Method and system for characterizing perforations in a tubular
Methods and system of characterizing perforations in a tubular. An imaging tool having an acoustic imaging probe transmits waves and receives acoustic reflections from the tubular. Image segments are processed from the acoustic data. A first trained neural network operates on the image segments to identify processed image segments containing a perforation. A second trained neural network operates on processed image segments containing a perforation to output the location of perforations along the tubular. Another trained neural network can calculate a geometric size of the perforations, such as their size or points along their contour.
1 . A method of characterizing perforations in a tubular, comprising:
generating processed image segments, based on a received acoustic data for acoustic reflections from the tubular;
convolving a first trained neural network with the processed image segments to identify processed image segments containing a perforation; and
convolving a second trained neural network on those processed image segments containing a perforation to output bounding boxes that locate the perforations along the tubular.
2 . The method of claim 1 , wherein the first trained neural network is a classification network that outputs a label estimating a probability of a given processed image segment containing a perforation.
3 . The method of claim 1 , wherein the first trained neural network comprises a Convolutional Neural Network and a Long Short Term Memory network.
4 . The method of claim 1 , wherein the first trained neural network comprises a UNET network.
5 . The method of claim 1 , wherein the first trained neural network comprises a first network part to output a feature for each of the processed image segments and a second network part that combines features for a sequence of the processed image segments to output a label that identifies perforations.
6 . The method of claim 1 , further comprising convolving a third trained neural network on a subset of the received acoustic data based on the bounding box to determine a geometric characteristic of the perforations.
7 . The method of claim 1 , wherein generating the processed image segments includes creating a single-channel image segment, where each region of the single-channel image segment includes a maximum measured intensity from a within a thickness of the tubular.
8 . The method of claim 1 , wherein generating the processed image segments includes creating a multi-channel image segment, where each region of the multi-channel image segment includes (i) a maximum measured intensity, (ii) a maximum measured intensity between an inner surface and an outer surface of the tubular, and (iii) a maximum measured intensity at the outer surface of the tubular at a corresponding azimuthal and axial location of the received acoustic data.
9 . The method of claim 1 , further comprising dividing the processed image segments into a plurality of smaller processed image segments prior to the step of determining the bounding box.
10 . The method of claim 1 , wherein the processed image segments have dimensions corresponding to azimuthal (φ) and axial (z) directions of the tubular.
11 . The method of claim 1 , wherein the processed image segments have dimensions corresponding to azimuthal (φ) and radial (r) directions of the tubular.
12 . The method of claim 1 , further comprising rendering and outputting an image from the received acoustic data based on least one of the identified processed image segments that contain a perforation.
13 . A method of characterizing a perforation in a tubular, comprising:
generating, based on a received acoustic data, an image segment containing a perforation and a corresponding bounding box around the perforation;
identifying key points of the perforation in the image segment using a trained neural network model; and
determining, based on the key points, geometric characteristics of the perforation.
14 . The method of claim 13 , generating processed image segments, based on a received acoustic data for acoustic reflections from the tubular.
15 . The method of claim 13 , wherein the image segment is a cross-view having dimensions corresponding to radial (r) and azimuthal (φ) directions of the tubular.
16 . The method of claim 13 , wherein the image segment is a tubeview having dimensions corresponding to axial (z) and azimuthal (φ) directions of the tubular.
17 . The method of claim 13 , wherein the geometric characteristics include a size of the perforation.
18 . The method of claim 13 , wherein the key points include a start point and an end point of the perforation.
19 . The method of claim 17 , wherein determining characteristics includes measuring the size of the perforation by spline fitting a start point and an end point.
20 . A system for characterizing a perforation in a tubular, comprising:
an acoustic probe comprising a plurality of ultrasonic sensors, each sensor capable of transmitting and receiving ultrasonic waves, wherein the acoustic probe is configured to generate acoustic data; at least one processor; and
at least one memory storing a first and a second trained neural network;
the at least one memory communicatively coupled with the at least one processor and storing machine-readable instructions that, when executed by the processor, cause the processor to:
generate processed image segments, based on the acoustic data;
convolve the first trained neural network with the processed image segments to identify processed image segments containing a perforation; and
convolve the second trained neural network on those processed image segments containing a perforation to output a bounding box that locates the perforation along the tubular.
21 . A system for characterizing a perforation in a tubular, comprising:
an acoustic probe comprising a plurality of ultrasonic sensors, each sensor capable of transmitting and receiving ultrasonic waves, wherein the acoustic probe is configured to generate acoustic data;
at least one processor; and
at least one memory storing a trained neural network;
the at least one memory communicatively coupled with the at least one processor and storing machine-readable instructions that, when executed by the processor, cause the processor to:
generate, based on a received acoustic data, an image segment containing a perforation;
identify key points of the perforation in the image segment using a trained neural network model; and
determine, based on the key points, geometric characteristics of the perforation.
22 . The system of claim 21 , the processor and the memory being located at a remote data center.
23 . The system of claim 21 , further comprising a user interface capable of displaying a rendered image and capable of receiving user prompts indicating which image segments contain a perforation.