Abstract: In order to enhance the sealing of car compartments, a commonly employed method in the past involved using a short-cycle bolt welding approach to secure bolts and reduce perforation count. However, the prevalent destructive testing method suffered from drawbacks such as low detection efficiency and an inability to identify welding joint issues during service. Addressing these issues, it proposed a bolt welding quality inspection scheme that combines dynamic welding process parameters with weld seam images. To overcome the insufficient number of unqualified samples in dynamic parameters, a distance-weighted data augmentation algorithm was introduced. Additionally, a deep learning model incorporating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) was designed to achieve bolt welding quality inspection. Experimental results validated the effectiveness of the proposed scheme, demonstrating a 95.33 % accuracy in defect bolt identification. Hence, this short-cycle bolt welding quality inspection method, based on dynamic parameters and weld seam images, showcases high detection precision and provides an effective quality inspection approach for practical engineering applications.

Key words: short-cycle stud welding, data augmentation, deep learning, quality inspection