Abstract
A Vector Network Analyzer (VNA) is used to reduce errors in equipment for Eddy Current Testing (ECT). After performing a standard 1-port calibration in the magnetic domain the measurements become practically free from drift and systematic errors and therefore easier to interpret. In addition, the linear equations used for VNA calibration is utilized to realistically randomize training data for machine learning. After training the AI algorithm accurately detects the thickness of duplex coatings used for corrosion protection of carbon steel substrates. The accuracy in terms of standard deviation relative to full scale is better than 13% without the need for tedious recalibration. The method becomes fast and well adapted for robot-assisted use. In addition, the method will be put to test for weld inspection, crack detection and wall thickness measurements.
