Local threshold identification and gray level classification of butt joint welding imperfections using robot vision system
This research is carried out be able to automatically identify the joint position and classify the quality level of imperfections for butt welding joint based on background subtraction, local thresholding and gray level approaches without any prior knowledge of the joint shapes. The background subtr...
Saved in:
| 总结: | This research is carried out be able to automatically identify the joint position and classify the quality level of imperfections for butt welding joint based on background subtraction, local thresholding and gray level approaches without any prior knowledge of the joint shapes. The background subtraction and local thresholding approaches consist of image pre-processing, noise reduction and butt welding representation algorithms. The approaches can automatically recognize and locate the butt joint position of the starting, middle, auxiliary and ending point according to the three different joint shapes; straight line, tooth saw and curved joint shapes. The welding process was done by implemented an automatic coordinate conversion between camera (pixels) and KUKA welding robot coordinate (millimeters) from the KUKA welding robot and camera coordinate ratio. The ratio was determined by a camera and three reference point (origin, x-direction and y-direction) taken around workpiece. Hence, the quality level of imperfection for butt welding joint was classified using Gaussian Mix Model (GMM), Multi-Layer Perceptron (MLP) and Support Vector Machine (SVM) classifiers according to their class of imperfection categories; good welds, excess welds, insufficient welds and no weld in each
welding joint shape. These classifiers introduced 72 characteristics of feature values of gray pixels taken from co-occurrence matrix. The feature values consist of energy, correlation, homogeneity and contrast combine with gray absolute histogram of edge amplitude including additional characteristic features with scaled image factor by 0.5. The proposed approaches were validated through experiments with a KUKA welding robot in a realistic workshop environment. The results show that the approaches introduced in this
research can detect, identify, recognize, locate the welding position and classify the quality level of imperfections for butt welding joint automatically without any prior knowledge of the joint shapes. |
|---|