Abstract
Real-time welding monitoring is of great importance for intelligent welding, and the recognition of penetration states based on acoustic signal has been a hot topic. However, previous studies have relied on expert knowledge to extract sound features and ignored that acoustic features have distinct effects on the penetration state. In this paper, based on convolution neuron networks (CNNs), an adaptively feature extraction model for arc sound has been proposed. The time-frequency spectrum of arc sound was obtained by short-time Fourier transform (STFT) and served as the input of proposed model. The fully connected layers are replaced with global average pooling (GAP) or global maximum pooling (GMP) to alleviate the overfitting. To research the interest of arc sound during GTAW process, the squeeze and excitation network (SE-Net) is embedded into CNNs to pay more attention to the region of interest in the time-frequency spectrum. The experimental results have demonstrated that the proposed model outperforms the conventional methods and yield high accuracy of 98.25 %. Furthermore, the interested region of arc sound under different penetration states is visualized by Grad-CAM and guided Grad-CAM. This study extends the application of attention mechanism in intelligent welding and achieved remarkable performance.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
