Automation of defect identification using large foundation models

Abstract

Presented embodiments provide a method for defect identification of a conduit used in a hydrocarbon recovery operation. In embodiments, a large base model is used to assist in automatic detection and characterization of defects, eliminating the need for expert analysis.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] none Technical Field

[0003] Aspects of the present disclosure relate to defect identification in pipelines. More specifically, aspects of the present disclosure relate to identifying defects in hydrocarbon recovery pipelines by using a large foundation model. Background Art

[0004] Steel pipes used to secure wells are susceptible to corrosion and wear factors that can have a serious impact on well stability and durability. To combat these potential effects, corrosion mapping tools (using ultrasonic or mechanical sensors – calipers) are designed to detect and capture corrosion defects of different shapes and sizes along the pipeline. However, methods still rely heavily on manual defect picking.

[0005] Identifying defects in the piping systems of hydrocarbon recovery systems is typically performed by domain experts. This is a lengthy, interpreter-dependent process. Accelerating or automating this task reduces operating costs and allows operators to focus on tasks that require more expertise.

[0006] Recently, methods using deep learning have emerged. However, the construction of this type of model is long: it requires extensive annotation of drill cuttings by domain experts (with consistency checks, etc..), design of the model, testing. These traditional methods are cumbersome and it is difficult to achieve a model that can be generalized to any type of formation (different geology) or a model that is robust to environmental changes (lighting, wet cuttings, etc.).

[0007] It would be desirable to provide an apparatus and method that is easier to operate than conventional apparatus and methods and provides defect identification, thereby minimizing the extensive use of expert analysis.

[0008] There is also a need to provide apparatus and methods that do not have the above-mentioned disadvantages (ie, long and complex review times and potential for human error).

[0009] There is also a need to reduce the economic costs of the above-mentioned conventional tool operations and equipment while providing rapid analysis of field conditions.

[0010] There is also a need to provide on-sight identification of defects without the lag time of acquiring the data and analyzing it in the laboratory. Summary of the invention

[0011] Thus, the manner in which the above-described features of the present disclosure may be understood in detail, a more specific description of the present disclosure, briefly summarized below, may be obtained by reference to the embodiments, some of which are illustrated in the accompanying drawings. It is to be noted that the accompanying drawings show only typical embodiments of the present disclosure; therefore, they should not be considered as limiting the scope thereof, as the present disclosure may allow for other equivalent embodiments that are not specifically described. Therefore, the following summary only provides a few aspects of the description and should not be used to limit the described embodiments to a single concept.

[0012] In one embodiment of the present disclosure, a method for automatic defect identification is disclosed. The method may include receiving a set of data related to pipeline formation. The method may also include preparing a visual representation of the data. The method may also include placing at least one positive data point on the visual representation. The method may also include placing at least two negative data points on the visual representation. The method may also include feeding the data of the at least one positive data point, the at least two negative data points, and the visual representation of the data into a large base model. The method may also include generating a mask with the large base model, wherein the mask identifies defects in the data.

[0013] In another example embodiment, an article of manufacture configured to include a set of instructions readable by a computer is disclosed. The article of manufacture is configured on a non-volatile memory. The set of instructions is configured to perform receiving a set of data related to pipeline formation and preparing a visual representation of the data. The set of instructions is also configured to place at least one positive data point on the visual representation. The set of instructions is also configured to place at least two negative data points on the visual representation. The set of instructions is also configured to feed the data of the at least one positive data point, the at least two negative data points, and the visual representation of the data into a large base model. The set of instructions is also configured to generate a mask using the large base model, wherein the mask identifies defects in the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] So that the manner in which the above-mentioned features of the present disclosure can be understood in detail, a more particular description of the present disclosure, briefly summarized below, can be obtained by reference to the embodiments, some of which are illustrated in the accompanying drawings. It should be noted, however, that the drawings illustrate only typical embodiments of the present disclosure; and therefore should not be considered limiting of its scope, as the present disclosure may admit to other equally effective embodiments.

[0015] Figure 1A and Figure 1B Thickness and radius plots of USIT tool and UCI tool with manufacturing patterns, respectively.

[0016] Figure 2A and Figure 2BThickness plots (along the long axis) of a USIT tool and a UCI tool with a weld, respectively.

[0017] Figure 3 is a graph of a pipeline in good condition with a manufacturing pattern.

[0018] Figure 4 is a diagram of the pipeline showing defects and manufacturing patterns.

[0019] Figures 5A-5D are examples of patches from the THBK image, where in 5B, three positive points are placed at three different depths, six corresponding negative points, and Figure 5C represents the mask drawn on the image, where Figure 5D is a manually created mask.

[0020] Figures 6A-6D are examples of blocks from the THBK image, where Figure 6B Three positive points are added at three different depths and 6 corresponding negative points, and among them Figure 6C A mask is drawn on the image, and Fig.6D A manually created mask was drawn on the image.

[0021] Fig. 7A and Figure 7B are two examples of THBK images, where Fig. 7A The original image is shown in the middle, the manual mask is presented in the center, and the pre-processed image is shown in the right column.

[0022] Figures 8A-8C Examples of THBK images are shown, respectively, with comparisons between manual masks and fine-tuned masks. Fig. 8A , Figure 8B and Figure 8C In , the left column corresponds to the original images, the middle column corresponds to the images with manual masks, and the right column corresponds to the images with fine-tuned masks.

[0023] Fig. 9 yes Figure 8C A magnified area of ​​the image.

[0024] Fig.10 is an example of THBK images, where the left column corresponds to the original image, the middle column corresponds to the image with boxed mask, and the right column corresponds to the image with fine-tuned mask.

[0025] Fig.11 is an example of a tool display with annotations.

[0026] Fig.12is a display after adding points to the image, where the points depict the masked area around the defect to be processed (dined).

[0027] Fig.13 This is what it looks like after creating a manual mask and the corresponding painted mask.

[0028] Fig.14 is the display after fine-tuning the mask, showing the comparison between the manually created mask and the fine-tuned mask.

[0029] Fig.15 This is the appearance of the tool with annotations.

[0030] Fig.16 is an example of a method of analyzing a pipeline in an example embodiment of the present disclosure.

[0031] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements in the drawings ("Figures"). It is contemplated that elements disclosed in one embodiment may be beneficially utilized on other embodiments without specific recitation. DETAILED DESCRIPTION

[0032] Reference is made below to embodiments of the present disclosure. However, it should be understood that the present disclosure is not limited to the specifically described embodiments. On the contrary, any combination of the following features and elements, whether or not related to different embodiments, is considered to enable and practice the present disclosure. In addition, although the embodiments of the present disclosure can achieve advantages over other possible solutions and / or prior art, whether a particular advantage is achieved by a given embodiment does not limit the present disclosure. Therefore, the following aspects, features, embodiments, and advantages are merely illustrative and are not considered to be elements or limitations of the claims unless expressly defined in the claims. Similarly, references to the "present disclosure" should not be interpreted as a summary of the inventive subject matter disclosed herein, and should not be considered to be elements or limitations of the claims unless expressly defined in the claims.

[0033] Although the terms first, second, third, etc. can be used to describe various elements, components, regions, layers and / or parts here, these elements, components, regions, layers and / or parts should not be limited by these terms. These terms are only used to distinguish an element, component, region, layer or part from another region, layer or part. Terms such as "first", "second" and other numerical terms, when used here, do not mean order or sequence, unless the context clearly indicates. Therefore, without departing from the teaching of the exemplary embodiment, the first element, component, region, layer or part discussed here can be referred to as the second element, component, region, layer or part.

[0034] When an element or layer is referred to as "on another element or layer", "engaged to", "connected to", or "coupled to" another element or layer, it can be directly on another element or layer, directly engaged, connected, or coupled to another element or layer, or there may be an intermediate element or layer. On the contrary, when an element is referred to as "directly on another element or layer", "directly engaged to", "directly connected to", or "directly coupled to" another element or layer, there may be no intermediate element or layer. Other words used to describe the relationship between elements should be interpreted in a similar manner. As used herein, the term "and / or" includes any and all combinations of one or more associated listed items.

[0035] Some embodiments will now be described with reference to the accompanying drawings. For consistency, similar elements in each figure will be represented by similar numbers. In the following description, many details are set forth to provide an understanding of various embodiments and / or features. However, it will be appreciated by those skilled in the art that some embodiments can be implemented without many of these details, and many changes or modifications to the described embodiments are possible. As used herein, the terms "above" and "below", "on" and "below", "upper" and "lower", "upwards" and "downwards", and other similar terms representing relative positions above or below a given point are used in this specification to more clearly describe certain embodiments.

[0036] In some aspects, a model is presented for evaluating data from a pipeline or pipeline system to identify defects in the pipeline system. In these embodiments, the evaluation can be done by an autonomous system that identifies defects and eliminates the need for domain experts to review logs obtained from field personnel. By eliminating high-cost experts, maintenance activities for pipeline systems can be completed in a fraction of the time and cost of traditional methods.

[0037] Aspects of the present disclosure are also superior to so-called "deep learning" algorithms, which have problems recognizing differences in very fine visually depicted features. Such deep learning algorithms have the significant problem of not being able to generalize to different types of formations, and therefore each deep learning algorithm model must be created for a specific purpose. Therefore, creating site-specific models is necessary for such "deep learning" models, resulting in additional funds being spent on each project. As a result of this necessary creation, such models are inferior to the methods proposed herein.

[0038] Various aspects of the present disclosure provide a model that allows defect identification and quantification. The model can be pre-trained in some aspects of the present disclosure. If the model does not produce the expected results, the model can be retrained. In some aspects, the present disclosure proposes an alternative method of using a large base model (LFM). This large base model does not require any training (zero shot learning) and enables active annotation to manage generalization. As can be appreciated, the model can be updated by retraining when necessary.

[0039] First, data can be obtained by using an ultrasonic transducer, for example, which sends and receives signals into the pipe. The sent and received data is recorded for analysis. In one embodiment, the data can be displayed visually. The visual display of the data can be, for example, along the axis of the extended pipe. In some embodiments, the input data includes a distribution map of the thickness or radius of the pipe around its respective mean value. The values ​​obtained directly reflect the amount of metal loss or excess in the pipe. The data can be presented as an image containing corrosion and other features, such as Figure 1A and 1B shown.

[0040] In one example, the data may include different features. In this non-limiting embodiment, the data includes:

[0041] - Macro features: or "background" represents the variations in wall thickening and thinning caused by the manufacturing process, e.g. machining: by e.g. Figure 1A and 1B Molded as shown or as Figure 2A and 2B Welding shown.

[0042] -Microscopic features:

[0043] o Collar: The intermediate area (or link) between joints that contains the extrema and is irregular (due to the manufacturing process).

[0044] oDefects: Corrosion and wear (metal loss) of pipes.

[0045] - Noise: In some high-resolution tools, the data is noisy - the image has additional features that have no physical meaning (just noise dots).

[0046] For ease of explanation, Figure 3 shows a scan of a pipe in good condition. Corrosion defects can have different shapes and sizes, from pits, local defects to axial grooves, such as Figure 4 shown.

[0047] Aspects of the present disclosure provide outputs of the disclosed methods that allow identification of anomalies. The disclosed results will be:

[0048] -First, these anomalies are detected and discriminated from all other features described previously and their locations are provided in the map.

[0049] - Secondly, calculate their dimensions: axial length, circumferential width and penetration depth.

[0050] The method steps involved will now be explained using examples. As will be appreciated, the method can be used not only to detect and identify anomalies associated with corrosion, but also to detect and identify other defects. Such defects may result from mechanical contact, such as wear, gouging and impact.

[0051] Annotating data for visualization with large-scale base models

[0052] In one embodiment, a large base model is used to generate a mask that depicts the groove area in the image. The large base model can be a pre-trained model. The pre-trained model can allow the analysis of visual data to generate a mask that depicts the groove area in the image using only a few mouse clicks. These masks are created for use in other deep learning applications and will serve as the ground truth of the model. In one embodiment, creating a mask involves the following steps:

[0053] 1) When you click with the left mouse button, positive points are added (yellow; positive points are points within the area to be segmented).

[0054] 2) Right click: Add Negative Points (red; negative points are points in the area outside the defect). Negative points can be used to restrict the area where the defect should be segmented. This provides local constraints for the analysis.

[0055] 3) Backspace key is used to delete all points (negative and positive).

[0056] 4) Enter key is used for splitting.

[0057] By using the above annotation method, most of the analysis tasks are completed. In the hypothetical method of the image, the amount of annotation will be 3×3=9 clicks (3 clicks per depth, because the visualization uses 1 positive point and 2 negative points). As mentioned above, Figures 5A-5D and Figures 6A-6D As shown in Figure 1, masks are generated for defect identification in images, which significantly reduces the manual annotation effort and time required to identify relevant pixels. Typically, this task is very sensitive and requires a lot of time to complete accurately. With the approach described here using a large base model, the same task is performed faster and with higher accuracy compared to manually created masks. Figures 6A-6DAn example large-scale base model is shown Figure 6C How the created mask compares to the manually created mask Fig.6D More accurate, especially at the bottom of the image where no defects exist. In summary, this technique offers several benefits, including faster and more efficient creation of masks for groove defects, reduced work and effort, and improved accuracy of defect detection.

[0058] Annotations using existing ground truth

[0059] As mentioned above, it is assumed that the logs are not interpreted. In another application of the method, the logs that have already been interpreted can be further enhanced. Such annotations can increase the consistency of the annotations for recurring defects in the existing database. Therefore, the entire part of the database containing the logs can be checked quickly and efficiently. Such activities are beneficial because the cost of inspection is expensive from a monetary and time perspective. Note that:

[0060] Annotation is user-dependent and requires significant expertise; and

[0061] The model is very sensitive to the quality of annotations.

[0062] In this example method, we examine imprecisely annotated data where we manually define masks for the images. The large base model is used to fine-tune these annotations. In these embodiments, the manual masks are used as a set of prior information for the locations of the grooves and will be used as starting points for creating the negative and positive points of the model. This is similar to the method described above. In this method, however, instead of manually selecting these points, the manual masks will determine the locations of the negative and positive points. The manual masks are used to pre-process the images to improve the performance and accuracy of the large base model. The details of the method are explained below.

[0063] In the preprocessing step, the magnified area around the defect is kept and the background (areas not detected in the manual mask) is replaced with its median value. This preprocessing step reduces the influence of the background intensity value on the defect area because this area can share intensity values ​​with the background. In the second step, the line containing the defect is normalized using a min-max normalizer. The second step locally highlights the defect and creates a larger gap between the intensity values ​​of the defect and the background. Fig. 7A and Figure 7B shows how this method can better highlight defects in an image, even if they are barely visible to the naked eye ( Fig. 7A ). This approach allows to create better contrast between defects and background and will help large base models detect grooves even in the most difficult cases.

[0064] The process of generating the mask can be done in segments or blocks. To achieve a compromise between computational time and performance, a block size of 50 rows is selected. A larger or smaller number of rows can be selected. This selection can be made by user input or can be done by artificial intelligence. To apply LFM, positive and negative points are next selected. In an embodiment, five positive points and ten negative points are selected. The number of positive and negative points can vary. The five positive points are selected from the rows with detections in the manual mask. In this row, the point with the lowest intensity value in the grayscale range is selected because the defect has the lowest intensity. The negative points are selected from the background area on the same row with a margin of 3 pixels from the end of the mask. The output of the model is shown below. Figures 8A-8C The created mask is a fine-tuned version of the manual mask and it represents the grooves in the image with higher accuracy.

[0065] When zoomed in Figure 8C and( Fig. 9 ), the masks showed significantly better accuracy than user-created masks, demonstrating the model's capabilities. This technique proved useful in minimizing the errors and biases typically associated with human-created annotations.

[0066] Manual selection box annotation

[0067] In some method steps or previous activities, a poor initial annotation may have occurred, such as from manual annotation. The disclosed aspects, however, are able to take a defective initial annotation and use a mask generated from a box created on the video screen to provide further annotation. In these embodiments, the mask is created using manually selected points around the defect, forming a polygon. Fig.10 As shown, the model is able to produce satisfactory results for detecting groove defects.

[0068] Markup tools

[0069] In an embodiment, the annotation tool combines all of the above techniques. It is a visualization tool that allows creating masks using a large base model at a specific depth and window size. Another option is to allow the user to inspect all depths of the image and create masks for defects by scrolling the image up and down.

[0070] In an embodiment, four main buttons and a visualization area are created for the tool. Fig.11 An example of a display as described above is presented.

[0071] Another feature is added to the display making it clickable. By clicking on this plot, the user can read and add points at the clicked location. These points are used to create an enlarged mask around the defect. The latter will serve as the initialization of the mask that will be fine-tuned using LFM, following the same procedure as described in the previous section, which is also the purpose of this tool.

[0072] In an embodiment, the user clicks points in a clockwise direction because the polygon is defined by ordered points. To facilitate the annotation process, the above four buttons are provided in the annotation tool:

[0073] o The Delete Point button allows the user to delete points if they want to start over or if they made a mistake when selecting points.

[0074] o The "Create Mask" button helps the user to create a mask from the clicked points, assuming there are at least four selected points, otherwise an error will occur. The corners of the mask will be defined by the selected points. Fig.13 As shown, the mask will be displayed superimposed on the image.

[0075] oBy clicking on the "Fine-tune Mask" button, the user can generate a fine-tuning mask using the model, following the procedure described previously. The display will show two separate plots. The right plot will show the image with the manual mask, and the left plot will show the image with the fine-tuning mask, as shown in Fig.14 shown.

[0076] oFinally, after scrolling along the image and creating the mask, the user can save the mask by clicking on the “Save Mask” button which saves the created mask in NumPy format. It will be saved in the same repository with the name “mask.npy”.

[0077] refer to Fig.16, a method 1600 according to an example embodiment of the present disclosure is shown. As will be appreciated, the method may be performed on a personal computer, a server, or a cloud-based computer device. The method includes receiving a set of data related to pipe formation at 1602. In some embodiments, as a non-limiting example, the received data may be ultrasonic data. In other embodiments, the data may be previously completed analytical logging that has been annotated by an expert. The set of data may be stored in a digital format. The storage may be performed locally, or the computer device may be connected to the Internet or a network for receiving data from a remote location. At 1604, the method continues by preparing a visual representation of the data received at 1602. In a non-limiting example embodiment, the visual representation may include a display on a computer monitor. At 1606, the method proceeds to placing a set of positive data points on the visual representation of the data. The set may be a single point or a plurality of points. At 1608, the method further includes placing a set of negative data points on the visual representation of the data. As will be appreciated, both method steps 1606 and 1608 may be performed by an operator examining the data of the visual representation. As described above, the points may be placed by using an annotation tool. In another embodiment, the placement of positive and negative data points can be accomplished using an artificial intelligence computer program. At 1610, the method can further include feeding the positive data points, negative data points, and the visual representation of the data into a large base model. At 1612, the method continues to generate a mask with the large base model, wherein the mask identifies defects in the data. The defects in the data may be related to corrosion or mechanical damage. As will be appreciated, the method can also include, for example, deleting points that have been placed on the visual representation if they are incorrectly positioned. The mask can be saved to a non-volatile memory for storage and further checked at the end of the method. The storage can be performed on different media, such as an optical disk, a universal serial bus device, a computer memory, a server memory, or other similar devices.

[0078] Other embodiments are possible with the method 1600 performed as described above, wherein the method also includes the ability to fine tune previously created masks to achieve better mask accuracy. For example, through experience, it has been determined that in certain designs, a particular defect, namely corrosion at a joint, occurs at a particular length of the pipe. Many previous masks may not be consistently accurate and show defects. The ability to fine tune and recreate a mask that will show defects would be of great value.

[0079] In an embodiment, the positioning of the negative point allows for the analysis to be limited. This limitation of the analysis can greatly speed up the processing of the overall results, where data points outside the boundaries generated by the negative point can be excluded from further analysis. To this end, since the total data set for analysis is limited, the placement of the negative point can substantially improve the efficiency of the calculation. Experience can be used to place the negative point. For example, in some cases, it is clear that in repeated data generation events, defects greater than 1 inch wide have never been encountered. To this end, the placement of the negative point can easily limit the total amount of data analyzed to one inch negative from the positive indicator point and one inch positive to the positive side of the positive data point.

[0080] Next, example embodiments of the present disclosure are presented. The example embodiments presented should not be considered restrictive. In one example embodiment, a method for automatic defect identification using a large base model is shown. The identified defects are located inside or on a pipeline system. In some non-limiting aspects, the pipeline system can be used as a means of transport for hydrocarbon materials. The embodiments can also be used for non-hydrocarbon fluids, such as water pipes or chemical fluid pipelines used in refineries and general industrial processes. In one embodiment of the present disclosure, a method for automatic defect identification is disclosed. The method may include receiving a set of data related to pipeline formation. The method may also include. The method may also include preparing a visual representation of the data. The method may also include placing at least one positive data point on the visual representation. The method may also include placing at least two negative data points on the visual representation. The method may also include feeding the data of at least one positive data point, at least two negative data points, and the visual representation of the data into a large base model. The method may also include generating a mask with the large base model, wherein the mask identifies defects in the data.

[0081] In another example embodiment, the method may be performed wherein the received data is ultrasound data.

[0082] In another example embodiment, the method may be performed where the data is a previously completed analytical well log that has been previously annotated.

[0083] In another example embodiment, the method may be performed wherein the data is in a digital format.

[0084] In another example embodiment, the method may be performed wherein the data is stored on one of a non-volatile device, a computer server, a personal computer, and a network.

[0085] In another example embodiment, the method may be performed wherein the visualization is represented on a computer monitor.

[0086] In another example embodiment, the method may be performed wherein at least one positive data point is placed or at least two negative data points are placed by a labeling tool.

[0087] In another example embodiment, the method may be performed wherein placing the at least one positive data point or placing the at least two negative data points is by an annotation tool, by using an artificial intelligence computer program.

[0088] In another example embodiment, the method may be performed wherein at least one positive data point is placed or at least two negative data points are placed by a labeling tool.

[0089] In another example embodiment, the method may be performed wherein the defects in the data are related to one of corrosion on the pipeline or mechanical damage to the pipeline.

[0090] In another example embodiment, the method may further include deleting at least one of the positive points and the negative points.

[0091] In another example embodiment, the method may further include saving the mask generated by the large base model.

[0092] In another example embodiment, the method may be performed wherein saving the masks generated by the large base model is saving to a memory device.

[0093] In another example embodiment, an article of manufacture configured to include a set of instructions readable by a computer is disclosed. The article of manufacture is configured on a non-volatile memory. The set of instructions is configured to perform: receiving a set of data related to pipeline formation, and preparing a visual representation of the data. The set of instructions is also configured to place at least one positive data point on the visual representation. The set of instructions is also configured to place at least two negative data points on the visual representation. The set of instructions is also configured to feed the data of the at least one positive data point, the at least two negative data points, and the visual representation of the data into a large base model. The set of instructions is also configured to generate a mask with the large base model, wherein the mask identifies defects in the data.

[0094] In another example embodiment, the article of manufacture may be configured wherein the set of instructions is further configured to cause processing to be performed if the received data is ultrasonic data.

[0095] In another example embodiment, the article of manufacture may be configured wherein the set of instructions is further configured wherein the data is a previously completed analytical log that has been previously annotated.

[0096] In another example embodiment, the article of manufacture may be configured wherein the set of instructions is further configured to use data in a digital format.

[0097] In another example embodiment, the article of manufacture may be configured wherein the set of instructions is further configured to use the data when the data is stored on one of a non-volatile device, a computer server, a personal computer, and a network.

[0098] The foregoing description of the embodiments is provided for the purpose of illustration and description. It is not intended to be exhaustive or to limit the present disclosure. The individual elements or features of a particular embodiment are generally not limited to that particular embodiment, but where applicable, they are interchangeable and can be used in selected embodiments even if not specifically shown or described. They can be varied in many ways. Such variations should not be considered as departing from the present disclosure, and all such modifications are intended to be included within the scope of the present disclosure.

[0099] Although embodiments have been described herein, those skilled in the art having benefit of this disclosure will recognize that other embodiments may be conceived without departing from the scope of this disclosure. Therefore, the scope of this claim or any subsequent claims should not be unduly limited by the description of the embodiments described herein.

Claims

1. A method for automatic defect identification, comprising: receiving a set of data related to pipeline formation; prepare visual representations of that data; Place at least one positive data point on the visualization; Place at least two negative data points on the visualization; feeding data of at least one positive data point, at least two negative data points, and a visual representation of the data into a large base model; A large base model is used to generate masks, where the masks identify defects in the data. The method of claim 1 , wherein the received data is ultrasonic data.

3. The method of claim 1, wherein the data is previously completed analytical logs that have been previously annotated.

4. The method of claim 1, wherein the data is in digital format.

5. The method of claim 4, wherein the data is stored on one of a non-volatile device, a computer server, a personal computer, and a network. The method of claim 1 , wherein the visual representation is on a computer monitor.

7. The method according to claim 1, wherein the placement of at least one positive data point or the placement of at least two negative data points is performed by a labeling tool.

8. The method of claim 1, wherein the placement of at least one positive data point or the placement of at least two negative data points is performed by a labeling tool and an artificial intelligence computer program.

9. The method according to claim 1, wherein the placement of at least one positive data point or the placement of at least two negative data points is performed by a labeling tool.

10. The method of claim 1, wherein the defects in the data are related to one of corrosion on the pipeline or mechanical damage to the pipeline.

11. The method of claim 1, further comprising deleting at least one of the positive points and the negative points.

12. The method of claim 1 , further comprising: saving a mask generated by the large base model. The method of claim 12 , wherein saving the mask generated by the large base model is saving to a memory device.

14. An article of manufacture configured to contain a set of computer-readable instructions, the article of manufacture being configured on a non-volatile memory, the set of instructions comprising a method written to: receiving a set of data related to pipeline formation; prepare visual representations of that data; Place at least one positive data point on the visualization; Place at least two negative data points on the visualization; feeding data of at least one positive data point, at least two negative data points, and a visual representation of the data into a large base model; A large base model is used to generate masks, where the masks identify defects in the data. The method of claim 14 , wherein the received data is ultrasound data.

16. The method of claim 14, wherein the data is a previously completed analytical well log that has been previously annotated.

17. The method of claim 14, wherein the data is in digital format.

18. The method of claim 17, wherein the data is stored on one of a non-volatile device, a computer server, a personal computer, and a network.

19. A method as shown and described.

20. Apparatus as shown and described.

Images