【摘要】 针对储罐底板超声导波检测数据集获取困难，数据质量难以保障等问题，提出一种基于W-GAN的超声导波数据增强方法，并使用ResNet验证其数据增强的性能。以损伤板材的导波检测数据为样本，包含无缺陷、针孔缺陷、裂纹缺陷和腐蚀缺陷4种类别。通过W-GAN对数据集进行增强，最终得到2 048条扩展数据，并对增强数据进行PSNR和Wilcoxon秩和检验。结果表明，由W-GAN生成的增强数据与原始数据具有较高的特征相似性和较强同分布性，原始数据和增强数据的分类准确率分别为96%,93%。测试结果表明，增强数据能够有效保留原始数据的损伤特征，在分类器中展现与原始数据相当的可识别性能。本研究可为储罐底板超声导波检测提供一种高效的数据增强手段。更多还原

【Abstract】 In light of the challenges in acquiring ultrasonic guided wave detection datasets for tank bottom plates and ensuring data quality,this study proposes a data enhancement method based on W-GAN(wasserstein generative adversarial network).The performance of the enhanced data is validated using ResNet.Taking the guided wave detection data of damaged plates as samples,the dataset includes four categories:non-defective,pinhole defects,crack defects,and corrosion defects.W-GAN is employed to augment the dataset,ultimately generating 2 048 extended samples,which are evaluated using PSNR and the Wilcoxon rank-sum test.The results show that the enhanced data generated by W-GAN exhibits high feature similarity and strong distribution,consistency with the original data.The classification accuracies for the original and enhanced datasets are 96% and 93% respectively.Testing demonstrates that the enhanced data effectively preserves the damage characteristics of the original data,achieving comparable recognizability in the classifier. This research provides an efficient data augmentation approach for ultrasonic guided wave detection of tank bottom plates.更多还原