US 11,939,858 B2
Identification of wellbore defects using machine learning systems
Peng Yuan, Houston, TX (US); Jaehyuk Lee, Houston, TX (US); and Feyzi Inanc, Houston, TX (US)
Assigned to Baker Hughes Oilfield Operations LLC, Houston, TX (US)
Filed by Baker Hughes Oilfield Operations LLC, Houston, TX (US)
Filed on Dec. 9, 2020, as Appl. No. 17/116,603.
Prior Publication US 2022/0178242 A1, Jun. 9, 2022
Int. Cl. E21B 47/005 (2012.01); G06N 20/00 (2019.01)
CPC E21B 47/005 (2020.05) [G06N 20/00 (2019.01); E21B 2200/20 (2020.05); E21B 2200/22 (2020.05)] 20 Claims
OG exemplary drawing
 
1. A method for detecting a wellbore defect, comprising:
receiving, at a trained machine learning system, input data, the input data corresponding to log data acquired from a radiation detector in a nuclear logging tool for a wellbore operation in a multi-barrier wellbore;
processing, using a first classifier of the machine learning system, the input data;
identifying, using the first classifier of the machine learning system, a feature of interest associated with the input data;
receiving, at a second classifier of the machine learning system, first output data from the first classifier and the input data, the first output data corresponding to a defect associated with the wellbore;
receiving, at a third classifier of the machine learning system, second output data from the second classifier and the input data, the second output data corresponding to a first property of the defect; and
providing a report corresponding to the defect, the first property of the defect, and a second property of the defect corresponding to third output data from the third classifier.