Localization of Thermal Wellbore Defects Using Machine Learning

Abstract Defect detection and localization are key to preventing environmentally damaging wellbore leakages in both geothermal oil/gas applications. In this study, a multistep, machine learning approach is used localize two types of thermal defects within model. This includes comsol heat transfer simulation generate base data, neural network classify defect orientations, algorithm synthesize sensor estimations into predicted location. A small-scale physical test bed was created verify the using experimental data. The classification results were quantified these all orientations correctly. location with an average root-mean-square error 1.49 in. core contributions study as follows: (1) overall architecture, (2) use centroid-guided mean-shift clustering for localization, (3) validation quantification performance.