An automatic welding defect location algorithm based on deep learning

Welding production has a pivotal role in the modern manufacturing industry. However, welding defects are frequently generated during the complex welding production process which will bring a certain effect to the welding quality. Therefore, the issue of welding defect detection has received considerable critical attention. However, traditional methods, based on handcrafted features or shallow-learning techniques could only detect welding defects under specific detection conditions or priori knowledge. In this paper, to serve the evaluation of the harmfulness of welding defects to different objects, based on the strong feature expression ability of deep learning, an automatic welding defect location method is proposed based on the improved U-net network from digital X-ray images which includes data augmentation and welding defect location. To acquire better location performance, the data augmentation is realized to enlarge the data set of welding defects to serve the network training. On the basis, a defect location method based on the improved U-net network is proposed to realize automatic and high-precision welding defect location. Experiments show that the proposed method could acquire the detection precision up to 88.4% on the public data set (GDXray Set) which shows a remarkable location performance compared with other related detection methods.