Detecting elbow erosion by percussion method with machine learning

Abstract

Elbow erosion monitoring is of great significance to the safety of pipeline systems and maintenance personnel. A percussion method is used to monitor pipe erosion. Under different erosion degrees, the mass reduction of the elbow is measured by high-precision electronic scale and the residual thickness of the elbow is detected by telescoping gauge. Machine learning algorithm, the K- Nearest Neighbor and recurrent network method are employed to identify the different wall thickness level and erosion mass level.

Description

DETECTING ELBOW EROSION BY PERCUSSION METHOD WITH MACHINE LEARNINGCROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional 63 / 534,079, filed August 22, 2023, which is hereby incorporated by reference as submitted in its entirety.FIELD

[0002] The present invention relates to detecting pipe erosion, more specifically a percussion approach and machine learning method for detecting elbow inner wall erosion.BACKGROUND

[0003] As material is transported through pipes, the inner walls of the pipes erode over time, affecting the integrity of the pipe. The inner wall of the pipe may erode, or thin, because of the types of liquid being transported, such as corrosive and / or abrasive liquids. A portion of the pipe, such as elbow portions or other curved portions, may be more susceptible to the effects of pipe erosion when compared to erosion of other portions, like straight pipe portions. Elbows are typically used to change the flow direction in pipeline systems. Pipeline systems are used in many industries, including oil and gas, energy, energy storage, etc. Studies have shown that the erosion of the elbow portion is about 50 times more serious than that of a straight portion. With the increase of service time, partial wall thickness of an elbow will become thinner due to erosion and wear, which may lead to piercing or bursting of the pipe, especially if the pipe is part of a high-pressure piping system.

[0004] In prior art systems, various nondestructive testing methods for pipeline defects have been used. Various nondestructive testing methods of the prior art include optical principal testing methods, radiography technology, magnetic flux leakage testing technology, and ultrasonic testing technology. However, the prior art systems have limitations, such as the undesirable requirement to install sensors or other equipment along pipelines.

[0005] What is needed in the art is an efficient and practical real-time monitoring method for the erosion of an elbow portion of a pipeline.BRIEF SUMMARY

[0006] Elbow wall thickness reduction and mass reduction caused by erosion is detected by apercussion method. Audio signals are collected while tapping pipe elbows, each of the pipe elbows having a different erosion rate. Collected sound signals are pre-processed and trained through a neural network (e.g., KNN and RNN) to create a predictive model. Erosion rates of pipe elbows are predicted using the predictive model with an accuracy above 90%. Audio signals are collected using a portable device, such as a smart phone.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] For a detailed description of numerous examples, reference will now be made to the accompanying drawings.

[0008] FIG. 1 is an exemplary system in accordance with one or more embodiments of the present disclosure.

[0009] FIG. 2 illustrates an exemplary network diagram in accordance with one or more embodiments of the present disclosure.

[0010] FIG. 3 illustrates a schematic diagram of elbow erosion by liquid-solid two-phase flow in accordance with at least one embodiment of the disclosure.

[0011] FIG. 4 illustrates an example diagram of a pipe elbow and its dimensions in accordance with one or more embodiments of the disclosure.

[0012] FIG. 5 illustrates an example environment showing an elbow and tools in accordance with one or more embodiments of the disclosure.

[0013] FIG. 6 illustrates an example table showing data of elbow erosion steps in accordance with one or more embodiments of the disclosure.

[0014] FIG. 7 illustrates an example environment in accordance with one or more embodiments of the disclosure.

[0015] FIG. 8 illustrates an example data pre-processing method and machine learning algorithm chart 800 in accordance with one or more embodiments of the disclosure.

[0016] FIG. 9 illustrates an example frequency plot diagram in accordance with one or more embodiments of the disclosure.

[0017] FIG. 10 illustrates an exemplary architecture of a recurrent neural network in accordance with one or more embodiments of the disclosure.

[0018] FIG. 1 1 illustrates an exemplary computing device in accordance with one or more embodiments of the disclosure.DETAILED DESCRIPTION

[0019] In the following description, numerous specific details are set forth to provide a thorough understanding of the various techniques. As part of this description, some of the drawings represent structures and devices in block diagram form. In this context, references to numbered drawing elements without associated identifiers (e.g., 100) refer to all instances of the drawing element with identifiers (e g., 100a and 100b). Further, as part of this description, some of this disclosure’s drawings may be provided in the form of a flow diagram. While the boxes in any flow diagram may be presented in a particular order, the flow of any flow diagram is depicted only to illustrate one exemplary embodiment. In other embodiments, any of the various steps depicted in the flow diagram may be omitted, or the steps may be performed in a different order or even concurrently. In addition, other embodiments may include additional steps not depicted as part of the flow diagram. Further, various steps described as being performed by particular modules or components may instead be performed by other other components than the ones described. The language used in this disclosure has been principally selected for readability and instructional purposes and may not have been selected to delineate or circumscribe the disclosed subject matter.

[0020] Before explaining at least one embodiment in detail, the concepts set forth herein are not limited in their application to the construction details or component arrangements set forth in the following description or illustrated in the drawings. It should also be understood that the phraseology and terminology employed herein are merely for descriptive purposes and should not be considered limiting.

[0021] It should further be understood that any one of the described features may be used separately or in combination with other features. Other embodiments of devices, systems, methods, features, and advantages described herein will be or become apparent to one with skill in the art upon examining the drawings and the detailed description herein. It’s intended that all such additional devices, systems, methods, features, and advantages be protected by the accompanying claims.

[0022] Elbows are widely used to change the flow direction in pipeline systems in many industries, including oil and gas, energy, energy storage, among others. Studies show that the erosion of theelbows is about 50 times more prevalent than the erosion of straight pipe. With the increase of the service time, the partial wall thickness of the elbow will become thinner due to erosion and wear, which may lead to piercing or bursting, particularly in high-pressure piping systems. This disclosed technology may be used to serve as a supplementary, efficient, and practical real-time monitoring method for the erosion of an elbow.

[0023] Embodiments of the disclosed technology and methods use a recording device or application (e.g., a smart phone having an app with audio recording capabilities) and a computing device (e.g., a laptop, etc.) to capture and process sound data along with a machine learning model trained to estimate pipe erosion as a function of the captured and processed sound data. Accordingly, embodiments of the disclosed technology and methods eliminate the need to install sensors (e.g., on the elbow of a pipe) to detect pipe erosion.

[0024] FIG. 1 is a simplified diagram 100 of an architecture of an elbow erosion detection system. In the embodiment of FIG. 1, the architecture includes a client device 102 in communication with a server 106 via a communications network 108. The client device 102 may be a multifunctional device, such as a mobile phone, tablet computer, personal digital assistant, wearable device, or any other electronic device that includes a media recording system that can record audio signals with a high degree of accuracy. Client device 102 may be used, for example, to record sounds generated when tapping a pipe section 104. While an elbow pipe is shown in the figure, the pipe section 104 may be a different shape, such as straight pipe section, a right angle or 90 degree pipe, a T-shaped pipe fitting, etc. Additionally, the construction of the pipe section 104 is not limited to any specific materials and may include, for example, PVC, plastic, copper, etc.

[0025] FIG. 2 is a simplified block diagram illustrating the elbow detection system 200, including the client device 102 in communication with the server 106 over the communications network 108.

[0026] The server 106 may include one or more hardware computer processors and non-transitory computer readable storage media, which may include various storage devices and may store various software modules. Although the server 106 is depicted as comprising various components in an exemplary manner, those components and the functionality described herein may be distributed across multiple devices (e.g., connected via the communications network 108), such as servers, network storage, etc. Further, additional components may be used and some combination of the functionality of any of the components may be combined. Generally, the server 106 mayinclude one or more memory devices 212, one or more storage devices 214, and one or more hardware computer processors 216, such as a central processing unit (CPU) or a graphical processing unit (GPU). Further, the server 106 may include multiple processors 216 of the same or different type. The storage 214 may include any non-transitory computer readable storage media, which may be located within a single location, or may be distributed across multiple locations, such as multiple servers 106. As described below, the server 106 may store media files, media file data, training data, historical data, audio clips, statistical data, etc. Additional data may include, for example, audio signal data, metadata, etc. The memory 212 may include one or more different types of memory, for example a cache, read only memory (ROM), and / or random-access memory (RAM). The memory 212 of the server 106 includes modules that include computer readable code executable by the processor 216 to cause the server 106 to perform various functions described herein. As depicted, the memory 212 may include a pre-processing and training module 202 and an erosion prediction module 204.

[0027] The communication network 108 may include any combination of public or private networks (e.g., local area networks, the internet, etc.), wired or wireless connections, etc.

[0028] The client device 102 may include one or more memory devices 242, one or more storage devices 244, and one or more processors 246, such as a central processing unit (CPU) or a graphical processing unit (GPU). The client device 102 may include multiple processors 246 of the same or different type. The memory 242 may include one or more different types of memory, for example a cache, read only memory (ROM), and / or random-access memory (RAM). The memory 242 may store various programming modules that include computer readable code, executable by the processor 246, to cause the client device 102 to perform various functions described herein. As depicted, the memory 242 may include a media player 252 and a recording module 254.

[0029] The recording module 254 of the client device 102 captures and records audio signal data, for example using one or more microphones included in the client device 102. The recording module 254 may be realized, for example, as a smartphone “voice record” application. Additionally or alternatively, the recording module 254 may provide a graphical user interface (GUI) to receive input from a user, for example pipe characteristics, measured erosion characteristics, metadata, etc. The graphical user interface may also provide functionality to upload audio signal data captured (e.g., captured by another device). The data collected and recorded bythe client device 102 may be transmitted to the server 106 for further processing, for example via the communications network 108, a direct (wired or wireless) communications link, removable storage, remote storage (e.g., cloud storage), etc.

[0030] The media player 252 may provide functionality to play audio content, such as the audio signal data captured by the recording module 254 of the client device 102 or received from another device.

[0031] The pre-processing and training module 202 extracts features used to generate a predictive model for predicting erosion characteristics of a portion of a pipe, such as an elbow portion. The predictive model may be created, for example, using artificial intelligence, machine learning, or a combination thereof. In one embodiment, a neural network may be used to create the predictive model. As described in detail below, the features used by the predictive model to predict the erosion characteristics may be extracted from the audio signal data captured by the recording module 254 of the client device 102. Additional features used by the predictive model may include pipe characteristics (e.g., pipe thickness, pipe shape, pipe type, pipe material) that characterize geometry and / or composition of the pipe section 104. To generate the predictive model, the preprocessing and training module 202 may be used collect training data, for example audio signal data captured while tapping on a plurality of pipe sections 104 (e.g., with known and varied pipe characteristics) having measured erosion characteristics. The training data may be captured using a portable device (e.g., the client device 102) and transmitted to the server 106 (e.g., via the communications network 108) as described above.

[0032] The erosion prediction module 204 uses the predictive model described above to predict the erosion characteristics of a pipe section 104 (e.g., residual thickness, erosion level, erosion rate, etc.) based at least in part on audio signal data captured while tapping on the pipe section 104 (as well as, in some instances, the pipe characteristics of the pipe section 104. The audio signal data (and / or the pipe characteristics) may be captured using a portable device (e.g., the client device 102) and transmitted to the server 106 (e.g., via the communications network 108) as described above.

[0033] Erosion is a serious threat to the normal use of a manifold elbow. Referring to diagram 300 of FIG. 3, fluid passes through the pipe section 104 from an inlet 304 to an outlet 308. The flow direction, as indicated by the arrows, changes sharply. Under the action of inertia force, solidparticles in the fluid impact intensively on local areas of the inner wall at a certain speed and angle, causing local material loss and local wall thickness thinning. In the example of FIG. 3, for instance, the fluid may impact an erosion area 306.

[0034] In an exemplary embodiment, percussion is performed on the pipe section 104 to assess the difference of wall thickness of pipe section 104 with K-nearest neighbor and a recurrent neural network method. In the example, a hammer 702 (e.g., as shown in FIG. 7) is used to repeatedly tap on the erosion area 306 of the pipe section 104. The sounds generated by the tapping hammer 702 are recorded, for example by the recording module 254 of the client device 102.

[0035] FIG. 4 is a diagram 400 illustrating an example pipe section 104 having example pipe characteristics. One type of pipe section 104 that may be used is an ASME B16. l l 90 degree, 1 inch SOCKET WELD END 3000# ASTM A105N, having an original wall thickness of 4.55 millimeters (or 0.179 inches). A cross-sectional view 404 of the pipe section 104 shows the dimensions of the pipe section 104 by referring to the dimensional chart 406. The example pipe section 104 shown of FIG. 4 is not meant to be limiting in any way, but merely an illustrative example. In other instances, the pipe section 104 may have differing pipe characteristics, such as differing materials (e.g., PVC, copper), sizes, wall thicknesses, curvatures, etc.

[0036] FIG. 5 is a diagram 500 illustrating exemplary measuring tools 502 that may be used to capture measured erosion characteristics in the example embodiment. A telescoping gauge and a caliper may be used to take measurements of the inner BORE ID of the pipe section 104. A plug may be used to plug into one end to set the reference original inner BORE ID. Additionally, a grinding bit may be used to grind wall thickness. A digital scale, such as a scale with 0.01g precision, may be used to measure the mass of the grinding powder. Table 600 of FIG. 6 shows measured erosion characteristics of six grindings and seven different wall thicknesses, including original wall thickness. The measured erosion characteristics shown in table 600 correlate to different levels of pipe erosion.

[0037] FIG. 7 is a diagram 700 illustrating the capture of audio signal data in the example embodiment. In the example of FIG. 7, audio signal data is captured (e.g., by the recording module 254 of the client device 102) while the pipe portion 104 is tapped by a steel hammer 702 (e.g., in a location that corresponds to the erosion area 306 described above with reference to FIG. 3).

[0038] To capture the training data used by the pre-processing and training module 202 to generate the predictive model as described above, audio signal data may be captured while tapping on pipe sections 104 (e.g., as described above with reference to FIG. 7) having various measured erosion characteristics (e.g., as described above with reference to FIGS. 5-6), for example a pipe section 104 at various levels of erosion. Additionally, audio signal data may be captured while tapping on numerous pipe sections 104 having various pipe characteristics (e.g., as described above with reference to FIG. 4) and various measured erosion characteristics. After the predictive model has been generated using the training data, the erosion prediction module 204 may use the predictive model to predict the erosion characteristics of a pipe section using audio signal data (and, in some instances the pipe characteristics of the pipe section) captured while tapping on the pipe section in a similar manner

[0039] FIG. 8 is an exemplary flow diagram 800 illustrating a data pre-processing process to create a training dataset for a machine learning algorithm. The exemplary flow diagram 800 may be implemented by the recording module 254 and the pre-processing and training module 202 of the server 106.

[0040] The flow chart begins at 802 with the tapping of a pipe section 104, for example by a hammer 702 as described above. At 804, audio signal data is recorded, for example by the recording module 254, to capture the sounds generated by the tapping at 802. As part of 804, the captured audio signal data may be transmitted from the client device 102 to the server 106, for example over the communications network 108.

[0041] In the embodiment of FIG. 8, the server 106 converts the captured audio signal data from a time domain to a frequency domain, for example using Fast Fourier Transform (FFT) at 806. Additionally, in the embodiment of FIG. 8, the frequency domain features are selected by identifying the energy levels across a number of sensitive frequency ranges (e.g., predetermined frequency ranges, frequency ranges having the highest energy levels, etc.).

[0042] At 810, measured erosion characteristics, pipe characteristics and the audio signal data (e.g., the frequency domain features) are used to train the predictive model, for example using a recurrent neural network to predict the measured erosion characteristics based on the pipe characteristics and the provided audio signal data. At 812, an accuracy percentage of the testing data may be calculated. The calculated accuracy may, for example, reflect how accurate thepredictive model is at predicting the measured erosion characteristics based on the pipe characteristics and the provided audio signal data.

[0043] FIG. 9 is a graph 900 illustrating the amplitudes of example audio signal data across the frequency domain. In this example, frequency domain features are selected by measuring the energy level across nine frequency ranges, for example 1500-2180, 2210-3120, 3250-4150, 4300- 4600, 4930-5200, 5250-5750, 5780-6300, 6500-7100, and 7400-8150. Accuracy is calculated for testing in Table I.Table I

[0044] Referring now to FIG. 10, an exemplary architecture is depicted of a recurrent neural network. In this example, a recurrent neural network may be applied. Additionally, Mel-frequency cepstral coefficients (MFCC) may be utilized as a feature representation. Each sample of the example may be represented by a matrix transformed by MFCC. Referring to FIG. 10, each row of the matrix correspond to time and the columns represent feature values. The architecture shown in FIG. 10 is merely an example implementation for a data pre-processing method including machine learning and is not meant to be limiting.

[0045] Example Case Study on Two Datasets A and B (Recurrent Neural Network Method)

[0046] In this study, two scenarios were implemented. One is to train dataset A with RNN model and test the trained model with dataset B. The other is to train dataset B with RNN model and test the trained model with dataset A. For both scenarios, a total of four experiments were implemented.The first is to keep all seven steps, the second is to use only four steps, which are SO, S2, S4, and S5 (See FIG. 6). The third is to use only three steps, which are SO, S3, and S6 (See FIG. 6). The fourth is to use only two steps, which are SO and S6 (See FIG. 6). Final results are shown in Table II.Table II

[0047] The experimental results in Table II show that for single dataset training via K-nearest neighbor method, the accuracy is above 90%. While for two dataset training via Recurrent neural network, the accuracy is only above 75%.

[0048] Another observation from two dataset training result is that accuracy is above 90% on the wall thickness different is larger than 37%. This indicates that this percussion method can be used as detecting critical wall thickness

[0049] Referring now to FIG. 11, a simplified functional block diagram of illustrative multifunction device 1100 is shown according to one embodiment. Multifunction electronic device 1100 may include processor 1105, display 1110, user interface 1115, graphics hardware 1120, device sensors 1 125 (e.g., proximity sensor / ambient light sensor, accelerometer and / or gyroscope), microphone 1130, audio codec(s) 1135, speaker(s) 1140, communications circuitry 1145, digital image capture circuitry 1150 (e.g., including camera system) video codec(s) 1155 (e.g., in support of digital image capture unit), memory 1160, storage device 1165, andcommunications bus 1170. Multifunction electronic device 1100 may be, for example, a personal computer, laptop, or a personal electronic device such as a smart phone, a personal digital assistant (PDA), mobile telephone, or a tablet computer.

[0050] Processor 1105 may execute instructions necessary to carry out or control the operation of many functions performed by device 1100 (e.g., the collection of audio signals during tapping and machine learning algorithms and modules for detecting wall thicknesses as disclosed herein). For instance, processor 1105 may execute instructions including, but not limited to, pre-processing and training module 202, erosion prediction module 204, media player 252, and / or recording module 254 of FIG. 2. Processor 1105 may, for instance, drive display 1110 and receive user input from user interface 1115. User interface 1 115 may allow a user to interact with device 1100. For example, user interface 1115 can take a variety of forms, such as a button, keypad, dial, a click wheel, keyboard, display screen and / or a touch screen. Processor 1105 may also, for example, be a system-on-chip such as those found in mobile devices and include a dedicated graphics processing unit (GPU). Processor 1105 may be based on reduced instruction-set computer (RISC) or complex instruction-set computer (CISC) architectures or any other suitable architecture and may include one or more processing cores. Graphics hardware 1120 may be special purpose computational hardware for processing graphics and / or assisting processor 1105 to process graphics information. In one embodiment, graphics hardware 1120 may include a programmable GPU.

[0051] Captured sound signals, predictive model data and training data may be stored in memory 1160 and / or storage 1165. Memory 1160 may include one or more different types of media used by processor 1105 and graphics hardware 1120 to perform device functions. For example, memory 1160 may include memory cache, read-only memory (ROM), and / or random access memory (RAM). Storage 1165 may store media (e.g., audio, image and video fdes), computer program instructions or software, preference information, device profile information, and any other suitable data. Storage 1165 may include one more non-transitory computer-readable storage mediums including, for example, magnetic disks (fixed, floppy, and removable) and tape, optical media such as CD-ROMs and digital video disks (DVDs), and semiconductor memory devices such as Electrically Programmable Read-Only Memory (EPROM), and Electrically Erasable Programmable Read-Only Memory (EEPROM). Memory 1160 and storage 1165 may be used to tangibly retain computer program instructions or code organized into one or more modules andwritten in any desired computer programming language. When executed by, for example, processor 1105 such computer program code may implement one or more of the methods described herein.

[0052] According to some embodiments, a processor or a processing element may be trained using supervised machine learning and / or unsupervised machine learning, and the machine learning may employ an artificial neural network, which, for example, may be a convolutional neural network, a recurrent neural network, a deep learning neural network, a reinforcement learning module or program, or a combined learning module or program that learns in two or more fields or areas of interest. Machine learning may involve identifying and recognizing patterns in existing data in order to facilitate making predictions for subsequent data. Models may be created based upon example inputs in order to make valid and reliable predictions for novel inputs.

[0053] According to certain embodiments, machine learning programs may be trained by inputting sample data sets or certain data into the programs, such as images, object statistics and information, historical estimates, or the like. The machine learning programs may utilize deep learning algorithms that may be primarily focused on pattern recognition and may be trained after processing multiple examples. The machine learning programs may include Bayesian Program Learning (BPL), voice recognition and synthesis, image or object recognition, optical character recognition, and / or natural language processing. The machine learning programs may also include natural language processing, semantic analysis, automatic reasoning, and / or other types of machine learning.

[0054] According to some embodiments, supervised machine learning techniques and / or unsupervised machine learning techniques may be used. In supervised machine learning, a processing element may be provided with example inputs and their associated outputs and may seek to discover a general rule that maps inputs to outputs, so that when subsequent novel inputs are provided the processing element may, based upon the discovered rule, accurately predict the correct output. In unsupervised machine learning, the processing element may need to find its own structure in unlabeled example inputs.

[0055] Reference in this disclosure to “one embodiment” or to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment, and multiple references to “one embodiment” or to “anembodiment” should not be understood as necessarily all referring to the same embodiment or to different embodiments.

[0056] The disclosed technology and methods include the following examples:

[0057] Example 1: A method for predicting erosion in a pipe section, the method comprising: receiving audio signal data captured while tapping on the pipe section; providing the audio signal data to a predictive model, generated by a machine learning algorithm trained on a dataset of audio signal data captured while tapping on pipe sections having one or more measured erosion characteristics; and predicting one or more erosion characteristics of the pipe section, by the predictive model, based on the audio signal data.

[0058] Example 2: The method of example 1, further comprising: converting the captured audio signal data from a time domain to a frequency domain; and selecting frequency features, wherein providing the audio signal data to the predictive model comprises providing the selected frequency domain features.

[0059] Example 3: The method of example 2, wherein selecting the frequency domain features comprises identifying, in the converted captured audio signal data, the energy levels across a plurality of frequency ranges.

[0060] Example 4: The method of example 1, wherein the audio signal data is captured using a portable electronic device.

[0061] Example 5: The method of example 1, further comprising: capturing the dataset of audio signal data while tapping on pipe sections having measured erosion characteristics; and training the machine learning algorithm on the captured dataset to generate the predictive model.

[0062] Example 6: The method of example 5, wherein the machine learning algorithm comprises a recurrent network.

[0063] Example 7: The method of claim 1, further comprising: providing one or more pipe characteristics characterizing the pipe section to the predictive model, wherein: the machine learning algorithm is trained on a dataset of audio signal data captured while tapping on pipe sections having known pipe characteristics, and the predictive model predicts the erosion characteristics of the pipe section based on the audio signal data and the one or more pipe characteristics of the pipe section.

[0064] Example 8: The method of claim 7, wherein the one or more pipe characteristics include pipe thickness, pipe shape, pipe type, or pipe material.

[0065] Example 9: The method of claim 1, wherein the erosion characteristics include an erosion rate of the pipe section.

[0066] Example 10: The method of claim 1, wherein the erosion characteristics include a residual thickness of the pipe section.

[0067] Example 11 : A system for predicting erosion in a pipe section, comprising: non-transitory computer readable storage media; and a hardware computer processor configured to: receive audio signal data captured while tapping on the pipe section; provide the audio signal data to a predictive model, generated by a machine learning algorithm trained on a dataset of audio signal data captured while tapping on pipe sections having one or more measured erosion characteristics; and predict one or more erosion characteristics of the pipe section, using the predictive model, based on the audio signal data.

[0068] Example 12: The system of example 11, wherein the hardware computer processor is further configured to providing the audio signal data to the predictive model comprises providing the selected frequency domain features by converting the captured audio signal data from a time domain to a frequency domain; selecting frequency features; and providing the selected frequency features to the predictive model.

[0069] Example 13: The system of example 12, wherein the hardware computer processor is configured to select the frequency domain features by identifying, in the converted captured audio signal data, the energy levels across a plurality of frequency ranges.

[0070] Example 14: The system of example 11, wherein the audio signal data is captured using a portable electronic device.

[0071] Example 15: The system of example 11, wherein the hardware computer processor is further configured to: capture the dataset of audio signal data while tapping on pipe sections having measured erosion characteristics; and train the machine learning algorithm on the captured dataset to generate the predictive model.

[0072] Example 16: The system of example 15, wherein the machine learning algorithm comprises a recurrent network.

[0073] Example 17: The system of example 11 , wherein: the hardware computer processor is further configured to provide one or more pipe characteristics characterizing the pipe section to the predictive model; the machine learning algorithm is trained on a dataset of audio signal data captured while tapping on pipe sections having known pipe characteristics; and the predictive model predicts the erosion characteristics of the pipe section based on the audio signal data and the one or more pipe characteristics of the pipe section.

[0074] Example 18: The system of example 17, wherein the one or more pipe characteristics include pipe thickness, pipe shape, pipe type, or pipe material.

[0075] Example 19: The system of example 11, wherein the erosion characteristics include an erosion rate of the pipe section.

[0076] Example 20: The system of example 11, wherein the erosion characteristics include a residual thickness of the pipe section.

[0077] While the disclosure has been described with respect to the figures, it will be appreciated that many modifications and changes may be made by those skilled in the art without departing from the spirit of the disclosure. Any variation and derivation from the above description and figures are included in the scope of the present disclosure as defined by the claims.

Claims

CLAIMSWe claim:

1. A method for predicting erosion in a pipe section, the method comprising: receiving audio signal data captured while tapping on the pipe section; providing the audio signal data to a predictive model, generated by a machine learning algorithm trained on a dataset of audio signal data captured while tapping on pipe sections having one or more measured erosion characteristics; and predicting one or more erosion characteristics of the pipe section, by the predictive model, based on the audio signal data.

2. The method of claim 1, further comprising: converting the captured audio signal data from a time domain to a frequency domain; and selecting frequency features, wherein providing the audio signal data to the predictive model comprises providing the selected frequency domain features.

3. The method of claim 2, wherein selecting the frequency domain features comprises identifying, in the converted captured audio signal data, the energy levels across a plurality of frequency ranges.

4. The method of any of the preceding claims, wherein the audio signal data is captured using a portable electronic device.

5. The method of claim 4, wherein the portable electronic device is a smartphone.

6. The method of any of the preceding claims, wherein the audio signal data is captured using a voice recording application of the smartphone.

7. The method of any of the preceding claims, further comprising: capturing the dataset of audio signal data while tapping on pipe sections having measured erosion characteristics; andtraining the machine learning algorithm on the captured dataset to generate the predictive model.

8. The method of claim 7, wherein the machine learning algorithm comprises a recurrent network.

9. The method of any of the preceding claims, further comprising: providing one or more pipe characteristics characterizing the pipe section to the predictive model, wherein: the machine learning algorithm is trained on a dataset of audio signal data captured while tapping on pipe sections having known pipe characteristics, and the predictive model predicts the erosion characteristics of the pipe section based on the audio signal data and the one or more pipe characteristics of the pipe section.

10. The method of claim 9, wherein the one or more pipe characteristics include pipe thickness, pipe shape, pipe type, or pipe material.

11. The method of any of the preceding claims, wherein the erosion characteristics include an erosion rate of the pipe section.

12. The method of any of the preceding claims, wherein the erosion characteristics include a residual thickness of the pipe section.

13. A system for predicting erosion in a pipe section, comprising: non-transitory computer readable storage media; and a hardware computer processor configured to perform the method of claim 1.

14. Non-transitory computer readable storage media storing instructions that, when executed by a hardware computer processor of a computing device, cause the computing device to perform the method of claim 1.

Images