{"title":"WeldNet: a lightweight deep learning model for welding defect recognition","authors":"Rongdi Wang, Hao Wang, Zhenhao He, Jianchao Zhu, Haiqiang Zuo","doi":"10.1007/s40194-024-01759-9","DOIUrl":null,"url":null,"abstract":"<div><p>Weld defect detection is an important task in the welding process. Although there are many excellent weld defect detection models, there is still much room for improvement in stability and accuracy. In this study, a lightweight deep learning model called WeldNet is proposed to improve the existing weld defect recognition network for its poor generalization performance, overfitting, and large memory occupation, using a design with a small number of parameters but with better performance. We also proposed an ensemble-distillation strategy in the training process, which effectively improved the accuracy rate and proposed an improved model ensemble scheme. The experimental results show that the final designed WeldNet model performs well in detecting weld defects and achieves state-of-the-art performance. Its number of parameters is only 26.8% of that of ResNet18, but the accuracy is 8.9% higher, while achieving a 24.2 ms inference time on CPU to meet the demand of real-time operation. The study is of guiding significance for solving practical problems in weld defect detection, and provides new ideas for the application of deep learning in industry. The code used in this article is available at https://github.com/Wanglaoban3/WeldNet.git.</p></div>","PeriodicalId":809,"journal":{"name":"Welding in the World","volume":null,"pages":null},"PeriodicalIF":2.4000,"publicationDate":"2024-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Welding in the World","FirstCategoryId":"88","ListUrlMain":"https://link.springer.com/article/10.1007/s40194-024-01759-9","RegionNum":4,"RegionCategory":"材料科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"METALLURGY & METALLURGICAL ENGINEERING","Score":null,"Total":0}
引用次数: 0
Abstract
Weld defect detection is an important task in the welding process. Although there are many excellent weld defect detection models, there is still much room for improvement in stability and accuracy. In this study, a lightweight deep learning model called WeldNet is proposed to improve the existing weld defect recognition network for its poor generalization performance, overfitting, and large memory occupation, using a design with a small number of parameters but with better performance. We also proposed an ensemble-distillation strategy in the training process, which effectively improved the accuracy rate and proposed an improved model ensemble scheme. The experimental results show that the final designed WeldNet model performs well in detecting weld defects and achieves state-of-the-art performance. Its number of parameters is only 26.8% of that of ResNet18, but the accuracy is 8.9% higher, while achieving a 24.2 ms inference time on CPU to meet the demand of real-time operation. The study is of guiding significance for solving practical problems in weld defect detection, and provides new ideas for the application of deep learning in industry. The code used in this article is available at https://github.com/Wanglaoban3/WeldNet.git.
期刊介绍:
The journal Welding in the World publishes authoritative papers on every aspect of materials joining, including welding, brazing, soldering, cutting, thermal spraying and allied joining and fabrication techniques.