Defect size evaluation in eddy current testing through wavelet transform and pseudo-color enhancement

Eddy current testing (ECT) is a method used to detect defects in conductive materials. The estimation of defect sizes from detected signals plays a crucial role in evaluating product quality and guiding production processes. However, the determination of defect sizes during testing is susceptible to variations in lift-off distances, electrical noise, and often relies on human expertise. This paper evaluates defect sizes using a convolutional neural network regression model based on the impedance signals of defects. First, a through-type eddy current testing system is constructed, resembling those used in industrial production, to obtain high-quality eddy current testing data. Secondly, data are partitioned to locate defect signals, which are then converted into time-frequency matrices using continuous wavelet transform. A pseudo-colour enhancement algorithm is utilized to transform matrices into RGB images and preserve discernibility of defect information. Finally, regression models are constructed to estimate defect sizes. Experimental results demonstrate that this method can accurately evaluate defect sizes with a resolution of 0.05mm, achieving an accuracy of 96.02% with an error margin of ±0.01mm. For defect sizes ranging from 0.3mm to 1.0mm, the mean absolute error in defect estimation is 3.0201×10^-3mm.