US-12626504-B2 - Machine learning model for measuring perforations in a tubular

Abstract

A method and instruction memory for processing acoustic images of a downhole casing to determine perforations of the tubular. The images may be acquired by an acoustic logging tool deployed into cased well. A Machine Learning model is trained to recognize regions of the acoustic images that are perforations or not, in order to calculate geometric properties of the perforation and overall casing. Renderings of the imaged casing may be overlaid with contours and properties of perforations to improve perforation, fracturing and producing operations.

Inventors

• Siavash Khallaghi

Assignees

• DARKVISION TECHNOLOGIES INC.

Dates

Publication Date

20260512

Application Date

20220602

Priority Date

20210624

Claims (20)

1. 1 . A method of identifying perforations in a downhole casing from ultrasound images, the method comprising: receiving an ultrasound image of the casing; determining sub-regions of the ultrasound image that each include one perforation; convolving corresponding pixels of each sub-region with a Perforation Segmentation Model to create a perforation mask that corresponds to the pixels and their probability of being a perforation within that sub-region expressed as a pixel perforation probability value being a range between a first value indicating a background pixel and a second value indicating a perforation pixel; calculating one or more geometric properties of each perforation from each perforation mask; storing the one or more geometric properties in a datastore; and rendering a visualization of the casing to a user from the received ultrasound image overlaid with at least one of: the perforation mask or calculated geometric properties from several perforations.
2. 2 . The method of claim 1 , further comprising thresholding the perforation mask to use pixels in the perforation mask above a threshold probability for calculating the one or more geometric properties.
3. 3 . The method of claim 1 , wherein one of the geometric properties is a contour that encapsulates the perforation, the contour being a 2D contour in coordinates of azimuth and axial position along the casing.
4. 4 . The method of claim 1 , further comprising imaging the casing using a ring-shaped phased-array of ultrasound transducers moved axially through the casing while capturing transverse image frames of the casing.
5. 5 . The method of claim 1 , wherein determining the sub-regions is performed manually via a User Interface displaying a 2D image of a portion of the ultrasound image and receiving locations of perforations or boundaries of sub-regions around perforations.
6. 6 . The method of claim 1 , wherein the geometric properties calculated is a diameter or volume of the perforation.
7. 7 . The method of claim 1 , wherein the ultrasound image comprises three-dimensional data provided in polar coordinates.
8. 8 . The method of claim 1 , wherein the Perforation Segmentation Model is a Semantic model corresponding to at least one of a UNet, UNet++, or Deeplab.
9. 9 . The method of claim 1 , further comprising assembling a geometric model of the casing from the geometric properties of hundreds of perforations.
10. 10 . A system for processing ultrasound images of a downhole casing to identify perforations comprising: a memory storing a Perforation Segmentation Model; one or more datastores storing an ultrasound image of the casing; and a non-transitory computer readable medium having instructions executable by a processor to perform operations comprising: receiving the ultrasound image of the casing; determining sub-regions of the ultrasound image that each include one perforation; convolving corresponding pixels of each sub-region with the Perforation Segmentation Model to create a perforation mask that corresponds to the pixels and their probability of being a perforation within the selected sub-region expressed as a pixel perforation probability value being a range between a first value indicating a background pixel and a second value indicating a perforation pixel; calculating one or more geometric properties of each perforation from each perforation mask; storing the one or more geometric properties in the one or more datastores; and rendering a visualization of the casing to a user from the received ultrasound image overlaid with at least one of: the perforation mask or calculated geometric properties from several perforations.
11. 11 . The system of claim 10 , further comprising a User Interface providing i) a 2D display of a portion of the ultrasound image and ii) input means for tagging locations of perforations or bounding sub-regions around perforations.
12. 12 . The system of claim 10 , the instructions further performing thresholding the perforation mask to apply pixels in the mask above a threshold probability to calculate the one or more geometric properties.
13. 13 . The system of claim 10 , wherein one of the geometric properties is a contour that encapsulates the perforation.
14. 14 . The system of claim 10 , further comprising a ring-shaped phased-array of ultrasound transducers for capturing transverse image frames of the casing.
15. 15 . The system of claim 10 , wherein the geometric properties calculated is a diameter or volume of the perforation.
16. 16 . The system of claim 10 , wherein the ultrasound image comprises three-dimensional data.
17. 17 . The system of claim 10 , wherein the Perforation Segmentation Model is a Semantic model.
18. 18 . The system of claim 10 , the instructions further performing assembling a geometric model of the casing from the stored geometric properties of hundreds of perforations.
19. 19 . The system of claim 10 , wherein the visualization corresponds to the received ultrasound image being overlaid by a plurality of contours that each encapsulate a respective perforation of a plurality of perforations, each contour being a 2D contour in coordinates of azimuth and axial position along the casing.
20. 20 . The method of claim 1 , wherein the visualization corresponds to the received ultrasound image being overlaid by a plurality of contours that each encapsulate a respective perforation of a plurality of perforations, each contour being a 2D contour in coordinates of azimuth and axial position along the casing.

Description

RELATED APPLICATIONS This application claims priority to GB Application No. 2109043.6, filed on Jun. 24, 2021, which is incorporated herein by reference in its entirety. FIELD OF THE INVENTION The invention relates generally to inspection of fluid-carrying tubulars, in particular using acoustic images and machine learning to identify perforations in downhole casings. BACKGROUND OF THE INVENTION Acoustic imaging is used to log tubulars, such as pipelines and wellbores. These images are manually inspected to identify conditions and damage in the tubular. In particular, cracks, corrosion, perforations, and bursts are of interest to operators. Reflections from these features are subtly different from the surrounding area. The existence and size of some of these features can lead to fluids leaking to the environment, so they must be regularly inspected and reported. Manual inspection is incredibly time-consuming task, as it involves viewing complex, noisy 3D images over many kilometers of pipe or casing. For each candidate feature, the operator may count, locate, or infer certain measurements. They may use digital calipers via the user-interface to estimate a diameter or area of the feature. This process is also prone to error due to judgment of the operator and from one operator to another. In many cases, the features are hard for human and image processing software to identify because the images are not camera images and not visually clear to people or usable with existing image software. SUMMARY OF THE INVENTION In accordance with a first aspect of the invention there is provided a method of identifying perforations in a downhole casing from ultrasound images. The method comprises: receiving an ultrasound image of the casing; determining sub-regions of the ultrasound image that each include one perforation; convolving each sub-region with a Perforation Segmentation Model to create a perforation mask that corresponds to pixels and their probability of being a perforation within that sub-region; calculating one or more geometric properties of each perforation from each perforation mask; and storing the one or more geometric properties in a datastore. In accordance with a second aspect of the invention there is provided a system for processing ultrasound images of a downhole casing to identify perforations comprising: a memory storing a Perforation Segmentation Model; a datastore storing an ultrasound image of the casing; and a non-transitory computer readable medium. The medium has instructions executable by a processor to perform operations comprising: receiving the ultrasound image of the casing; determining sub-regions of the ultrasound image that each include one perforation; convolving each sub-region with the Perforation Segmentation Model to create a perforation mask that corresponds to pixels and their probability of being a perforation within the selected sub-region; calculating one or more geometric properties of each perforation from each perforation mask; and storing the one or more geometric properties in a datastore. Aspects may further comprises thresholding the perforation mask to use pixels in the perforation mask above a threshold probability for calculating the one or more geometric properties. One of the geometric properties may be a contour that encapsulates the perforation, preferably a 2D contour in coordinates of azimuth and axial position along the casing. Aspects may further comprises a ring-shaped phased-array of ultrasound transducers movable axially through the casing for capturing transverse image frames of the casing. Aspects may determine the sub-regions manually via a User Interface displaying a 2D image of a portion of the ultrasound image and receiving locations of perforations or boundaries of sub-regions around perforations. The geometric properties may be a diameter or volume of the perforation. The ultrasound image may comprises three-dimensional data, preferably provided in polar coordinates. The Segmentation Model may be a Semantic model, preferably a UNet, UNet++ or Deeplab. Aspects may assemble a geometric model of the casing from the geometric properties of hundreds of perforations. Aspects may render a visualization of the casing to a user from the received ultrasound image overlaid with the perforation mask and/or calculated geometric properties from several perforations. Aspects may comprise a User Interface providing i) a 2D display of a portion of the ultrasound image and ii) input means for tagging the locations of perforations or bounding sub-regions around perforations. Further aspects of the invention are set out below and in the appended claims. Thus preferred embodiments of the invention enable the device to automatically identify voids in tubulars, such as perforations and cracks, and output geometric measurements of them. BRIEF DESCRIPTION OF THE DRAWINGS Various objects, features, and advantages of the invention will be apparent from the following descri