Quantitative detection method for phased array ultrasonic defect images based on U-Net image segmentation algorithm

The results of phased array ultrasonic testing require the evaluation by inspectors, which poses some issues such as strong subjectivity, low efficiency, and poor reliability. To address these issues, an intelligent quantitative method for defect detection in phased array ultrasonic testing based on image segmentation algorithms was proposed. Firstly, defect images of flat-bottom holes were collected and enlarged to form a training database. Secondly, a U-Net intelligent defect segmentation model was constructed and trained to automatically segment defects from the background in the inspection images. Thirdly, a quantitative method based on the post-segmentation binary image was proposed for measuring defect dimensions. Finally, the quantitative detection capability of the U-Net model for phased array defects was verified. The results showed that the average quantitative detection error based on U-Net defect segmentation model was less than 6%, which could achieve the same quantitative detection ability as the -6 dB method, and had the advantages of high efficiency, intelligence and easy operation.