has to undergo extensive training and have substantial experience to be able to accurately estimate the arrival time. Even a well-trained and experienced human user still tends to make biased and inconsistent classifications, which undermines the reliability and consistency of the determined signal arrival times. These drawbacks are addressed by providing a machine learning model trained to determine the arrival time. In particular, the determination of the signal arrival time can be performed more quickly and accurately than by a human user in an unbiased and consistent manner. Thus, the disclosed method provides a more reliable and consistent estimate of the signal arrival time, which means that it can be relied upon with a higher degree of confidence for the subsequently derived signal velocity and related geographical data parameters. This, in turn, can improve the planning and safety of infrastructure and construction projects.

[0012] This method may advantageously involve analyzing two or more audio datasets. For example, the method may include first and second audio data representing two signal recordings performed at two different recording locations. As will be described in more detail below, the two signal recordings can be performed by two separate receivers or by a single receiver sequentially positioned at different locations within a ground volume. By analyzing the signals at the two receiving locations, the time it takes for the signal to traverse between the two receiving locations can be determined, as will be described in more detail below. This provides an accurate and reliable method for measuring the average speed of the signal over that distance between the two recording locations (also referred to as the "interval speed"). This, in turn, allows for the accurate and reliable determination of various geographic data parameters of the ground volume surrounding the two receiver receiving locations.

[0013] One way to supply first audio data and second audio data for analysis is to estimate the arrival times of the same single signal at two receivers. As described above, this provides a reliable and accurate mechanism for analyzing the behavior (particularly the velocity) of signals within the ground volume surrounding the receivers. Thus, in this embodiment, the method can include obtaining second acoustic data, where the second acoustic data represents acoustic signals recorded by a second receiver provided within a subsurface ground volume. In other words, the second receiver detects the same signal as the first receiver at different times (i.e., generally any receiver further from the transmitter will detect the signal sequentially). There are cases where the second receiver is part of the same probe as the first receiver, or alternatively where the two probes are independent. Similar to the signal data recorded at the first receiver, the signal data recorded at the second receiver can be used within a machine learning context to train a machine learning model to estimate signal arrival times. In particular, the method can include obtaining a target arrival time of an acoustic signal at the second receiver and supplying the second acoustic data to the machine learning model to obtain an output of the machine learning model, where the output of the machine learning model includes a determined arrival time of the acoustic signal at the second receiver, and adjusting parameters of the machine learning model to reduce an error between the determined arrival time and the target arrival time for the second receiver. In other words, the second acoustic signal data is treated similarly to the first acoustic signal data and contributes to the training of the machine learning model. Thus, the considerations and available implementations described above for the first acoustic data also apply to the second acoustic data.

[0014] As already mentioned, another method for acquiring the first and second acoustic data is to use a single receiver and measure the arrival of multiple signals using the receiver at two or more locations. This provides an alternative mechanism for acquiring multiple sets of acoustic data without requiring multiple receivers. This approach is particularly suitable for conducting a Seismic Cone Penetration Test (SCPT), in which a signal transmitter is located on the ground surface and a receiver in a penetration probe inserted into the target soil volume detects the signal. Thus, in such embodiments where a single receiver is present, the method may further include acquiring second acoustic data, the second acoustic data representing second acoustic signals recorded by the first receiver when the first receiver is located at different locations within the subsurface soil volume. For example, the first receiver can be placed at different depths. Similar to the implementations described above, the second acoustic data (recorded in this example by the first receiver at a different location, rather than by the second receiver) can be used in a machine learning context to train a machine learning model to estimate the signal arrival time. In particular, this method may include obtaining the target arrival time of a second acoustic signal at a first receiver, supplying the second acoustic data to a machine learning model to obtain the output of the machine learning model, wherein the output of the machine learning model includes the determined arrival time of the second acoustic signal at the first receiver, and adjusting the parameters of the machine learning model to reduce the error between the determined arrival time and the target arrival time for the second acoustic signal at the first receiver. In other words, the second acoustic signal data also contributes to training the machine learning model, as in the previous implementation. Therefore, the considerations and available implementations described above for the first and second acoustic data also apply to the second acoustic data in such a single-receiver implementation.

[0015] Advantageously, in implementations where the first and second audio data are recorded (by two receivers or by a single receiver at two locations), the method may include feeding the first and second audio data as a combined input to a machine learning model. In other words, the machine learning model is trained simultaneously on both the first and second audio data. As a result, the machine learning model is trained to handle the first and second audio data similarly, or more precisely, to label the first and second data similarly. The combined input of the first and second audio data can be thought of as a two-channel input, similar to how a machine learning model might take three input channels (red, green, and blue) when processing an RGB image. The inventors determined that this strategy of feeding the first and second audio data as a combined input improves the training process and results in a more reliable and accurate determination of the signal arrival time, as the model effectively learns to recognize patterns and trends common to both signal recordings. Since both the first and second acoustic data represent either the same single signal formed at an active signal source (in the case of two receivers) or two similar signals recorded by the same receiver (in the case of a single receiver), it is beneficial to train a machine learning model to treat the resulting acoustic data as related. This is achieved by supplying the acoustic data as combined inputs. In one implementation, more than two receivers, e.g., three or more receivers, can be used.

[0016] As described above, the investigation of acoustic signals traveling through underground media generally involves the detection and classification of compression-related components (commonly referred to as P-waves) and shear-related components (commonly referred to as S-waves) of the signal. Therefore, in the method of this disclosure, the first acoustic data and / or the second acoustic data may include data representing the compression-related signal component or the shear-related signal component.

[0017] The shear-related component relates to the characterization of shear waves, which can be true shear (bulk) waves or bent surface waves that can determine the properties of the shear wave. For example, if an active signal transmitter is placed on the surface of the ground and a receiver measures the transmitted value of the signal from the surface to the subsurface volume, the shear-related component may be a true shear body wave. Alternatively, if the signal transmitter is placed in the same borehole as the receiver, as part of the same probe as an optional means, the detected shear-related component represents a bent surface wave traveling along the inner surface of the borehole. These bent waves are not true shear waves. However, they can be used to derive approximations of shear-related properties such as shear wave velocity. Therefore, these bent waves can still be considered shear-related signal components.

[0018] In an implementation in which the first and / or second acoustic data include data representing shear-related components, the first and / or second acoustic data may advantageously include data representing two shear-related components that are polarized in opposite directions, for example, two bending waves. In other words, an acoustic signal transmitter can form an acoustic signal having two oppositely polarized shear components, and from these components, form two oppositely polarized bending waves in a borehole. These oppositely polarized bending waves are then detectable by a receiver. Supplying two oppositely polarized signal components in this way makes it easier to determine the time of arrival of the signal at the receiver, as will be explained in more detail below. When the signal includes data representing two oppositely polarized components, these components can be supplied to a machine learning model as a combined input in the same manner as described above.

[0019] A machine learning model can determine the arrival time of an acoustic signal in any suitable way. In one favorable implementation, for each input acoustic data (i.e., the first acoustic data described above and, as an optional means, a second acoustic data), the machine learning model is configured to determine the arrival time of the acoustic signal by separating the acoustic data into a noise portion and a signal portion, determining the transition point between the noise portion and the signal portion consisting of the target signal, other signals, and background noise, and outputting the transition point as the estimated arrival time of the acoustic signal. Separating the acoustic data into a noise portion and a signal portion can be done in any suitable way, and the specific implementation chosen is determined by the type of machine learning model used. In one implementation, the model is a segmentation model that divides the signal into a signal portion and a noise portion. In another implementation, the model is an image recognition model that learns to identify the transition of signal traces between signal and noise through an image recognition process. Thus, as can be understood, the model can be an image recognition model or a signal processing model. Other available approaches include classification strategies in which the model classifies whether a sample in the trace is signal or noise, or alternatively, estimation methods in which the model determines where transition points are located. In some cases, the machine learning model directly outputs the velocity or section velocity of the acoustic signal. In one implementation, the model is a combination of a segmentation model and a classification model.

[0020] In another aspect of this disclosure, a method is provided for obtaining an estimate of the arrival time of an acoustic signal at an underground receiver using a machine learning model trained according to one of the methods described herein. Thus, this method can be considered to represent the use phase of the trained machine learning model. The method includes obtaining input data including first acoustic data, wherein the first acoustic data represents an acoustic signal recorded by a first receiver of a probe located in an underground ground volume, and said acoustic signal is formed by an active signal transmitter, and applying the input data to a machine learning model trained according to one of the methods disclosed herein in order to obtain the output of the machine learning model as an estimate. In other words, acoustic signal data recorded by a signal receiver (such as one of the receivers described above) is supplied as input to a trained machine learning model, and the machine learning model outputs an estimated arrival time of the acoustic signal at the receiver as its output. As described above, this provides an improved mechanism for determining the arrival time of an acoustic signal at a receiver, which provides an estimate that is faster, more accurate, more reliable, and more reproducible than is possible when a human interpreter is used. As mentioned above, this then makes it possible to determine important geographic data parameters related to subsurface volume more quickly and with greater confidence.

[0021] Similar to training the machine learning model described above, during the usage phase, the input data may include two or more sets of acoustic data. For example, as mentioned above, the input data may include acoustic data recorded by two or more receivers, or acoustic data recorded by a single receiver at multiple locations.

[0022] In one implementation, the input data supplied to the machine learning model may further include second acoustic data representing acoustic signals recorded by a second receiver located within the underground ground volume. In other words, the second receiver detects the same signal as the first receiver at a different time. Therefore, in this case, the output of the machine learning model may further include an estimate of the time of arrival of the acoustic signal at the second receiver. The above discussion regarding the details of the first and second receivers also applies here, in particular, that the receivers may be part of the same underground probe or different underground probes.

[0023] In an alternative implementation, as described above, a single receiver can be used, and therefore the second acoustic data may include acoustic data recorded by the same first receiver at different locations in response to subsequent similar acoustic signals. In this case, the input data further includes the second acoustic data, which in this example represents the second acoustic signal recorded by the probe's first receiver when the probe was located at different locations within the subsurface volume. In this case, the output of the machine learning model further includes an estimate of the arrival time of the second acoustic signal at the first receiver.

[0024] As with training, when using a machine learning model in the context of a first and second set of acoustic data, the first and second sets of acoustic data can be advantageously supplied to the machine learning model as a combined input. This has the same advantages as described above for the training phase, namely, the machine learning model applies consistent processing to both records, regardless of whether the records represent recordings of the same signal at the first and second receivers or recordings of different signals at the first receiver. This approach is beneficial given that in either implementation, these records relate to either the same single signal formed by the transmitter (in the case of two receivers) or two similar signals formed by the transmitter at different times (in the case of a single receiver). Treating the two records as a related combined input ensures that the arrival times of the signals estimated by the machine learning model are more reliable.

[0025] The method further includes determining the section velocity of one or more components of the acoustic signal detected by the first receiver and / or the second receiver. The term "section velocity" indicates that the determined velocity is the average velocity over the separation distance between the two locations where the first and second acoustic data were recorded, respectively. If the recorded acoustic signal data represents both compression-related and shear-related signal components, separate section velocities can be determined for each component. The method of determining the section velocity varies depending on whether a multi-receiver setup or a single-receiver setup is used.

[0026] In one implementation, if there are multiple receivers and the first and second acoustic data represent a single signal recorded in the first and second receivers, the interval velocity can be determined based on the estimated arrival times of the acoustic signal in the first and second receivers and the known distance between the first and second receivers. Therefore, this method may include determining the interval velocity of one or more components of an acoustic signal based on the estimated arrival times of the acoustic signal in the first and second receivers and the known distance between the first and second receivers. In other words, based on the known distance between the two receivers and the determined arrival times of the signal in each receiver, it is possible to determine the time it took for the signal to travel between the first and second receivers, and therefore the speed at which the signal was traveling during this time. Thus, the term “interval velocity” in this example indicates that the determined velocity is the average velocity over the separation distance between the two receivers.

[0027] In a single-receiver implementation, the receiver records data representing a first signal at a first location within the ground volume. The receiver is then moved to a different second location within the ground volume and records data representing a second signal at that new second location. Both the first and second signals are similar signals transmitted by a signal transmitter. In this example, the section velocity can be determined based on the estimated arrival times of the acoustic signals at the first and second locations and the known distance between the first and second locations. Thus, in this implementation, the method may further include determining the section velocity of one or more components of the second acoustic signal based on the estimated arrival times of the first and second acoustic signals at the first receiver and the known distance between the locations within the ground volume where the first receiver detected the first and second acoustic signals, respectively.

[0028] Further details on the precise method for calculating the section speed in each of the above implementations are presented below.

[0029] Regardless of the implementation used and how the section velocity is calculated, the method may further include calculating one or more parameters related to the ground volume based on the determined section velocity of one or more components of a first or second acoustic signal. These parameters may represent useful geographic data that can provide insights into the structure, stability, or other properties of the ground volume in which the receiver is located. These insights and parameters can further provide information for important infrastructure and engineering decisions, for example, in relation to the installation and construction of buildings or facilities. Some examples of one or more parameters include the small-strain shear modulus related to the ground volume, the small-strain Poisson's ratio related to the ground volume, the small-strain Young's modulus related to the ground volume, and the bulk modulus related to the ground volume. Each of these parameters can be determined more accurately and reliably through the use of the method and system of the disclosure, in particular, because the calculation of each of these variables relies on determining the time of arrival of the signal at the receiver and, from this time of arrival, determining the velocity of one or more components of the signal passing through the ground volume in question.

[0030] The disclosure method may further include presenting the estimated arrival times of the acoustic signals at the first and / or second receivers on a display. Advantageously, the estimated arrival times can be superimposed on a visual display of the signal traces of the acoustic signals recorded by the receivers. This facilitates inspection and interpretation of the estimated arrival times, enabling improved quality control. When inspection is performed, the user can provide an input to confirm whether the arrival times estimated by the machine learning model are acceptable, for example, based on whether the estimated arrival times appear accurate to the user. This can also function as a form of quality control or audit to ensure the proper functioning of the machine learning model, which can be particularly useful in the early stages of training when the model is still at a rudimentary level. When such quality control inspections are performed by the user, the method may further include receiving an input representing confirmation of the estimated arrival times of the acoustic signals at the first and / or second receivers. For example, the user can provide an input confirming that the arrival times estimated by the machine learning model appear accurate. Alternatively, the method may include receiving an input representing correction and / or rejection of the estimated arrival times of the acoustic signals at the first and / or second receivers. This can happen if the user decides that the estimated arrival time output by the machine learning model is incorrect.

[0031] As described above, a probe including a first receiver and, optionally, a second receiver, can be installed in an underground borehole. In one particular embodiment, the probe comprises a P&S suspension logging (PSSL) tool or a seismic cone penetration test (SCPT) tool. The functionality and usefulness of both of these tools, inserted into the ground to analyze the ground volume, are enhanced by incorporating the methods disclosed herein. Optionally, the probe may include an active signal transmitter. This is more relevant to PSSL-type setups typically used, where a single probe typically includes two or more receivers and signal transmitters. This reduces the number of separate components that need to be used and simplifies the geographic data acquisition process.

[0032] The machine learning model used may be any suitable machine learning model. In one particular implementation, the machine learning model may include a convolutional neural network (CNN), which is a type of network particularly well suited to the methods of this disclosure.

[0033] As described above, this method relates to acquiring geographic data using one or more receivers located below the surface within a subsurface volume. Therefore, acquiring input data for the disclosed method may include inserting a probe equipped with a receiver into a subsurface volume, forming an acoustic signal using an active signal transmitter, and recording the acoustic signal in the receiver to generate acoustic data. This applies equally to a first and second receiver and any corresponding first and second acoustic data recorded by these receivers. The same approach applies even if there are three or more receivers or signals.

[0034] Advantageously, the disclosure method is repeatable at both a single depth and / or multiple depths below the Earth's surface. This applies to both single-receiver realizations and multiple-receiver realizations. Thus, the disclosure method may further include repeating any of the disclosure methods at the same depth, or moving the probe to a new depth and repeating the method at the new depth.

[0035] Another aspect of this disclosure provides a system comprising one or more processors and one or more memories storing computer-readable instructions configured to cause one or more processors to execute any of the methods disclosed herein.

[0036] In yet another aspect of this disclosure, when executed by one or more data processing devices, one or more computer-readable media are provided that contain instructions causing one or more data processing devices to perform any of the methods disclosed herein.

[0037] According to yet another aspect of this disclosure, a machine learning model is provided which is stored in one or more computer-readable media, and the model is trained according to one of the training methods disclosed herein.

[0038] To illustrate how the above and other advantages and features of this disclosure can be obtained, a more detailed description of the basic scheme briefly described above is given by reference to its specific embodiments shown in the accompanying drawings. It should be understood that these drawings only illustrate exemplary embodiments of this disclosure and should therefore not be considered to limit the scope of this disclosure. The basic scheme in this specification is described and explained more specifically and in detail with reference to the accompanying drawings as examples illustrating aspects of this disclosure. [Brief explanation of the drawing]

[0039] [Figure 1]This figure shows an exemplary probe equipped with a suitable receiver and an active acoustic signal transmitter for implementing the ground analysis method of this disclosure. [Figure 2] This figure shows a compression-related signal trace representing the acoustic signal data recorded in the probe's first receiver in response to the detection of an acoustic signal. [Figure 3] This figure shows a compression-related signal trace representing the acoustic signal data recorded in the probe's second receiver in response to the detection of an acoustic signal. [Figure 4] This figure shows a shear-related signal trace representing the acoustic signal data recorded in the probe's first receiver in response to the detection of an acoustic signal. [Figure 5] This figure shows a shear-related signal trace representing the acoustic signal data recorded in the probe's second receiver in response to the detection of an acoustic signal. [Figure 6] This diagram schematically illustrates an approach to training a machine learning model to estimate the arrival time of an acoustic signal at a receiver based on recorded acoustic data. [Figure 7] This figure schematically illustrates a more detailed approach to training a machine learning model to estimate the arrival time of an acoustic signal, as disclosed herein. [Figure 8] This figure schematically illustrates the method for using a trained machine learning model to estimate the arrival time of an acoustic signal, as disclosed herein. [Figure 9] This figure shows a shear-related signal trace representing the acoustic signal data recorded in the probe's first receiver in response to the detection of an acoustic signal, where the signal contains two mutually oppositely polarized shear components. [Figure 10] This figure shows a shear-related signal trace representing the acoustic signal data recorded in the probe's second receiver in response to the detection of an acoustic signal, where the signal contains two mutually oppositely polarized shear components. [Figure 11] This is a block diagram of computing devices that can be used to implement the disclosure method.

[0040] Detailed explanation The following is a description of a specific embodiment of the present invention, given only as an example with reference to the drawings.

[0041] Various embodiments of this disclosure are described in detail below. While specific embodiments are described, it should be understood that this is for illustrative purposes only. Those skilled in the art will recognize that other components and configurations may be used without departing from the spirit and scope of this disclosure. Therefore, the following descriptions and drawings are illustrative and should not be construed as limiting. Numerous specific details are described to provide a complete understanding of this disclosure. However, in certain examples, well-known or conventional details are omitted to avoid obscuring the description. References to embodiments in this disclosure may refer to the same embodiment or any other embodiment. Therefore, such references relate to at least one of the embodiments described herein.

[0042] The terms used herein generally have their respective ordinary meanings in the art within the context of this disclosure and in the specific context in which each term is used. Alternative languages ​​and synonyms may be used for one or more of the terms discussed herein, and no special meaning should be placed on whether a term is detailed or discussed herein. In some cases, synonyms for certain terms are presented. The listing of one or more synonyms does not preclude the use of other synonyms. The use of examples anywhere in this specification, including examples of any term discussed herein, is illustrative only and is not intended to further limit the scope and meaning of this disclosure or any exemplary term. Similarly, this disclosure is not limited to the various forms of implementation given herein.

[0043] This disclosure describes an improved system and method for determining the material properties of a subsurface volume, particularly by using machine learning to estimate the arrival time of an acoustic signal at one or more receivers within the subsurface volume. The determined arrival times can be used to compute useful geographic data parameters related to the subsurface volume surrounding the receivers, including material parameters useful for infrastructure and construction planning. Beginning with Figure 1, an exemplary probe that can be used in the context of the disclosed method is described. Acoustic signal data recorded by one or more receivers of such probes is described with reference to Figures 2–5, particularly in the context of determining the arrival time of the acoustic signal at the receivers. A method for training a machine learning model to perform the disclosed method is described with reference to Figures 6 and 7. Next, a method for using the machine learning model trained according to the disclosed method to determine the arrival time of an acoustic signal within the subsurface volume is disclosed with reference to Figure 8. Deformation forms in which the formed signal shear components have shear components polarized in opposite directions are described with reference to Figures 9 and 10. Finally, computing devices available for performing the disclosed method are described with reference to Figure 11.

[0044] As described above, this disclosure generally relates to forming an acoustic signal (also referred to as an “audio signal”) using an active signal transmitter, such as dropping a hammer or loading explosives. The acoustic signal then passes through a subsurface volume below the ground surface and is detected and recorded as acoustic data (sometimes referred to as “audio data”) by one or more acoustic receivers (also referred to as “audio receivers”) of probes placed within the subsurface volume. Based on the characteristics of the recorded signal, particularly the signal velocity, geographic data parameters useful for engineering and infrastructure planning purposes can be determined.

[0045] The velocity of a signal can be determined from the basic relationship V = d / (t-t0), where V is the (average) velocity of the signal, d is the known distance the signal traveled during the measurement period, t is the time the signal arrived at the receiver, and t0 is the time the signal was formed. t-t0 represents the time the signal traveled, i.e., the time it took the signal to travel distance d. The parameters d and t0 can be set to known values ​​at the time the analysis is performed. Then, by performing the method of this disclosure, t (i.e., the time the signal arrived at the receiver) can be determined. From this, V is calculated and can subsequently be used to determine useful geographic data parameters that are important for engineering and infrastructure planning.

[0046] From the above, it will be clear that accurately and reliably determining the time t at which the signal arrives at one or more receivers of the underground probe is extremely important, since the measurement of t enables the determination of the signal velocity V. Therefore, methods for improving the accuracy and reliability of determining the time t at which the signal arrives represent a major focus of this disclosure.

[0047] Next, with reference to Figure 1, an exemplary experimental setup for conducting an investigation to determine the arrival time t of an acoustic signal traveling through a target ground volume is described. It should be understood that this setup is merely an example, and other experimental setups can be used. In particular, any setup that allows the arrival time of a formed signal to be determined at a receiver within the ground volume is applicable to the method of this disclosure.

[0048] Referring to Figure 1, an exemplary probe 100 that can be used to carry out the method of disclosure is shown. The probe 100 can be inserted into a borehole in the target soil volume to enable analysis of the soil volume. In this example, the probe includes a first receiver 102 and a second receiver 104, but in some examples, the probe 100 may include one receiver or three or more receivers. Receivers 102, 104 can be any audio receiver configured to detect and record acoustic signals in the subsurface environment. In this example, the probe 100 also includes an active signal transmitter 106, which may also be referred to as an active acoustic signal source. The active signal transmitter 106 is configured to form an acoustic signal, which propagates through the soil volume and can subsequently be detected by receivers such as the first receiver 102 and the second receiver 104 of the probe 100. In the example of Figure 1, the active signal transmitter 106 includes a hammer configured to strike a plate to form an acoustic signal. Other mechanisms, such as those based on explosive loading, are also possible for forming acoustic signals. In some realizations, the active signal transmitter 106 does not have to be provided as part of the probe 100, but instead may be provided as part of a different probe located in the same borehole as probe 100. Alternatively, the active signal transmitter may be located elsewhere, such as on the ground surface.

[0049] It should be noted that the term “active” in the context of “active signal transmitter” simply means that the signal transmitter 106 is configured to form an acoustic signal in a controlled (“active”) manner in response to operator control, such as the activation of a trigger or the transmission of a command signal. Such active signal formation is in contrast to “passive” acoustic signals, which can be caused by uncontrolled background noise. Such passive acoustic signals may be recorded and analyzed in some geographic data acquisition contexts, but this is not relevant to the methods of this disclosure. In the illustrated example, probe 100 includes acoustic isolators 108 to isolate receivers 102, 104 from each other and from the signal source 106. However, these isolators are not mandatory.

[0050] The operation of the probe 100 in Figure 1 is as follows: First, a borehole is drilled within the target soil volume. The borehole has a diameter sufficient to allow the probe 100 to be fitted into it. Next, the probe 100 is lowered into the borehole so that a receiver is positioned at a desired depth within the soil volume. If the probe includes two or more receivers, as in the example in Figure 1, the depth of the probe may be measured as the midpoint between the receivers, in this case the midpoint between the first receiver 102 and the second receiver 104. Once positioned, the active signal source 106 is triggered, forming an acoustic (sound) signal. This signal is radiated outward in wave form from the signal source 106 into the soil volume surrounding the borehole. The signal spreads through the soil volume and re-enters the borehole. Some components of the signal also travel along the inner surface of the borehole, as will be described in more detail below. The signal is then detected in one or more receivers of the probe. The speed of a signal can be determined based on the time it arrives at the receiver and the known distance it has traveled. This general basic method applies regardless of where the active signal transmitter is located or the number of receivers provided. Next, to aid understanding, we will describe three examples of how the signal speed can be determined.

[0051] single receiver In some examples (not shown in FIG. 1), there may be only a single receiver within the probe 100. In this case, the arrival time t of the signal at the receiver is determined. In this case, the travel time of the signal is the difference between the arrival time t and the time t0 when the signal was formed at the signal source. Assuming that the distance between the signal transmitter and the receiver is known from here, the speed of the signal can be determined based on V = d tra_rec / (t - t0), where V is the signal speed and d tra_rec is the distance between the signal transmitter and the receiver, t is the arrival time of the signal at the receiver, and t0 is the time when the signal was formed at the transmitter.

[0052] Alternatively, in a single receiver implementation, the measurements can be made at a first depth and a second depth within the ground volume. In other words, the receiver can detect a first signal at the first depth. Then, the receiver is moved to a different depth and can detect a second signal at its new second depth. In this case, the difference in the time it takes for the same acoustic signal to reach the receiver at each depth can be determined. Based on that difference, the interval (i.e., average) speed over the distance between the first depth and the second depth can be determined. In other words, the signal travel time to the first depth is t1 - t 0_1 and the signal travel time of the same signal to the second depth is t2 - t 0_2 In this case, the travel time of the signal here between the first depth and the second depth is (t2 - t 0_2 ) - (t1 - t 0_1 ). t 0_1 is the start time of the first signal and t 0_2 is the start time of the second signal. Then, the interval speed of the signal over the distance d 1_2 between the first depth and the second depth is V = d 1_2 / ((t2 - t 0_2 ) - (t1 - t 0_1This is a single-receiver approach at multiple depths, particularly relevant to implementations where the signal transmitter is on the ground surface and a single receiver probe is inserted into the ground volume. An example of such an approach is the SCPT probe.

[0053] When a single receiver approach is used to detect the first and second signals in this manner, the first and second signals are configured to be similar signals having similar or identical initial frequencies and amplitudes. Naturally, there will be a slight difference in the time it takes for the signals to reach the receiver between the first and second locations, because the signals are traveling through different distances of soil volume. As a result, small changes in signal characteristics can be observed as a result of differences in soil conditions and distance. However, in practice, these changes are determined to be negligible so that the first and second signals can be treated as identical. In that case, as mentioned above, only the arrival time of each signal at each depth needs to be determined.

[0054] Multiple receivers When a probe includes two or more receivers, as in the case of probe 100 in Figure 1, the calculation of the signal speed can be simplified and made more reliable by considering only the distance between two receivers, such as the first receiver 102 and the second receiver 104. More specifically, in this case, since the first receiver 102 is located closer to the signal source 106, the signal reaches the first receiver 102 before reaching the second receiver 104. Thus, the first receiver 102 and the second receiver 104 record different arrival times t1 and t2, respectively. Then, the difference in arrival times between the two receivers t2-t1 and the known distance d between the receivers are used. 1_2 The signal speed can be determined based on this. In other words, in this example, V=d 1_2 / (t2-t1), where V is the signal velocity, and d 1_2t1 is the distance between the first receiver 102 and the second receiver 104, t1 is the time the signal arrives at the first receiver 102, and t2 is the time the signal arrives at the second receiver 104.

[0055] General Considerations As mentioned above, regardless of the specific experimental setup and the number of receivers used, determining the signal velocity through a particular portion of the ground volume is extremely useful because, as will be explained in more detail below, this velocity can be used to calculate various important geographic data parameters.

[0056] Those skilled in the art will understand that, for the sake of simplification, the above description refers to a single signal with a single signal velocity V, but in reality, an acoustic signal formed in or within a ground volume actually forms multiple signal components, each with a corresponding component signal velocity. As mentioned above, an acoustic signal traveling through a ground volume generally has two main components: a compression-related component and a shear-related component. A compression wave, more commonly referred to as a P-wave, causes particles in a volume to vibrate back and forth in a compressive motion parallel to the direction of wave propagation. A shear wave, more commonly referred to as an S-wave, causes particles in a volume to vibrate back and forth in a transverse motion perpendicular to the direction of wave movement. Both compression-related and shear-related waves can be detected by a receiver using the setup described above. The velocity at which each of these signal components travel again depends on the material through which each signal component is traveling, and the compression wave velocity v p and shear wave velocity v s Both of these can be determined using the basic method described above.

[0057] Regarding shear wave velocity, the way this is actually determined varies slightly depending on the experimental setup. In particular, if the active signal transmitter is located on the ground surface or in a borehole separate from the receiver, the shear-related component received at the receiver is the true shear (bulk) wave that has traveled through the main body of the ground volume before reaching the receiver. Therefore, in this case, the determination of the velocity of the shear-related signal component directly leads to v s This can be obtained. Alternatively, if the signal transmitter is located in the same borehole as the receiver (as in the example in Figure 1), v s The most effective method for determining the true shear wave velocity is to detect bending waves that propagate upward along the inner surface of the borehole rather than moving through the main body of the ground volume. In this case, the velocity of these bending waves can be obtained by determining the signal velocity of the shear-related components using the method described above. Then, the true shear wave velocity v s This can be determined from the bending velocity using methods known in the art. More specifically, a bending wave has the inherent property that, in the low-frequency region, its propagation speed is equal to the shear wave velocity. The bending wave is subject to a dispersion effect that depends on the frequency that changes the wave speed. The dispersion is minimized when the wavelength is at least three times the borehole diameter and is in the low-frequency range (<2 kHz). For these reasons, it is reasonable to assume that the bending velocity is equal to the true shear wave velocity.

[0058] Ultimately, the distinction between true shear waves and bending waves is not important for understanding the main method of this disclosure and is presented here solely for the sake of completeness and to aid understanding. In either case, the final result is the compressed wave velocity v of the acoustic signal. p and shear wave velocity v s The methods of this disclosure relating to determining the arrival time of a signal (component) are applied in the same manner regardless of which signal component (compression, true shear, or flexural shear) is being analyzed.

[0059] Acoustic data recording Next, referring to Figures 2 to 7, traces representing different types of signal components that can be detected and recorded by the receiver of this disclosure are presented. The amplitude of the received signal is shown on the Y axis, and the elapsed time is shown on the X axis. The signal traces in Figures 2 to 7 are presented in the context of probe 100 of Figure 1, which has two receivers, and therefore two signal traces are recorded for each type of signal. However, it will be understood that the concepts described apply regardless of how many receivers there are.

[0060] Figure 2 shows the compression-related component (i.e., P-wave) detected by the receiver, in this case the first receiver 102 of the probe 100. Figure 3 shows the same compression-related signal detected by the second receiver 104 of the probe. Figures 2 and 3 share the same time axis, and as can be seen, the signal arrives at the second receiver 104 later than the first receiver 102, as described above. The arrival time of the compression signal component at the first receiver 102 is also denoted by t1 and is shown by a dashed line in Figure 2. The arrival time of the compression signal component at the second receiver 104 is similarly denoted by t2 and is similarly shown by a dashed line in Figure 3. As described above, determining both t1 and t2 makes it possible to determine the velocity of the compression signal component (P-wave) based on the known distance between the first receiver 102 and the second receiver 104.

[0061] Figures 4 and 5 are identical to Figures 2 and 3, except that they relate to the shear-related component of the acoustic signal. In this case, since probe 100 is used, the active signal transmitter 106 is located in the same borehole as receivers 102 and 104. Therefore, for the reasons stated above, the detected shear-related signals shown in Figures 4 and 5 are bent waves that have traveled across the inner surface of the borehole before being detected. However, from the velocity of these bent waves, the true shear wave velocity v sGiven that it is still possible to calculate the shear wave, the waves in Figures 4 and 5 can be simply referred to as "shear waves" or "S-waves" for simplicity, despite the fact that they are strictly speaking bent waves. The signal trace in Figure 4 shows the S-wave component of the formed acoustic signal detected by the first receiver 102. The signal trace in Figure 5 similarly shows the S-wave component of the same formed acoustic signal detected by the second receiver 104 a short time later. The arrival time of the shear-related signal component in the first receiver 102 is again represented by t1 and is shown by a dashed line in Figure 4. The arrival time of the shear-related signal component in the second receiver 104 is similarly represented by t2 and is again shown by a dashed line in Figure 5.

[0062] By determining t1 and t2 for each signal component (compression and shear), the velocity of that component can be determined by the method described above. In particular, the compressed wave velocity v p and shear wave velocity v s These can be determined. From these speeds, many useful geographic data parameters can be determined that characterize the ground volume through which the signal traveled, with the receiver installed. Some exemplary parameters are as follows:

number

[0063] As is clear, the determination of each of these geographic data parameters depends on the accurate and reliable determination of the arrival time of each acoustic signal component in a given receiver, which is the compression rate (v p ) and shear rate (v sThis value makes it possible to determine the time of arrival of the signal at the receiver. Therefore, in the example above using probe 100 in Figure 1, it is important to accurately determine t1 (the time of arrival of the signal component at the first receiver 102) and t2 (the time of arrival of the signal component at the second receiver 104) for both the P-wave and S-wave components. Unfortunately, existing methods for determining the time of arrival of a signal at a receiver are rudimentary and prone to errors and biases. This is because the determination of the time of arrival is currently an entirely manual process, with highly trained geographic data scientists or engineers looking at the signal trace (such as one of the traces shown in Figures 2-5) and manually determining when the acoustic signal reached the receiver. Moreover, making such a determination is extremely difficult and requires a high level of expertise. This is because audio receivers are constantly detecting background noise. As a result, it is often extremely difficult to determine the point in the signal trace where the noise ends and the signal of interest (i.e., the signal formed by the active signal transmitter 106) begins. Consider, for example, the traces shown in Figures 2-5. In Figure 2, determining the arrival time is not particularly difficult because the background noise is relatively low, making it very easy to find the point at which the target signal arrives, as shown by the dashed line t1. However, the remaining signals are far more difficult. For example, in Figure 3, the background noise amplitude is higher than in Figure 2 and more closely approximates the initial amplitude of the target signal, meaning that determining the transition point between the noise and the signal (i.e., arrival time t2) for the signal trace in Figure 3 is extremely difficult. Engineers need to be experienced and well-trained to accurately determine the precise arrival time. Referring to Figures 4 and 5, determining the precise arrival time in these cases is even more difficult because the background noise is extremely severe. Only highly experienced and well-trained experts were able to accurately select the arrival time, i.e., the transition from noise to the target signal, in these traces.

[0064] As a result of these difficulties, determining signal arrival times is currently an extremely difficult and slow task that can only be performed by a small number of experienced specialists. In addition, even expert analysts are susceptible to human error and bias, meaning that one specialist may tend to characterize the signal in a different way than another. This reduces the reliability, reproducibility, and consistency of manually determined arrival times, even among specialists. All of these factors result in significant shortcomings in the availability, reliability, accuracy, and consistency of determined signal arrival times. This means that, in this case, the useful geographic data parameters mentioned above often cannot be determined with particularly high accuracy, and cannot be determined within a reliably satisfactory time frame. Therefore, there is a great need for an improved method for determining signal arrival times in underground receivers. Accordingly, the remainder of this disclosure presents an improved approach to determining signal arrival times that addresses some of the aforementioned shortcomings associated with existing approaches.

[0065] An improved approach that uses machine learning to determine the time of signal arrival. The inventors have identified that the shortcomings identified above can be effectively addressed by employing a trained machine learning model to determine the arrival time of a given signal component in a receiver. This approach has been shown to improve the reliability, accuracy, and consistency of the determined signal arrival times, and further, to lead to a more reliable and accurate determination of geographic data parameters. Next, a method for implementing this improved approach will be described with reference to Figures 6 to 8.

[0066] Referring to Figure 6, a schematic diagram illustrates how to train a machine learning model to identify the arrival time of an acoustic signal in an underground receiver. This training method can be used in the context of the ground analysis described above; in other words, a machine learning model trained using the method in Figure 6 can be used to output the arrival time of a signal component, which can further be used to determine the signal component velocity (especially v s and v p) and other relevant geographic data parameters, such as those mentioned above, can be used to determine these parameters.

[0067] The method in Figure 6 begins in step 602 by acquiring input acoustic signal data. This is recorded acoustic data experimentally obtained in an underground receiver through the recording of acoustic signals formed by an active signal transmitter. For example, the input acoustic signal data may include data representing recorded acoustic signal traces of the type shown in Figures 2 to 5.

[0068] In step 604, the method proceeds by acquiring target signal arrival time data. The target signal arrival time data includes labels or indices indicating the signal arrival time associated with each audio signal contained in the input audio signal data supplied in step 602. For example, if the input audio signal data includes data related to the audio trace in Figure 2, the target signal arrival time data includes an indication of t1 for this trace, i.e., the signal arrival time at the receiver. The labels here may be supplied by a professional analyst or another machine learning network. The target data (labeled arrival times) can then be used within the training method as "validated data" for each audio signal. The validated data arrival times are what the machine learning model is trying to replicate.

[0069] In step 606, the method trains a machine learning model using the input acoustic signal data supplied in step 602 and the target signal arrival time data supplied in step 604. In particular, data related to the acoustic signal recording is supplied to the machine learning model. The machine learning model then predicts the arrival time of the acoustic signal at the receiver based on the input data. Then, in step 604, it is evaluated whether the determined (predicted) signal arrival time matches the actual arrival time specified in the respective target arrival time data supplied for the acoustic signal.

[0070] Training can be performed, as is known in the art, by adjusting the model parameters of a machine learning model to reduce the error between the model output and the target data. This process is repeated until a sufficiently well-trained model is obtained, as measured by a convergence criterion, an error criterion, or a stopping criterion such as a predetermined number of epochs. Numerous programming languages ​​and libraries are available to implement this process using a wide variety of machine learning models. Those skilled in the art will understand that an appropriate model can be selected based on considerations such as available data and computation, as well as the type of data to be processed based on everyday considerations. The exact details of the training process in step 606 will vary based on the details of these implementations, as will be understood by those skilled in the art. The final result of the method in Figure 6 is a machine learning model trained to identify the arrival time of an acoustic signal at a receiver based on input acoustic signal data recorded by that receiver. Thus, a machine learning model trained in this manner can be used to carry out the method described above, with the ultimate objective of determining the velocity of a particular signal component and associated geographic data parameters.

[0071] Referring to Figure 7, an implementation of the method in Figure 6 is shown in more detail in the context of a receiver, such as one of the receivers of probe 100 in Figure 1. The training method in Figure 7 was developed by the inventors and provides particularly effective training for enabling a machine learning model to determine the arrival time of an acoustic signal in an underground receiver.

[0072] The training method in Figure 7 begins in step 702 by acquiring acoustic data representing acoustic signals recorded by a receiver of a probe placed within the underground ground volume. This data is used as input data in the subsequent training process. Thus, it will be understood that step 702 corresponds to step 602 in Figure 6. As described with reference to Figure 6, the acoustic data may be data recorded by, for example, a first receiver 102 or a second receiver 104 in the probe 100 in Figure 1.

[0073] In step 704, the process obtains the target arrival time of the acoustic signal at the receiver. The target arrival time is used as target or validated data in the training process. Thus, it will be understood that step 704 corresponds to step 604 in Figure 6. Steps 706-712 then provide a more detailed explanation of the training process of step 606 in Figure 6.

[0074] Starting in step 706, the acoustic data acquired in step 702 is fed to the machine learning model. In response to this input, the output of the machine learning model is obtained in step 708. In particular, the output of the machine learning model includes a determined arrival time of the acoustic signal at the receiver, which is determined based on the input data. In other words, the machine learning model takes acoustic data recorded by a receiver as its input and generates an estimated arrival time of the acoustic signal at that receiver as its output. For example, the machine learning model may output a predicted arrival time in milliseconds from a known start time, or on a given time scale.

[0075] In step 710, the error between the determined signal arrival time and the target signal arrival time is determined. This can be done in any suitable way by comparing the output obtained in step 708 with the target data supplied in step 704. In response to this determination, the method proceeds to step 712, where the parameters of the machine learning model are adjusted so that the error determined in step 710 is reduced.

[0076] The process in Figure 7 can then be repeated for further sets of acoustic signal data until a stopping criterion is met. For example, this method can be repeated N times for a predetermined number of epochs or until a convergence criterion is met. In implementations with two or more receivers, the process in Figure 7 can be performed for each receiver that has recorded the acoustic signal formed by the acoustic signal transmitter. The process in Figure 7 can also be performed for each signal component formed by an active signal transmitter; for example, this process can be performed for the compression-related components of the signal (e.g., P-waves) and / or the shear-related components of the signal (bending waves or S-waves).

[0077] Once trained in this way, the machine learning model can be used to determine the arrival time of signals in audio data recordings that have not yet been observed or labeled, thereby replacing the need for human experts to label these signals. This means that the signals can be analyzed more quickly, and the accuracy, reliability, and consistency of arrival time determination are improved for the reasons mentioned above. Next, with reference to Figure 8, we will describe how to obtain an estimate of the arrival time of an acoustic signal in an underground receiver using the trained machine learning model.

[0078] In step 802, input data including acoustic data is acquired. In this case, the acoustic data represents acoustic signals recorded by the receiver of the probe located within the subsurface volume, for example, one of the first receiver 102 and the second receiver 104 of the probe 100 in Figure 1. Thus, step 802 corresponds to step 602 in Figure 6 and step 702 in Figure 7, but since the method in Figure 8 relates to the usage phase rather than the training phase, in this case the acquired acoustic data is acoustic data that has not been previously observed and has not yet been labeled to indicate the time of signal arrival. The acoustic data can be acquired directly from the receiver or database by any suitable wired or wireless means. As described in relation to steps 602 and 702, the acquired acoustic data in this case also represents a recording of acoustic signals formed by an active signal transmitter, such as the active signal transmitter 106 of the probe 100 described in relation to Figure 1. The acoustic signal data may include data representing recorded acoustic signal traces of the type shown in Figures 2 to 5.

[0079] In step 804, the input data (i.e., the acoustic data acquired in step 802) is supplied as input to the trained machine learning model. If the model has been trained according to the methods of this disclosure (particularly the methods in Figures 6 and 7 described above), the model can proceed to determine the arrival time of the signal at the receiver based on the input acoustic data. This is done in step 806, and the machine learning model outputs an estimated arrival time of the acoustic signal at the receiver (e.g., t1 or t2 described above).

[0080] As an optional means, the method may then include further steps relating to processing the output of a machine learning model. In one example, the method proceeds to step 808, in which the velocity of the signal represented by the acquired acoustic data is determined. As described above, the acoustic data may represent specific components of an acoustic signal formed by an active signal transmitter, such as compression-related components or shear-related components. Based on the determined arrival time, the velocity of the signal component being analyzed can be determined, v p or v s The output is obtained. These calculated velocities may represent the velocity of the signal (component) between the transmitter and receiver, or the average velocity of the signal over the distance between the first receiver and the second receiver or between the first recording position / depth and the second recording position / depth in the case of multiple receivers or multiple recording positions / depths (also known as the "section velocity"). Step 808 is, for example, the shear-related velocity v s and compression-related speed v p To obtain both, this can be done for each signal component.

[0081] As an optional means, in step 810, the method may further include calculating one or more geographic data parameters based on the signal velocity determined in step 808. These geographic data parameters may be, for example, the small-strain Poisson's ratio ν and the small-strain shear modulus G. max , Young's modulus E of small strain max , and the bulk modulus K can be included. Each of these parameters is v s and formula v pTherefore, it can be determined using the above formula. Each determined parameter corresponds to the material properties of the ground volume in which the receiver is installed, and thus provides useful geological insights. More specifically, in the case of two receivers in a downhole probe 100 of the type shown in Figure 1, the determined geodata parameters typically represent the material properties of the cylindrical shell volume over the distance between the first receiver 102 and the second receiver 104, up to about 10 cm outside the borehole wall. The same is true for realizations in which a single receiver is used to record acoustic data at two recording locations or depths. In the case of a single-receiver system at a single depth, the calculated geodata parameters represent the material properties of the ground volume between the receiver and the transmitter. In each realization, useful geodata is provided regarding the ground volume of some object in the vicinity of the receiver.

[0082] The method in Figure 8 can be repeated as many times as necessary to take into account the processing of multiple acoustic signal components (e.g., compression and shear-related components) and multiple signals (from multiple receivers such as the first receiver 102 and second receiver 104 of probe 100 in Figure 1, or from a single receiver recording data at multiple positions / depths, etc.).

[0083] All methods of disclosure are computer-implementable using a computing system, for example, an exemplary system or network described below with reference to Figure 11. It will be understood that all steps in Figures 6, 7, and 8 can be performed locally at the location where the probe 100 and / or transmitter 106 are provided, remotely at a different location (e.g., using a remote server or computing resources), or a combination of both local and remote locations. Referring to Figure 8, steps 808 and 810 as optional means can be performed by the same computing system or network that performs steps 802-806, or, as optional means, can be computed separately by different parties or entities that performed steps 802-806 using different computing resources. Also, the training steps (i.e., Figures 6 and 7) can be performed by different entities or parties at a different time or place than the usage steps (i.e., Figure 8). In other words, one party can train a machine learning model and then provide the trained model to another party for use.

[0084] Transformation form The detailed description above illustrates various exemplary configurations and methods for determining the arrival time of an acoustic signal in an underground receiver. However, the described configurations and methods are merely illustrative, and it will be understood by those skilled in the art that various modifications can be made without departing from the scope of the attached claims. Some of these modifications are briefly described here, however, this list of modifications should not be considered exhaustive, and other modifications should also be obvious to those skilled in the art.

[0085] As described above, in some implementations, two or more sets of acoustic signal data are generated by multiple receivers or by a single receiver performing signal recording at multiple locations / depths. Regardless of how the acoustic dataset is generated, the inventors have found that in such implementations, it is beneficial to feed the acoustic dataset from both (or all) receivers or locations / depths as a combined input to the machine learning model. This is true both during training (as described, e.g., in relation to Figures 6 and 7) and during use (as described, e.g., in relation to Figure 8) of the machine learning model. In other words, in this implementation, the input data supplied to the machine learning model in either step 706 or 804 may include a combined input that includes acoustic signal data from two or more receivers, or acoustic signal data recorded by a single receiver at two or more locations or depths. As described above, supplying the data as a “combined input” means that the acoustic data from each receiver or location is supplied to the machine learning model as part of the same input that the machine learning model processes uniformly. In other words, the machine learning model uniformly analyzes and labels both sets of acoustic data when assigning arrival times to each set of acoustic data. As described above, the inventors determined that this strategy of supplying acoustic data from multiple receivers as combined inputs improves the training or analysis process and leads to a more reliable and accurate determination of signal arrival times, as the model effectively learns to recognize common patterns and trends in signals recorded at both the first and second receivers, or at both the first and second positions / depths. Since each set of acoustic data represents either the same single signal or similar signals formed at an active signal source, it is beneficial to train the machine learning model to treat the resulting acoustic data as relevant.

[0086] The above discussion, particularly the signal traces in Figures 2-5, focused for simplification on waves having a single compression (i.e., P-wave) related component and a shear (i.e., S-wave) related component. However, the inventors found it advantageous to configure the active signal transmitter to form two shear-related components polarized in opposite directions. As a result, the recorded acoustic signal data includes data representing two bent waves polarized in opposite directions. Importantly, the inventors found this makes it easier for machine learning models to assign arrival times to the signals. An example of such acoustic data is shown in Figures 9 and 10. Figure 9 shows the shear-related component (in this case, the bent wave) detected by a receiver, in this case the first receiver 102 of probe 100. Figure 10 shows the same shear-related signal detected by the second receiver 104 of the probe. Thus, Figures 9 and 10 are similar to Figures 4 and 5 described above, except that each set of acoustic data includes two shear-related signals, the bent wave in this example. Since the bent wave is polarized in opposite directions, its peaks and troughs correspond approximately to each other, making it easier to identify the signal arrival time. Thus, the formation of an acoustic signal with two mutually oppositely polarized shear components allows machine learning models to perform the decisions in steps 708 and 806 described above with greater accuracy. The signal prior to the arrival of the shear-related signal in question contains noise. If the shear-related wave is the signal in question, the noise may contain compression-related components, and vice versa.

[0087] Any suitable machine learning model may be used to perform the method of disclosure, and it will be understood that the choice of machine learning model will affect how the determination of the signal arrival time is performed. However, generally, most approaches involve the machine learning model analyzing the acoustic signal (components) represented in the input acoustic signal data and determining the arrival time of the acoustic signal by determining the point in the recorded signal where ambient background noise is replaced by the recorded signal of interest or transitions to the recorded signal of interest. Thus, the transition point represents the arrival time of the signal of interest. In other words, the machine learning model of this disclosure may be configured to determine the arrival time of an acoustic signal by separating the acoustic data into noise and signal portions, determining the transition point between the noise and signal portions, and outputting the transition point as the estimated arrival time of the acoustic signal. This approach may be applied during training and / or use.

[0088] Separating acoustic data into noise and signal portions can be done using any suitable method, such as a segmentation-based approach or an image recognition-based approach. In other words, the model can be a signal processing model or an image recognition model. In either case, a suitable machine learning model architecture is employed. More generally, any suitable machine learning model can be used to perform the disclosure method. One particularly suitable model is a convolutional neural network (CNN). Long short-term memory (LSTM) networks, and more generally recurrent neural networks (RNNs), can also be used.

[0089] In several implementations, the method of disclosure can be implemented in a way that facilitates effective quality control by human evaluators or experts. In particular, the system can be configured to display the estimated arrival time of the acoustic signal at the receiver, determined by a machine learning model. This allows a human evaluator to evaluate, confirm, or provide corrective inputs for the determined arrival time output by the machine learning model. For example, the system may receive input from the evaluator confirming the arrival time determined by the machine learning model. Alternatively, the system may receive input from the evaluator rejecting or correcting the arrival time determined by the machine learning model. This allows a human evaluator to interact with the machine learning model's decisions and act as a filter or check on the output. This is particularly useful and important during the training phase, as it can allow evaluators to easily determine how well the machine learning model is performing and whether the training is progressing properly. This feature also allows for easy and effective auditing of previously trained models, ensuring that they are still performing as expected. In one particularly advantageous configuration, presenting the estimated arrival time of an acoustic signal at a receiver, determined by a machine learning model, may include presenting a signal trace representing the acoustic signal (such as one of the traces shown in Figures 2-5 or 9-10) and indicating the estimated arrival time on the trace, for example, using visual indicators or markers. This allows for a particularly simple and intuitive analysis of the determined arrival time by the evaluator.

[0090] As previously stated, the receiver of this disclosure can be installed below the ground surface in any suitable manner. However, as previously stated, one particularly useful embodiment includes incorporating the receiver into a probe, such as probe 100 in Figure 1, and inserting the probe into the ground, for example, into a borehole. In one example, the probe may comprise an acoustic logging tool, one example being a P&S suspension logging (PSSL) tool, a known type of tool particularly well suited for detecting compression-related signal components (P-waves) and shear-related signal components (S-waves) within a subsurface volume. In another embodiment, the probe may include an earthquake cone penetration test (SCPT) tool, a known type of tool for investigating subsurface ground properties, but which can be readily modified to incorporate an acoustic receiver for performing the method disclosed herein.

[0091] As described above, an active signal transmitter can be any suitable form of transmitter configured to form an acoustic (i.e., audio) signal in a controlled manner so that the signal passes through the target ground volume and is recorded by a suitable receiver.

[0092] The specific mechanism and method for recording acoustic signal data varies depending on the implementation and type of probe used. However, the method as a whole includes inserting a probe equipped with one or more receivers into the target subsurface volume, forming an acoustic signal using an active signal transmitter, and recording the acoustic signal in one or more receivers to generate acoustic data for each of the one or more receivers that can be input into a machine learning model as described above. All of the methods disclosed herein can be performed multiple times at a particular depth and / or at multiple different depths to improve the reliability of the obtained data. The recorded acoustic data can be stored locally or remotely on any suitable computer storage component or server.

[0093] Computer system Next, with reference to Figure 11, a computing device 1100 suitable for performing the above-described method will be explained. Figure 11 shows a block diagram of one implementation of a processing system 1100 in the form of a computing device, in which a set of instructions for causing the computing device to perform any one or more of the methods discussed herein is executable. In alternative implementations, the computing device can be connected to (e.g., networked) other machines in a local area network (LAN), intranet, extranet, or internet. The computing device may operate as a server or client machine in a client-server network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The computing device may be a personal computer (PC), tablet computer, set-top box (STB), personal digital assistant (PDA), mobile phone, web appliance, server, network router, switch or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) specifying actions to be performed by that machine. Furthermore, although only a single computing device is shown, the term “computing device” should also be interpreted to include any set of machines (e.g., computers) that individually or collectively execute a set (or set) of instructions for performing any one or more of the methods described herein.

[0094] An exemplary processing system 1100 includes a processor 1102, main memory 1104 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM) or rhombus DRAM (RDRAM), etc.), static memory 1106 (e.g., flash memory, static random access memory (SRAM), etc.), and secondary memory (e.g., data storage device 1118), all of which communicate with each other via a bus 1130.

[0095] Processor 1102 represents one or more general-purpose processors, such as microprocessors and central processing units. More specifically, processor 1102 may be a composite instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing another instruction set, or a processor implementing a combination of instruction sets. Processor 1102 may be one or more dedicated processors, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), or a network processor. Processor 1102 is configured to execute processing logic (instructions 1122) for performing the operations and steps discussed herein.

[0096] The processing system 1100 may further include a network interface device 1008. The processing system 1100 may also include a video display unit 1110 (e.g., a liquid crystal display (LCD) or cathode ray tube (CRT)), a character / number input device 1112 (e.g., a keyboard or touchscreen), a cursor control device 1114 (e.g., a mouse or touchscreen), and an audio device 1016 (e.g., a speaker).

[0097] It will be clear that some features of the processing system 1100 shown in Figure 10 may be unnecessary. For example, there may be cases where the processing system 1100 does not require a display device 1110 (or any associated adapter). This could be the case, for example, of a specific server-side computer device used solely for its processing power and not needing to display information to the user. Similarly, the user input device 1112 may not be necessary. In its simplest form, the processing system 1100 comprises a processor 1102 and main memory 1104.

[0098] The data storage device 1118 may include one or more machine-readable storage media (or more specifically, one or more non-temporary computer-readable storage media) 1128 in which one or more sets of instructions 1122 for embodying any one or more of the methods or functions described herein are stored. The instructions 1122 may also be entirely or at least partially present in the main memory 1104 and / or the processor 1102 during their execution by the processing system 1100, and the main memory 1104 and the processor 1102 also constitute the computer-readable storage media 1128.

[0099] The various methods described above can be executed using computer programs. A computer program may include computer code configured to instruct a computer to perform one or more of the functions of the various methods described above. Computer programs and / or code for performing such methods can be provided to a computer or other device on one or more computer-readable media, or more generally on a computer program product. The computer-readable media may be temporary or non-temporary. One or more computer-readable media may be, for example, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, or a propagation medium for transmitting data to download code, for example, over the Internet. Alternatively, one or more computer-readable media may take the form of one or more physical computer-readable media, such as semiconductor or solid-state memory, magnetic tape, removable computer diskettes, random-access memory (RAM), read-only memory (ROM), rigid magnetic disks, and optical disks such as CD-ROMs, CD-R / Ws, or DVDs.

[0100] A computer program is executable by the processor 1102 to perform the functions of the system and method described herein.

[0101] In one implementation, the modules, components, and other features described herein may be implemented as individual components or integrated into the functionality of hardware components such as ASICs, FPGAs, DSPs, or similar devices.

[0102] A “hardware component” is a tangible (e.g., non-transient) physical component (e.g., a set of one or more processors) capable of performing a specific operation, and may be configured or arranged in a specific physical manner. A hardware component may include dedicated circuitry or logic permanently configured to perform a specific operation. A hardware component may be, or include, a dedicated processor such as a field-programmable gate array (FPGA) or ASIC. A hardware component may also include programmable logic or circuitry temporarily configured by software to perform a specific operation.

[0103] Therefore, the term “hardware component” should be understood to encompass tangible entities that are physically constructed, permanently configured (e.g., hardwired), or temporarily configurable (e.g., programmable) to operate in a particular manner or to perform a particular operation as described herein.

[0104] Furthermore, modules and components may be implemented as firmware or functional circuits within hardware devices. Additionally, modules and components may be implemented in any combination of hardware devices and software components, or solely in software (e.g., code stored in or otherwise embodied in machine-readable or transmission media).

[0105] Unless otherwise specified, as is evident from the following explanation, any use of terms such as “receive,” “determine,” “compare,” “activate,” “calculate,” “identify,” “analyze,” “estimate,” and “supply” throughout the explanation is understood to refer to the actions and processes of a computer system or similar electronic computing device that manipulate data represented as physical (electronic) quantities in the registers and memory of a computer system and convert it into other data similarly represented as physical quantities in computer system memory or registers or other such information storage, transmission, or display devices.

[0106] It should be understood that the above description is illustrative and not restrictive. Many other embodiments will be apparent to those skilled in the art upon reading and understanding the above description. While this disclosure has described with reference to specific exemplary embodiments, it should be recognized that this disclosure is not limited to the embodiments described and can be implemented with modifications and changes within the spirit and scope of the appended claims. Therefore, this specification and drawings should be considered illustrative, not restrictive. Accordingly, the scope of this disclosure should be determined with reference to the appended claims, along with the entire scope of the equivalent granted.

[0107] While at least one exemplary embodiment was presented in the detailed description above, please understand that a vast number of variations exist, and only a few of them have been described above. Please also understand that one or more exemplary embodiments are merely examples and are not intended to limit the scope, applicability, or configuration of this disclosure. Rather, the detailed description above provides a convenient roadmap for carrying out one or more exemplary embodiments. Please understand that various modifications to the function and configuration of the elements can be made without departing from the scope of this disclosure as set forth in the appended claims and their legal equivalents.

Claims

1. A method for training a machine learning model to identify the arrival time of an acoustic signal in an underground receiver, The method involves acquiring first acoustic data, wherein the first acoustic data represents an acoustic signal recorded by a first receiver of a probe located within the underground ground volume, and the acoustic signal is formed by an active signal transmitter. To obtain the target arrival time of the acoustic signal in the first receiver, The first acoustic data is supplied to the machine learning model to obtain the output of the machine learning model, wherein the output of the machine learning model includes the determined arrival time of the acoustic signal at the first receiver. Adjusting the parameters of the machine learning model to reduce the error between the determined arrival time and the target arrival time. Methods that include...

2. The above method further, The method involves acquiring second acoustic data, wherein the second acoustic data represents the acoustic signal recorded by a second receiver located within the underground ground volume. To obtain the target arrival time of the acoustic signal in the second receiver, The process involves supplying the second acoustic data to the machine learning model in order to obtain the output of the machine learning model, wherein the output of the machine learning model includes the determined arrival time of the acoustic signal at the second receiver. Adjusting the parameters of the machine learning model to reduce the error between the determined arrival time and the target arrival time for the second receiver. Includes, As an optional means, the probe includes the second receiver. The method according to claim 1.

3. The above method further, The acquisition of second acoustic data, wherein the second acoustic data represents a second acoustic signal recorded by the first receiver when the first receiver is located at a different position within the underground ground volume. To obtain the target arrival time of the second acoustic signal in the first receiver, The process involves supplying the second acoustic data to the machine learning model in order to obtain the output of the machine learning model, wherein the output of the machine learning model includes the determined arrival time of the second acoustic signal at the first receiver. Adjusting the parameters of the machine learning model to reduce the error between the determined arrival time of the second acoustic signal in the first receiver and the target arrival time. The method according to claim 1, including the method described in claim 1.

4. The method according to claim 2 or 3, wherein the first acoustic data and the second acoustic data are supplied to the machine learning model as a combined input.

5. The method according to any one of claims 1 to 4, wherein the first acoustic data and / or the second acoustic data includes data representing compression-related signal components and / or shear-related signal components, and as an optional means, the first acoustic data and / or the second acoustic data includes data representing two bent waves polarized in opposite directions.

6. For each input acoustic data, the machine learning model determines the arrival time of the acoustic signal at each receiver. The aforementioned acoustic data is separated into a noise portion and a signal portion. The transition point between the noise portion and the signal portion is determined. The transition point is output as the estimated arrival time of the acoustic signal. It is configured to be determined by, The method according to any one of claims 1 to 5.

7. A method for obtaining an estimated arrival time of an acoustic signal in an underground receiver, The method involves acquiring input data including first acoustic data, wherein the first acoustic data represents an acoustic signal recorded by a first receiver of a probe located within the underground ground volume, and the acoustic signal is formed by an active signal transmitter. In order to obtain the output of the machine learning model as the estimated value, the input data is applied to a machine learning model trained according to any one of claims 1 to 6. Methods that include...

8. The input data further includes second acoustic data, the second acoustic data representing the acoustic signal recorded by a second receiver located within the underground ground volume, The output of the machine learning model further includes an estimate of the arrival time of the acoustic signal at the second receiver. The method according to claim 7.

9. The method according to claim 8, further comprising determining the interval velocity of one or more components of the acoustic signal based on the estimated arrival times of the acoustic signal at the first receiver and the second receiver and a known distance between the first receiver and the second receiver.

10. The input data further includes second acoustic data, the second acoustic data representing a second acoustic signal recorded by the first receiver of the probe when the probe is located at different locations within the underground ground volume, The output of the machine learning model further includes an estimate of the arrival time of the second acoustic signal at the first receiver. The method according to claim 7.

11. The method according to claim 10, further comprising determining the section velocity of one or more components of the second acoustic signal based on the estimated arrival times of the first acoustic signal and the second acoustic signal in the first receiver and the known distance between the locations in the ground volume where the first receiver detected the first acoustic signal and the second acoustic signal, respectively.

12. The method further includes calculating one or more parameters related to the ground volume based on the determined interval velocity of one or more components of the acoustic signal. As an optional selection means, one or more of the above parameters are The small strain shear modulus related to the aforementioned ground volume, The small strain Poisson's ratio related to the aforementioned ground volume, The small strain Young's modulus related to the aforementioned ground volume, or One or more bulk moduli related to the aforementioned ground volume including, The method according to claim 9 or 11.

13. The method according to any one of claims 8 to 12, wherein the first acoustic data and the second acoustic data are supplied to the machine learning model as a combined input.

14. The method further includes displaying the estimated arrival times of the first and / or second acoustic signals in the first and / or second receivers on a display. The above method further includes, as an optional means, Receiving an input in the first receiver and / or the second receiver that indicates confirmation of the estimated arrival time of the first acoustic signal and / or the second acoustic signal, or Receiving an input in the first receiver and / or the second receiver that represents correction and / or rejection of the estimated arrival time of the first acoustic signal and / or the second acoustic signal. including, The method according to any one of claims 7 to 13.

15. The probe is provided in an underground borehole, and / or The probe is equipped with a P&S suspension logging (PSSL) tool or an earthquake cone penetration test (SCPT) tool, and / or The probe includes the active signal transmitter and / or The aforementioned machine learning model includes a convolutional neural network (CNN). The method according to any one of claims 1 to 14.

16. To obtain input data including the first acoustic data, Inserting the probe equipped with the first receiver into the underground ground volume, The acoustic signal is formed using the aforementioned active signal transmitter, To generate the first acoustic data, the acoustic signal in the first receiver is recorded. The method includes, and further, as an optional means, Moving the probe to a new underground depth, Repeat the above method at the new depth. including, The method according to any one of claims 7 to 15.

17. It is a system, One or more processors, and one or more memories storing computer-readable instructions configured to cause the one or more processors to perform an operation including the steps described in any one of claims 1 to 15, or When executed by one or more data processing devices, one or more computer-readable media containing instructions for causing the one or more data processing devices to perform an operation including the steps described in any one of claims 1 to 15, or A machine learning model stored in one or more computer-readable media, which is trained according to the method described in any one of claims 1 to 6, or according to claim 15 if dependent on any one of claims 1 to 6. A system that includes these features.

Images