Butt welding joints recognition and location identification by using local thresholding

Manual detection and identification of butt weld joints through welding image in real-time by human observation is subjective in nature, requires experience and can be biased at times. Furthermore, since that most of the welding robots application are programmed by teach and play means that they need to be reprogrammed each time they deal with new task. This is time consuming, plus welding parameters also need to be refined for every new program. Hence, this research aims to tackle the aforementioned issues by suggesting an alternative method that can automatically recognize and locate the butt welding position at starting, middle, auxiliary and end point under three scenarios which are; (1) straight, (2) saw tooth and (3) curve joint. This is done without any prior knowledge of shapes involved. A new approach known as local thresholding is proposed in the segmentation process which consists of image pre-processing, noise reduction and edge region points generation for butt welding joint identification. Region points for butt welding seam path are selected by shape selection generated from contour points according to the three scenarios. Each point is located by 2D coordinates that is usually used by robot controller for path planning. Experimental results showed that the local thresholding approach has managed to fulfill the research objective in detecting, identifying and locating the butt welding joint position in the three different scenarios. When compared with other methods such as background subtraction, local thresholding narrowly loses out in terms of less mismatch error produced. However, it gives the best results in the detection of the butt welding joint edges. Besides having the advantage of the ability to perform identification without prior knowledge from an image, local thresholding also showed that it can work just fine even in the presence of imperfections such as scratches on the surface of mild steel.