Automatic Identification of Multi-Type Weld Seam Based on Vision Sensor With Silhouette-Mapping
Yingzhong Tian,Hongfei Liu,Long Li,Guangjie Yuan,Jiecai Feng,Yanbin Chen,Wenbin Wang
DOI: https://doi.org/10.1109/jsen.2020.3034382
IF: 4.3
2021-02-15
IEEE Sensors Journal
Abstract:Automatic identification of weld seam types by welding robot is a key link in intelligent welding as some adjustment scheme (e.g., welding trajectory planning, initial welding position, welding parameters) vary with the weld seam types. However, the variable welding environment and various weld seam profile omnifarious affect the robustness of weld seam types identification. To overcome the challenges derived from the weld seam diversity, in this paper, the silhouette-mapping was selected as the weld seam intermedium and a multi-type weld seam automatic identification system based on vision sensor was introduced. Two different laser sources were adopted to obtain robust silhouette-mapping features in proper gestures. Based on the silhouette-mapping data (stripe-mapping and spot-mapping), the related image processing algorithms were carried out to achieve automatic identification. Specifically, the bidirectional deviation search method was proposed to locate the spot-mapping area based on the stripe-mapping image. Aiming at the characteristics of the spot-mapping image, a carefully designed CNN (convolutional neural network) model was used to classify types. Experimental results prove that the silhouette-mapping and CNN are an effective combination for the multi-type weld seam identification, and a total of 97.6% of weld seam types were correctly predicted. Some weld-related studies include welding features extraction, and welding quality detection may improve its accuracy on the basis of determining weld seam types.