Deep learning methods for wellbore pipe inspection

1. A method for inspecting tubular integrity comprising:

conveying an electromagnetic pipe inspection tool inside an innermost tubular of multiple nested tubulars, wherein the electromagnetic pipe inspection tool has one or more transmitters and one or more receivers;

taking measurements of the multiple nested tubulars with the electromagnetic pipe inspection tool;

inverting the measurements for a tubular integrity property of each individual tubular of the multiple nested tubulars to provide inverted tubular integrity properties;

arranging the inverted tubular integrity properties into an inverted image representative of an estimated tubular integrity property of each individual tubular; and

feeding the inverted image to a pre-trained deep neural network (DNN) to produce a corrected image,

wherein the DNN comprises at least one convolutional layer, and

wherein the corrected image comprises a representation of a corrected tubular integrity property of each individual tubular of the multiple nested tubulars.

2. The method of claim 1 , wherein the inversion is based on a radial one-dimensional (R1D) inversion model.

3. The method of claim 2 , further comprising correcting, via the DNN, inversion artifacts introduced by the R1D inversion model.

4. The method of claim 1 , wherein taking measurements of the multiple nested tubulars with the electromagnetic pipe inspection tool comprises

transmitting electromagnetic fields at one or more frequencies with the one or more transmitters; and

measuring at least one of a real-part, an imaginary-part, an absolute, an amplitude, and a phase of a received signal at the one or more frequencies with the one or more receivers.

5. The method of claim 1 , wherein taking measurements of the multiple nested tubulars with the electromagnetic pipe inspection tool comprises

exciting the multiple nested tubulars with pulsed electromagnetic fields with the one or more transmitters; and

measuring a decay response of the pulsed electromagnetic fields in the time domain with the one or more receivers.

6. The method of claim 1 , wherein the tubular integrity property comprises a cross-sectional thickness, a magnetic permeability, an electrical conductivity, or a combination thereof.

7. The method of claim 1 , wherein a value assigned to each pixel in the corrected image is proportional to a percentage change of the tubular integrity property of each of the individual tubulars of the multiple nested tubulars from a nominal tubular integrity property of each of the individual tubulars of the multiple nested tubulars.

8. The method of claim 1 , wherein feeding the inverted image to the pre-trained DNN comprises splitting the inverted image into sections based on depth.

9. The method of claim 1 , wherein the pre-trained DNN further comprises at least one of a concatenation layer, a summation layer, a max pooling layer, an up-sampling layer, and a dense layer.

10. The method of claim 1 , further comprising training the DNN to provide the pre-trained DNN,

wherein training the DNN comprises building a database by using at least one of measurements of known cases and simulation,

wherein the database includes a plurality of samples, and

wherein each sample of the plurality of samples comprises a true image of the tubular integrity property of each of the individual tubulars of the multiple nested tubulars and a corresponding inverted image.

11. The method of claim 10 , wherein training the DNN further comprises finding optimum network parameters to minimize a misfit between output images produced by the DNN and corresponding true images according to an error metric.

12. The method of claim 1 , further comprising training the DNN to provide the pre-trained DNN,

wherein training the DNN comprises

building a database by using at least one of measurements of known cases and simulation, and

finding optimum hyperparameters for inversion,

wherein the database includes a plurality of samples, and

wherein each sample of the plurality of samples comprises a true image of the tubular integrity property of each of the individual tubulars of the multiple nested tubulars and corresponding raw measurements of the multiple nested tubulars.

13. The method of claim 1 , wherein taking measurements of the multiple nested tubulars with the electromagnetic pipe inspection tool comprises taking azimuthal measurements of the multiple nested tubulars using the electromagnetic pipe inspection tool.

14. The method of claim 13 , wherein the inverted image comprises a three-dimensional (3D) inverted image, and

wherein the 3D inverted image has depth as a first dimension, azimuth as a second dimension, and the inverted tubular integrity property as a third dimension.

15. The method of claim 13 , wherein the corrected image comprises a 3D representation of the tubular integrity property of each the individual tubulars of the multiple nested tubulars.

16. The method of claim 13 , wherein the convolutional layer comprises a convolutional filter, and wherein the convolutional filter is 3D filter.

17. One or more non-transitory computer-readable media comprising program code for inspecting tubular integrity, the program code to:

initiate measurements of multiple nested tubulars with an electromagnetic pipe inspection tool conveyed inside an innermost tubular of the multiple nested tubulars;

perform an inversion on the measurements for a tubular integrity property of each individual tubular of the multiple nested tubulars to provide inverted tubular integrity properties;

arrange the inverted tubular integrity property into an inverted image representative of an estimated tubular integrity property of each individual tubular; and

feed the inverted image to a pre-trained DNN to produce a corrected image,

wherein the DNN comprises at least one convolutional layer, and

wherein the corrected image comprises a representation of a corrected tubular integrity property of each individual tubular of the multiple nested tubulars.

18. The computer-readable media of claim 17 , wherein the tubular integrity property comprises a cross-sectional thickness, a magnetic permeability, an electrical conductivity, or a combination thereof.

19. The computer-readable media of claim 17 , wherein a value assigned to each pixel in the corrected image is proportional to a percentage change of the tubular integrity property of each of the multiple nested tubulars from a nominal tubular integrity property of each of the multiple nested tubulars.

20. A system comprising:

an electromagnetic pipe inspection tool disposed inside an innermost tubular of multiple nested tubulars;

a pre-trained DNN comprising at least one convolutional layer;

a processor; and

a computer-readable medium having program code executable by the processor to:

initiate measurements of the multiple nested tubulars with the electromagnetic pipe inspection tool conveyed inside the innermost tubular;

perform an inversion on the measurements for a tubular integrity property of each individual tubular of the multiple nested tubulars to provide inverted tubular integrity properties;

arrange the inverted tubular integrity properties into an inverted image representative of an estimated tubular integrity property of each individual tubular; and

feed the inverted image to the pre-trained DNN to produce a corrected image,

wherein the corrected image comprises a representation of a corrected tubular integrity property of each individual tubular of the multiple nested tubulars.