2018 Volume 59 Issue 8 Pages 1348-1353
This study proposes an inverse algorithm to evaluate the depth of full circumferential wall thinning from low-frequency electromagnetic field signals. Finite element simulations were performed and revealed that normalizing the magnetic fields enabled the elimination of the effects of the length and the position of wall thinning and the dimensions of the pipe under test. Subsequently, this study attempted to evaluate the residual wall thickness on the basis of the maximum normalized magnetic field using Gaussian process regression. Forty-eight signals obtained by numerical simulations were used to develop a regression model. Eighteen further signals obtained by numerical simulations and nine signals obtained by experiments using a carbon steel pipe (diameter 101.6 mm and thickness 5.7 mm) were used for validation. The root mean squared error of the validation was 0.07 mm for the simulated signals and 0.23 mm for the signals obtained by the experiment. The 95% confidence intervals of the predictions, which would contribute to probabilistic risk analysis, were approximately 0.5 mm.
