Study on the improvement of defect localization accuracy of electromagnetic ultrasonic inspection system based on CNN

Electromagnetic Ultrasonic Thickness Measurement Technology has great potential in corrosion detection for pressure equipment of special devices due to its non-contact nature and high sensitivity. However, in practical applications, when dealing with complex geometries or abrupt thickness variations (such as weld seams and step junctions), electromagnetic ultrasonic thickness measurement may fail to produce results or exhibit significant errors. To address this issue, this study employs a Convolutional Neural Network (CNN) model, using the waveform data from an electromagnetic ultrasonic thickness gauge as input. The model is trained with thickness labels derived from 3D scanned data of test blocks, with the output being the thickness corresponding to the probe's center point. Experimental results demonstrate that the trained CNN model can provide more accurate thickness measurements in cases of complex geometries and abrupt thickness variations. The average measurement error is less than 0.1 mm, significantly improving the positioning accuracy of metal defect detection using electromagnetic ultrasonic testing.