Multi-Task Learning in Eddy Current Testing of Spherical Tank Arrays: Defect Detection and Classification

With the development of industrial testing technology, eddy current testing of spherical tank arrays has been widely used as a non-destructive testing method for identifying and evaluating material defects. Traditional eddy current testing methods still face many challenges in terms of robustness and detection efficiency. To address these issues, this article proposes a multi-task learning (MTL)-based eddy current detection method aimed at simultaneously improving the performance of defect detection and classification. The strategy of shared feature learning and joint training can effectively capture feature information of different defect types. The detection accuracy and classification accuracy of this method on the dataset are significantly better than traditional techniques, especially in the recognition of small defects. The model exhibits strong robustness, reaching up to 92.9%, and maintains stable detection performance under different environmental conditions. Through an automated detection process, this method effectively reduces the interference of human judgment and improves overall detection efficiency, reaching up to 86%. The eddy current detection method based on MTL provides new ideas for the development of intelligent detection technology in the future, and has important practical application value and promotion potential.