Visual Feature Detection For Lap Weld Of Thin Steel Plate

With the third industrial revolution booming, modern machinery is more and more widely applied in production area. To meet the demand for high quality and high efficiency production in steel welding industry, the intelligent welding robots are becoming more and more popular, which can enhance the flexibility and intelligence of welding process that is still a challenging problem. This thesis focuses on the machine vision methods and system for welding robot, developing intelligent recognition of the seam position system. These results advantage to alleviate labor intensity of welding workers, improve their work condition in harsh environments, which will bring some important social value and economic value.A reasonable vision system for the feature recognition of thin lap weld seam is designed by depending on the visual spatial constraints and vision measurement principle. And the designed vision system includes a CMOS camera made by Daheng, a line laser light projector with wavelength, a Computar lens, a narrow-band filter class, an attenuation filter and an image computer.The feature extraction of weld seam image includes a region of interest area in real-time(ROI)、an image filtering、a differential search algorithm to get the accurate extraction of structured light stripe skeleton, gaining the image feature points and interpolation filter of skeleton points. Then the Hough transform is utilized to get the main line of laser stripe by using its skeleton points. By differentiating the distance between skeleton characteristic points to the main line of laser stripe, the accurate feature points can be obtained. Then the feature points’ position will deliver to the PLC controller through Modbus communication protocol as soon as the feature’s point request from PLC controller is received. Finally, the motor will drive the welding torch to complete the error rectification and welding process.For image filtering process, this thesis proposes an innovative pseudo color transform filtering method based on energy, the burr and the weaker energy splash around the laser stripe can be reliably filtered and help the feature extraction in next step.To realize the accurate prediction of feature points under high disturbance welding environments, the thesis uses a random sample consensus algorithm to analysis probabilistic of image feature historical data and build the local model of weld features point. And the reliable and intelligence detection of weld feature points are remarkably improved.The proposed algorithms in the thesis can be tested in two typical industrial robot, such as a three degree-of-freedom Cartesian robot platform in the laboratory and gantry robot platform in the factory. Some experimental results proved that the visual system and visual processing methods are precision and high reliability. The image processing algorithm can extract the characteristic points of the lap weld position effectively and reliably in severe welding disturbance environment with the arc light and spatter.