Modeling eddy current analysis data to determine depth of weld penetration

The Applied Engineering Technology Group in the Engineering Sciences and Applications Division at Los Alamos National Laboratory is currently providing the design, engineering, assembly, and testing of an eddy current instrument for weld inspection. This instrument is designed to provide an in situ weld depth measurement of nuclear weapons primary components during fabrication. The goal of this effort is to improve the accuracy and reliability of the measurement of the weld joint depth and provide a mechanism for inspection without removing the part from the fixture. This feature is essential to accommodate the re-welding process if the inspection fails. The production system consists of a commercially available eddy current instrument and eddy current probe connected to a portable PC. The objective of the system software is acquire and analyze voltage and phase angle data to produce a near-real time estimate of weld depth. The data obtained from the instrument are perfectly suited for analysis by a neural network technique. This paper compares the effectiveness of a neural network application with traditional mathematical models for the analysis of weld depth information.