Skip to main content
placeholder image

Mask R-CNN-Based Welding Image Object Detection and Dynamic Modelling for WAAM

Chapter


Abstract


  • As a new emerging technology, wire arc additive manufacturing (WAAM) has attracted extensive interests from both academia and industry during recent years. WAAM uses welding arc as an energy source to fuse metal wire and deposit layer by layer, which provides the advantages of freeform deposition. In order to improve its manufacture precision, stability and repeatability, it is necessary to develop sensing and control strategy for WAAM process. This research develops a passive visual sensing system for a robotic WAAM system. A new deep learning technique (Mask R-CNN) is proposed to detect and segment the melt pool area, and the width of melt pool can be measured based on the coordinate of the bounding rectangle. The pseudo-random ternary (PRT) signals were used to stimulate the WAAM process, and the corresponding width can be measured by the Mask R-CNN. Based on the width data and corresponding PRT input, a dynamic model of adaptive neuro-fuzzy inference system was built for the WAAM process.

Publication Date


  • 2020

Citation


  • Xia, C., Pan, Z., Zhang, S., Polden, J., Li, H., Xu, Y., & Chen, S. (2020). Mask R-CNN-Based Welding Image Object Detection and Dynamic Modelling for WAAM. In Transactions on Intelligent Welding Manufacturing (pp. 57-73). doi:10.1007/978-981-15-7215-9_4

Scopus Eid


  • 2-s2.0-85115245895

Web Of Science Accession Number


Book Title


  • Transactions on Intelligent Welding Manufacturing

Start Page


  • 57

End Page


  • 73

Abstract


  • As a new emerging technology, wire arc additive manufacturing (WAAM) has attracted extensive interests from both academia and industry during recent years. WAAM uses welding arc as an energy source to fuse metal wire and deposit layer by layer, which provides the advantages of freeform deposition. In order to improve its manufacture precision, stability and repeatability, it is necessary to develop sensing and control strategy for WAAM process. This research develops a passive visual sensing system for a robotic WAAM system. A new deep learning technique (Mask R-CNN) is proposed to detect and segment the melt pool area, and the width of melt pool can be measured based on the coordinate of the bounding rectangle. The pseudo-random ternary (PRT) signals were used to stimulate the WAAM process, and the corresponding width can be measured by the Mask R-CNN. Based on the width data and corresponding PRT input, a dynamic model of adaptive neuro-fuzzy inference system was built for the WAAM process.

Publication Date


  • 2020

Citation


  • Xia, C., Pan, Z., Zhang, S., Polden, J., Li, H., Xu, Y., & Chen, S. (2020). Mask R-CNN-Based Welding Image Object Detection and Dynamic Modelling for WAAM. In Transactions on Intelligent Welding Manufacturing (pp. 57-73). doi:10.1007/978-981-15-7215-9_4

Scopus Eid


  • 2-s2.0-85115245895

Web Of Science Accession Number


Book Title


  • Transactions on Intelligent Welding Manufacturing

Start Page


  • 57

End Page


  • 73