Bayesian data fusion of eddy current testing for flaw characterization with uncertainty evaluation

Abstract

This study proposed a data fusion method based on Bayesian estimation for flaw characterization using eddy current signals and evaluated its applicability using measured signals. The proposed method can fuse several measured data to define the likelihood function as multiplication of probabilistic distributions of eddy current signals and calculate posterior distribution based on the assumption that all signals are independent. Rectangular slits on austenitic stainless-steel plates were measured by plus-point and uniform eddy current probes with three frequencies: 50 kHz, 200 kHz, and 400 kHz. All combinations of probes and frequencies were utilized to estimate flaw size distribution. The results indicate that the proposed method can accurately evaluate the flaw size from the eddy current signals and probabilistically evaluate the error. Furthermore, variations in the importance of each data were found in the comparison of all combinations of probes and frequencies; therefore, the probabilistic sizing method could be used to determine the proper measurement conditions from the viewpoints of uncertainty of the evaluation.